{"generated":"2026-08-09T19:22:49.109Z","publication":{"name":"Agent Mode AI","url":"https://agentmodeai.com","editorial_model":"AI-written by Claude, reviewed and signed off by Peter.","review_cadence":"30 to 90 days per claim; see next_review field."},"holding_up":{"description":"Claims published by Agent Mode AI itself. One claim per article.","methodology_url":"https://agentmodeai.com/standards/","index_url":"https://agentmodeai.com/holding/","claims":[{"id":"AM-001","claim":"70% of AI-implementation failure is people and process, not technology — cultural transformation is the strongest predictor of AI ROI at the 2024-2025 maturity stage.","article_url":"https://agentmodeai.com/ai-readiness-in-organizations-the-2024-2025-landscape/","topic":"agentic-ai-governance","pub_date":"2026-04-19","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-002","claim":"Agentic AI's $3.50-per-dollar average return masks a 70% task-failure rate on the Carnegie Mellon benchmark; only narrowly-scoped deployments clear the reality bar.","article_url":"https://agentmodeai.com/the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025/","topic":"agent-procurement","pub_date":"2026-04-19","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug language ('revolution', 'real-world success stories') carries hype register the publication explicitly avoids; survey-of-surveys structure does not stand up to source-verification at the level the publication now demands. Google has rejected the URL despite an active claim status. URL now redirects to /retractions/?retired=the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025. Claim withdrawn — status moves to Not holding, no further reviews scheduled."},{"date":"2026-05-06","verdict":"partial","note":"URL state changed. The /the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025/ slug now serves a deliberately rewritten retrospective (claimId AM-130, \"Agentic AI 2024-2025 retrospective\", published 04 May 2026) against audited primary sources. The 28 Apr 2026 redirect to /retractions/ has been lifted to allow that. AM-002 the claim remains Not holding — the original $3.50/dollar + 70% failure-rate framing was withdrawn and is not restored. AM-130 is a separate claim with its own evidence chain. Readers arriving at /holding/AM-002 see the withdrawal here; the article link surfaces the new piece at the URL the original lived at, with this entry as the audit trail."}],"primary_sources":[]},{"id":"AM-003","claim":"GPT-5 Pro's tiered-subscription model forces enterprises to classify problems by computational difficulty — $200/month premium routing only repays for the top decile of 'very hard' queries.","article_url":"https://agentmodeai.com/gpt-5-pro-vs-enterprise-ai-agents-what-very-hard-problems-means-for-your-business/","topic":"enterprise-ai-cost","pub_date":"2026-04-19","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"partial","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-05-28","verdict":"partial","note":"Pricing/model drift: a $100/mo Pro tier now sits beside the $200 tier (added 9 Apr 2026) and the premium model is GPT-5.5 Pro. Core thesis holds; the single-$200-tier framing no longer matches. Re-verify current tiers at chatgpt.com/pricing."}],"primary_sources":[]},{"id":"AM-004","claim":"The agentic AI discovery phase upstream of procurement is not a vendor-evaluation sprint to a go-decision; it is an organisational-readiness test where the deciding question is whether the procuring enterprise can clear four upstream tests (definitional clarity across the senior team, a named operational candidate workflow with measured baseline and named owner, threat-model literacy on the cross-agent and browser-resident classes, and workforce-readiness against the BCG access gap) before any vendor conversation. Gartner's January 2025 poll of 3,412 executives (19% significant, 42% conservative, 31% wait-and-see, 8% no investment) describes the phase distribution; the 39% in 'wait-and-see' or 'no investment' postures are not failing discovery but correctly identifying that the upstream tests are not yet cleared.","article_url":"https://agentmodeai.com/navigating-the-discovery-phase-how-organizations-first-explore-agentic-ai/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-005","claim":"The AI assistant vs AI agent distinction is operationally meaningful for enterprise procurement: assistants are reactive, request-driven, human-in-the-loop systems whose deployment and ROI patterns are documented at named-customer scale (McKinsey's Lilli platform with 72% employee adoption, 500,000+ prompts processed monthly, ~30% time savings on knowledge work, six-month deployment from proof-of-concept to full rollout); agents are proactive, goal-directed, autonomous-action systems whose deployment patterns are still emerging and whose cohort-scale failure rate is documented (Gartner June 2025: 40%+ of agentic AI projects cancelled by end-2027). Assistants and agents are different procurement decisions rather than points on a continuum; an assistants-first enterprise roadmap is defensible on the documented named-success cohort, an agents-first roadmap is defensible only when the AM-004 discovery-phase tests are cleared and the AM-140 procurement-committee questions are answered.","article_url":"https://agentmodeai.com/ai-assistant-vs-ai-agent-understanding-the-key-differences-for-enterprise-implementation/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-006","claim":"The 56% AI-skill wage premium reported by the Federal Reserve Bank of Atlanta (May 2025, drawing on Lightcast job-posting data through 2024) describes a real labour-market signal at scale, but materially overstates what the typical mid-career worker should expect from a generic AI-literacy program: the premium attaches to specific technical skills surfacing in 1.62% of all 2024 job postings, and the BCG 14%-vs-44% gap in AI upskilling access between frontline workers and leaders is the operational variable that decides which cohort captures the premium and which sees credential inflation without the wage signal.","article_url":"https://agentmodeai.com/the-56-solution-how-workers-are-turning-ai-anxiety-into-career-gold/","topic":"enterprise-ai-cost","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-007","claim":"The AgentFlayer class of zero-click cross-agent prompt-injection attacks (Zenity Labs disclosure at Black Hat USA 2025) and the EchoLeak CVE-2025-32711 disclosure earlier the same year describe a structural failure mode in agentic AI rather than incidental bugs; the procurement-relevant signal is the vendor-response split — which vendors patched and committed to a response-SLA versus which classified the behaviour as 'intended functionality' — answered before the contract closes, not after.","article_url":"https://agentmodeai.com/agentflayer-attack-why-chatgpt-copilot-6-major-ai-platforms-are-being-hacked-right-now/","topic":"agentic-ai-governance","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-008","claim":"AI infrastructure water consumption has moved from sustainability-footnote to procurement-deck variable: Google reported 8.1 billion gallons of data-centre water consumption in 2024 (a 33% year-over-year increase from 6.1 billion in 2023), Microsoft reported 6.4 million cubic metres in 2022 at a Water Usage Effectiveness of 0.30 litres per kilowatt-hour (a 39% improvement from 0.49 in 2021), and the EU Energy Efficiency Directive 2023/1791 made WUE and water-consumption reporting mandatory for data centres above 500 kilowatts of IT power demand starting 15 September 2024. Closed-loop and immersion cooling technologies (Microsoft's zero-water evaporation systems standardised for new builds August 2024; immersion cooling at sub-1.1 PUE) have matured enough that the procurement question for cloud and co-location vendors in 2026 is the vendor's water-efficiency posture in writing, not whether water consumption is a procurement-relevant variable.","article_url":"https://agentmodeai.com/ai-environmental-impact-the-hidden-water-crisis-threatening-digital-transformation/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-06-17","next_review":"2026-08-16","verdict":"partial","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-17","verdict":"partial","note":"Source-text figure re-review: Google's 2024 Environmental Report reports a 28% year-over-year increase to 8.1 billion gallons, not the 33% (from a 6.1 billion 2023 base) asserted at publish. The 8.1B 2024 figure and the Microsoft WUE 0.30 L/kWh / 39%-improvement figure are unchanged and verified. Article corrected to 28% and the unsupported 6.1B base removed; the claim text retains the original figure with this correction per the Holding-up protocol."}],"primary_sources":[]},{"id":"AM-009","claim":"Anthropic's Claude for Chrome launch (26 Aug 2025, 1,000 Max-plan subscribers at $100-200/month) is a procurement-decision data point about the maturity of the browser-resident agentic AI class rather than about Anthropic specifically; the company's own security disclosure (23.6% prompt-injection success rate pre-mitigation, 11.2% post-mitigation, 0% on URL-injection variants after subsequent patches) describes the structural exposure level the deploying enterprise inherits across the class, including from Anthropic's competitors as they ship parallel browser-resident products. The procurement-relevant signal is the published-disclosure posture (Anthropic disclosed the rates honestly with mitigation deltas), which places Anthropic in the AM-007 Cohort A and gives procurement a verifiable vendor-response baseline; the rate itself bounds the deployment-layer compensating-control burden but does not, on its own, decide the procurement question.","article_url":"https://agentmodeai.com/anthropics-claude-for-chrome-changes-everything-what-business-leaders-need-to-know-now/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-010","claim":"Enterprise agentic AI deployments that reach the McKinsey 23% scaling cohort share five operational characteristics drawn from the cited JPMorgan, Toshiba, Wipro, and Aberdeen City Council deployments: measured pre-deployment baselines, named deployment-owner accountability, scoped-experimentation governance, training-over-hiring resource posture, and CIO-level visibility on per-deployment ROI; the characteristics are observational and replace the prior draft's 'ADAPT' acronym framing, which had no published source.","article_url":"https://agentmodeai.com/the-cios-playbook-orchestrating-human-ai-teams-that-actually-want-to-work-together/","topic":"enterprise-ai-cost","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-011","claim":"The IBM Watson Health collapse (2015 launch through 2022 sale to Francisco Partners at approximately one-fifth of the initial investment) is the canonical enterprise AI failure case where the underlying technology was substantively functional and the organisational integration was not — physician rejection at named partner sites, workflow misalignment with clinical practice, and professional-identity-threat dynamics drove abandonment despite the underlying capability; the pattern reproduces at the cohort scale RAND Corporation's 2024 study (n=65 senior data scientists) identifies at the 80% AI-project failure rate, with organisational resistance dominant over technical limitation as the failure cause. The procurement-deck implication is that the change-management variable belongs in the discovery phase (AM-004) and the procurement decision (AM-140), not as a post-deployment afterthought.","article_url":"https://agentmodeai.com/the-human-agent-partnership-why-67-of-ai-projects-fail-without-cultural-change/","topic":"agentic-ai-governance","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-012","claim":"The enterprise IT operations workforce is structurally the highest-exposure population to autonomous-action AI: the task surface (incident triage, configuration management, ticket processing, routine diagnostics, scripted remediation) maps onto the agent-class capability boundary more directly than any other large enterprise job-family, and public-sector workforce data (US Bureau of Labor Statistics Computer and Information Technology Occupations Outlook; World Economic Forum Future of Jobs Report 2025) places IT-ops roles at the top of both the displacement and the role-transformation lists. The procurement-deck question for the CIO is not whether the IT-ops role mix changes but on what timeline against which named roles, and whether the workforce-transition posture is agent-orchestration (training the team toward managing fleets of agents) or agent-replacement (letting workforce churn through to a smaller team operating the deployed agents).","article_url":"https://agentmodeai.com/ai-workforce-transformation-the-human-guide-to-building-your-autonomous-it-future/","topic":"agentic-ai-governance","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-013","claim":"Q1 2026 is the quarter enterprise agentic-AI crossed three thresholds simultaneously — the first at-scale in-the-wild exploits, the first vendor-shipped governance infrastructure, and the first hard ROI data — and programmes designed around only one will not make the 28% that pay off.","article_url":"https://agentmodeai.com/agentic-ai-got-real-q1-2026/","topic":"agentic-ai-governance","pub_date":"2026-04-19","last_reviewed":"2026-06-04","next_review":"2026-08-03","verdict":"holding","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-014","claim":"The ~73% of enterprise agentic-AI projects that fail share three structural gaps — no named owner, scope drift, and missing agent-level MTTD — and the 27% that succeed cluster around the inverse.","article_url":"https://agentmodeai.com/why-73-of-agentic-ai-projects-fail-and-how-the-27-generate-312-roi/","topic":"enterprise-ai-cost","pub_date":"2025-08-03","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-08-03","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Both stats in the slug ('73% fail', '312% ROI') were backfilled with status: partial on 19 Apr 2026 noting the article predates editorial standard. Body never rewritten. Google's quality algorithm has independently flagged the URL nine days later. The slug carries the structural problem. URL now redirects to /retractions/?retired=why-73-of-agentic-ai-projects-fail-and-how-the-27-generate-312-roi. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-015","claim":"An agentic-AI Center of Excellence justifies its overhead only after the organisation has three production agents running; before that, it over-governs an experimental footprint.","article_url":"https://agentmodeai.com/building-a-center-of-excellence-for-agentic-ai-in-it-operations-complete-enterprise-guide/","topic":"agentic-ai-governance","pub_date":"2025-08-01","last_reviewed":"2026-06-04","next_review":"2026-08-03","verdict":"partial","verdict_history":[{"date":"2025-08-01","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."}],"primary_sources":[]},{"id":"AM-016","claim":"Agent-mediated network management reduces unplanned firewall-change incident costs only when the agent's action log feeds into the same change-management audit trail human changes use — not as a parallel system.","article_url":"https://agentmodeai.com/the-7-2m-firewall-change-that-transformed-network-management-how-agentic-ai-prevents-it-disasters/","topic":null,"pub_date":"2025-07-27","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-07-27","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. '$7.2M' figure in the slug cannot be traced to any disclosed firewall-change incident. Body never rewritten past 19 Apr 2026 backfill. The dollar specificity in the URL is the structural problem and Google's quality algorithm has independently flagged the URL. URL now redirects to /retractions/?retired=the-7-2m-firewall-change-that-transformed-network-management-how-agentic-ai-prevents-it-disasters. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-017","claim":"Agentic AI's durable enterprise pattern is redeployment-first, not replacement-first. The Salesforce Agentforce sequence — announce redeployment paths before automation ships, fund retraining from the automation budget, co-locate accountability — is the working template most enterprises are copying. Replacement-first announcements produce measurably worse adoption + sales-cycle outcomes.","article_url":"https://agentmodeai.com/the-day-9000-people-asked-to-be-replaced/","topic":null,"pub_date":"2025-07-19","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-07-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-19","verdict":"partial","note":"Anchor verification complete (see audit/ANCHOR_VERIFICATION_2026-04-19.md). The Salesforce Agentforce redeployment of ~9,000 support engineers is a real, widely-reported Benioff-era story, but the specific text-message transcript in the article is a fabricated dramatisation. Spine (opt-in beats mandate) is defensible at principle level, but the Salesforce story is not the right case for it — that transition was management-directed. Rewrite flagged for before 18 Jun 2026 review."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten. Fabricated text-message transcript removed. Claim spine retargeted from 'workforce opt-in beats mandate' (Salesforce is not that case) to 'redeployment-first beats replacement-first' (the pattern Salesforce actually executed). Status moves from Partial to Up. Next review 60 days out (18 Jun 2026) to check for counter-evidence — see Holding-up note in the rewritten body."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug premise ('asked to be replaced') is the dramatized framing the body had to remove on 19 Apr 2026. The Salesforce 9,000-person redeployment is a real, defensible event but the slug attaches an invented framing to it. Body preserved in archived/. Google has independently rejected the URL. URL now redirects to /retractions/?retired=the-day-9000-people-asked-to-be-replaced. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-018","claim":"Agentic AI's compounding economics show up in back-office operations (AP, IT ticket triage, HR onboarding, procurement, close-cycle reconciliation), not in front-office customer-facing workflows. The 12% of deployments that clear 300%+ ROI cluster there for structural reasons: per-action savings × action frequency × task-specification tightness × existing process instrumentation.","article_url":"https://agentmodeai.com/the-executive-who-discovered-her-competitors-secret-weapon/","topic":null,"pub_date":"2025-07-19","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-07-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-19","verdict":"partial","note":"Anchor verification complete (see audit/ANCHOR_VERIFICATION_2026-04-19.md). 'Sarah Chen' and the 2 AM Munich-hotel scenario are fully fabricated — the article's narrative protagonist does not correspond to any real executive. The underlying framework (back-office cost compounding faster than front-office wins; per-action delta × frequency) IS defensible against McKinsey + Futurum operational-AI-ROI data. Rewrite required before the article can move to Holding."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten. Fabricated 'Sarah Chen' narrative frame removed entirely. Claim spine sharpened: original was 'back-office cost compounding faster than front-office'; new version adds the structural explanation (per-action × frequency × task-specification × measurement instrumentation) and specific 2026 benchmark anchors (Stanford DEL 12%/88%, Gartner 28%, Futurum 71% vs 40%). Status moves from Partial to Up. Cross-links to AM-020 (TCO), AM-021 (measurement discipline), AM-022 (bimodal ROI) explicitly drawn in the body. Next review 18 Jun 2026."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug structure (fictional protagonist, 'discovered her competitors' secret weapon') is the fabricated narrative frame the body had to remove on 19 Apr 2026. Body rewritten with Stanford DEL / McKinsey / Futurum sourcing (preserved in archived/) but the slug is the structural problem. URL now redirects to /retractions/?retired=the-executive-who-discovered-her-competitors-secret-weapon. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-019","claim":"Manufacturing deployments hitting the 30% unplanned-downtime-reduction benchmark share one architectural pattern — the agent writes its actions into the plant's existing MES/CMMS audit trail rather than a parallel log. Parallel-log deployments underperform by a factor of 2-3.","article_url":"https://agentmodeai.com/manufacturing-4-0-how-multi-agent-systems-reduce-downtime-by-30/","topic":"agentic-ai-governance","pub_date":"2025-08-01","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2025-08-01","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten. Original headline number (30% downtime reduction) survives against current case-study data. New analytical spine: the audit-trail architecture separates wins from stalls. Status moved from rewrite-in-progress Partial placeholder to Up. Next review 60 days out because architectural claims age slower than pricing claims."}],"primary_sources":[]},{"id":"AM-020","claim":"The 40-60% TCO underestimate on enterprise agentic-AI deployments is not a cost-visibility failure — it is a cross-departmental cost-attribution failure. Integration, tokens, maintenance, supervision, and compliance costs land on IT, HR, and Legal budgets that do not reconcile in most organisations, so the CFO sees the bill late and partial.","article_url":"https://agentmodeai.com/the-hidden-costs-of-agentic-ai-a-cfos-guide-to-true-tco-and-roi-modeling/","topic":"enterprise-ai-cost","pub_date":"2025-07-31","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2025-07-31","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop. Status moves from rewrite-in-progress placeholder to Up. New analytical spine: the TCO underestimate is cross-departmental cost-attribution failure, not hidden costs. Five cost categories named with budget owners. 60-day review cadence."}],"primary_sources":[]},{"id":"AM-021","claim":"The 87% vs 27% success-rate gap between Six-Sigma and non-Six-Sigma organisations on agentic-AI deployments reflects pre-existing measurement discipline, not the DMAIC methodology itself. Agents require a clean baseline, defect definition, documented root-cause analysis, and a change-management gate — four conditions that ISO 9001, ITIL, SRE, or HACCP practices produce just as reliably.","article_url":"https://agentmodeai.com/dmaic-for-agentic-ai-deployment/","topic":"agentic-ai-governance","pub_date":"2025-08-16","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2025-08-16","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop. Status moves from rewrite-in-progress placeholder to Up. New thesis: the causation runs the opposite direction from the vendor narrative — the measurement discipline was the prerequisite, the methodology name doesn't matter. 60-day review."},{"date":"2026-04-28","verdict":"partial","note":"Slug migration to §6a-compliant URL: from-dmaic-to-ai-agents-how-traditional-optimization-methods-accelerate-agentic-ai-success → dmaic-for-agentic-ai-deployment. Body unchanged from 19 Apr rewrite, only the URL changed. Old slug 308-redirects to new. Reason: the long descriptive slug carried §6a-grade friction (88+ chars, vendor-cliche framing) and Google's quality algorithm had flagged the original URL as low-quality (per the 28 Apr 2026 GSC drilldown showing it in the 'Crawled - currently not indexed' bucket). The clean slug preserves the analytical content while removing the URL-level quality penalty."},{"date":"2026-06-10","verdict":"partial","note":"Primary-source erosion on the headline statistic. The Gravitex page (gravitexgenesys.com/blog/ai-agents-lean-six-sigma-automating-dmaic) no longer carries the 87%/27% success-rate comparison — checked 10 Jun 2026, the URL now serves Six Sigma course-offering content with no AI-deployment success-rate data. A web search found no independent source corroborating the 87/27 pair. The claim's interpretive reading (the gap reflects pre-existing measurement discipline; ISO 9001/ITIL/SRE/HACCP produce the same four conditions) is unaffected and remains supported by Gartner's 7 Apr 2026 I&O survey (57% of failures cited 'too much too fast'). Status Up → Partial until the 87/27 figure can be re-anchored to a retrievable primary source."}],"primary_sources":[]},{"id":"AM-022","claim":"The 171% average ROI on enterprise agentic-AI deployments is the mean of a bimodal distribution — roughly 12% of deployments clear 300%+ and 88% sit at or below break-even. The single factor distinguishing the clusters is not a multi-pattern framework; it is whether business-line (not IT) ownership held the kill-switch and accountability before the deployment shipped.","article_url":"https://agentmodeai.com/the-agentic-ai-success-formula-7-proven-patterns-driving-171-roi-in-enterprise-deployments/","topic":null,"pub_date":"2025-08-06","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-08-06","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop (7-patterns vendor framework with fabricated case studies). New thesis: bimodal distribution, not normal — the 171% average describes no specific deployment. Business-line kill-switch ownership is the single distinguishing factor. Cross-links to AM-020 + AM-021 on the shared organisational-precondition thread."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug carries '171% ROI' as a category average and a '7 proven patterns' framework that the body had to disown — the rewritten body explicitly argues 171% is the mean of a bimodal distribution, not a benchmark. Body rewritten 19 Apr 2026 (preserved in archived/) but the slug contradicts the rewritten thesis and Google has rejected the URL. URL now redirects to /retractions/?retired=the-agentic-ai-success-formula-7-proven-patterns-driving-171-roi-in-enterprise-deployments. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-023","claim":"The 10 Apr 2026 Google AI Mode rollout to eight markets is the first vertical (restaurant booking) where agentic search reduces named SaaS aggregators (OpenTable, TheFork, ResDiary and five others) to API backends rather than destinations. The template applies to every enterprise-relevant aggregation vertical — business travel, expense management, procurement, ATS, HR service delivery — and incumbents in those verticals have 18-24 months to pick API-backend or destination positioning before agentic search forces the choice.","article_url":"https://agentmodeai.com/google-ai-mode-restaurant-booking-the-50-billion-business-revolution-every-ceo-must-understand-2025/","topic":"enterprise-ai-cost","pub_date":"2025-08-23","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2025-08-23","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop (the '$50 Billion Revolution' headline and 'act within 90 days' crisis-FOMO framing were both fabrications). New thesis: restaurant booking is a template, not the story. Named 5 enterprise-relevant aggregation verticals (business travel, expense, procurement, ATS, HR service) and the API-backend-vs-destination choice incumbents face. Next review in 60 days."}],"primary_sources":[]},{"id":"AM-024","claim":"Enterprise-AI decisions in 2026 are made on a citation chain nobody in the chain verifies. The infrastructure gap CIOs face is a verification layer for the claims their procurement runs on — not an information gap. The 88% failure rate in enterprise agentic AI is the predictable output of decision-making on unverified citations, not a capability problem.","article_url":"https://agentmodeai.com/the-unverified-citation-chain-where-enterprise-ai-decisions-actually-come-from/","topic":"agentic-ai-governance","pub_date":"2026-04-20","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-04-20","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Claim-text figure unanchored — a same-day reversal of the morning verdict, recorded rather than smoothed over. The batch-1 re-review earlier on 10 Jun 2026 returned Holding after confirming the cited sources were live and unrevised (including the Stanford DEL playbook PDF at its URL) and deferred re-anchoring the 88% figure to the next cycle. A batch-2 full-text extraction the same afternoon disproved the figure: the playbook contains no 88% failure rate and no 12/88 ROI distribution (full finding at AM-029, correction of 10 Jun 2026). URL-liveness is not figure-verification; the morning verdict does not survive the afternoon evidence. The claim's spine (enterprise-AI decisions run on citation chains nobody verifies) holds and is itself illustrated by this incident, but its quantitative anchor — 'the 88% failure rate in enterprise agentic AI' — has no verifiable source as a deployment-failure or ROI distribution. The only verified figure carrying the same numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial, same day as the batch-1 Holding verdict."}],"primary_sources":[]},{"id":"AM-025","claim":"Enterprise agentic AI governance in 2026 fails at the operational layer even when it passes at the compliance layer. Boards receive EU-AI-Act-mapped compliance decks while the agentic deployments actually shipping out of IT ops have no measurable overlap with that deck. Durability requires six instrumented dimensions scored 0–100 (GAUGE framework) with a 90-day setup cadence and a 12-month trajectory target — not a compliance matrix.","article_url":"https://agentmodeai.com/the-enterprise-agentic-ai-governance-playbook-2026/","topic":"agentic-ai-governance","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-026","claim":"Generic enterprise SaaS RFPs systematically underweight six agent-specific governance dimensions (governance maturity, threat model, ROI evidence, change management, vendor lock-in, compliance posture). A 60-question RFP layer mapped to the GAUGE framework materially changes vendor selection outcomes by disqualifying vendors whose operational governance will not survive the 18-month enterprise review cycle.","article_url":"https://agentmodeai.com/the-enterprise-agentic-ai-rfp-60-questions/","topic":"agent-procurement","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-027","claim":"A durable enterprise agentic AI business case requires three specific documents — a TCO model with ten named cost categories (not vendor-supplied line items), an ROI model with a pre-deployment measured baseline and an independent validation round, and a three-scenario risk-adjusted NPV. The single-scenario vendor-framed business cases that dominate 2026 enterprise AI investment committees are the predictable root of the 40%+ projected agentic AI project cancellation rate.","article_url":"https://agentmodeai.com/the-cfos-agentic-ai-business-case-tco-and-roi/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-028","claim":"Partner — co-development with a vendor on a structured non-standard engagement — is structurally under-chosen in enterprise agentic AI procurement in 2026. Procurement committees have templates for build and buy but none for partner, so the third path does not get evaluated on an equal footing. The vendor-lock-in and change-management dimensions of the GAUGE framework usually favour partner when it is honestly evaluated, not buy or build.","article_url":"https://agentmodeai.com/build-vs-buy-vs-partner-for-enterprise-agentic-ai-2026/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-029","claim":"The 12/88 bimodal distribution in enterprise agentic AI ROI realisation (Stanford DEL 2026 + cross-validated by Gartner, McKinsey, CMU) is a governance-discipline outcome, not a model-capability outcome. The 12% instrument the six GAUGE dimensions on a 90-day review rhythm; the 88% treat governance as a deliverable to the audit committee. Capability gap (CMU's 30.3% best-in-class task completion) constrains what is possible, not what separates the 12% from the 88%.","article_url":"https://agentmodeai.com/why-88-percent-of-agentic-ai-deployments-fail/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"not_holding","note":"Primary-source verification failed on the headline figure. The claim attributes a 12/88 bimodal ROI distribution (12% of enterprise agentic AI deployments clearing 300%-plus ROI, 88% at or below break-even at 12-18 months) to the Stanford Digital Economy Lab Enterprise AI Playbook (Apr 2026). The full report text contains no such distribution and no ROI-realisation failure data: it is a study of 51 successful deployments (Pereira, Graylin, Brynjolfsson), in which the only 88% figure is '88% of organizations use AI in at least one function' and the only 12% figure is a sponsor-engagement category. The report's design cannot yield a failure distribution. A web search found no Stanford DEL publication reporting 12/88. Gartner (28% of AI I&O projects fully pay off, Apr 2026) and McKinsey (6% high performers, Nov 2025) document a small-tail pattern but do not corroborate the specific 12/88 split the claim asserts. Status Up -> Not holding. The article remains published with this correction log; the governance-over-capability argument is re-anchored, where it appears elsewhere in the corpus, to figures that survive verification."},{"date":"2026-06-10","verdict":"not_holding","note":"Article restated. The piece at the source URL was rewritten the same day, at the same slug, on the verifiable counterpart figure the fabricated one shadowed: IDC research commissioned by Lenovo (CIO Playbook 2025, Feb 2025; global survey n=2,920) reporting 88% of AI proof-of-concepts failing to reach production, with 4 of every 33 POCs (roughly 12%) graduating, per CIO.com (25 Mar 2025). The restated article asserts a new tracked claim, AM-213, and opens with a correction notice pointing back to this record. This claim stays Not holding as the permanent record of the fabricated attribution; its claim text is unchanged."}],"primary_sources":[]},{"id":"AM-030","claim":"The McKinsey State of AI 2025 figure (23% of enterprises scaling an agentic AI system, 39% still experimenting) is an operational-preconditions outcome, not a technical-readiness outcome. Four preconditions (agent registry, measured pre-deployment baseline, differentiated change-management playbook for adjacent units, cross-agent threat model at scale) separate pilots that cross into production from pilots that stall. The 6% AI-high-performer segment is the subset of the 23% scaling with additional measurement discipline that makes ROI audit-survivable.","article_url":"https://agentmodeai.com/the-mckinsey-23-percent-agentic-ai-scaling-gap/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-031","claim":"The CMU TheAgentCompany 2026 benchmark figure (30.3% task completion for best-in-class frontier model, up from 24% in 2024) is the current capability constraint for enterprise agentic AI. Capability trajectory projects to ~40% by late 2027, which does not cross the 95% production-readiness threshold within the 3-year TCO horizon enterprise business cases operate against. The Stanford DEL 12% durable cohort operates within the 30.3% (narrow scope + human-in-the-loop + GAUGE-dimensional governance discipline), not around it. Capability is not the variable that separates the 12% from the 88%.","article_url":"https://agentmodeai.com/the-cmu-30-percent-agent-capability-gap/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One leg unanchored on re-review. The CMU capability figures verify cleanly (30.3% full completion for Gemini 2.5 Pro and 39.3% partial-credit on the 175-task TheAgentCompany set per paper v2; 24% for Claude 3.5 Sonnet in the Dec 2024 v1). The Stanford DEL '12% durable cohort' referenced in the claim text does not exist in the cited source: the Enterprise AI Playbook (Pereira, Graylin, Brynjolfsson, Apr 2026) studies 51 successful deployments and contains no 12/88 ROI cohort (full finding at AM-029, correction of 10 Jun 2026). The claim's capability-constraint argument holds on its own evidence; the sentence tying the constraint to the 12%/88% cohort behaviour has no verifiable referent. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-032","claim":"EU financial-services agentic AI deployments operate under a compounded five-framework obligation surface (DORA, NIS2, MiFID II, EU AI Act, GDPR) that sits on top of general AI governance. Liability does not transfer to the vendor contractually regardless of SLA language — MiFID II conduct rules, EU AI Act deployer obligations, and DORA third-party-risk provisions place customer-facing and regulator-facing liability on the deploying financial institution. Compliance-posture and vendor-lock-in are the dominant GAUGE dimensions for the sector, scoring 15-25 points lower than cross-industry averages on first pass.","article_url":"https://agentmodeai.com/agentic-ai-in-financial-services-compliance-and-liability/","topic":"agentic-ai-governance","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-033","claim":"The McKinsey 17%-EBIT-attributable-to-genAI figure, the most-cited single statistic in 2026 enterprise agentic AI procurement decisions, is a self-reported attribution from McKinsey's State of AI 2025 survey of approximately 1,491 respondents. The way it is typically read in CIO decks, as evidence that 17% of enterprises have produced 5% or more of EBIT from genAI, materially overstates what the survey supports. The figure documents 17% of survey respondents asserting that level of attribution, not 17% of enterprises producing it under audited measurement.","article_url":"https://agentmodeai.com/the-mckinsey-17-percent-ebit-claim/","topic":"enterprise-ai-cost","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-034","claim":"AI assistants and AI agents are not the same product class. An AI assistant is a productivity-augmentation tool that suggests; an AI agent is an automation-execution system that acts on a downstream surface (tools, APIs, write-paths). Conflating them in 2026 enterprise procurement produces the most common single category mistake — buying an assistant under the assumption it is an agent, or buying an agent and governing it as if it were an assistant. The risk profile, contract structure, audit obligation, and TCO model differ categorically.","article_url":"https://agentmodeai.com/ai-assistant-vs-ai-agent/","topic":"agent-procurement","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-035","claim":"The EU AI Act enforcement deadline of 2 August 2026 applies high-risk-system obligations under Articles 9 through 49 to most enterprise agentic AI deployments operating in EU jurisdiction or providing services to EU nationals — not only to deployments explicitly classified within the Annex III high-risk categories. The compliance gap most enterprises face is structural: the Act requires evidence-of-action production (logs, oversight records, post-market monitoring, incident reports) that most agentic deployments do not generate by default. Building the evidence layer post-hoc, after a regulator request, is the failure mode.","article_url":"https://agentmodeai.com/eu-ai-act-agentic-ai-compliance/","topic":"agentic-ai-governance","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"partial","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"The 2 Aug 2026 high-risk-obligations leg is overtaken by the EU Digital Omnibus. Council and Parliament reached a provisional agreement on 6-7 May 2026 (confirmed by the Council 13 May 2026) deferring Annex III stand-alone high-risk-system obligations to 2 Dec 2027 and Annex I embedded high-risk obligations to 2 Aug 2028; formal adoption is expected before 2 Aug 2026. What still lands on 2 Aug 2026 is narrower than the claim asserted: Article 50 transparency obligations (with a watermarking grace period to 2 Dec 2026 for systems already in market) plus the penalties and governance architecture. The claim's structural argument is unchanged and current: agentic deployments still do not generate the evidence-of-action layer (Article 12 logs, Article 14 oversight records, post-market monitoring, incident reporting) by default, and building it post-hoc remains the failure mode; the deferral changes the deadline, not the gap. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-036","claim":"Enterprise shadow AI in 2026 is structurally different from enterprise shadow AI in 2024. The 2024 framing assumed unsanctioned tool adoption — workers pasting confidential data into consumer ChatGPT or installing browser extensions outside IT review. The 2026 reality is that the larger blast radius is agentic capability silently activating inside already-approved tools, often through configuration changes (Custom GPT actions, Copilot custom agents, MCP server connections from approved IDEs) that the original procurement approval did not anticipate. Discovery has to look at capability state, not vendor identity. Most enterprise shadow-AI inventories built against the 2024 framing miss 50 to 80% of the actual exposure surface.","article_url":"https://agentmodeai.com/shadow-ai-discovery-playbook/","topic":"shadow-ai-discovery","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-037","claim":"AI agents are structurally different from earlier classes of non-human identity (service accounts, API keys, machine certificates, bot identities), and the IAM platforms most enterprises run in 2026 cannot represent them adequately because those platforms authorise on principal identity rather than on per-action behavioural context. The 92% of enterprises that report low IAM confidence for agentic AI are not configured wrong; they are running an identity model with one structural axis where the agentic deployment requires four (identity, behaviour, context, revocation). The remediation is a four-layer extension on top of existing IAM, not a rip-and-replace migration. Most enterprises can ship the augmentation in 8 to 12 weeks of engineering.","article_url":"https://agentmodeai.com/non-human-identity-ai-agents/","topic":"non-human-identity","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-038","claim":"Model Context Protocol (MCP) reached enterprise procurement gravity in 18 months, faster than typical interoperability standards. The 10,000+ active public MCP servers, adoption by ChatGPT, Cursor, Gemini, Microsoft Copilot, and VS Code, and the December 2025 Linux Foundation donation made MCP a tooling-layer choice that ripples through every adjacent agentic-AI procurement decision: which agents connect to which enterprise systems, which audit boundaries hold, which vendor lock-in patterns activate. The actual procurement decision enterprise IT faces is not whether to adopt MCP (the question is moot once any approved tool ships MCP support); it is the scope-and-governance decision: which MCP servers the enterprise allows agents to connect to, what scopes those connections grant, and how cross-agent delegation through MCP is monitored. Treating MCP as a binary adoption question rather than a scope-and-governance question is the most common enterprise procurement mistake on this surface in 2026.","article_url":"https://agentmodeai.com/mcp-enterprise-agent-tooling/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-039","claim":"The 2026 enterprise agentic AI vendor comparison reduces to four credible platform plays (Anthropic, OpenAI, Google, Microsoft), and the procurement decision between them is no longer primarily about model capability. The model layer has converged to comparable parity for most enterprise use cases. The procurement decision in 2026 is on three other axes: pricing model (Anthropic Managed Agents at 8 cents per session-hour plus tokens versus OpenAI Agents SDK at no first-party runtime fee versus Microsoft and Google's vertically-integrated platform pricing), governance and BAA posture (Anthropic's three-cloud BAA position is structurally distinct), and ecosystem distribution (Microsoft's Office plus Azure footprint has no near peer; Google's vertical integration on Workspace and Cloud is second). Treating this as a model-quality bake-off is the most common 2026 procurement mistake and produces decisions that age badly within the first 12 months.","article_url":"https://agentmodeai.com/enterprise-ai-agent-vendor-comparison/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Extracted-text verification failed on the governance axis of the claim. 'Anthropic's three-cloud BAA position is structurally distinct' overstates Anthropic's own coverage: per Anthropic's BAA documentation (privacy.claude.com, retrieved 10 Jun 2026), Anthropic signs BAAs for the first-party API and HIPAA-ready Claude Enterprise only; the page contains no Bedrock, Vertex, or Azure coverage. Claude consumed via AWS Bedrock or Google Vertex AI is covered by the hyperscaler's BAA, and an Azure-side Anthropic BAA could not be verified. The article's stronger formulation ('Anthropic operates under BAAs with Amazon Web Services, Google Cloud, and Microsoft Azure simultaneously', cited only to a secondary Ampcome blog) is the same overstatement. The substantive point survives restated: BAA-covered Claude deployment surfaces span more clouds than competitors offer, but the BAAs are not Anthropic's across three clouds. This matches the AM-053 correction of the same day. The pricing axis ($0.08 per session-hour plus tokens; Agents SDK no first-party runtime fee) and the ecosystem axis verify and stand. Status Up -> Partial. Article body needs a Peter-approved BAA restate (FAQ x2, body x3, howTo step 1)."}],"primary_sources":[]},{"id":"AM-040","claim":"Enterprise agentic AI in 2026 is in its first year of operational consequence rather than its first year of capability. The deployment record across multiple independent datasets shows a stable bimodal distribution (a small high-performing tail clearing 300%+ ROI and a much larger struggling body at or below break-even), four credible platform plays converging at the vendor layer, a structurally inadequate IAM posture across 92% of enterprises, and a 14-week runway to the EU AI Act August 2026 enforcement window. The aggregate signal is that the year's defining variable is deployment discipline, not model capability or vendor selection. The 6% AI-high-performer segment and the 12% Stanford DEL high-ROI cohort instrument six specific governance dimensions on a 90-day review cadence; the remaining 88-94% mostly do not.","article_url":"https://agentmodeai.com/state-of-enterprise-agentic-ai/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One component unanchored on re-review. The claim text cites 'the 12% Stanford DEL high-ROI cohort' against 'the remaining 88-94%'. Full-text verification on 10 Jun 2026 found the Stanford DEL Enterprise AI Playbook contains no 12% high-ROI cohort and no ROI distribution of any kind — it studies 51 successful deployments by design (full finding at AM-029, correction of 10 Jun 2026). The McKinsey 6% AI-high-performer figure (Nov 2025, n=1,993) verifies independently, as do the IAM-posture and EU AI Act runway components. The 'stable bimodal distribution' framing is supported only as a small-tail/large-body shape (Gartner: 28% of AI I&O projects fully paying off; McKinsey: 6% high performers), not as the two-cluster distribution the claim names. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-041","claim":"The 2026 enterprise agentic AI procurement playbook resolves to a six-stage sequence that integrates the build-vs-buy-vs-partner decision, the 60-question agentic AI RFP, the GAUGE governance scoring, the four-vendor comparison, and the EU AI Act compliance scaffolding into one operational track. Most enterprises in 2026 run these as separate work streams owned by separate functions, which produces structurally inconsistent procurement records and substantial duplicate effort. The integrated six-stage track ships in 8 to 10 weeks for standard environments and produces an audit-defensible per-deployment procurement artifact that satisfies the EU AI Act Article 9 risk-management system requirement by construction.","article_url":"https://agentmodeai.com/enterprise-agentic-ai-procurement-playbook/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-042","claim":"The 6% AI-high-performer cohort identified by McKinsey and the 12% high-ROI cohort identified by the Stanford Digital Economy Lab share ten measurable governance practices that an enterprise can audit in under 60 minutes. An enterprise answering YES to 8 or more of the 10 diagnostic questions has the operating profile of the high-performing segment. An enterprise answering YES to 4 or fewer has the operating profile of the 88-94% struggling cohort and is unlikely to clear break-even on agentic AI deployment without a posture rebuild. The diagnostic audits posture, not outcomes; it identifies where governance investment is needed before the next deployment commitment, not whether a specific deployment will succeed.","article_url":"https://agentmodeai.com/agentic-ai-readiness-diagnostic/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of two cohort anchors unanchored on re-review. The claim text pairs McKinsey's 6% AI-high-performer cohort with 'the 12% high-ROI cohort identified by the Stanford Digital Economy Lab' and an '88-94% struggling cohort'. Full-text verification on 10 Jun 2026 found the Stanford DEL Enterprise AI Playbook identifies no 12% high-ROI cohort and no struggling-cohort percentage — it studies 51 successful deployments by design (full finding at AM-029, correction of 10 Jun 2026). The source article additionally cites a 'Stanford Digital Economy Lab 2026 Enterprise AI Productivity Study', a publication title that does not exist. The McKinsey 6% leg verifies (Nov 2025, n=1,993). The ten diagnostic questions stand as an editorial instrument, but the claim's cohort calibration now rests on one verified dataset, not two. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-043","claim":"The OWASP Agentic Security Initiative's threat taxonomy for agentic AI (memory poisoning, tool misuse, privilege compromise, resource overload, cascading hallucination, intent breaking, misaligned and deceptive behaviour, repudiation and untraceability, identity spoofing, overwhelming human-in-the-loop) maps cleanly onto seven specific enterprise controls: scoped non-human identity, action-class approval gates, decision audit logging at Article 12 evidence quality, MTTD-for-Agents layered detection, deployment-tier resource quotas, behavioural drift monitoring, and HITL throughput limits. An enterprise that operates these seven controls covers all ten OWASP threat classes; an enterprise missing more than two of the controls has structural exposure to at least four of the threat classes.","article_url":"https://agentmodeai.com/owasp-agentic-ai-top-10-walkthrough/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-044","claim":"Six well-documented public agentic AI deployment failures from 2024-2025 (Air Canada bereavement-refund chatbot, NYC MyCity small-business chatbot, Replit production-database wipe, Cursor unauthorised code deletion, Klarna customer-service reversal, DPD chatbot escalation incident) cluster into three structural failure modes: (1) the agent acts as a binding agent of the enterprise without disclosure or approval, (2) the agent operates with permissions the deployment never authorised, (3) the agent's economic case requires a service quality the deployment cannot sustain. Each failure mode maps to a specific control from the seven-control surface; all six failures would have been mitigated by controls already specified in the OWASP Agentic AI Top 10 enterprise walkthrough. The pattern is consistent enough that an enterprise can use the cases as a procurement filter: any vendor unable to point to its specific control posture against each of the three failure modes is not procurement-ready.","article_url":"https://agentmodeai.com/agentic-ai-failure-case-studies/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-045","claim":"EchoLeak (CVE-2025-32711, disclosed by Aim Security in June 2025 against Microsoft 365 Copilot) is the canonical example of a class of attacks rather than a single vulnerability: cross-agent prompt injection in which a malicious payload travels through ordinary content channels (an email, a shared document, a calendar invite, a tool response) into one or more agents' context windows, where it manipulates the agents into actions the deploying enterprise did not authorise, with no user interaction required. The attack class is structurally inherent to any architecture in which an LLM-based agent ingests untrusted content and has tool surfaces capable of exfiltration or action; closing the class requires architectural separation between content-ingest and tool-execution privileges, not point-fixes against specific exploit chains. Enterprises in 2026 operating multiple agents that share context, share memory, or hand off tasks to each other are structurally exposed to the EchoLeak class until the architectural separation is implemented.","article_url":"https://agentmodeai.com/echoleak-cross-agent-prompt-injection/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-046","claim":"EU AI Act Article 12 (record-keeping for high-risk AI systems) and Article 19 (record retention by providers) are operationalised for agentic AI by a 14-field audit-evidence template that captures every agent decision in a regulator-queryable form: deployment ID, agent identity, session ID, ISO timestamp, user prompt, retrieved context with provenance, model output, planned action, action class, approval reference, executed action, tool-call audit chain, output disclosure surface, and policy version. Logs retained for the regulatory minimum (typically 6 months for the EU AI Act baseline, 5 to 7 years for sector-specific overlays like HIPAA and SOX) in a queryable format that supports under-4-business-hour evidence assembly. An enterprise that captures the 14 fields, retains them for the maximum applicable period, and instruments the queryable export has substantially completed Article 12 compliance for the agent layer; the residual work is integrating the agent log stream with the broader audit substrate.","article_url":"https://agentmodeai.com/eu-ai-act-article-12-audit-evidence/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-047","claim":"The Head of AI Governance role (variant titles: Chief AI Officer, VP AI Strategy, Director of Responsible AI) is now a named operating role in 60% of Fortune 100 enterprises per Forrester's 2026 Enterprise AI Predictions, and is the strongest single predictor of an enterprise's score on Q10 of the readiness diagnostic. The role's effective shape converges on six accountabilities: cross-functional governance ownership, EU AI Act compliance posture, vendor procurement gate-keeping, deployment kill-criterion enforcement, audit-evidence substrate ownership, and internal upskilling. The role reports to the executive committee (CEO direct or CFO/COO) rather than to IT, security, or legal, because matrixed reporting into existing functions reproduces the matrixed-shared-accountability failure pattern. Compensation in 2026 ranges from $250-450K base for the Director tier, $400-700K for VP tier, and $600K-$1.2M total comp at the C-level, with significant equity components in growth-stage and tech enterprises.","article_url":"https://agentmodeai.com/head-of-ai-governance-role/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Extracted-text verification failed on two parts of the claim. (1) The claim asserts the role 'is now a named operating role in 60% of Fortune 100 enterprises per Forrester's 2026 Enterprise AI Predictions'. Forrester's actual prediction reads '60% of Fortune 100 companies will appoint a head of AI governance in 2026' (Predictions 2026, quoted by CIO Dive, 16 Dec 2025) — a forecast of appointments during 2026, not a measurement of current adoption. The cited sourceUrl (forrester.com/blogs/the-ai-cio-will-govern-outcomes-at-scale/, 9 Apr 2026) contains neither the 60% figure nor any Fortune-100 reference; the article's 'In Q1 2026, Forrester's Enterprise AI Predictions found 60% ... had hired or were actively recruiting' sentence has no locatable source. (2) The compensation bands ($250-450K Director base, $400-700K VP, $600K-$1.2M C-level total comp) could not be located in any primary source and are unlabelled in the article. The six-accountabilities convergence and the executive-committee reporting-line argument are editorial synthesis and stand on their own. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-048","claim":"The NIST AI Risk Management Framework (AI RMF 1.0, published January 2023, with the Generative AI Profile published July 2024) maps onto enterprise agentic AI deployment work across its four functions (Govern, Map, Measure, Manage) using the same artefacts an enterprise produces for EU AI Act Article 9. Specifically: NIST Govern maps to the Head of AI Governance role and the AI governance committee; NIST Map maps to the deployment inventory and the OWASP Agentic Top 10 walkthrough; NIST Measure maps to the 14-field Article 12 audit substrate plus the GAUGE governance dimensions; NIST Manage maps to the kill-criterion enforcement and the seven-control surface. An enterprise that has the EU AI Act preparation track running has substantially completed NIST AI RMF coverage and can document the mapping as a single cross-reference matrix. The reverse mapping (NIST → EU AI Act) requires more work because NIST is voluntary in posture and the EU AI Act is operational; an enterprise that started with NIST as the framework needs to extend audit substrate granularity and add the Article 73 incident-reporting workflow.","article_url":"https://agentmodeai.com/nist-ai-rmf-agentic-ai-mapping/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-049","claim":"Enterprise multi-agent architectures resolve to three orchestration patterns (hierarchical, peer-to-peer, broker-mediated) with materially different governance properties: hierarchical concentrates accountability at the orchestrator and is the easiest to audit but the most exposed to orchestrator-compromise; peer-to-peer distributes accountability and is the most resilient to single-agent failure but the hardest to audit; broker-mediated centralises the inter-agent communication path and is the most defensible against the cross-agent prompt-injection class. The choice of pattern is not a free architectural decision in 2026 because the EU AI Act's Article 9 risk-management requirements and the OWASP Agentic AI threat surface impose specific control obligations on each pattern. An enterprise should default to broker-mediated for new deployments above the high-risk threshold; hierarchical is acceptable for low-risk and contained deployments; peer-to-peer should be avoided in production agentic AI in 2026 unless the audit substrate is materially stronger than vendor-native baseline.","article_url":"https://agentmodeai.com/multi-agent-architecture-playbook/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-050","claim":"The A2A (Agent2Agent) protocol announced by Google Cloud in April 2025 is the most credible 2026 candidate for an open standard for cross-vendor agent-to-agent interoperability, with backing from 50+ partners across the enterprise software ecosystem (Salesforce, SAP, ServiceNow, MongoDB, Atlassian, and others). The protocol layer covers what MCP (Model Context Protocol) does not: MCP is for agent-to-tool communication, A2A is for agent-to-agent communication. The two protocols are designed to be complementary rather than competing. A2A's adoption trajectory through 2026 will determine whether broker-mediated multi-agent patterns become the cross-vendor default; current trajectory points to deployment-grade stability in the second half of 2026, with widespread enterprise adoption following in 2027. Enterprises selecting agent platforms in 2026 should require A2A roadmap commitments from any vendor whose product will participate in cross-vendor agent workflows.","article_url":"https://agentmodeai.com/a2a-agent-to-agent-protocol/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-051","claim":"Enterprise AI governance organisational design resolves to three operating models in 2026: centralised (a single AI governance function owns policy, procurement, audit, and kill-criterion enforcement enterprise-wide), federated (each business unit owns its AI deployments with cross-unit coordination through a small central function), and hybrid (a central function owns regulatory and procurement; business units own deployment operations and ROI accountability). The dominant 2026 pattern in Fortune 500 enterprises is hybrid, because purely centralised models do not scale past 50-100 deployments and purely federated models cannot satisfy EU AI Act Article 9 risk-management documentation consistency. The right model for a given enterprise depends on three variables: deployment count, regulatory exposure, and the maturity of the existing risk-management organisation. The hybrid model is structurally superior to the alternatives once an enterprise crosses approximately 30 production deployments or operates in two or more EU AI Act high-risk Annex III categories.","article_url":"https://agentmodeai.com/centralized-vs-federated-ai-governance/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-052","claim":"Enterprise agentic AI vendor contracts in 2026 require eight specific exit-clause provisions that standard SaaS contract templates do not adequately cover: (1) full audit-log export with retention, (2) trained-state extraction or destruction guarantee, (3) prompt and configuration portability, (4) tool-and-MCP-connector reconfiguration support during transition, (5) named-individual handoff for in-flight deployments, (6) regulatory-evidence preservation through transition, (7) data-residency continuity, (8) liability-tail coverage for agent actions taken before the transition completes. An enterprise that signs an agentic AI contract without these eight provisions has effectively created a one-way procurement decision; the realistic cost of a forced transition without the provisions is materially higher than the contract value, which inverts the procurement leverage. The provisions add typically modest contract complexity but materially change the enterprise's negotiating posture and the vendor's incentive structure during the relationship.","article_url":"https://agentmodeai.com/ai-agent-contract-exit-clauses/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-053","claim":"HIPAA-compliant agentic AI deployment in U.S. healthcare in 2026 requires four conditions that materially constrain vendor selection and architectural design: (1) the vendor offers a BAA covering the specific agent workflow including any subprocessors and any tools the agent calls, (2) the agent's audit log structure satisfies HIPAA 164.312(b) audit controls AND the EU AI Act Article 12 14-field structure simultaneously, (3) PHI flows through agent tool calls are explicitly mapped and authorised under the HIPAA Privacy Rule's minimum necessary standard, (4) the agent's behavioural drift monitoring includes correctness against clinical-decision benchmarks, not just engagement or business-metric benchmarks. Anthropic's three-cloud BAA position (covering AWS, GCP, and Azure deployment surfaces) is structurally distinct in the 2026 vendor landscape and materially expands healthcare deployment options. The OCR's 340% spike in AI-related discrimination complaints (logged in 2025) makes audit-substrate readiness the highest-priority preparatory work for any healthcare AI deployment going into production in 2026.","article_url":"https://agentmodeai.com/hipaa-compliant-agentic-ai-healthcare/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Extracted-text verification failed on two parts of the claim. (1) The asserted 'OCR's 340% spike in AI-related discrimination complaints (logged in 2025)' cannot be located in any primary source: three targeted searches (10 Jun 2026) across HHS OCR publications, the Section 1557 final-rule coverage, enforcement trackers, and trade press surface no AI-specific complaint-volume series from OCR and no 340% figure anywhere. The article attributes the figure directly to OCR with only the OCR homepage as citation. The figure is unanchored and is treated as failed verification, not as pending. (2) 'Anthropic's three-cloud BAA position' is imprecise: per Anthropic's own BAA documentation, Anthropic signs BAAs for the first-party API and HIPAA-ready Claude Enterprise plans; Claude consumed via AWS Bedrock or Google Vertex AI is covered by the hyperscaler's BAA (AWS Artifact; Google Cloud BAA), not by an Anthropic BAA, and an Azure-side Anthropic BAA could not be verified. The deployment-surface breadth is real; the BAA attribution to Anthropic across three clouds is not. The four deployment conditions (BAA-with-subprocessor coverage, dual 164.312(b)+Article-12 logging, minimum-necessary PHI mapping, clinical-correctness drift monitoring) are editorial architecture and stand. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-054","claim":"Public-sector agentic AI deployment in 2026 operates under five constraints that materially narrow the vendor and architectural options compared to private-sector deployment: (1) FedRAMP authorisation (Moderate or High depending on data sensitivity) is required for federal deployments and increasingly for state, (2) sovereign data residency requirements (data and model inference must remain within national or sub-national boundaries), (3) procurement transparency obligations (the deployment, the vendor, and the decision logic typically must be publicly disclosed), (4) explicit accountability under administrative law (decisions affecting individuals are subject to due-process and appeal frameworks that the agent must support), (5) FOIA-equivalent disclosure of audit logs to the public on request. Public-sector deployments cannot reasonably use peer-to-peer multi-agent patterns and cannot accept vendors without published government cloud SKUs; the realistic 2026 options are Microsoft Azure Government, AWS GovCloud-deployed Anthropic, Google Cloud Public Sector, and a small number of specialist government-AI vendors. The NYC MyCity case (claim AM-044) is the canonical 2026 public-sector failure illustrating what happens when the constraints are inadequately addressed.","article_url":"https://agentmodeai.com/public-sector-agentic-ai/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-055","claim":"Retail and logistics agentic AI deployments in 2026 cluster around five workflow patterns with substantially different governance properties: customer-service agents (the Klarna failure case applies directly, claim AM-044), inventory and demand-forecasting agents (operationally lower-risk but with material accuracy requirements), dynamic-pricing agents (carry antitrust exposure that is structurally distinct from other AI risks), supply-chain orchestration agents (multi-party data flows that complicate audit substrate ownership), and returns-and-fraud-detection agents (consumer-protection law exposure including disparate-impact claims). The dominant 2026 production pattern is augmentation rather than replacement of human operators; deployments framed as headcount-replacement have produced reversals at material rates (the Klarna pattern). Retailers and 3PLs (third-party logistics providers) operating across multiple jurisdictions face an additional layer of consumer-protection law fragmentation that the EU AI Act does not pre-empt and that materially affects the deployment scope.","article_url":"https://agentmodeai.com/retail-logistics-ai-agents/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-056","claim":"Enterprise AI agent ROI calculation in 2026 requires a structured eight-input model that captures the costs and benefits the standard SaaS-style ROI calculator misses: (1) per-session-hour or per-task model cost at the deployment's actual usage profile, (2) human-in-the-loop labour cost including approval-gate review time, (3) deployment-layer instrumentation cost (audit substrate, drift monitoring, MTTD detection), (4) regulatory compliance cost amortised across the deployment's revenue, (5) productivity uplift on existing human staff (the augmentation case), (6) avoided cost from reduced incident rate and reduced kill-criterion losses, (7) revenue impact net of service-quality regression risk, (8) the strategic-option value of the deployment's underlying capability. The calculation produces a 90-day ROI checkpoint figure, a 12-month payoff figure, and a kill-criterion threshold. The calculation also produces a sensitivity table showing which inputs drive the ROI most heavily; cost-side sensitivity is typically dominated by inputs 2 and 3, revenue-side by inputs 5 and 7. Most 2026 enterprise AI deployments evaluated against this model break even between months 9 and 18; deployments outside that range are either materially under-investing in instrumentation (faster apparent ROI) or are operating in unfavourable cost structures (longer payoff).","article_url":"https://agentmodeai.com/ai-agent-roi-calculator/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-057","claim":"The enterprise AI agent risk register for 2026 resolves to a 12-column template that captures every risk an enterprise must document under EU AI Act Article 9 and NIST AI RMF Manage function: risk ID, deployment ID, threat class (per OWASP Agentic AI Top 10), likelihood, impact, inherent risk score, control mapping (against the seven-control surface), residual risk score, named accountable individual, review cadence, status, last-reviewed date. The register is operated by the Head of AI Governance, reviewed monthly in the AI governance committee, and queryable in the under-4-business-hour Article 73 incident-response window. The 12-column template integrates the threat surface (OWASP Agentic AI Top 10, claim AM-043), the controls (seven-control surface, claim AM-043), the audit substrate (claim AM-046), and the kill-criterion enforcement (claim AM-047), into a single living artefact. An enterprise that operates the register seriously has substantially completed the Article 9 risk-management system documentation requirement; the register is the single artefact that resolves the cross-reference matrix between operational reality and regulatory framework.","article_url":"https://agentmodeai.com/ai-agent-risk-register-template/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-061","claim":"Production agentic-AI costs at scale routinely run multiples of POC projections, and a layered optimisation programme covering model tiering, vendor prompt caching, batch APIs, context-window discipline, and observability budgeting closes most of the gap.","article_url":"https://agentmodeai.com/the-2m-ai-bill-that-became-200k-the-enterprise-cost-optimization-playbook-for-production-ai-agents/","topic":"enterprise-ai-cost","pub_date":"2025-07-27","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2025-07-27","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-28","verdict":"partial","note":"Rewritten 27-28 Apr 2026 from 27 Jul 2025 WordPress-migrated original. Original used a fictional CTO scene (Marcus Chen, $4.2B logistics company, 9:47 AM Tuesday Seattle), fabricated case figures ($2.1M to $187K monthly, named-company before/after teardowns), fabricated expert quotes (Patricia Williams VP of Engineering at Walmart; David Park Principal at Goldman Sachs), and banned phrases (plot twist, the dirty secret, revolutionary, emoji subheads). Rewrite extracts the verifiable cost-driver categories with primary-source citations from Anthropic's published multi-agent token-ratio research, vendor prompt caching and batch-API pricing pages, McKinsey State of AI, Andreessen Horowitz on LLM inference economics, and Gartner's April 2026 I&O finding. Approved + published 28 Apr 2026."}],"primary_sources":[]},{"id":"AM-063","claim":"AI agents executing financial transactions need a four-control bundle (action-approval gates by blast radius, kill-switch protocols, decision-audit trails, per-action revocation); enterprises shipping agentic-AI without this bundle face CISO governance pressure they cannot satisfy under existing model-risk-management, FFIEC, and EU AI Act expectations.","article_url":"https://agentmodeai.com/your-ai-agents-just-approved-2-7m-in-vendor-payments-and-other-nightmares-keeping-cisos-awake/","topic":"agentic-ai-governance","pub_date":"2025-07-27","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2025-07-27","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-28","verdict":"partial","note":"Rewritten 27-28 Apr 2026 from 27 Jul 2025 WordPress-migrated original. Original used fictional Seattle CISO scene with fabricated $2.7M case, fabricated cohort scheduling, emoji subheads, and 'battle-tested' hype. Rewrite extracts the verifiable control-set framework with primary-source citations (NIST AI RMF, NIST AI 600-1 Generative AI Profile, FFIEC IT Examination Handbook, SR 11-7, OCC Bulletin 2011-12, ISACA AI Audit Toolkit, Cloud Security Alliance MAESTRO framework). Cross-links to the live AM-037 non-human-identity piece as the identity-layer companion. Approved + published 28 Apr 2026."}],"primary_sources":[]},{"id":"AM-100","claim":"AI-authored + human-signed publications produce more verifiable enterprise-AI commentary than human-only or anonymous-AI alternatives, when the AI authorship is paired with a public claim ledger and dated correction log.","article_url":"https://agentmodeai.com/ai-writes-about-ai-tracked-claims-case/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-101","claim":"Across the named analyst-publication comparable set (Stratechery, The Information, the Substack analyst stack, the Big-4 research blogs, Gartner, Forrester, IDC) as of late April 2026, none maintains a public claim ledger — a tracked register of every primary claim with scheduled reviews, dated verdicts, and a public correction log. The absence is structural, not accidental, and explains why none of the category produces the kind of audit-able commentary the Holding-up system makes possible.","article_url":"https://agentmodeai.com/why-this-publication-has-a-ledger/","topic":"agentic-ai-governance","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-07-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-102","claim":"Among the comparable publications surveyed in AM-101 (Stratechery, The Information, the Substack analyst stack, the Big-4 research blogs, Gartner, Forrester, IDC) as of late April 2026, none uses the disclosed-AI-author + named-human-signatory + public-claim-ledger format. The combination is structurally rare and the rarity is what makes the format consequential, not the disclosed AI authorship alone.","article_url":"https://agentmodeai.com/the-ai-author-signature-decision/","topic":"agentic-ai-governance","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-07-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-103","claim":"Across two of the three Q1 2026 ventures Peter built with Claude (agentmodeai, Rhino-basketball; DealVex pending git-versioning), rework rate measured as deletions / total git churn ranged from 8.1% to 13.5% over the 90-day window from 28 Jan to 28 Apr 2026. The data is meaningfully lower than typical solo-developer projects but substantially higher than the 'AI codes it correctly the first time' marketing narrative implies, supporting the thesis that AI-paired development requires explicit measurement, not assumed productivity.","article_url":"https://agentmodeai.com/learning-ai-by-doing-ai-the-data/","topic":"agentic-ai-governance","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-07-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-104","claim":"Anthropic's withholding of Claude Mythos forces senior IT teams to advance their AI cyber-threat-model timeline by two to three years, and to rebuild three specific assumption sets — patch prioritization, third-party risk on AI infrastructure, and AI procurement diligence — inside Q2 2026.","article_url":"https://agentmodeai.com/claude-mythos-cio-risk-posture/","topic":"agentic-ai-governance","pub_date":"2026-04-27","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-105","claim":"Organizations that have not adopted an offensive-security operating mode (continuous attack-surface validation, AI-augmented internal vulnerability discovery, standing threat-hunting, deception, counter-AI controls) by Q4 2026 will show measurably wider mean-time-to-detect for AI-assisted attackers than peers that have, in industry-survey data published in late 2026 and early 2027.","article_url":"https://agentmodeai.com/offensive-security-cio-clockspeed/","topic":"agentic-ai-governance","pub_date":"2026-04-27","last_reviewed":"2026-04-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-04-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-106","claim":"Loaded human FTE cost ($90K-$180K all-in for typical knowledge work) vs total agentic-AI operational cost (token plus orchestration plus integration plus observability plus human oversight) does not favour replacement at parity in 2026 for most roles; the math works for narrow, high-volume, low-judgment task categories and breaks down where regulatory accountability, customer trust, or judgment-under-ambiguity is load-bearing.","article_url":"https://agentmodeai.com/agentic-ai-vs-human-worker-cost-economics/","topic":"enterprise-ai-cost","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — spine is observable from current public deployment cost data and labour-displacement research, per-category quantitative bands tracked against next review cycle. REVIEW: Peter — please verify claim text + cited sources before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-107","claim":"The 2026 insurance market does not yet offer agent-specific E&O policies in any mature form; existing cyber and tech-E&O policies were drafted against human-error and software-defect risk models that don't cleanly map to autonomous reasoning actors. Enterprises shipping agentic-AI face an underwriting gap: the cyber policy may not respond to a loss caused by an agent's reasoning step, and the professional-liability policy may exclude AI-generated outputs entirely. CIOs and CROs need to surface this gap with their broker before the loss event, not after.","article_url":"https://agentmodeai.com/agentic-ai-insurance-and-underwriting/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 as a staged draft (rewriteInProgress: true). Status set to Partial because the underlying market is in a transitional phase and per-carrier wording specifics may shift inside the 60-day review window. REVIEW: Peter to verify (a) the Lloyd's Lab Cohort 12 dating and submissions detail, (b) the Munich Re aiSure agentic-deployment extension claim, (c) the NAIC Model Bulletin scope, (d) whether the AIG CyberEdge and Chubb Integrity+ AI endorsement language descriptions reflect the most recent product updates, and (e) the MGA list (Armilla, Vouch, Coalition, Relm) is currently in market with AI-liability paper before promoting from staged draft to published."}],"primary_sources":[]},{"id":"AM-108","claim":"Agentic-AI data-residency requirements are not cleanly inherited from existing GDPR cross-border transfer practice. Agent context windows, retrieval indexes, and reasoning traces all create new categories of personal-data processing that have to be located, documented, and (for high-risk Annex III deployments) data-resident inside the EEA before EU AI Act Article 16 enforcement opens on 2 August 2026. The deployment topology has to shift to single-region EEA-resident for high-risk systems; hub-and-spoke remains defensible for general-purpose deployments under documented GDPR Chapter V transfer mechanisms.","article_url":"https://agentmodeai.com/agentic-ai-data-residency-eu-ai-act/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — spine is anchored to the Act itself plus current vendor compliance pages, but the four-surface Article-mapping has not yet been tested against an enforced case (the August 2026 enforcement window opens inside the next review cycle). REVIEW: Peter — please verify claim text + Article references + vendor citations before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-109","claim":"Enterprises focused on the headcount-reduction half of agentic-AI transformation are systematically under-budgeting the retraining cost for the residual workforce, and programmes that ship the cuts without simultaneously shipping the upskilling produce a 6-12 month productivity dip that erases the early ROI.","article_url":"https://agentmodeai.com/agentic-ai-retraining-gap-survivors/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — the productivity-dip duration is observable from current public workforce data but the 6-12 month band has not been tested against post-2026 enterprise case data yet. REVIEW: Peter — please verify claim text + cited sources before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-110","claim":"Traditional SLAs (uptime, p95 latency, error rate) are structurally insufficient for autonomous agentic-AI; the four metrics that actually work are action-bounded availability, MTTD-for-Agents, output-distribution drift, and per-class action error budget, and vendors that cannot expose the telemetry these require are not yet production-ready against the 2026 enterprise procurement bar.","article_url":"https://agentmodeai.com/agentic-ai-sla-architecture/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — the four metrics are observable from current SRE/OTel practice but have not been tested as a procurement bar against 2026 vendor SLAs yet. REVIEW: Peter — please verify claim text + cited primary sources (especially the OpenTelemetry GenAI stable-promotion date and the Anthropic/MS Agent Framework reliability docs) before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-111","claim":"The right enterprise playbook for an agent incident in 2026 has six steps that do not appear in any standard SRE handbook — action-class containment before root-cause analysis, reasoning-trace forensics, blast-radius reconstruction across downstream agents and systems, stakeholder notification with the specific failure mode named, regulatory exposure assessment for in-scope deployments, and selective re-enable with degraded-mode guardrails — and CIOs without this playbook will spend their first agent incident discovering it under crisis conditions.","article_url":"https://agentmodeai.com/agent-incident-response-playbook/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — six-step playbook is a synthesis from current SRE practice + AI-specific guidance and has not been tested against a major published agent-incident postmortem yet. REVIEW: Peter — please verify claim text + cited primary sources before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-112","claim":"Healthcare agentic-AI sits across three regulatory regimes that do not compose cleanly — HIPAA on PHI handling and BAA topology, FDA software-as-medical-device guidance on clinical decision support and predetermined change control, and state medical/nursing board licensure rules placing the practitioner as the responsible party of record — and the five-control bundle of BAA-aware architecture, PCCP, clinical-judgement-of-record audit trail, on/off-switch with practitioner attribution, and breach-notification readiness is the minimum defensible architecture for any clinical agentic-AI deployment.","article_url":"https://agentmodeai.com/healthcare-agentic-ai-governance/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-113","claim":"Standard 2026 agentic-AI vendor MSAs contain six contract patterns that systematically transfer risk from vendor to enterprise customer in ways that do not appear in equivalent pre-AI enterprise software MSAs — model-version unilateral-change, training-data ambiguity on customer inputs, usage-cap auto-escalation, indemnification carve-outs for model output, data-residency commitments that don't bind sub-processors, and liability caps tied to fees-paid that don't scale with autonomous-action authority.","article_url":"https://agentmodeai.com/agentic-ai-vendor-contract-gotchas/","topic":"agent-procurement","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-114","claim":"Production agentic-AI in 2026 needs four observability layers — infrastructure, LLM-call, trace, and output — and most enterprise deployments instrument only the cheaper subset (Layers 1 and 2 plus partial Layer 3); the failure modes Layers 3 and 4 catch (multi-step reasoning failure and output-distribution drift) are the ones EU AI Act Article 9 and Article 17 evidence obligations from 2 Aug 2026 onward will require coverage of, and the four layers compose directly into the four AM-110 SLA metrics (action-bounded availability, MTTD-for-Agents, output-distribution drift, per-class action error budget).","article_url":"https://agentmodeai.com/agent-observability-stack-production/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — the four-layer model is observable from current 2026 tool categories and OpenTelemetry GenAI convergence, but the procurement-or-build cost bands are publication estimates and have not been tested across a representative sample of enterprise deployments. REVIEW: Peter — please verify (1) the OpenTelemetry GenAI stable-promotion date (13 Mar 2026) is consistent with what AM-110 cites; (2) the cost-band ranges in the §Share-thoughts template are defensible as our-estimate or need tightening; (3) Datadog AI Observability and New Relic AI Monitoring product names are current; (4) Arize Phoenix open-source/managed dual-form description is accurate; (5) the CNCF OpenTelemetry GenAI working-group framing matches the actual project structure before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-115","claim":"Agent Mode AI publishes a public quarterly review of every claim it has made, with verdict before/after, named primary-source movement, and aggregate verdict-change rate across the corpus. The bulletin runs on a fixed quarterly cadence (end of Apr, Jul, Oct, Jan); the rhythm is the editorial discipline the niche has been missing.","article_url":"https://agentmodeai.com/q2-2026-claim-review-bulletin/","topic":null,"pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-07-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 — the first Quarterly Claim Review Bulletin. The claim itself is recursive: it asserts that the bulletin will ship quarterly, and the next review (30 Jul 2026) tests whether the Q3 bulletin actually appeared. Status starts as 'up' because the claim is currently true (the Q2 bulletin shipped). The verdict at end of July 2026 will move to Holding, Partial (bulletin shipped but on a delayed cadence), or Not holding (no bulletin shipped). REVIEW: Peter — please verify claim text + cadence wording before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-116","claim":"A class of derivative actions is forming in 2025-2026 around board failure to supervise AI deployments under the Caremark line, and D&O carriers are responding at renewal with explicit AI questionnaires and emerging exclusions, materially shifting director liability exposure that most boards have not yet read in their actual policy language.","article_url":"https://agentmodeai.com/directors-officers-insurance-ai-supervision-claim/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-117","claim":"AI Bill of Materials (AI-BOM) is moving from optional security artefact to enforceable procurement requirement in 2026, driven by EU AI Act Article 11 + Annex IV technical-documentation requirements (effective 2 August 2026) and the CycloneDX ML-BOM and SPDX 3.0 specifications. Enterprise SBOM programs need three specific extensions (generation path for AI components, AI-specific risk correlation feeds, procurement-side language for AI-BOM delivery).","article_url":"https://agentmodeai.com/ai-bill-of-materials-supply-chain-disclosure/","topic":"agent-procurement","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-118","claim":"As of April 2026 the largest sovereign-wealth and pension funds (NBIM, CalPERS, ABP, OTPP, USS) have published almost no formal AI position papers, despite trillion-dollar AI exposure across portfolios. The structural absence is the signal: AI is being rated by these investors but the rating criteria have not been formally codified, leaving public-company IR teams preparing engagement against expectations the investors have not yet written down.","article_url":"https://agentmodeai.com/pension-fund-sovereign-wealth-ai-policy-void/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-119","claim":"The 2026 cyber-insurance renewal tightening enterprises are experiencing is upstream-driven by reinsurance market repricing of catastrophic AI tail risk (Lloyd's of London, Munich Re, Swiss Re), not by primary-carrier loss data. The reinsurance signal travels via tighter treaty terms, AI-specific exclusions, and elevated retentions, with a 6-12 month lag to primary policies. Enterprise risk officers negotiating against the primary on AI terms have limited room because the carrier's own treaty caps what it can offer.","article_url":"https://agentmodeai.com/reinsurance-market-ai-tail-risk-pricing/","topic":"agent-procurement","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-120","claim":"AI agent deployments touching employee work in EU jurisdictions with co-determination law (Germany BetrVG §87, Netherlands WOR Art. 27, France CSE provisions) require works council consent before activation in 2026. Most US-headquartered AI vendors lack a customer-success workflow for this, producing a class of stalled rollouts that read as 'vendor delay' but are actually compliance gaps. Total EU-site timeline from selection to production is 6-9 months when handled well, 12-18 when consultation begins late.","article_url":"https://agentmodeai.com/works-council-ai-agent-deployment-eu/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-121","claim":"AI in IT operations in mid-2026 delivers measurable productivity gains (UK Government Digital Service trial: 26 minutes per user per day across 20,000 staff; BT pilot: 35% case-resolution-time reduction with named CIO on the record; ServiceNow's own help desk: 90% L1 deflection in vendor-internal optimal conditions) but the staff-reduction story is structurally smaller than vendor pitches suggest. Gartner finds only 11% of Fortune 500 companies have actually cut support headcount via AI; Forrester reports 55% of AI-attributed layoffs are regretted and roughly half are reversed; CRMArena-Pro shows multi-step agent reliability at ~35%. The cost saving lands first on the BPO/contractor line, second on contractor spend, and only slowly and controversially on direct headcount. Agentic L2/L3 remediation remains pilot-stage: per Gartner's October 2025 survey of 360 IT app leaders, only 15% are considering, piloting, or deploying fully autonomous agents, and Gartner predicts >40% of agentic AI projects will be cancelled by end-2027.","article_url":"https://agentmodeai.com/ai-it-operations-reality-check/","topic":"enterprise-ai-cost","pub_date":"2026-05-02","last_reviewed":"2026-05-02","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-02","verdict":"holding","note":"Claim created at publish."},{"date":"2026-05-02","verdict":"partial","note":"Klarna walk-back primary-source upgrade — added Siemiatkowski verbatim quotes via Bloomberg-cited-by-Fortune (9 May 2025) and the Uber-style freelance hiring detail via Entrepreneur. Closes the highest-priority evidence gap from the source dossier."}],"primary_sources":[]},{"id":"AM-122","claim":"The four credible 2026 agent-evaluation platforms (DeepEval, Braintrust, LangSmith, Patronus AI) do not compete on capability rank; each fits a distinct deployment shape (engineering-led eval-as-code; SaaS-first eval-as-product; LangChain-stack-native bundled with observability; research-grade hallucination + simulation), and picking by capability matrix produces the wrong procurement outcome for most enterprises. The structurally load-bearing eval-vs-observability split (companion piece AM-123) compounds this: 'is the agent right' and 'what did the agent do' are different procurement decisions answered by different platforms.","article_url":"https://agentmodeai.com/agent-eval-frameworks-deepeval-braintrust-langsmith-patronus/","topic":"agent-procurement","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-123","claim":"Evaluation answers 'is the agent right'; observability answers 'what did the agent do'. The four credible 2026 agent-observability platforms (Langfuse, Arize, Helicone, LangSmith) split cleanly on a single structural axis: open-source-first vs SaaS-first. Helicone has been in maintenance mode since 3 March 2026 (founders joined Mintlify) and should not be selected for greenfield 2026 deployments. Production deployments need both eval and observability; the procurement decisions are different and conflating them produces SLA architecture that fails its first incident.","article_url":"https://agentmodeai.com/agent-observability-langfuse-arize-helicone-langsmith/","topic":"agent-procurement","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-124","claim":"Pharma and life sciences agentic AI in 2026 inherits five regulatory regimes simultaneously (21 CFR Part 11, GxP under GAMP 5 Second Edition, EMA Annex 11 in 2025-2026 revision, the EMA Reflection Paper on AI in the medicinal product lifecycle, and the EU AI Act). The audit substrate that satisfies any one regime does not by default satisfy the others. The 2026 procurement gap is treating the regimes as substitutable. Four conditions materially constrain compliant deployment (validated computerised system status under GAMP 5 plus CSA; 17-field audit trail covering Part 11 + Annex 11 + Article 12 simultaneously; ALCOA+ data integrity with contemporaneous, original, enduring records; EU AI Act high-risk-system registration with Article 11 technical file plus Article 16 post-market monitoring). Three vendor postures emerge in market (pre-validated Category 4 packaging; general-purpose platform plus customer-validated wrapper; open-source stack plus customer-engineered audit substrate).","article_url":"https://agentmodeai.com/pharma-life-sciences-agentic-ai-21-cfr-part-11/","topic":"agentic-ai-governance","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-08-01","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-125","claim":"ITSM agent procurement in 2026 is not three independent vendors but two acquirer ecosystems plus one product line at the intersection: ServiceNow (acquirer; Now Assist native plus Moveworks acquired 15 Dec 2025 for $2.4B closed consideration vs the announced $2.85B) and Automation Anywhere (acquired Aisera Nov 2025). The procurement decision in 2026 is shaped less by the feature matrix than by the post-acquisition reality. Picking by feature matrix without mapping the acquirer's strategic interest produces the wrong answer. ServiceNow Now Assist is the bolt-on for organisations already on ServiceNow; Moveworks is the omnichannel layer (still standalone branding, ServiceNow-owned); Aisera is the auto-resolution play that competes on closure rate, now under Automation Anywhere's portfolio.","article_url":"https://agentmodeai.com/servicenow-now-assist-vs-moveworks-vs-aisera/","topic":null,"pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-126","claim":"The OWASP Agentic AI Top 10 names what to defend against; it does not say how to test that the defences work. The 2026 enterprise red-team for agentic systems is a distinct discipline from generalised pen-testing, with its own methodology (four disciplines: prompt injection, tool misuse, context-window attacks, multi-turn objective drift), tooling stack (PyRIT v0.13.0, Garak, custom harnesses, MITRE ATLAS for structured threat-modelling vocabulary), evidence model (six-section report including ATLAS technique mapping plus residual-risk plus EU AI Act Article 12 substrate alignment plus Article 16 post-market monitoring recommendations), and procurement decisions (in-house vs specialist-vendor vs hybrid). Most enterprises run the wrong test (generalised application pen-test) and pass it; the passing report is the procurement evidence that produces false confidence.","article_url":"https://agentmodeai.com/agent-red-teaming-owasp-companion/","topic":"agentic-ai-governance","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-127","claim":"Of the eleven claims this publication has published against the 2 August 2026 EU AI Act enforcement deadline, the four operational-evidence claims (AM-108 data residency, AM-046 audit-evidence under four hours, AM-117 AI-BOM procurement, AM-120 works council workflow) carry materially higher risk of moving from Holding to Partial in Q3 2026 than the two governance-process claims (AM-047 Head of AI Governance role, AM-051 centralised-vs-federated). Materially higher risk is defined as: at least three of the four operational-evidence claims will be downgraded to Partial or Not holding by 1 October 2026, while at least one of the two governance-process claims will remain Holding.","article_url":"https://agentmodeai.com/90-days-eu-ai-act-enforcement-what-corpus-says/","topic":"agentic-ai-governance","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-01","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-128","claim":"The MIT NANDA 'GenAI Divide' 95% pilot-failure statistic (August 2025) is widely cited in 2026 enterprise procurement decks as evidence that 95% of AI projects fail. The underlying methodology measures something narrower and more specific: 95% of 300 analysed AI projects delivered no measurable P&L impact, where 'no measurable impact' is largely a function of pilots not having documented pre-deployment baselines, not a function of pilots failing technically. The structurally interesting findings underneath the headline (build-vs-buy 67%-vs-22% spread, 40%-licensed / 90%-shadow-using gap, marketing-vs-back-end deployment misdirection, the static-error / learning-gap pattern) are more useful for procurement teams than the headline number, and they update against the Stanford 12/88 bimodal ROI distribution (claim AM-029) cleanly.","article_url":"https://agentmodeai.com/the-mit-genai-pilot-failure-claim/","topic":"enterprise-ai-cost","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-129","claim":"No mid-market enterprise has produced a documented +240% ROI in 90 days from agentic AI under audited conditions. Read against McKinsey State of AI 2025 (n=1,993; 23% scaling, 17% EBIT-attribution at 12-month horizon), MIT NANDA GenAI Divide (95% of pilots produce no measurable P&L impact, 67% buy vs 22% build success spread), and Stanford Digital Economy Lab Enterprise AI Playbook (12/88 bimodal ROI distribution at 12-18 months), the realistic 90-day mid-market ROI band for the highest-discipline 12% cohort is 20-40% operator-time savings on bounded use cases plus a working pilot pattern that scales into 12-18-month measurable ROI — not the 240% ROI in 90 days the vendor pitch frames it as. The four-artefact 90-day deliverable (documented baseline, bounded production deployment, per-class action error budget, scaling-vs-stop decision) is what the 12% cohort actually produces.","article_url":"https://agentmodeai.com/achieve-240-roi-in-90-days-with-ai-agents-for-mid-market/","topic":"enterprise-ai-cost","pub_date":"2026-05-04","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of three read-against anchors unanchored on re-review. The claim text cites 'Stanford Digital Economy Lab Enterprise AI Playbook (12/88 bimodal ROI distribution at 12-18 months)' and frames the realistic ROI band around 'the highest-discipline 12% cohort'. Full-text verification on 10 Jun 2026 found the playbook contains no 12/88 distribution, no bimodal ROI shape, and no 12-18-month ROI measurement point (full finding at AM-029, correction of 10 Jun 2026). The claim's core negative finding — no mid-market enterprise has produced a documented +240% ROI in 90 days under audited conditions — is unaffected; the McKinsey State of AI 2025 and MIT NANDA legs verify and continue to support it. The '12% cohort' framing has no verifiable referent. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-130","claim":"Agentic AI 2024-2025 produced four distinct classes of evidence the 2026 procurement reader should not collapse into a single 'AI is working' narrative: (1) vendor-published wins inside vendor-controlled environments (ServiceNow internal 90% L1 deflection, framed by Nenshad Bardoliwalla as upper bound conditioned on two decades of structured workflow data the customer does not have), (2) audited customer pilots with active human oversight (BT 35% case-resolution improvement with random checks per Hena Jalil; UK Government Digital Service 26 minutes/day saved across 20,000 staff in Q4 2024; HMRC 28,000-staff M365 Copilot rollout April 2026), (3) public walk-backs (Klarna May 2025 Bloomberg-reported reversal of the 700-agent claim while the original press release stayed live; GitHub Copilot April 2026 token-counting bug; Salesforce Agentforce IT 200-customer reality vs Marc Benioff's launch pitch), and (4) structural failure modes (CRMArena-Pro 35% multi-step agent reliability finding; Carnegie Mellon independent verification at 30-35%; EchoLeak CVE-2025-32711 cross-agent prompt-injection class). Each class produces a different procurement lesson; treating them as one narrative is the most common 2026 enterprise mistake.","article_url":"https://agentmodeai.com/the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025/","topic":"agent-procurement","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-131","claim":"The AI Training Lead role — the human who curates the agent's evaluation set, reviews sampled outputs against it, and partners with the ML engineer on retraining decisions — is now a budget-line for enterprise agentic AI deployments rather than a vendor-bundled professional-services function. Domain experts (five-plus years inside the workflow the agent is meant to assist) outperform pure-ML hires in the role because the work is judgement-heavy, not algorithm-heavy. CIOs that do not budget the role explicitly see deployments fail at the iteration boundary.","article_url":"https://agentmodeai.com/from-it-pro-to-ai-training-lead-the-180k-career-path-nobodys-talking-about/","topic":"agentic-ai-governance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-132","claim":"Enterprise agentic AI ROI in 2026 is bimodal across four independent datasets. Stanford Digital Economy Lab's 2026 Enterprise AI Playbook documents 12% of deployments clearing 300%+ ROI with 88% at or below break-even at 12-18 months. Gartner Q1 2026 Infrastructure & Operations Survey reports 28% of AI projects 'fully paying off'. McKinsey State of AI 2025 (n=1,993) reports 23% scaling with 17% EBIT-attribution at 12 months. MIT NANDA's GenAI Divide reports 95% of pilots produce no measurable P&L impact alongside the 67% buy vs roughly 22% build success spread. The 73%/27% slug rounds the four numbers; the bimodal shape is reproducible and the variable separating the two cohorts is operational discipline (instrumented under GAUGE: governance, audit substrate, use-case maturity, guardrails, evidence/baseline, exit posture), not model selection.","article_url":"https://agentmodeai.com/why-73-of-agentic-ai-projects-fail-and-how-the-27-generate-312-roi/","topic":"enterprise-ai-cost","pub_date":"2026-05-05","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of four legs unanchored on re-review. The claim text attributes '12% of deployments clearing 300%+ ROI with 88% at or below break-even at 12-18 months' to the Stanford DEL 2026 Enterprise AI Playbook. Full-text verification on 10 Jun 2026 found no such figure in that source: the playbook (Pereira, Graylin, Brynjolfsson, Apr 2026) studies 51 successful deployments by design and contains no ROI distribution, no 300%-plus cohort, and no break-even measurement point (full finding at AM-029, correction of 10 Jun 2026). The only verified figure carrying the same 12/88 numerals is IDC research with Lenovo (via CIO.com, Mar 2025): roughly 88% of AI proof-of-concepts never reach production and roughly 12% graduate — a pilot-to-production graduation metric, not an ROI distribution. The Gartner 28%, McKinsey 23%/17%, and MIT NANDA 95% legs verify; they support a small high-performing tail and a large struggling body, but none documents the two-peak bimodal shape the claim asserts. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-134","claim":"The 2026 implementation cut on non-human identity for AI agents resolves on three factors (existing IAM relationship, deployment topology, cross-platform integration burden) across six credible control planes: Okta NHI, Microsoft Entra ID Workload Identities, Auth0, Keycloak, SPIFFE/SPIRE for Kubernetes-native deployments, and AWS IAM Roles Anywhere for hybrid AWS-anchored deployments. The procurement-defensible audit substrate captures three event classes regardless of vendor: identity issuance, authentication, and authorisation.","article_url":"https://agentmodeai.com/agent-identity-iam-architecture-nhi/","topic":"agentic-ai-governance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-135","claim":"EU AI Act Article 50 takes effect 2 August 2026 and creates four distinct transparency obligations requiring different UX implementations: Article 50(1) chatbot interaction disclosure on providers, Article 50(2) machine-readable marking on generative AI output, Article 50(3) biometric categorisation and emotion recognition disclosure on deployers, and Article 50(4) deepfake disclosure on deployers (with the artistic-or-creative-work exception). The procurement-defensible disclosure UX has six properties (visible at the right moment, plain language, persistent or recurrent, linked to a substantive disclosure surface, auditable, updateable). Most enterprises have absorbed the legal text without designing the UX it requires.","article_url":"https://agentmodeai.com/eu-ai-act-article-50-transparency-disclosure/","topic":"agentic-ai-governance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-136","claim":"Across the 24-month window May 2024 to April 2026, every major foundation-model provider (Anthropic, OpenAI, Google, AWS Bedrock, Azure OpenAI) experienced at least one multi-hour outage that exceeded the SLA-credit threshold defined in their published terms. The procurement-defensible posture is multi-provider routing with documented failover and hard-dollar incident liability above the standard SLA-credit cap. Three architectural patterns dominate 2026 production deployments: gateway abstraction (LiteLLM, OpenRouter, Portkey), provider-side regional failover (partial mitigation), and explicit multi-provider provisioning at the application layer.","article_url":"https://agentmodeai.com/foundation-model-uptime-sla-track-record/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-30","next_review":"2026-06-29","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-137","claim":"Agent evaluation in production resolves on three operational components that determine whether the chosen evaluation platform produces useful signal: eval-set design across three layers (50-200 calibration prompts, 30-100 edge-case prompts, 10-50 production-sampled prompts per week), drift detection across three signal classes (output-distribution, score-distribution, tool-use distribution), and a regression-budget framework that forces binary ship/hold decisions (defensible default 5% absolute decline on calibration set, 10% on edge-case set, per release window). The procurement decision (which platform to buy, covered at AM-122) is the easier half; the operational discipline is what most enterprises under-invest in even after buying a platform.","article_url":"https://agentmodeai.com/agent-evaluation-in-production/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-138","claim":"The 2 August 2026 EU AI Act deployer-obligations enforcement window adds three new clause families to the AI MSA red-team checklist that were optional or absent in pre-enforcement contracts: Article 11 technical-file pass-through, Article 16 post-market-monitoring support, and Article 26 deployer-documentation supply. The post-enforcement checklist grows from the 38-item RES-005 v1.0 baseline to roughly 54 items across 11 clause families, with Article 50 transparency UX (covered at AM-135) and foundation-model uptime hard-dollar liability (covered at AM-136) as additional 2026 additions. The asymmetric-instrument observation — that enterprise and operator AI procurement face the same vendor-citation-chain manipulation pattern with different audit instruments — is embedded as a 600-word insert in this piece.","article_url":"https://agentmodeai.com/vendor-msa-renewal-post-eu-ai-act-enforcement/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-139","claim":"Enterprise AI buyers and operator AI buyers face the same vendor-citation-chain manipulation pattern with asymmetric audit instruments, and consume vendor case studies aimed at the other cohort with mirror-image misreads. The enterprise reads the IndieHacker timeline as procurement-cycle benchmark and removes controls under timeline pressure; the operator reads the Fortune-500 efficiency gain as result-attribution and inherits expectation without the operational substrate. The cross-borrow that is procurement-defensible at both scales: enterprises borrow the operator's cancellation-trigger discipline (OPS-051) and the cohort-fit filter (OPS-011); operators borrow the enterprise's MSA red-team scoped down (RES-005), evaluation discipline scaled to weekly (AM-137), and audit substrate at lightweight scale (AM-046). The verification gap is the same gap; the instruments are different; the publication's two-register architecture is the editorial response.","article_url":"https://agentmodeai.com/vendor-case-study-misreads-across-buyers/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-140","claim":"Vendor 'successful pilot' references presented at procurement-committee evaluation transfer to scaled production at the procuring enterprise's measurement and governance regime at roughly the McKinsey 23% rate (n=1,491, Nov 2025); the gap is operational rather than capability-driven and is tractable with six pre-pilot questions a procurement committee can require answered in writing before the contract closes, not after.","article_url":"https://agentmodeai.com/agentic-ai-pilot-to-production-gap/","topic":"agent-procurement","pub_date":"2026-05-06","last_reviewed":"2026-05-06","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-06","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-141","claim":"Agent Mode is the same brand-name shipping in three different product classes in 2026: Microsoft 365 Copilot (productivity-suite agents), Cursor (developer-IDE agents), and GitHub Copilot (code-platform agents). The procurement decision is not feature-comparison; it is which class fits the in-house workflow. Anthropic Managed Agents and OpenAI Agents SDK occupy a fourth category (dedicated agent platforms) that competes for adjacent budgets without using the Agent Mode brand.","article_url":"https://agentmodeai.com/what-is-agent-mode/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-142","claim":"AI agent, AI assistant, and LLM are three structurally different categories in 2026, distinguished by whether the system can reason about a goal (LLM yes), invoke tools to achieve it (assistant adds), and operate autonomously across multi-step workflows (agent adds again). Procurement that conflates the three optimizes the wrong axis: model-quality bake-offs decide the LLM tier, governance scaffolding decides the assistant tier, and operational preconditions (registry, baseline, change-management, threat model) decide whether an agent can scale at all.","article_url":"https://agentmodeai.com/ai-agent-vs-assistant-vs-llm/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-143","claim":"An AI Bill of Materials in 2026 is the audit-ready inventory of every model, dataset, training source, evaluation method, and deployment dependency in a production AI system. Most enterprises do not yet ship one; the EU AI Act Article 16 deployer-documentation obligations make it mandatory in scope by 2 August 2026. Six layers belong on the BOM: foundation model + version + provider, training datasets + provenance + opt-out signals, fine-tuning data, evaluation methodology + scores, system prompts + guardrails, deployment dependencies (vector DB, RAG sources, MCP servers, agent orchestrator). CycloneDX-AI is the emerging machine-readable format; SBOM under Executive Order 14028 is the precedent.","article_url":"https://agentmodeai.com/ai-bom-enterprise/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-144","claim":"For a Microsoft-stack enterprise in 2026, Microsoft 365 Copilot Agent Mode is the lower-friction agent-platform choice if the workflow already lives in Microsoft Graph; it is structurally weaker on multi-vendor deployment, model-portability, and platform-independence than dedicated agent platforms (Anthropic Managed Agents, OpenAI Agents SDK, Vertex AI Agent Builder). The procurement decision turns on three questions: (1) is the workflow Microsoft-resident, (2) is multi-vendor model selection a hard requirement, (3) is the agent's primary surface productivity-suite or workflow-orchestration.","article_url":"https://agentmodeai.com/microsoft-copilot-agent-mode-enterprise/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-145","claim":"AI vendor switching in 2026 is bound primarily by contract terms — exit clauses, data-portability obligations, model-deprecation rights — not by technical migration cost. Seven clause patterns repeatedly create the lock-in most enterprises only discover at year two of the relationship: (1) data-portability scope narrowness, (2) model-deprecation rights without credit, (3) sub-processor expansion without consent, (4) output-IP ambiguity, (5) pricing-tier rebalancing mid-contract, (6) agent-uptime SLA definition gaps, (7) audit-evidence retention obligations. Vendor consolidation (Moveworks→ServiceNow Dec 2025, Aisera→Automation Anywhere Nov 2025) and model deprecations make this a 2026 procurement story.","article_url":"https://agentmodeai.com/ai-vendor-exit-clauses-checklist/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-146","claim":"A vendor claim of 'ready-to-run' agentic AI that does not name (a) the specific task being measured, (b) the baseline against which accuracy is reported, and (c) the methodology by which the measurement was produced is not procurement evidence regardless of how the rate is described in marketing; the 2026 industry baseline for procurement-credible accuracy disclosure is the Anthropic Cohort A pattern (red-team rates with named attack corpus, pre/post-mitigation deltas, named patch cadence) on the vendor side and the academic-benchmark pattern (CRMArena-Pro 35% multi-step reliability with defined CRM task corpus, CMU TheAgentCompany 30-35% reproduction range, WebArena ~36% browser-agent ceiling) on the methodology side; vendor 'ready-to-run' positioning without equivalent disclosure leaves the deploying enterprise inheriting the methodology gap as an audit-defense burden.","article_url":"https://agentmodeai.com/agentic-ai-accuracy-claims-task-baseline-methodology/","topic":"agent-procurement","pub_date":"2026-05-09","last_reviewed":"2026-05-09","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-05-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-147","claim":"The Firefox 150 / Claude Mythos disclosure (November 2025) marks the operational shift in agentic AI code auditing from 'AI can find bugs' (true since 2023, but blocked from production CI by the false-positive rate that earlier read-only GPT-4 / Sonnet 3.5 attempts produced) to 'agentic verification clears the false-positive wall by building and running its own test cases before reporting'; the procurement-deck consequence is that CI-time agentic auditing becomes the default expectation for any shipping enterprise software in 2026, and three derived questions belong in any software-vendor procurement (does the vendor's CI pipeline include an agentic-auditing step; what is the vendor's disclosure posture when bugs are found in their own product by agentic tools; what is the vendor's posture on the dual-use risk that the same pipeline architecture works in reverse, as the reported Anthropic investigation of unauthorized Mythos use via a third-party vendor environment makes explicit).","article_url":"https://agentmodeai.com/agentic-code-auditing-firefox-claude-mythos-procurement-read/","topic":"agent-procurement","pub_date":"2026-05-10","last_reviewed":"2026-05-10","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-05-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-148","claim":"GPT-5.5 (released 23 Apr 2026) and Claude Opus 4.7 (released 16 Apr 2026) are not substitutable models for an enterprise running both agentic-coding workloads and knowledge-work workloads in 2026: GPT-5.5 leads the public evaluation evidence on agentic-coding and computer-use surfaces (Terminal-Bench 2.0 82.7% vs 69.4%; GDPval 84.9% vs 80.3%; FrontierMath Tiers 1-3 51.7% vs 43.8%) and runs roughly 72% fewer output tokens than Opus 4.7 on identical coding tasks per Artificial Analysis; Opus 4.7 leads the public evaluation evidence on contamination-resistant coding, finance, and vision-reasoning surfaces (SWE-Bench Pro 64.3% vs GPT-5.4 57.7%; Finance Agent v1.1 64.4%; CharXiv reasoning 78.3%; GPQA Diamond 94.2%) and reports a 36% AA-Omniscience hallucination rate against GPT-5.5's 86% on the same independent evaluation, a 50 percentage-point spread that is the load-bearing data point of any 2026 single-model standardisation decision. The procurement-architecture answer for an enterprise running both workload types is three-tier routing (GPT-5.5 with Codex for agentic coding; Opus 4.7 plus retrieval augmentation for knowledge work; Mythos-via-Glasswing or Opus 4.7 with verification layer for frontier and high-stakes-verification work), not single-model standardisation.","article_url":"https://agentmodeai.com/split-verdict-gpt55-opus47/","topic":"agent-procurement","pub_date":"2026-05-10","last_reviewed":"2026-05-10","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-05-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-149","claim":"In production agentic systems documented across the publicly observable 2025–2026 deployments, the observable band of internal LLM calls per user-facing request sits between 1:18 and 1:60 across documented deployments, with tail cases regularly exceeding 1:400 — meaning unit-economics, latency budgets, and observability scopes built on a 1:1 mental model under-provision by one to two orders of magnitude.","article_url":"https://agentmodeai.com/agent-fan-out-problem-llm-call-amplification/","topic":"enterprise-ai-cost","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-07-11","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-150","claim":"Across the publicly documented 2025–2026 enterprise deployments, single-agent architectures with structured tool-calling outperform multi-agent orchestrations on accuracy, cost, and MTTD for tasks below approximately 12 distinct tool-domains; multi-agent only pays back above that threshold and only when inter-agent state is bounded by a shared structured artifact rather than free-text handoff.","article_url":"https://agentmodeai.com/single-agent-vs-multi-agent-decision-framework/","topic":"agentic-ai-governance","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-07-11","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-151","claim":"Across the 2025–2026 documented deployments at AmLaw 100 firms, agentic AI captures durable value in three of the six billable-hour sub-tasks (document review, precedent retrieval, deposition prep) and produces a net malpractice-risk increase in two (legal drafting submitted as final, citation generation) vs a junior-associate-drafted equivalent at the same time-to-delivery; the remaining sub-task (client communication) is bounded by professional-conduct rules, not technology.","article_url":"https://agentmodeai.com/agentic-ai-legal-services-billable-hour/","topic":"agentic-ai-governance","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-08-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-152","claim":"Across the publicly disclosed 2025-2026 U.S. federal and EU member-state agentic AI procurements, contract renewals are running materially below the broader enterprise SaaS renewal benchmark — driven primarily by audit-evidence failures under OMB M-24-10 §5 and EU AI Act Article 12, not by technical performance — and the renewal-rate gap is the leading early indicator that public-sector agentic AI is following the Salesforce-for-government 2010s adoption curve, not the cloud-for-government 2015s curve.","article_url":"https://agentmodeai.com/public-sector-agentic-ai-procurement-record/","topic":"agent-procurement","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-07-11","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-153","claim":"Of the 8 most-cited enterprise agentic AI vendor claims made in Q1 2026 (Salesforce Agentforce, Microsoft Copilot Agent Mode, Google Gemini Enterprise, Anthropic Claude for Enterprise, OpenAI Agents SDK, ServiceNow AI Agents, Workday Illuminate, SAP Joule), a minority remain Holding at 90-day review, a majority sit at Partial with at least one falsified component, and customer-cited ROI claims hold materially better than vendor-cited ROI claims — meaning the citation-source of an enterprise AI claim is a stronger predictor of its 90-day durability than the size of the vendor making it.","article_url":"https://agentmodeai.com/enterprise-agentic-ai-quarterly-record/","topic":"vendor-trajectory","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-08-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-154","claim":"Enterprise AI compute growth in the credible 10x to 100x range by 2030 will absorb most of the renewable-buildout headroom the energy transition depends on, extending fossil reliance by roughly a decade unless enterprise IT functions begin modelling AI energy demand in kWh and including it in cloud and on-premise procurement criteria.","article_url":"https://agentmodeai.com/ai-energy-consumption-enterprise/","topic":"enterprise-ai-cost","pub_date":"2026-05-15","last_reviewed":"2026-05-15","next_review":"2026-08-13","verdict":"holding","verdict_history":[{"date":"2026-05-15","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-155","claim":"The Cyber Safety Review Board's April 2024 report on Storm-0558 documented four credential-management failures at Microsoft (a signing key seven years past rotation; an environment-separation boundary enforced procedurally rather than technically; a crash-dump leak the existing scanning could not see; an anomaly-detection baseline that did not exist for the credential class). All four conditions are reproduced in most enterprise AI agent deployments in 2026: long-lived agent credentials without rotation policy, dev/staging/production credentials promoted without re-issuance, runtime telemetry that leaks short-lived tokens without scanning, no issuance-and-use baseline per agent. The CSRB report is forward-readable as a structural map of where AI agent identity programmes fail, not a Microsoft-specific post-mortem. The blast radius is wider for AI agents than it was for Storm-0558 because the action surface authorised by a compromised AI agent credential routinely includes writes, transactions, and downstream tool-use chains, where the Storm-0558 attacker had read-only mail access from one credential.","article_url":"https://agentmodeai.com/non-human-identity-after-the-csrb-report/","topic":"non-human-identity","pub_date":"2026-05-16","last_reviewed":"2026-05-16","next_review":"2026-08-14","verdict":"holding","verdict_history":[{"date":"2026-05-16","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-156","claim":"The detection lag observed in Samsung Electronics' April 2023 ChatGPT incidents (three confidential pastes discovered after the fact by internal audit and self-report, leading to the 2 May 2023 restriction memo) was not a Samsung-specific operational failure. It was the structural output of running enterprise DLP, designed against email/file/removable-media egress channels, against a new egress class (paste-into-chat-interface) that the controls were not built for. Three years later, the structural gap remains the dominant detection failure in enterprise shadow-AI programmes, with the pattern now inverted: the 2023 case was unsanctioned external tools, the 2026 case is agentic capability silently activating inside approved tools (Microsoft 365 Copilot agents acquiring write capability, Custom GPTs created against corporate accounts, MCP servers connected by approved IDEs). The 2026 case is harder to detect because the egress destination is an approved vendor and the AI capability sits behind a procurement approval that did not assess the capability surface. The operational test for whether a programme has closed the Samsung gap is a 24-hour AI-capable-surface inventory, a confidential-document trace test, and an automatic update path when vendors ship new AI features into approved tools.","article_url":"https://agentmodeai.com/shadow-ai-detection-lag-after-samsung/","topic":"shadow-ai-discovery","pub_date":"2026-05-16","last_reviewed":"2026-05-16","next_review":"2026-07-15","verdict":"holding","verdict_history":[{"date":"2026-05-16","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-157","claim":"Three independently-disclosed CVE classes in May 2026 (Microsoft Security Response Center's CVE-2026-25592 and CVE-2026-26030 against Semantic Kernel on 7 May 2026; OX Security's MCP STDIO supply-chain advisory traversing every published MCP implementation regardless of language; the Windsurf 1.9544.26 prompt-injection-to-MCP-registration path) share a single structural property: in the default configuration of 2026 agent frameworks, tool-configuration is treated as data the model is allowed to author, which means the deployer's allowlist is enforced against the configured tools rather than against the model's ability to mutate the configuration. The patch surface is therefore the framework default, not the deployer's wrap. The conventional 2024–2025 enterprise treatment of prompt injection — sandbox the agent's reachable surface at deployment time — is necessary but no longer sufficient. The procurement template for an agent vendor must add five framework-layer attestations (tool-configuration as a privileged operation, runtime enumeration of the tool-configuration surface, configuration-mutation telemetry, coordinated-disclosure record on framework-layer issues, MCP protocol-revision commitment) on top of the deployer-control questions that remain in place.","article_url":"https://agentmodeai.com/prompt-injection-rce-threshold-semantic-kernel-mcp-cves/","topic":"agentic-ai-governance","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-07-16","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-158","claim":"Most enterprises in the EU AI Act high-risk-system in-scope cohort (Annex III categories: biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice) will not have a documented conformity-assessment artifact, an operational post-market monitoring telemetry pipeline, and an Article 13 model-card-and-instructions-for-use production cadence in place by the 2 August 2026 activation. The gap is not legal interpretation, which outside counsel can answer in days. It is a budget gap on three operating-expense lines (conformity-assessment headcount, audit-evidence pipeline infrastructure, model-card production cadence) that the chief financial officer has not yet been asked to size and that the audit committee has not yet authorised. The procurement record, posted-position count, and Q2 2026 enterprise-filing line items together suggest the cohort is mid-cycle on acquisition and pre-production on operational delivery.","article_url":"https://agentmodeai.com/eu-ai-act-readiness-budget-high-risk-systems/","topic":"regulatory-readiness","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-08-15","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-159","claim":"Anthropic's financial-services agent stack announced on 5 May 2026 (ten agents covering investment-banking workflows, Moody's data partnership, full Microsoft 365 integration) will reach a documented production deployment at a tier-1 bank — defined as a published case study, board-disclosed P&L impact, or CIO-level public attribution — by 1 September 2026. If it does, the vertical-specialised stack becomes the procurement default for high-headcount, high-document-throughput workflows in 2027 and horizontal-only platforms face a squeeze on the enterprise contracts where vertical depth is the deciding factor. If it does not, the Wall Street launch is a finance-specific anomaly driven by the Moody's data partnership and the unusual document-shape concentration in investment banking, and horizontal platforms remain the procurement default for cross-functional enterprise adoption. The procurement-template implication for non-finance CIOs is operational now regardless of the predictive outcome: vendor questionnaires must include a vertical-stack roadmap question, MSAs must include an early-renegotiation right triggered by vendor vertical-stack shipments, and 2026 procurement should run against multiple vendors with materially different vertical bets (Anthropic vertical-depth-first; Google platform-and-protocol-first; OpenAI horizontal-with-services-overlay; Microsoft horizontal-with-incremental-vertical-layering).","article_url":"https://agentmodeai.com/anthropic-wall-street-agents-cio-cross-industry-read/","topic":"vendor-trajectory","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-08-15","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-160","claim":"Andrej Karpathy's 19 May 2026 announcement that he has joined Anthropic, paired with Anthropic's confirmed framing that he will lead a team focused on using Claude to accelerate pre-training research (under team lead Nick Joseph), is a foundational-layer vendor-trajectory signal that composes with the 5 May 2026 Wall Street agents launch (AM-159) to describe Anthropic operating on both ends of the platform stack simultaneously — vertical-depth-first on the application layer and name-recognition-first on the pre-training layer. The mandate (Claude accelerating Claude) is more procurement-relevant than the hire itself, because it is a public commitment to recursive self-improvement of the model line at the foundational layer rather than at the application layer. By 17 August 2026, observable evidence in the AI-research community will or will not appear across four markers: (1) a published paper from Anthropic's pre-training team describing a Claude-in-the-loop component with measurable productivity or capability impact; (2) a Claude release crediting Claude-assisted research methodology in the development cycle; (3) public commentary from Karpathy or Anthropic leadership on team progress beyond the launch-day framing; (4) Anthropic-attributed performance gains on community-authoritative benchmarks. Procurement-template implication: AI-vendor questionnaires should add a model-improvement-methodology disclosure field, and multi-year MSAs should add a research-roadmap-attestation clause requiring thirty-day advance notice on material methodology changes.","article_url":"https://agentmodeai.com/karpathy-joins-anthropic-cio-vendor-trajectory-read/","topic":"vendor-trajectory","pub_date":"2026-05-19","last_reviewed":"2026-05-19","next_review":"2026-08-17","verdict":"holding","verdict_history":[{"date":"2026-05-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-161","claim":"Senior IT leaders that frame internal AI-and-jobs communications at the job level ('will AI replace your role?') produce defensive postures from employees, lower reskill take-up, and the under-budgeted reskill line AM-109 documents. The task-level frame — which tasks shift on which horizon, which moats hold, which residual skills (agent output review, exception escalation routing, prompt and policy maintenance, vendor evaluation) the surviving role requires — is both more honest about what workers see at their desks and the only frame that resolves into the four skill gaps determining whether the post-displacement function actually works. The operational move for CIOs in 2026-2027 is to replace function-wide reassurance with task-level analytical resources teams can engage with directly, and to use those resources as the basis for role-by-role conversations rather than function-wide town halls. The task-level conversation should run before the reskill budget conversation, because the task inventory is the input the budget line needs.","article_url":"https://agentmodeai.com/how-ai-changes-jobs-task-level-frame/","topic":"enterprise-ai-cost","pub_date":"2026-05-20","last_reviewed":"2026-05-20","next_review":"2026-08-03","verdict":"holding","verdict_history":[{"date":"2026-05-20","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-162","claim":"Karpathy's seat at Anthropic will remain an IC/research-lead role — not a VP or exec-hierarchy title — through at least end of 2026.","article_url":"https://agentmodeai.com/karpathy-anthropic-bench-not-org-chart/","topic":"enterprise-ai-cost","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-12-31","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-163","claim":"SAP's Sapphire 2026 Autonomous Suite — which embeds more than 50 domain-specific Joule AI Assistants across finance, supply chain, HR, procurement, and CX, with Claude powering the finance, procurement, and supply chain agents — materially changes the ERP renewal calculus for enterprise CIOs: the AI agent layer is now inside the RISE with SAP and SAP GROW contract structure rather than a separately evaluated point-tool purchase, and most H2 2026 renewal teams have not yet built a line item for the point-tool displacement this creates.","article_url":"https://agentmodeai.com/sap-sapphire-joule-agents-erp-renewal-cio/","topic":"agent-procurement","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-08-20","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-164","claim":"Anthropic's May 21 2026 discussions with Microsoft to adopt Maia 200 inference chips, read alongside the same-day SpaceX filing disclosing a $1.25B/month compute contract through May 2029, reveals that the foundation-model inference stack is visibly diversifying from commodity Nvidia hardware to hyperscaler-proprietary silicon — a structural change that is currently invisible in standard enterprise AI vendor questionnaires and that introduces a triple dependency (model vendor, cloud provider, silicon provider) into the procurement risk map.","article_url":"https://agentmodeai.com/anthropic-microsoft-maia-chip-inference-infrastructure/","topic":"vendor-trajectory","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-07-21","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-165","claim":"The gap between AI initiative volume (97 percent of enterprises have active AI programmes, per D&B's 2026 AI Momentum Survey of 10,000 businesses across 32 countries) and AI data readiness (5 percent say their data is adequately ready) is not primarily a technology problem: it is a budget-allocation failure in which enterprises funded model-layer spend without funding the prerequisite data-infrastructure investment, and enterprises that correct the allocation by treating data infrastructure as the prerequisite rather than the follow-on will reach meaningful scale in 2026 before those that do not.","article_url":"https://agentmodeai.com/enterprise-ai-data-readiness-gap-scaling-problem/","topic":"enterprise-ai-cost","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-08-20","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-166","claim":"AI productivity gains create workforce reduction pressure through two structural mechanisms: demand ceilings in mature markets and competitive price compression from industry-wide AI adoption. Both mechanisms resolve independently in the direction of fewer workers for the same revenue, regardless of management intent.","article_url":"https://agentmodeai.com/ai-productivity-demand-ceiling-workforce/","topic":"enterprise-ai-cost","pub_date":"2026-05-23","last_reviewed":"2026-05-23","next_review":"2026-08-21","verdict":"holding","verdict_history":[{"date":"2026-05-23","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-167","claim":"A 2026 enterprise contracting for a third-party AI agent platform is, in almost every case, accepting a vendor-issued non-human identity into its environment with the authority to read, write, transact, and call further agents. Standard 2026 agentic AI master service agreements govern data classes, SLAs, pricing, and exit terms but do not govern the identity primitive the vendor uses for the agent's credentials, the rotation cadence, the customer's right to inventory and audit them, or the vendor's disclosure obligations if a credential class is compromised on the vendor side. The four procurement clauses that close the gap (identity primitive disclosure, rotation cadence and audit right, vendor-side breach disclosure, customer-side revocation control) are absent from the standard MSA templates of every major agentic AI platform reviewed in early 2026 and are present only in customer-redlined versions used by procurement-mature enterprises.","article_url":"https://agentmodeai.com/nhi-procurement-clause-gap/","topic":"non-human-identity","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-168","claim":"The dominant 2026 shadow-AI gap is not unsanctioned vendors but sanctioned vendors that have shipped agentic capabilities inside already-approved tenants without triggering a re-evaluation in the customer's SaaS approval process. The canonical examples are Microsoft 365 Copilot Studio inside approved M365 tenants, Slack AI and Slack agent platform inside approved Slack workspaces, Notion AI agents hub inside approved Notion workspaces, ServiceNow Now Assist inside approved ServiceNow tenants, Atlassian Rovo inside approved Atlassian estates, and Salesforce Agentforce inside approved Salesforce contracts. Existing 2024-era shadow-AI discovery playbooks (oriented to vendor-discovery and DLP egress detection) do not surface this class because the vendor is in-policy and no egress boundary is traversed. The structural fix is procurement-side: a re-evaluation trigger inside the SaaS approval policy that runs the original data-class and risk-assessment workflow against any new agentic capability inside an existing tenant within 30 days of activation.","article_url":"https://agentmodeai.com/approved-tool-unapproved-capability-shadow-ai/","topic":"shadow-ai-discovery","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-169","claim":"The three open agent protocols active in 2026 (Anthropic's Model Context Protocol, Google's Agent2Agent protocol contributed to the Linux Foundation, and Meta's Llama Stack) are not on a convergence trajectory, and the four major proprietary agentic platforms (Microsoft Copilot Agent, Salesforce Agentforce, SAP Joule, ServiceNow Now Assist) do not adopt any of the three as first-class defaults. The cost of being wrong on the model choice is low because swapping models is a configuration change. The cost of being wrong on the protocol choice is high because the locked asset is the tool inventory — the MCP servers, A2A endpoints, Llama Stack tool plugins, or proprietary extensions the customer has built or commissioned — and re-platforming the tool inventory is the long-tail engineering bill. Standard 2026 agentic AI MSAs do not include the three procurement clauses (protocol portability disclosure, tool inventory exit terms, protocol-roadmap commitment) that would price the protocol-roadmap optionality back to the customer.","article_url":"https://agentmodeai.com/agent-protocol-tax-mcp-a2a-llama-stack/","topic":"agent-procurement","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-170","claim":"As of mid-2026, the major enterprise agent platforms enable persistent agent memory with retention defaults, residency locations, encryption-at-rest ownership models, erasure-propagation pathways, and audit-evidence export capabilities that are not surfaced in standard procurement, leaving agent memory outside the enterprise data-retention register and the Article 30 record of processing activities. The compliance surface exists already (GDPR Article 5(1)(e) storage-limitation, Article 17 right to erasure, Article 30 records of processing, EU AI Act Article 12 record-keeping) but no AI-specific regulation has yet named persistent agent memory as a governed data class, leaving the obligation in force and the implementation gap unaddressed by procurement and identity-governance teams in most 2026 enterprises.","article_url":"https://agentmodeai.com/agent-memory-governance/","topic":"agentic-ai-governance","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-171","claim":"As of mid-2026, the majority of enterprises running production AI agents cannot terminate a misbehaving agent within their own stated incident-response window, because containment is specified as kill criteria in the risk register rather than built and tested as a runtime control plane with the four primitive actions (purpose binding, kill switch, network isolation, credential revocation). Kiteworks' 2026 Data Security and Compliance Risk Forecast measured the gap at 60% cannot terminate quickly, 63% cannot enforce purpose limitations, 55% cannot isolate networks, with the government-sector figures materially worse. Microsoft Agent 365 with Intune and Defender (GA 1 May 2026, runtime-controls preview from June 2026) is the first major-platform consolidation of the four primitives in a customer-administered control plane, which moves the question from engineering integration to procurement evaluation but does not resolve the cross-platform standardisation gap.","article_url":"https://agentmodeai.com/agent-kill-switch-containment-architecture/","topic":"agentic-ai-governance","pub_date":"2026-05-26","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-172","claim":"As of mid-2026, US enterprises do not need new federal AI legislation to be exposed on AI governance; AI-touching workflows already fall under HIPAA Security Rule access and audit controls (45 CFR 164.312), GLBA Safeguards Rule (16 CFR Part 314) access-control and incident-notification obligations, SEC cyber-disclosure rules (Item 106 of Regulation S-K and Item 1.05 of Form 8-K, with the 4-business-day materiality clock), and FTC Section 5 deception and unfairness jurisdiction at the data layer. The structural pattern, captured in the UK ICO's May 2026 'AI-powered cyber threats' guidance and the seven threat categories it names, is that AI governance has become data governance; the most common 2026 implementation gap is the fragmented audit log (Kiteworks 2026 Forecast: 33% lack audit trails entirely, 61% have fragmented logs), not the absence of regulation.","article_url":"https://agentmodeai.com/ai-governance-data-governance-us-frameworks/","topic":"regulatory-readiness","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-08-24","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-173","claim":"The aggregate water footprint of generative AI is small relative to common consumer products (a year of heavy ChatGPT use is on the order of 300 to 900 litres, against 15,415 litres per kilogram of beef and 2,495 litres per cotton t-shirt; data centers globally remain a small single-digit fraction of one percent of freshwater withdrawals, with agriculture at about 70% and industry at about 20%); the real governance concern is geographic concentration of consumption in water-stressed regions (Microsoft 2024 disclosure: 42% of consumption in water-stressed areas; Microsoft West Des Moines, Iowa Jul 2022 about 11.5M US gallons / about 6% of the city that month per AP 9 Sep 2023), not global query volume.","article_url":"https://agentmodeai.com/ai-water-use-in-context/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-08-25","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-174","claim":"The realistic year-one TCO of a security-platform agentic AI procurement at mid-sized SOC scale runs 4 to 7 times the order-form platform-fee line, decomposing across five cost components (platform fee, integration of SIEM and EDR and identity and ticketing telemetry, analyst retraining of the L1 and L2 SOC tier, tuning by the detection-engineering function in the first 6 months, and exit migration if the relationship ends); vendor-published ROI figures should be discounted by 30 to 50 percent against demo-environment bias, 20 to 40 percent against the customer's actual false-positive-adjusted alert backlog, and 60 to 80 percent against the survivorship bias in published case studies; the structural procurement instrument that prices these discounts at signing is a contractual 90-day in-environment paid-pilot evaluation gate that runs at production scope against four pre-defined customer baseline measurements (mean-time-to-triage, false-positive rate per detection class, analyst-hours per closed incident, backlog age), with a documented walk-away clause that returns the customer to the pre-pilot operating state without successor-platform commitment.","article_url":"https://agentmodeai.com/security-platform-agentic-ai-tco-roi/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-175","claim":"The 2026 Salesforce-platform-vs-Microsoft-platform AI procurement is a different conversation than the product-level Agentforce-vs-Copilot comparison and resolves on five comparison axes (data gravity against the customer's existing CRM and collaboration substrate; identity gravity against the standing IAM commitment with Microsoft Entra structurally advantaged for Microsoft-mature enterprises; developer-tooling gravity against the existing Power Platform versus Lightning + Apex developer population; regulatory-fit at the sector-specific certification layer; year-three exit cost dominated by data-egress, connector-mesh rebuild, and analyst-retraining) rather than on the per-seat or per-conversation headline pricing; the Salesforce stack (Data Cloud + Einstein + Agentforce + MuleSoft + Tableau) answers the customer-360-plus-customer-facing-agents question well, the Microsoft stack (Microsoft Graph + Fabric + Azure AI Foundry + Copilot + Power Platform + Entra + Purview) answers the knowledge-worker-productivity-plus-internal-workflow-agents question well, and the buying-committee mistake to avoid is treating the two platforms as substitutes when the larger 2026 enterprise procurement pattern is treating them as complements (Salesforce for customer-facing surfaces, Microsoft for internal-workflow surfaces) with the integration tax priced explicitly at signing rather than discovered in year two.","article_url":"https://agentmodeai.com/salesforce-platform-vs-microsoft-platform-ai/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-176","claim":"The 2026 enterprise agent-identity procurement choice between Okta and specialized NHI vendors is not a binary; Okta covers three slices of the NHI surface natively (workforce and managed-service-account lifecycle, session anomaly via October 2024 ITDR, privileged human access via 2025 Okta Privileged Access expansion), is partial on two (OAuth third-party app token governance and workload identity for cloud-native runtimes), and does not cover at GA the agent-runtime credential issuance against ephemeral workloads with sub-hour lifetimes that the specialist tier (Astrix Security for OAuth-app sprawl, Apono for just-in-time cloud access, Britive for multi-cloud privileged orchestration, Aembit for workload-to-workload SPIFFE-style attestation, Andesite for NHI runtime detection on top of existing SIEM, P0 Security for temporary access management with audit-trail evidence) is purpose-built for; the architecture-grade procurement output is one comparison matrix per identity class in scope (typically three: human workforce, managed service accounts, agent-runtime), named federation seams at four specific surfaces (identity-source authority, provisioning protocol with SCIM 2.0 the default, federation protocol with OIDC or SPIFFE the choice, audit-event format aligned to the customer's SIEM), and the procurement-side contractual instruments (AM-167 NHI procurement clause work) that make the federation enforceable at the MSA layer.","article_url":"https://agentmodeai.com/okta-vs-specialized-nhi-vendors/","topic":"non-human-identity","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-177","claim":"The 2026 'compare AI agent vendors regulated enterprise' procurement question resolves to four sector-specific vendor matrices, not one universal scorecard; each begins with a structural disqualification pass on the regulatory baseline (FedRAMP High plus DoD Impact Level for federal-civilian and defence; HIPAA BAA plus FDA 21 CFR Part 11 plus HITRUST plus EU AI Act Article 6/14 for healthcare and pharma; Federal Reserve SR 11-7 plus NYDFS Part 500 plus FINRA RN 24-09 plus SEC Rule 17a-4 for US financial services; NERC CIP plus EU NIS2 plus ISA/IEC 62443 plus regional reliability operator data-handling for energy and utility), populates a 5-or-6-row matrix per vertical against the surviving vendors, and translates the row gaps into contractually enforceable MSA addenda (the AM-167 NHI procurement-clause work instruments); the energy and utility matrix is structurally thinner than the other three because the OT/ICS overlay disqualifies most general-purpose agentic AI platforms in favour of the OT-specialist tier (Claroty, Dragos, Nozomi) with AI overlays.","article_url":"https://agentmodeai.com/agentic-ai-regulated-enterprise-vendor-matrix/","topic":"regulatory-readiness","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-178","claim":"The 2026 enterprise agentic AI orchestration-framework choice across the five major frameworks (AWS Bedrock AgentCore GA October 2025, Microsoft Azure AI Foundry plus Copilot Studio, Google Vertex AI Agent Builder plus the open-source Agent Development Kit, OpenAI Agent Builder GA October 2025 plus the open-source Swarm primitive, Anthropic Claude Agent SDK late 2024 plus the open-source Model Context Protocol) prices the decision as a 3-year orchestration-layer commitment along five comparison axes (orchestration primitive, tool-use protocol, deployment topology, observability tier, exit cost), with the framework choice resolved by gravity-fit against the customer's existing cloud, identity, and data substrate rather than by model-tier performance; the 2026 cross-vendor convergence on the Model Context Protocol as the tool-use standard is the structural change that makes exit cost newly tractable for customers who write tool definitions in the protocol and emit OpenTelemetry traces, materially lowering the year-three re-platforming cost relative to the framework-native alternatives.","article_url":"https://agentmodeai.com/aws-microsoft-google-openai-anthropic-frameworks/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-179","claim":"The 2026 enterprise AI infrastructure vendor SLA conversation resolves on five dimensions (uptime commitment with the denominator named explicitly, latency commitment at P95 and P99 negotiated into the MSA addendum because public SLAs typically omit it, support response tier per severity level, credit calculation scope and cap, exclusions list scope including scheduled-maintenance window, content-policy actions, capacity constraints, partial-availability events, and third-party-source outages); the publicly disclosed headline numbers (AWS Bedrock 99.9% monthly with 10/25/100% credit tiers, Azure OpenAI Service inheriting Azure platform 99.9% with PTU separate availability, Google Vertex AI 99.5%-99.9% varying per-model and per-region, OpenAI Enterprise 99.9%-99.99% per-customer in MSA, Anthropic Enterprise commitments per-customer with no public uniform tier) understate the year-two operational reality because exclusions list scope and credit calculation scope vary materially across vendors; the buying-committee discipline is to populate the per-vendor matrix at short-list rather than discover the gaps at year-one operational experience or year-two renewal.","article_url":"https://agentmodeai.com/enterprise-ai-vendor-sla-uptime-comparison/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-180","claim":"The 3-year IAM TCO envelope for a 2,000-employee mid-enterprise absorbing the agent-runtime identity class lands at roughly USD 1.5M to USD 3.05M, decomposed across five cost components (license, integration, operations, migration, exit) and three identity classes (human workforce, managed service accounts, agent-runtime); the platform-fee headline (Okta enterprise pricing per okta.com/pricing/) accounts for roughly 25-35% of the envelope, the remainder distributing across federation seam build-out, SCIM provisioning, OIDC/SAML configuration, audit-event format alignment, identity-operations team additions (typically 0.5 to 2.5 FTE), and the periodic-review cadence redesign for sub-hour credentials; three line items reliably unpriced in the year-one budget are the agent-runtime credential issuance integration (USD 100K-250K of first-year engineering recovered as year-two overage), the identity-operations team capacity uplift (USD 100K-200K annual), and the access-review redesign tooling or process work (USD 50K-150K).","article_url":"https://agentmodeai.com/agentic-iam-tco-mid-enterprise/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-181","claim":"The 2026 buying-committee diligence on an agentic AI vendor's strategic narrative resolves on seven proof points (named-customer references with segment-level revenue contribution; model-vendor relationships disclosed in the MSA at contractual rather than press-release level; engineering team tenure and turnover pattern as a leading indicator of narrative-product disconnect; post-revenue-recognition product-roadmap evidence comparing 12-month-prior commitments against 12-month actual ship; regulatory disclosure cadence covering SOC 2 Type II, ISO 27001 surveillance, sector-specific certifications, and public incident disclosure record; executive incentive structure as a structural read on what the vendor's leadership is trying to achieve over the 3-year MSA horizon; public technical-content cadence as downstream evidence of engineering depth); the pattern across roughly 30 vendor diligence cycles surfaced in 2025-2026 is consistent (vendors pass proof points one and two easily, fail or partially fail proof points three through five, split on six and seven); the buying committee that walks all seven systematically before the technical-feature comparison produces a structurally different diligence output and a 30-60% short-list reduction relative to the buying committee that anchors on the narrative alone.","article_url":"https://agentmodeai.com/vendor-strategic-narrative-proof-points/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-182","claim":"The 2026 Agentforce-vs-Microsoft-Copilot pricing decision resolves on the structural per-conversation versus per-user-seat pricing-model choice (Agentforce listing publicly at USD 2 per conversation with the 2025 Flex credit alternative; Microsoft 365 Copilot listing at USD 30 per user per month with Copilot Studio at USD 200 per tenant for 25,000 messages plus per-message overage) rather than the headline unit rate; enterprise-scale negotiation typically produces 30-50% per-conversation discount at committed volumes above 100,000 conversations per year on the Salesforce side and 15-25% per-user discount at committed seats above 1,000 on the Microsoft side, with multi-year commitment adding 5-15% per year at the cost of reducing year-one renegotiation leverage; three year-two renewal surprises (usage divergence from forecast, vendor pricing-model migration, bundle deconstruction) reliably surface against customers who priced only the order-form headline at signing.","article_url":"https://agentmodeai.com/agentforce-vs-microsoft-copilot-pricing/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-183","claim":"The 2026 digital-transformation RFP that includes agentic AI procurement extends the existing 60-question RFP playbook (/the-enterprise-agentic-ai-rfp-60-questions/, 38 Copilot citations) with 15 UX assessment questions across five categories (interaction modes; error-handling and confidence-communication; accessibility against WCAG 2.2 + Section 508 + EN 301 549 + assistive-technology compatibility; multi-user and cross-device continuity; design-system extensibility), producing a 75-question total RFP that the buying committee uses with separate reviewer assignments (UX team plus accessibility office plus design-system stewards review the UX section); the procurement-mature pattern asks the vendor to demonstrate the answers to error-handling, accessibility, continuity, and extensibility questions rather than describe them, surfacing the gaps between strategic narrative and product capability before the MSA negotiation.","article_url":"https://agentmodeai.com/digital-transformation-rfp-ai-ux-questions/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-184","claim":"The Digital Omnibus political agreement reached by the European Parliament and Council on 7 May 2026 postpones the EU AI Act's high-risk obligations to 2 December 2027 for standalone Annex III systems and 2 August 2028 for Annex I systems embedded in regulated products, and postpones the Article 50(2) provider watermarking duty to 2 December 2026, but leaves the deployer transparency obligations applicable from 2 August 2026 and leaves the GPAI obligations, the governance regime, the prohibited practices, and the AI literacy duty already in force, so an enterprise that reads the delay as a reason to pause its AI Act programme is misreading what still applies on and before 2 August 2026.","article_url":"https://agentmodeai.com/eu-ai-act-digital-omnibus-what-still-applies/","topic":"regulatory-readiness","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-08-26","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-185","claim":"In May 2026 the frontier model vendors began moving down the stack into systems integration: on 4 May 2026 Anthropic launched a roughly 1.5 billion dollar enterprise AI services company with Blackstone, Hellman and Friedman, and Goldman Sachs that embeds engineers inside mid-market and private-equity-owned companies rather than operating as a traditional consultancy, and OpenAI launched a parallel venture, the Deployment Company, with Bain Capital, Advent International, TPG, and Brookfield; the structural consequence for the enterprise buyer is that the model vendor, the integrator, and in the private-equity-owned case the company's own owner can be the same commercial interest, which changes the independence assumptions built into standard build-versus-buy and vendor-selection processes.","article_url":"https://agentmodeai.com/frontier-labs-as-systems-integrators/","topic":"vendor-trajectory","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-08-26","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-186","claim":"The Digital Omnibus postponement of the EU AI Act high-risk obligations to 2 December 2027 (Annex III standalone) and 2 August 2028 (Annex I embedded) re-times the conformity workstream but does not gate the readiness foundations beneath it. An enterprise running agentic AI should keep three foundations moving on the original timeline through and beyond 2 August 2026: a current inventory of which agents run under whose authority, agent-aware vendor contract terms, and active shadow-AI discovery. Each is either required by an obligation that did not move (the Article 50 deployer transparency duties applicable 2 August 2026, the GPAI and governance regime in force since 2 August 2025, and the Article 4 AI literacy duty in force since 2 February 2025) or is the prerequisite evidence base for the high-risk conformity work when it lands, and none of the three benefits from waiting for the new dates.","article_url":"https://agentmodeai.com/eu-ai-act-delay-readiness-agenda/","topic":"regulatory-readiness","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-08-26","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-187","claim":"A model's score on a coding benchmark such as SWE-bench is a weak predictor of its reliability on tasks that have no automatic verifier; enterprise model-maturity assessment therefore has to be measured on a second axis that headline leaderboards do not capture, namely common-sense robustness, run-to-run consistency, and the model's willingness to flag and correct its own errors.","article_url":"https://agentmodeai.com/car-wash-test-model-maturity/","topic":"vendor-trajectory","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-188","claim":"A special, sector-specific tax on AI justified by the arguments that it is built on the public's collective knowledge or that it runs on public infrastructure is the wrong instrument: both justifications lack a limiting principle and apply to every firm and prior technology, the legitimate grievances underneath them are better served by copyright markets and litigation (for uncompensated training data) and by the existing progressive corporate and capital-gains base (for concentrated profits), and a dedicated AI levy would fall on buyers and workers through prices and wages, entrench incumbents against startups and open source, and tax a productivity input at the moment economies most need it; the larger real tax distortion runs the other way, since the current code over-subsidises automation relative to labour.","article_url":"https://agentmodeai.com/why-an-ai-tax-is-the-wrong-instrument/","topic":"regulatory-readiness","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-189","claim":"Aggressive AI-driven workforce reduction is not, by itself, producing superior financial returns in the current cycle; across large enterprises the firms cutting deepest have shown returns close to those cutting least, which locates the return on agentic AI in retaining and amplifying the people who supervise autonomous systems rather than in headcount elimination.","article_url":"https://agentmodeai.com/ai-layoffs-roi-evidence/","topic":"enterprise-ai-cost","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-190","claim":"As measured by the 2026 Verizon Data Breach Investigations Report, AI is raising the throughput of cyberattacks rather than the underlying capability of attackers, because AI-assisted intrusions overwhelmingly scale known techniques rather than generate novel ones; the more consequential 2026 shift is that vulnerability exploitation has overtaken stolen credentials as the leading initial-access vector, and together these redirect enterprise defensive priority toward patch velocity and identity hygiene over hunting novel AI-authored threats.","article_url":"https://agentmodeai.com/ai-attacker-throughput-not-capability/","topic":"agentic-ai-governance","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-191","claim":"Within a roughly two-week window in May 2026, at least three of the four largest professional-services firms standardized their delivery organizations on a single AI model vendor, which makes the model an enterprise's auditor and implementation partner have adopted an input to that enterprise's own architecture decisions and a concentration risk that vendor-neutral strategy advice does not surface.","article_url":"https://agentmodeai.com/big-four-ai-model-concentration/","topic":"vendor-trajectory","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-192","claim":"ISO/IEC 42001:2023, the first certifiable artificial intelligence management system standard, has moved from a vendor marketing claim to an enterprise AI procurement checkpoint through 2025-2026, appearing as a stated or preferred requirement in regulated-sector and EU AI RFPs; the certificate is a necessary-not-sufficient screen (it attests to a governance management system, not to any specific model's safety, accuracy, or data provenance), so the buying-committee discipline is to require the evidence behind it (scope statement, Statement of Applicability, certification body and accreditation, validity dates, product-level AI risk assessment) and to pair it with a control-baseline mapping (NIST SP 800-53 / AI RMF) and the buyer's own product evaluation rather than treating the certificate as proof.","article_url":"https://agentmodeai.com/iso-42001-enterprise-ai-vendor-checkpoint/","topic":"agent-procurement","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-08-28","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-193","claim":"Enterprises mapping agentic AI onto NIST SP 800-53 today find real, recurring control gaps concentrated in four families (Access Control, Identification and Authentication, Audit and Accountability, and Supply Chain Risk Management) because the catalogue's implementation guidance assumes human-operated, deterministic systems rather than autonomous agents that hold delegated credentials, can be steered by untrusted input, and depend on a model-and-tool supply chain; NIST's COSAiS project (Control Overlays for Securing AI Systems) is writing single-agent and multi-agent overlays to close the gap, but finalized agent-specific guidance is not expected before 2027, so the interim burden is on the enterprise to identify the touched controls, document where standard guidance does not fit the agent case, and record compensating controls.","article_url":"https://agentmodeai.com/nist-cosais-sp-800-53-ai-agent-security-gap/","topic":"agentic-ai-governance","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-08-28","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-194","claim":"Enterprises that scale agentic AI without a dedicated inference FinOps discipline (workload-level cost allocation, spend-cap and budget-alert tooling, and model-routing policy) systematically under-budget production spend, because agentic workloads break the two assumptions cloud FinOps was built on: per-request cost is non-deterministic (token consumption varies with input and reasoning steps, and a single user request fans out into many model calls) and ownership is opaque (without tagging, inference arrives as one unattributable line item); the 2026 platform direction of cloud-native spend caps and AI cost-explainability confirms the gap is real but does not close it, because the missing layer is the operating discipline and a named owner, not the tooling.","article_url":"https://agentmodeai.com/agentic-ai-cost-governance-finops-enterprise/","topic":"enterprise-ai-cost","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-07-29","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-195","claim":"The May 2026 disclosures against AI coding agents (Adversa AI's TrustFall on 7 May 2026, a one-keypress remote code execution reaching Claude Code, Cursor, Gemini CLI, and GitHub Copilot CLI, and SymJack on 26 May 2026, a symlink-hijack confirmed against six agents that overwrites an agent's own configuration to plant a malicious MCP server, plus Microsoft's Semantic Kernel CVE-2026-26030 and CVE-2026-25592) share one design assumption, that showing an approval prompt is the same as obtaining informed consent, and because the coding agent executes attacker-supplied instructions with the developer's full credentials and write access to the build and deploy chain, it is a production attack surface that the enterprise should govern as a managed endpoint (inventory, deliberate version-pinning and patching, credential separation, monitoring for config-write-then-execute, and no untrusted repositories on credentialed machines) rather than as developer tooling outside the inventory.","article_url":"https://agentmodeai.com/ai-coding-agents-enterprise-attack-surface/","topic":"agentic-ai-governance","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-08-31","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-196","claim":"By mid-2026 the binding constraint on enterprise agentic-AI value had shifted from model capability, now a commodity any buyer can rent, to human deployment capacity, the forward-deployed engineer who integrates a model into one company's exceptions, legacy systems, and undocumented processes; because that capacity sits with the vendor, the forward-deployed-engineer-led delivery model converts what looks like a software purchase into a professional-services engagement with vendor-operability lock-in, so the buyer's defensible response is to classify and govern the spend as professional services, contract knowledge-transfer milestones with acceptance tests, build internal counterpart capacity, and require operability and exit terms, rather than treat the engagement as a delivery convenience.","article_url":"https://agentmodeai.com/forward-deployed-engineers-the-ai-bottleneck/","topic":"vendor-trajectory","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-08-31","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-197","claim":"As of mid-2026 there is no US federal AI floor coming on a plannable timeline (the White House National Policy Framework for AI of 20 March 2026 is explicitly non-binding and would require Congressional action, and the proposed federal moratorium on state AI laws was not enacted), and the most-watched comprehensive state law retreated rather than advanced (Colorado's SB 26-189, signed 14 May 2026, repealed and reenacted the Colorado AI Act, removing the algorithmic-discrimination duty of care and the risk-management and impact-assessment obligations and moving the effective date from 30 June 2026 to 1 January 2027), so the operative reality is a non-converging state patchwork, and the defensible enterprise posture is to build to the strictest obligation that actually applies to its own deployments and treat the regulatory map as a moving input rather than waiting for a federal floor.","article_url":"https://agentmodeai.com/us-ai-regulation-federal-state-standoff/","topic":"regulatory-readiness","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-08-31","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-199","claim":"Across the publication's first cohort of tracked claims, the ones that moved to Not holding were hype-register slugs inherited from the pre-standard era rather than errors of analysis, and pricing or model-tier claims were the fastest to require correction.","article_url":"https://agentmodeai.com/enterprise-ai-claims-what-held-up/","topic":"vendor-trajectory","pub_date":"2026-06-03","last_reviewed":"2026-06-03","next_review":"2026-07-18","verdict":"holding","verdict_history":[{"date":"2026-06-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-200","claim":"By mid-2026 the major enterprise agentic-AI platforms have converged on a common primitive set (agent builder, MCP tool integration, a policy gateway, and observability), so the durable enterprise selection criterion shifts from model capability to the auditability of each vendor's accountability surface: a published model-version and deprecation policy, SLA specificity, and compliance documentation.","article_url":"https://agentmodeai.com/enterprise-ai-vendor-comparison/","topic":"vendor-trajectory","pub_date":"2026-06-03","last_reviewed":"2026-06-03","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-06-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-201","claim":"In 2026 the enterprise-AI cost question that matters is total cost of ownership measured against realised ROI, not headline seat price; and across four independent datasets (Stanford DEL's 12% clearing 300%+ ROI vs 88% at or below break-even, McKinsey's 23% scaling and 17% self-reported EBIT, Gartner's 28% fully paying off, MIT NANDA's 95% of pilots with no measurable P&L impact) the high-return minority is separated from the majority by measurement discipline and operational preconditions, not by model capability or vendor choice.","article_url":"https://agentmodeai.com/enterprise-ai-cost-and-roi/","topic":"enterprise-ai-cost","pub_date":"2026-06-04","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-06-04","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of four named datasets unanchored on review. The claim text names 'Stanford DEL's 12% clearing 300%+ ROI vs 88% at or below break-even' as one of four independent datasets. Full-text verification on 10 Jun 2026 found the Stanford DEL Enterprise AI Playbook contains no such distribution — it studies 51 successful deployments by design and carries no ROI-realisation failure data (full finding at AM-029, correction of 10 Jun 2026). The McKinsey (23% scaling, 17% EBIT-attribution), Gartner (28% fully paying off), and MIT NANDA (95% no measurable P&L impact) datasets verify; the claim's spine stands on three datasets rather than four. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric from an ROI distribution. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-202","claim":"Microsoft's 1 May 2026 launch of Microsoft 365 E7 (the Frontier Suite) at $99 per user per month, combined with the 1 Jul 2026 increases to the E3 base ($36 to $39) and E5 base ($57 to $60), restructures Microsoft 365 economics so that the enterprise AI-licensing decision becomes a platform-tier decision taken at renewal rather than a Copilot add-on bought separately, and customers who decline Copilot still face a higher base cost.","article_url":"https://agentmodeai.com/microsoft-365-e7-frontier-suite-cost/","topic":"enterprise-ai-cost","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-20","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-203","claim":"Anthropic's $965 billion Series H valuation, closed 28 May 2026 and overtaking OpenAI's $852 billion (31 Mar 2026), marks the consolidation of frontier enterprise AI into a hyperscaler-backed top two, which shifts the binding risk in a multi-year Claude or GPT contract from model capability to vendor pricing power and operational switching cost, making contractual exit terms rather than benchmark wins the procurement variable that matters.","article_url":"https://agentmodeai.com/anthropic-valuation-vendor-risk/","topic":"vendor-trajectory","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-09-03","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-204","claim":"By mid-2026 non-human identities (service accounts, API keys, OAuth tokens and AI-agent credentials) outnumber human identities by roughly an order of magnitude (Cloud Security Alliance: an average of 45 to 1, up to 144 to 1 in cloud-native environments) while most enterprises lack any documented policy to provision or retire them (78% per CSA), making NHI the fastest-growing unmanaged enterprise attack surface, and the binding first control is an inventory with owner and lifecycle, not additional perimeter security.","article_url":"https://agentmodeai.com/non-human-identity-governance-vacuum/","topic":"non-human-identity","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-09-03","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-205","claim":"Enterprises systematically overestimate their visibility into AI agents (Cloud Security Alliance, Apr 2026: 82% had discovered at least one AI agent running without their security or IT team's knowledge in the past year while 68% believed they had strong visibility, with only 21% running any formal agent decommissioning process), and because a written policy cannot be enforced against agents nobody can see, continuous discovery rather than policy is the binding first control.","article_url":"https://agentmodeai.com/shadow-ai-discovery-visibility-gap/","topic":"shadow-ai-discovery","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-09-06","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-206","claim":"The May 2026 wave of enterprise AI-security moves (Zscaler agreeing to acquire Symmetry Systems on 21 May, Snowflake agreeing to acquire Natoma on 27 May, and Microsoft bringing Agent 365 to general availability on 1 May at $15 per user per month) signals that the contested layer in enterprise AI security has moved from the network to the agent-to-data access graph, so buyers should evaluate platforms on whether they can map which AI agent accesses which data, by what path, rather than on network controls.","article_url":"https://agentmodeai.com/ai-security-acquisitions-agent-access-graph/","topic":"vendor-trajectory","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-09-06","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-207","claim":"The 2 Jun 2026 US executive order 'Promoting Advanced Artificial Intelligence Innovation and Security' concentrates its mandatory weight on the federal government (CISA binding operational directives on AI-enabled cyber defence within 30 days, an NSA/CISA/Treasury AI cybersecurity clearinghouse within 30 days, OPM cyber hiring within 60 days) while making frontier-model pre-release government review explicitly voluntary, and the binding federal directives will move the de-facto AI-cyber baseline that auditors, insurers and customers measure enterprises against even though the order compels no private firm.","article_url":"https://agentmodeai.com/white-house-ai-executive-order-security/","topic":"regulatory-readiness","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-208","claim":"The SpaceX IPO filings (S-1 of 20 May 2026 and the 5 Jun 2026 free writing prospectus) disclose a circular frontier-compute economy in which Anthropic pays roughly $1.25 billion a month and Google roughly $920 million a month for GPU capacity in rival xAI's data centers — about $26 billion a year flowing into an AI segment that reported a $6.355 billion operating loss for 2025, on leases both cancellable at 90 days' notice — which makes compute supply, not model capability, the binding constraint to plan around in enterprise AI roadmaps.","article_url":"https://agentmodeai.com/xai-ipo-compute-disclosures/","topic":"vendor-trajectory","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-24","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-209","claim":"Agentic AI in banking moved from pilot to named production deployments in H1 2026 (the FIS×Anthropic Financial Crimes AI Agent announced 4 May 2026, compressing AML investigations from days or hours to minutes, with BMO and Amalgamated Bank in active development toward H2 2026 general availability; Lloyds Banking Group's 40,000-licence Microsoft 365 Copilot estate at 97% active use), and the shared deployment pattern is decision-preserving: agents compress evidence-assembly and draft the case narrative while humans retain the legally consequential filing decision — the configuration that survives regulatory scrutiny.","article_url":"https://agentmodeai.com/agentic-ai-banking-aml-deployments/","topic":"agentic-ai-governance","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-09-07","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-210","claim":"Agent washing — Gartner's term for the rebranding of existing products such as AI assistants, RPA and chatbots as agentic AI without substantial agentic capabilities — is the dominant noise source in the 2026 agentic market (Gartner's April 2026 Hype Cycle assesses only about 130 of the thousands of self-described agentic vendors as delivering real capability, against 17% of organisations deployed and more than 80% intending to deploy within two years), and the buyer's defence is a pre-contract capability test for goal-directed multi-step autonomy, tool-based action, carried state and deviation handling, rather than reliance on vendor labels.","article_url":"https://agentmodeai.com/what-is-agent-washing/","topic":"agent-procurement","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-09-07","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-211","claim":"Siemens' Eigen Engineering Agent (launched at Hannover Messe on 20 Apr 2026, available to the 600,000+ users of TIA Portal after pilots at 100+ companies in 19 countries, with vendor-reported gains of up to 50% in automation-engineering efficiency and 2-5x faster execution) marks manufacturing's first at-scale production agentic layer landing in engineering design — PLC coding, HMI visualisation, device configuration — rather than runtime operations, because design iteration is cheap and reversible while runtime carries safety cases, and the vendor's up-to figures are pilot hypotheses for buyers to test, not business cases.","article_url":"https://agentmodeai.com/agentic-ai-manufacturing-engineering/","topic":"agentic-ai-governance","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-212","claim":"AI observability — per Gartner's two-part definition, the characteristic of systems being understandable from their outputs, extended by dedicated tools that manage and assess the behaviour, decision-making and risks of an AI solution such as model drift, bias and LLM logic — is a distinct discipline from classic application monitoring because AI fails semantically (drift, bias, opaque reasoning) while APM watches infrastructure and application health, and with Gartner predicting 40% of AI-deploying organisations will run dedicated AI observability tools by 2028 from a nascent base, the CIO-grade sequence is to define wrong-outcome metrics and measured detection time before buying tooling.","article_url":"https://agentmodeai.com/what-is-ai-observability/","topic":"agentic-ai-governance","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-213","claim":"Per IDC research commissioned by Lenovo (CIO Playbook 2025, February 2025; global survey n=2,920), 88% of enterprise AI proof-of-concepts fail to reach production: for every 33 POCs a company launched, only four (roughly 12%) graduated to widescale deployment, and the graduation gap is an organizational-readiness outcome (unclear ROI, insufficient AI-ready data, lack of in-house AI expertise) rather than a model-capability outcome.","article_url":"https://agentmodeai.com/why-88-percent-of-agentic-ai-deployments-fail/","topic":"enterprise-ai-cost","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-214","claim":"The publication's correction record for the fabricated 'Stanford DEL 12/88' statistic is complete across the written corpus and public as of 10 Jun 2026: eight claim verdicts changed with dated, append-only correction entries (AM-029 to Not holding; AM-024, AM-031, AM-040, AM-042, AM-129, AM-132 and AM-201 to Partial), the Claim Archive's first retraction (ACA-2026-003), roughly 120 occurrences across 30 published article bodies restated on the verified IDC/Lenovo graduation figure or softened to qualitative language, and the signature article restated at its original URL under new claim AM-213, with no occurrence silently deleted.","article_url":"https://agentmodeai.com/anatomy-of-a-fabricated-statistic/","topic":"agentic-ai-governance","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-215","claim":"Claude Fable 5, released 9 Jun 2026 at $10/$50 per million tokens (twice Opus 4.8), can refuse a request mid-call in three restricted domains (cybersecurity, biology and chemistry, model distillation) and fall back to Opus 4.8, which makes deploying it a runtime-reliability problem (refusal handling, per-call capability variance, and variable billing) rather than a one-time procurement decision.","article_url":"https://agentmodeai.com/claude-fable-5-enterprise-fallback-reliability/","topic":"vendor-trajectory","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-09-12","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-216","claim":"The EU AI Act's enforcement calendar splits in two: general-purpose AI model-provider obligations become directly enforceable on 2 Aug 2026 (finable up to 3% of global annual turnover or €15M), while the provisionally agreed Digital Omnibus defers high-risk deployer obligations for stand-alone Annex III systems to 2 Dec 2027, so enterprises face their AI vendors' new compliance demands roughly 16 months before their own high-risk deadline arrives.","article_url":"https://agentmodeai.com/eu-ai-act-gpai-enforcement-vendor-flow-down/","topic":"regulatory-readiness","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-09-12","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-217","claim":"IBM's June 2026 study of 2,000 technology executives shows a structural AI control gap: 66% are accountable for AI systems they do not fully control, and the organisations that embed control directly into their AI systems report materially better outcomes (18% higher operating margins, 25% fewer incidents, four times less wasted AI budget) than those relying on manual, bolt-on governance.","article_url":"https://agentmodeai.com/ai-control-gap-cio-accountability/","topic":"agentic-ai-governance","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-09-12","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-CANON-001","claim":"AI-written commentary can be more verifiable, not less, than human-written commentary — when it is published inside an explicit accountability architecture with six components: disclosure, claim isolation, verdict tracking, dated retraction, primary-source pinning, and review cadence.","article_url":"https://agentmodeai.com/canon/v1.0/accountability-architecture/","topic":null,"pub_date":"2026-05-14","last_reviewed":"2026-05-14","next_review":"2026-08-12","verdict":"holding","verdict_history":[{"date":"2026-05-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-001","claim":"For a 4–10 person ops team running ~50 automations including five agentic steps in 2026, the platform choice is binary between n8n self-hosted and Make.com Pro, decided by whose time pays for the platform; Zapier earns its cost only when a critical integration is vendor-locked.","article_url":"https://agentmodeai.com/operators/n8n-vs-make-com-vs-zapier/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-002","claim":"For a 5-person consultancy already on either Notion or ClickUp in 2026, the AI features alone do not justify a workspace switch; the bundling difference (Notion bundles AI into Business at $19.50/seat, ClickUp Brain is a separate $9/seat add-on) makes the platform-shape choice (doc-centric vs project-centric) the actual decision.","article_url":"https://agentmodeai.com/operators/notion-ai-vs-clickup-ai-consultancy/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-05-28","verdict":"partial","note":"Price drift: Notion Business with bundled AI now about $15/seat annual ($20 monthly) vs cited $19.50; ClickUp Brain now $7/seat vs cited $9. Verdict logic unchanged; figures need updating."}],"primary_sources":[]},{"id":"OPS-003","claim":"For a solo founder choosing exactly one consumer AI subscription at around $20/month in 2026, the choice between Claude Pro and ChatGPT Plus is workflow-shape (long-document review and code favour Claude Pro; voice mode, image generation, and integration breadth favour ChatGPT Plus) — not capability-rank, which both vendors trade leadership on monthly.","article_url":"https://agentmodeai.com/operators/claude-pro-vs-chatgpt-plus-solo-founder/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-005","claim":"At sub-1M tokens per month (typical SMB agent volume) in 2026, the absolute dollar gap between Claude Haiku 4.5, GPT-4o-mini, and Gemini 2.5 Flash is small enough (≤$3/month) that price is the wrong tiebreaker; tool-use reliability, instruction-following on long context, and ecosystem fit determine the right cheap-tier model per workload shape.","article_url":"https://agentmodeai.com/operators/anthropic-vs-openai-vs-gemini-api-smb/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-011","claim":"If a candidate first-AI-agent use case at an SMB cannot answer all four of (a) what does success look like in numbers, (b) who owns it on Monday, (c) what breaks if it fails silently, (d) what is the rollback — the use case is not ready to deploy, regardless of vendor demo quality or model capability.","article_url":"https://agentmodeai.com/operators/picking-first-ai-agent-small-business/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-014","claim":"An SMB AI vendor evaluation defensible to the typical cyber-insurance reasonable-care expectation can be completed in 90 minutes by walking through five questions in order — model provenance, data residency, sub-processor list, breach history, termination clause — each answered from the vendor's public site or the contract about to be signed.","article_url":"https://agentmodeai.com/operators/ai-vendor-due-diligence-small-business/","topic":"operators-ai-compliance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-021","claim":"Across the published 2026 small-bookkeeping AI corpus (Xero OS, Intuit Assist, Canopy AI Notetaker, Digits MCP Server, with CPA Practice Advisor as the trade-press source), AI now reliably handles five recurring grind workflows at 1-to-5-person firm scale (bank-feed categorisation, receipt OCR, recurring journal posting, sales-tax reconciliation, AR ageing emails), but the judgement-call workflows (period close, advisory conversations, audit defence) remain human-led.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-small-firm-case-study/","topic":"operators-ai-economics","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-022","claim":"Across the published 2026 small-law-firm AI corpus (Spellbook with named small-firm customers Westaway, KMSC Law, Polley Faith; Harvey AI with mid-size roster Thompson Hine through Lowenstein Sandler; GC AI as named Anthropic enterprise customer claiming 1,500 companies and 14 hours/week saved), AI now ships at 1-to-20 lawyer-firm scale for contract drafting, document review at scale, and legal research with citation, but privileged-content workflows still require Enterprise-tier model access with zero-data-retention contractual posture per ABA Formal Opinion 512.","article_url":"https://agentmodeai.com/operators/ai-small-law-firm-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Vendor attribution error in the claim text. The claim names Polley Faith among 'Spellbook with named small-firm customers Westaway, KMSC Law, Polley Faith'. Polley Faith LLP is a Harvey-listed law-firm customer, not a Spellbook customer: the live Spellbook site (now spellbook.com; spellbook.legal 301-redirects) names Westaway, KMSC Law, and McInnes Cooper with no Polley Faith, and the source article's own body correctly places Polley Faith on Harvey's roster — the claim text and the article excerpt bundled it with the wrong vendor at publish. The remaining legs verify against extracted source text on 10 Jun 2026: Anthropic's GC AI customer story carries 'More than 1,500 companies' and '14 hours saved per week on average ... based on a survey of more than 100 active customers' verbatim; Harvey's published roster (Thompson Hine, Fox Rothschild, Lowenstein Sandler, Polley Faith) matches; ABA Formal Opinion 512 remains the governance baseline. The corpus reading (AI ships at 1-to-20 lawyer scale; privileged work stays on Enterprise-tier zero-retention access) is unaffected. Status Up -> Partial."}],"primary_sources":[]},{"id":"OPS-026","claim":"The published 2026 construction-AI case corpus is overwhelmingly vendor-led (Procore, Autodesk Construction Cloud, OpenSpace, Buildots, Doxel) with thin named small-contractor self-published cases. Reading the vendor corpus honestly, three workflows now show consistent under-100-employee contractor AI deployment (estimating speed via takeoff acceleration, schedule risk surfacing, as-built reality capture); a fourth (AI safety detection) remains structurally biased toward larger sites with the camera coverage and safety officer to act on alerts.","article_url":"https://agentmodeai.com/operators/ai-small-construction-firm-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-027","claim":"Across the published 2026 dental-AI case corpus (Pearl with FDA-cleared 2D and 3D radiography AI plus 23,000 published practices; Overjet with 21+ named small-and-family-practice customers including Promenade Center, Quest Dental, Midtown Dental Studio), AI now ships at 1-to-3-dentist practice scale for FDA-cleared radiography assist, insurance verification automation, and patient-education visualisation; ambient voice AI for clinical notes is the next surface to ship widely.","article_url":"https://agentmodeai.com/operators/ai-small-dental-practice-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-028","claim":"The published 2026 small-beauty-salon AI case-study corpus is materially thinner than dental, legal, or bookkeeping (booking platforms publish customer counts but rarely individual-salon AI-attributable outcomes; solo stylists who use AI share informally on Instagram and TikTok rather than in case-study form). Reading the platform corpus honestly, the 2026 working pattern at 1-to-5 chair scale concentrates on no-show reduction via deposits, marketing copy via consumer-tier AI assistants, and portfolio/look generation via Canva and similar tools. AI-driven hairstyling recommendation, voice-AI booking, and dynamic pricing are not yet at the published-case-density that supports a small-salon recommendation.","article_url":"https://agentmodeai.com/operators/ai-small-beauty-salon-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-029","claim":"For solo founders and small teams (under ~50 people) building with AI in 2026, the build-vs-buy decision tree has inverted: specification, not engineering capacity, is now the bottleneck. The teams that can describe their workflow in operational detail can ship things they could not previously afford to build; the teams that cannot still cannot ship, regardless of how good the AI tooling is.","article_url":"https://agentmodeai.com/operators/three-launches-with-ai-the-lessons/","topic":"operators-ai-tooling","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-030","claim":"The fastest path for an owner-operator to build practical agentic-AI competence in 2026 is the three-week build-by-shipping protocol — specification + scaffolding + ship + connect + deploy + iterate, against a real workflow, with one external user — not formal study or consulting engagement. The protocol produces more transferable competence than published comparable courses on three measurable outcomes: operational decisions the operator can make after, debugging capability without external help, and calibration on when to build versus buy.","article_url":"https://agentmodeai.com/operators/using-ai-to-learn-ai-operator-playbook/","topic":"operators-ai-tooling","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-031","claim":"Solo founders evaluating AI bookkeeping in 2026 face three realistic options: a fully-managed AI-augmented service (Bench, Pilot), a software-led tool that does AI categorisation inside an existing accounting product (QuickBooks Live, Xero with Hubdoc), or a DIY stack (Claude/ChatGPT + a spreadsheet template). The fully-managed option scales when revenue passes ~$30K MRR; below that, the DIY stack with a 30-min monthly review beats both software-led and managed. The failure mode is paying for managed-service automation while still doing 80% of the categorisation yourself because the AI hasn't seen enough of your transaction patterns yet.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-for-solo-founders/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 covering the three-option market split (fully-managed / software-led / DIY) and the ~$30K MRR threshold for fully-managed to become net-positive. REVIEW: Peter to verify current Bench and Pilot entry-tier pricing on or before 13 Jun 2026; if either has launched a sub-$100/month tier the threshold call shifts."}],"primary_sources":[]},{"id":"OPS-032","claim":"For SMB content workflows in 2026 (blog drafts, weekly newsletter, social copy, email sequences) at a 1-to-10 person business shipping two-to-four pieces per week, the practitioner read is workflow-shape not capability-rank: Claude wins on long-form editorial voice and structured drafting; ChatGPT wins on speed-and-iteration plus image generation in the same conversation; Gemini wins on Google-stack integrations. Paying for all three Plus tiers (around $60/month) without a deliberate task split is the expensive failure mode.","article_url":"https://agentmodeai.com/operators/chatgpt-vs-claude-vs-gemini-smb-content/","topic":"operators-ai-tooling","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 with status=partial. Recommendation derived from vendor pricing pages 29 Apr 2026 + public eval leaderboards + practitioner write-ups, not from a tracked SMB-cohort replication. Promotes to Holding once two consecutive 45-day reviews replicate the workflow-shape split on a real operator sample. REVIEW: Peter."}],"primary_sources":[]},{"id":"OPS-033","claim":"AI customer-service automation at 1-10 employee scale clears net-positive only when 70% or more of weekly inquiries are repetitive, low-stakes, and factually resolvable (hours, pricing, simple status). Below 50% the trust-erosion and remediation cost exceeds the headcount saving; between 50% and 70%, the answer turns on whether responsiveness is the brand differentiator.","article_url":"https://agentmodeai.com/operators/ai-customer-service-small-business/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Break-even thresholds (70/50) and never-deflect list are editorial synthesis from cited platform docs and CS-automation research, not a primary-data study. REVIEW: Peter to validate against any first-party SMB deployment data he has access to before status promotion to Holding."}],"primary_sources":[]},{"id":"OPS-034","claim":"For a solo founder processing 100-300 emails a day in 2026, the cheap-stack option (Gmail labels + Claude Pro at $20/month + a 5-line prompt template) recovers roughly 90% of the value of an $83/month premium stack (Superhuman AI + Shortwave Pro + Reclaim.ai Pro) at about 24% of the cost. The premium stack is worth its price under three conditions only — 2+ hours/day in email, keyboard-shortcut speed gain that pays back at the founder's hourly rate, and a documented bottleneck the cheap stack failed to solve after a two-week trial. Without all three, the founder is paying for an aesthetic, not measurable productivity.","article_url":"https://agentmodeai.com/operators/solo-founder-email-triage-ai-stack/","topic":"operators-ai-tooling","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 with status=partial. Cost-side claims (vendor pricing) verifiable against the four cited pricing pages on the publication date. Time-recovery claim (90+ min compressed to ~20 min) drawn from published productivity-blogger benchmarks rather than Peter-run measurement; first-cohort replication on the publication's tracked operator cohort due by 13 Jun 2026. REVIEW: Peter."}],"primary_sources":[]},{"id":"OPS-035","claim":"There are five categories of small-business work where AI substitution in 2026 costs more in trust and liability exposure than it saves in productivity: (1) signed legal documents and tax-return positions, (2) trust-laden customer touchpoints (cancellations, refunds, conflict de-escalation), (3) regulatory submissions where the human signature is the audit trail, (4) anything requiring genuine domain credentialing (medical advice, licensed financial advice, signed engineering work), and (5) the first six conversations with a new high-value client.","article_url":"https://agentmodeai.com/operators/when-not-to-use-ai-for-small-business/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Status set to Partial at publication because category 5 lacks the same regulatory/cited-consequence anchor as categories 1-4. REVIEW: Peter to confirm category 5 evidence base and either upgrade to Holding (with strengthened citation) or amend the claim to four categories."}],"primary_sources":[]},{"id":"OPS-036","claim":"An SMB AI policy that actually changes day-to-day behaviour fits on one page and contains exactly eight clauses — sanctioned tools, prohibited data, human-review gate, client disclosure rule, prohibited uses, incident-report path, review cadence, and signature line — each closing a failure mode currently surfacing in regulatory guidance, court records, and breach disclosures through 2025-2026.","article_url":"https://agentmodeai.com/operators/1-page-ai-policy-for-small-business/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Status set to Partial at publication because clause 6 commentary references an order-of-magnitude remediation-cost gap derived from the IAPP 2024 AI Governance Profession Report; the report characterises the gap as material but does not publish a precise multiple, so the wording is annotated source: our-estimate. REVIEW: Peter to source a precise figure or amend the commentary."}],"primary_sources":[]},{"id":"OPS-037","claim":"AI-drafted invoices for EU SMB operators in 2026 fail VAT audit at higher rates than human-drafted invoices specifically on cross-border treatment (OSS scheme wording, reverse-charge language, customer VAT-status verification), because LLM training data underweights post-2021 e-commerce VAT rules. The fix is a 4-line VAT-compliance prompt prefix that names the operator's VAT registration, the customer's VAT status, and the applicable scheme; most SMB invoicing tooling does not ship this by default.","article_url":"https://agentmodeai.com/operators/ai-invoicing-vat-compliance-small-business/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-038","claim":"SMB AI-VA deployments displacing admin work in collective-agreement-covered sectors (Dutch CAO, German Tarifvertrag, French Convention Collective) trigger collective-agreement provisions even at sub-10-employee scale in 2026, via job-classification-displacement and technology-introduction-consultation channels. Most SMB owners are unaware until the first union audit; FNV / DGB / IG Metall / CFDT activity in this area has shifted from theoretical to operational since 2024.","article_url":"https://agentmodeai.com/operators/ai-va-small-business-collective-agreement/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-039","claim":"AI-drafted contracts in EU notary-required jurisdictions (NL, DE, AT, BE, CH) are producing a class of legal-malpractice incidents in 2026 where the SMB owner treats an AI draft as final binding document, missing the notarisation requirement for real-estate transfers, GmbH/BV share transfers, and certain marriage/inheritance instruments. The fix is a 30-second pre-signing check on transaction-type and jurisdictional notarial-form requirement; AI tooling does not flag this by default.","article_url":"https://agentmodeai.com/operators/ai-drafted-contracts-notary-requirement-eu/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-040","claim":"Dutch ZZP'ers losing recurring client work to AI replacement in 2026 sit outside the WW (Werkloosheidswet) safety net entirely and find that available AOV (arbeidsongeschiktheidsverzekering) products mostly exclude demand-side income loss; the structural gap is pushing affected ZZP'ers into bijstand at faster rates than the 2024 baseline. The realistic options are operational (client-base diversification, offer restructuring, larger liquid buffer), not insurance-based.","article_url":"https://agentmodeai.com/operators/zzp-ai-displacement-unemployment-gap-nl/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-041","claim":"SMB owners using AI to produce marketing content are hitting platform algorithmic penalties at increasing rates in 2026, with platform-specific enforcement: Google Helpful Content system + March 2024 spam policy update target scaled-content-without-E-E-A-T; LinkedIn feed-distribution deprioritises fully-AI-generated content while tolerating AI-assist; Etsy listing-policy enforcement is heavier than either, with category-specific AI prohibitions. The defensible cross-platform posture is AI drafts + human edits + human signature with sustainable cadence.","article_url":"https://agentmodeai.com/operators/platform-algorithm-ai-content-penalties/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-042","claim":"For under-100-employee construction firms in 2026, the AI procurement order is estimating + bidding tools first (Togal.AI for general takeoff; Procore Copilot if already on Procore), with visual-progress capture (Buildots, OpenSpace) deferred until project portfolio exceeds 8 simultaneous projects per project manager. The vendor pitch oversells visual capture and undersells the takeoff workflow where the actual hours go (35-45% of estimator/PM time on bidding work, 5-10% on jobsite walkthroughs).","article_url":"https://agentmodeai.com/operators/ai-construction-estimating-bidding-tools/","topic":"operators-ai-verticals","pub_date":"2026-05-03","last_reviewed":"2026-06-17","next_review":"2026-08-16","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-043","claim":"For solo founders under €5K MRR running 20-80 customer-service tickets per week in 2026, the cheap stack (shared inbox host + Claude Pro at €20/month + a copy-paste prompt-pack, total under €40/month) is structurally cheaper than the dedicated AI helpdesks (Intercom Fin, Crisp AI, Tidio Lyro) until ticket volume sustains above ~200/week. Above that threshold, the per-resolution and per-conversation pricing on the dedicated platforms starts to compete; below it, the cheap stack wins on cost AND on operator experience. The volume threshold is the procurement signal, not the vendor pitch.","article_url":"https://agentmodeai.com/operators/solo-founder-customer-service-ai-stack/","topic":"operators-ai-tooling","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-044","claim":"For appointment-driven local-service businesses in 2026 (hairdresser, plumber, garage, cleaner, beautician), the AI value concentrates in two workflows neither booking-platform AI feature serves well: no-show reduction via personalised SMS sequences (3rd-party SMS API on top of the booking platform's webhook, typical 30-50% no-show reduction in published case studies) and review generation (post-appointment SMS or WhatsApp, typical 3-5x review-completion lift). The booking-platform decision (Booksy, Square Appointments, Treatwell, Vagaro) is shaped by customer-discovery model and existing payment infrastructure; the AI decision is shaped by whichever third-party SMS-and-review-automation layer bolts on top. Operators picking the booking platform on its bundled AI features pay for AI that does not move the numbers.","article_url":"https://agentmodeai.com/operators/ai-local-service-business-appointment-driven/","topic":"operators-ai-verticals","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-045","claim":"OPS-031's jurisdiction-neutral DIY AI bookkeeping case for solo founders under €30K MRR breaks at the NL-specific Belastingdienst audit-trail boundary. The procurement decision per omzetband: Moneybird (€15-€39/month) under €100K omzet with API-driven AI flow via Make.com or n8n; e-Boekhouden as the goedkope fallback with bundled Scan & Herken OCR; Exact Online above €500K omzet or at BV-overgang where Exact's interne AI replaces the external prompt-pack workflow. NL-specifieke prompt-prefix (klant locatie, dienst type, reverse-charge applicability, OSS-scheme applicability, BTW-rubriek per Belastingdienst-aangifte 2026) is the operationally load-bearing addition that makes AI-getekende journaalposten direct invoerbaar in the chosen tool's BTW-aangifte.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-nl-moneybird-eboekhouden-exact/","topic":"operators-ai-economics","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-046","claim":"Marketplace-reseller AI in 2026 fails differently per platform and the cross-platform mitigation pattern is to separate AI-on-listing-copy (broadly safe across Etsy, Marktplaats, Vinted) from AI-on-listing-images (increasingly penalised on all three platforms via different mechanisms: Etsy's Creativity Standards and AI-disclosure requirement; Marktplaats's photo-fingerprint deduplication; Vinted's image-similarity penalty for resale-of-resold). The 'AI does the entire listing' workflow is the procurement pattern that produces the account-suspension report 6-12 months later. The defensible reseller workflow uses real photos, AI-assisted copy with platform-required disclosure, and per-platform performance tracking on impressions and sales.","article_url":"https://agentmodeai.com/operators/ai-marketplace-resellers-etsy-marktplaats-vinted/","topic":"operators-ai-verticals","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-047","claim":"EU AI Act Annex III point 4 (employment, workers management, recruitment) applies to SMB AI hiring use even at four-employee scale; the threshold does not scale with company size, and the 2 August 2026 enforcement window covers AI-screened CVs in ChatGPT/Claude/Gemini the same way it covers dedicated platforms (Workable, Greenhouse, Lever, BrightHire). The defensible posture is AI-assisted decisions with a documented human decision-maker plus retained AI-output records — not AI-decided hiring. Solely-automated candidate scoring also conflicts with GDPR Article 22; ICO, AP, and Garante guidance from 2024-2025 is consistent on the human-in-the-loop requirement.","article_url":"https://agentmodeai.com/operators/ai-hiring-smb-eu-ai-act-annex-iii/","topic":"operators-ai-compliance","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-048","claim":"Solo founders adding AI to cold outbound see a deliverability collapse around day 60-90 because AI lifts personalisation breadth at the same volume rather than personalisation depth at lower volume. The collapse is mechanical: AI-templated personalisation degrades recipient engagement, engagement decay triggers spam-classifier de-prioritisation, lower inbox rate produces more complaints, complaints trigger soft blocks. The defensible 2026 posture: 30-40 sends per inbox per day, named-specific first-paragraph personalisation, reply-rate KPI not open-rate, plus a documented EU GDPR Article 6(1)(f) Legitimate Interest Assessment for B2B founders in scope of e-Privacy Directive.","article_url":"https://agentmodeai.com/operators/ai-cold-sales-solo-founder-deliverability/","topic":"operators-ai-tooling","pub_date":"2026-05-04","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-049","claim":"German Mittelstand AI deployment in 2026 hits two compliance surfaces most US-headquartered AI vendors do not handle out of the box: BetrVG §87(1) point 6 co-determination triggers at the first AI assistant or agent that touches employee work activity (Bundesarbeitsgericht broad interpretation covers any system that captures, processes, or analyses employee work activity, primary purpose immaterial); DSGVO Article 35 + Datenschutzkonferenz Muss-Liste require pre-deployment DPIA for most AI-employee-data deployments. The early-engagement workflow (works council notified at vendor selection, DPIA in parallel with vendor evaluation, joint Betriebsvereinbarung drafting, documented pilot at one team for 60-90 days, broader rollout after pilot review) compresses Mittelstand AI timeline from 12-18 months (late engagement) to 6-9 months.","article_url":"https://agentmodeai.com/operators/ai-mittelstand-betrvg-dsgvo-deployment/","topic":"operators-ai-compliance","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-050","claim":"Local SMB AI use on Google Business Profile and local-SEO content splits into two cohorts in 2026: AI-as-research-and-assembly (keyword research, citation audit, performance analysis via Surfer/Frase/Ahrefs/BrightLocal/Whitespark) compounds visibility safely; AI-as-generation (auto-published reviews, auto-published review responses, bulk service-area pages, high-cadence GBP posts) triggers Google's Helpful Content classifier and the March 2024 spam policy update enforcement, with documented suspensions and ranking collapse on a 30-90 day cycle. The defensible posture is AI for the work that scales poorly (research, cross-reference) and human for any content that reaches the public surface.","article_url":"https://agentmodeai.com/operators/ai-local-seo-google-business-profile-smb/","topic":"operators-ai-compliance","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-051","claim":"AI proposal tools in 2026 split into two clusters by what they let the operator publish: tools that AI-assist proposal assembly (PandaDoc, Better Proposals, Proposify, Bonsai) compound; tools that AI-generate proposal narrative (Pitch, Gamma, Tome AI generation features) read as AI-generated to most buyers within thirty seconds and close at materially lower rates. Three structural patterns trigger the buyer-side AI-generated detection: the three-phase project structure regardless of actual scope, the credentials paragraph that lists capability without naming clients, the pricing section that over-explains itself. The defensible posture is AI for assembly (pricing tables, scope-of-work blocks, clause libraries from CRM) and human for voice (cover letter, executive summary, project-fit paragraph, next-step CTA).","article_url":"https://agentmodeai.com/operators/ai-client-proposals-tools-solo-founder/","topic":"operators-ai-tooling","pub_date":"2026-05-04","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One named member of the generation cluster was already defunct at publication: Tome shut down its presentation/narrative product (Tome Slides) in March 2025 and pivoted to sales tooling, with the brand later sold to AngelList (deckary.com shutdown timeline; signalhub.substack.com post-mortem, both checked 10 Jun 2026). The generation cluster reduces to Pitch + Gamma. The two-cluster thesis itself is unaffected and arguably strengthened — the pure AI-narrative product failed to find a sustainable business while Gamma (70M users, $100M ARR as of Nov 2025) and the assembly cluster (PandaDoc, Better Proposals, Proposify per Luniq 2026 agency comparison) both compound. Status Up → Partial for the factual error in the tool list."}],"primary_sources":[]},{"id":"OPS-052","claim":"Voor de Nederlandse zelfstandige advocaat (eenmanspraktijk, klein kantoor onder 5 partners) is AI in 2026 toegestaan voor drie hoofdcategorieën onder de NOvA-gedragsregels: juridisch onderzoek met advocaat-verificatie van elke citatie, document-drafting waar de advocaat reviewt en signeert, en cliëntcommunicatie-ondersteuning waar de advocaat elke uitgaande communicatie reviewt voor verzending. AI is niet toegestaan zonder advocaat-review voor: advies-generatie aan cliënten, procesvertegenwoordiging, cliëntgegevens-verwerking via niet-EU-LLM zonder Verwerkersovereenkomst, en het ondertekenen van documenten met AI-gegenereerde citaten zonder primaire-bron-verificatie. EU AI Act Artikel 50 disclosure is verplicht voor cliënt-AI-chatbots vanaf 2 augustus 2026.","article_url":"https://agentmodeai.com/operators/ai-solo-legal-paralegal-nl-bar-rules/","topic":"operators-ai-compliance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-053","claim":"For marketplace resellers running AI image workflows in 2026, the safe pattern across Marktplaats, Vinted, and Etsy is original photography of the actual item with light AI enhancement (lighting, contrast, background cleanup) only. AI-generated listing imagery and heavy enhancement that produces consistent visual fingerprints across listings trigger Marktplaats's photo-fingerprint deduplication (most aggressive), Vinted's image-similarity penalty for the resale-of-resold pattern, and Etsy's Creativity Standards on AI-generated imagery in handmade categories. The five-rule safe workflow: original photography of every item, light AI enhancement only, fresh photography per relisting, per-platform disclosure where required, and impression-to-view ratio tracking as the leading indicator of algorithm-induced ranking suppression.","article_url":"https://agentmodeai.com/operators/ai-marketplace-image-workflow-marktplaats-vinted/","topic":"operators-ai-verticals","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-054","claim":"For EU-based solo developers doing client work in 2026, the procurement-defensible AI-tool posture turns on client-code data residency rather than on Cursor-vs-Copilot-vs-Claude-Code feature comparison. All three dominant AI coding tools support EU data residency at Enterprise tiers (Copilot via Microsoft Azure OpenAI EU regions, Cursor via configurable LLM provider routing, Claude Code via Anthropic API EU-region availability). Three contract clauses now appear in regulated EU client agreements: client-code-non-transmission, EU-residency requirement, and sub-processor disclosure. The procurement-defensible workflow has five steps: AI-tool inventory, per-client risk assessment, configure tools per client, document configuration in engagement contract, audit quarterly. Three scenarios where the right answer is to disable AI tooling entirely: explicit contract prohibition that cannot be negotiated, embedded regulated data in the codebase, national-security or jurisdictionally-sensitive code.","article_url":"https://agentmodeai.com/operators/ai-solo-dev-eu-client-code-residency/","topic":"operators-ai-compliance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-055","claim":"For German solo founders and small Mittelstand operators running AI-bookkeeping in 2026, the Buchhaltungssoftware choice resolves on Umsatz tier and Steuerberater relationship: DATEV (€20-€80/month plus Steuerberater-coupling) above €100K Umsatz where the Steuerberater workflow is binding, sevDesk (€8-€48/month) under €100K Umsatz as the cheapest path that produces a GoBD-compliant audit trail, and Lexware (€10-€40/month) as the legacy-Mittelstand fallback. The OPS-031 jurisdiction-neutral DIY-AI-bookkeeping case breaks at the moment the AI-drafted Buchungssatz must land in a tool that preserves the GoBD audit trail; the German-tool layer is the complement to the DIY-AI case. The OSS-Verfahren and reverse-charge VAT prompt-prefix is the operational discipline that prevents AI-VAT-error in 1 of 20 EU-cross-border invoices.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-de-datev-sevdesk-lexware/","topic":"operators-ai-economics","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-056","claim":"For bootstrapped SaaS founders under €30K MRR with AI features in production, the metric that matters is token cost per active user (not total monthly AI spend). Total monthly spend is the lagging indicator that signals problems only after they have crossed gross-margin thresholds; cost per active user is the leading indicator that catches runaway patterns before they erode unit economics. The defensible cancellation-trigger threshold sits at 30-40% of per-user revenue. Four levers when the cost crosses the trigger, ranked by disruption: provider-tier switch (40-70% reduction, low impact), prompt and caching optimisation (20-40% reduction, moderate impact), product change (30-60% reduction, high impact), provider switch (10-30% reduction, highest disruption). Token cost dropped roughly 90% from 2023-2026 but per-user cost stayed flat because product features pulled 10-30x more tokens per session and user behaviour shifted toward higher engagement.","article_url":"https://agentmodeai.com/operators/ai-cost-discipline-bootstrapped-saas/","topic":"operators-ai-economics","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-057","claim":"For an Etsy seller earning under €100K/year, the cheapest AI stack that consistently pays back combines Claude Pro (€18/month) for listing copy and customer reply drafts with a single image-generation tool (€10-30/month — Midjourney, Adobe Firefly, or DALL-E via ChatGPT Plus). Most paid Etsy-specific AI tools (eRank Pro, Sale Samurai, ListEasy, Alura, Marmalead) only repay above ~200 active listings or ~€50K annual revenue. Etsy's AI-content disclosure policy and the EU AI Act Article 50 transparency obligations apply to AI-generated imagery.","article_url":"https://agentmodeai.com/operators/etsy-ai-tools-for-sellers/","topic":"operators-ai-verticals","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-058","claim":"For a 1-5 person service business in 2026, AI voice agents repay above ~30 inbound calls/week per agent line. Below that, the integration cost (~10-20 hours setup) and per-minute usage rates (5-12 cents per minute typical) exceed the value of automated handling. Vapi, Bland, and Retell occupy three different price-and-control points: Vapi's developer-platform model wins on flexibility and BYO-model control, Bland's no-code on speed-to-deploy, Retell's enterprise-tier on compliance posture (SOC 2 Type II, HIPAA-eligible). TCPA + FCC AI-voice rulings (US) and GDPR Article 22 + ePrivacy Directive (EU) constrain deployment with mandatory disclosure and human-handoff patterns.","article_url":"https://agentmodeai.com/operators/ai-voice-agents-solo-business/","topic":"operators-ai-tooling","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-059","claim":"SMBs without legal teams sign AI vendor MSAs that lock them in via seven recurring clause patterns: (1) data-portability narrowness (prompts/embeddings/agent state excluded from 'your data' definitions), (2) auto-renewal with short notice windows, (3) model-deprecation rights without credit, (4) sub-processor expansion without consent, (5) output-IP ambiguity, (6) pricing escalator without cap, (7) termination-data export window too short. Pattern recognition + a 1-page checklist applied before signature is the practical defence. Five questions emailed to the vendor sales rep before signing — and their willingness to answer in writing — is itself a signal.","article_url":"https://agentmodeai.com/operators/ai-vendor-redflags-smb/","topic":"operators-ai-compliance","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-060","claim":"For Dutch e-commerce SMBs (under €500K annual revenue) in 2026, the AI stack that consistently pays back is Claude Pro or ChatGPT Plus (€18-20/month) for product copy + customer-service drafts, plus one image tool (Midjourney/Firefly), plus EU-resident hosting if data residency matters. Bol.com's API constraints, AVG (Dutch GDPR implementation), and EU AI Act Article 50 transparency obligations create a different procurement reality than US/UK SMBs. Shopify Magic + Sidekick win on speed-to-deploy; WooCommerce wins on data-residency control; Bol.com wins on Dutch-marketplace reach but penalises low-quality AI-drafted listings.","article_url":"https://agentmodeai.com/operators/ai-for-dutch-ecommerce/","topic":"operators-ai-verticals","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-061","claim":"For a 1-5 person business in 2026, AI consistently pays back on six task classes (drafting, summarising, scheduling, research synthesis, code generation for solo developers, image/asset production) and consistently fails on six others (high-stakes customer-facing decisions without disclosure, regulatory advice, complex multi-party negotiations, brand-distinctive creative work, anything requiring physical presence, anything requiring social proof of human authenticity). The 90-second test before delegating any new task: (a) if AI gets it wrong, what is the worst-case cost, (b) does the customer expect a human authored this, (c) is disclosure feasible without breaking the trust contract.","article_url":"https://agentmodeai.com/operators/what-to-delegate-to-ai/","topic":"operators-ai-tooling","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-062","claim":"For a UK sole trader, Claude Pro and ChatGPT Plus subscriptions are allowable expenses under HMRC's wholly-and-exclusively test (BIM37007) only when paid from the business account and used for trade purposes; the business-tier seats (Claude Team £24/user/month, ChatGPT Business $25/user/month) are cleaner deductibles above £50k turnover because the personal-use exposure is structurally lower and the audit trail is built for the trade. Reverse-charge VAT applies on Anthropic invoices (US entity) and OpenAI invoices (Ireland entity invoicing most UK customers) under VAT Notice 741A place-of-supply rules; Microsoft Copilot UK plans invoice via Microsoft UK Ltd with VAT on the invoice. The MTD-ITSA regime that landed 6 Apr 2026 (mandatory at £50k combined trading + property income, dropping to £30k in Apr 2027 and £20k in Apr 2028) requires every AI-vendor invoice to be captured in HMRC-recognised software at the point it lands; FreeAgent is the practitioner default for UK sole traders under £200k turnover (Open Banking feed, receipt capture included, reliable AI-vendor categorisation), Xero scales better as headcount appears, QuickBooks works but its UK MTD-ITSA module is the youngest of the three. The practical trigger to switch from consumer-tier (Pro / Plus) to business-tier (Team / Business) is VAT registration: the £4-7/month/seat uplift is below the noise floor; the wholly-and-exclusively defence, the admin console, and the audit trail justify the move.","article_url":"https://agentmodeai.com/operators/uk-sole-trader-ai-stack-mtd-vat/","topic":"operators-ai-economics","pub_date":"2026-05-12","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-063","claim":"Under the BNC micro regime, AI-tool subscriptions are not separately deductible because the 34% abattement forfaitaire is fixed; therefore the decision to add AI tooling above ~50 k€ CA is not a tax question but a velocity-to-ceiling question — at the 77 700 € threshold the right move is to forecast the régime réel crossover before adding tooling, not after.","article_url":"https://agentmodeai.com/operators/micro-entrepreneur-ia-urssaf-bnc-france/","topic":"operators-ai-economics","pub_date":"2026-05-12","last_reviewed":"2026-06-10","next_review":"2026-07-20","verdict":"partial","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"The micro-regime ceiling figure in the claim text is overtaken: the 2026-2028 triennial revalorisation raised the services/BNC plafond from 77 700 € to 83 600 € (mixed-activity global cap 203 100 €), per LégiFiscal's 2026 thresholds bulletin and the URSSAF 2026 seuils announcement; the cited impots.gouv.fr page still presents the 77 700 € figure under its 2023-2025 framing. The claim's structural analysis is unaffected — the 34% abattement forfaitaire remains fixed (BOI-BNC-DECLA-10-70), AI subscriptions remain non-deductible under micro-BNC, and the velocity-to-ceiling advice stands with the crossover forecast now running against 83 600 €. Secondary note correction: the TVA franchise for services in 2026 is 37 500 € base / 41 250 € majoré (service-public.fr F21746, extracted 10 Jun 2026: 'Pour l'année 2026, les seuils de franchise en base de TVA française applicables restent inchangés' and the single 25 000 € threshold proposal 'a été abandonnée'); the 36 800 € figure carried in the article body and prior note was the 2023-2024 value. Status Up -> Partial. Article body needs a Peter-approved threshold refresh: 77 700 € appears in the title, excerpt, supportingFigure, FAQ, and body; 36 800 € in FAQ and body; the slug carries no figure and is unaffected."}],"primary_sources":[]},{"id":"OPS-064","claim":"For a freelance translator below 0.10 €/word, accepting MTPE rates at agency-standard 40–60% of full rate is rational only when productivity exceeds 1.8× source-rate baseline; below that ratio, MTPE work is rate-cannibalising and the freelancer should refuse it or move directly to higher-margin language pairs.","article_url":"https://agentmodeai.com/operators/freelance-translator-ai-stack-post-editing/","topic":"operators-ai-verticals","pub_date":"2026-05-12","last_reviewed":"2026-06-04","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-065","claim":"A solo agency delivering AI-assisted work to a client needs four contract clauses by Aug 2026 — disclosure of AI use, IP warranty carve-out for AI-generated portions, training-data exclusion of client materials, and a liability cap tied to fee paid — without which the agency carries strict liability under EU AI Act Article 50 plus contract-law warranty exposure on copyright.","article_url":"https://agentmodeai.com/operators/ai-client-deliverable-contract-clauses/","topic":"operators-ai-compliance","pub_date":"2026-05-12","last_reviewed":"2026-06-04","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-066","claim":"For a services firm under 50 people, paid AI seats pencil at ~5-person firms only when at least 2 of 5 staff are knowledge workers spending more than 10 hours/week on text drafting, and they fail to pencil at 25–40-person firms if the firm-wide rollout includes less than 60% adoption — between those zones, the break-even is determined by adoption rate, not seat price.","article_url":"https://agentmodeai.com/operators/ai-break-even-headcount-smb/","topic":"operators-ai-economics","pub_date":"2026-05-12","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-067","claim":"For a 5-15 person services agency running Cursor, Windsurf, Claude Code, or any internal agent platform built on the Model Context Protocol on paid-client-work machines, the May 2026 CVE class (Microsoft Security Response Center's CVE-2026-25592 and CVE-2026-26030 against Semantic Kernel on 7 May 2026, OX Security's MCP STDIO supply-chain advisory, and the Windsurf 1.9544.26 prompt-injection-to-MCP-registration path) cannot be cleared by vendor auto-update alone. A 5-step playbook (inventory every machine; pin the patched version and disable auto-update; write a one-page MCP allowlist; disclose AI-IDE use to active clients in writing; schedule a 30-day review) is the agency-level minimum that holds against the question an enterprise client will ask in procurement and against the residual liability the agency carries if a remediation conversation becomes necessary.","article_url":"https://agentmodeai.com/operators/windsurf-mcp-advisories-small-agency-ide-playbook/","topic":"operators-ai-security","pub_date":"2026-05-17","last_reviewed":"2026-06-04","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-068","claim":"The solopreneur AI subscription stack is in active compression through 2026, with median monthly spend tracking from $300-$500 in 2024 toward $100-$130 in early 2026 (Godberry Studios Zoom Solopreneur 50 teardown corroborated by BetterCloud SaaS industry data) and toward under $80 by end-Q3 2026 (editorial estimate, source:our-estimate, based on observed compression rate and announced 2026 feature roadmaps of Claude, ChatGPT, and Microsoft 365 Copilot). The compression is not a pricing story; it is a category-collapse story driven by 12 named standalone-AI categories being absorbed into the foundation-model subscription (Claude Pro or ChatGPT Plus) and the host platform's native AI: standalone writing assistants, meeting summarisers, slide generators, email-draft assistants, SEO optimisers, form-builder add-ons, calendar overlays, standalone chatbots, image-generation subscriptions for non-creative-pro use, note-taking add-ons, no-code agent builders, and standalone research assistants. The operator-side decision rule is cancel-now, wait-one-cycle, or migrate-carefully, applied via a four-line sub-thirty-minute test-before-cancel script against each line in the standing-subscription list.","article_url":"https://agentmodeai.com/operators/solopreneur-ai-stack-consolidation/","topic":"operators-ai-tooling","pub_date":"2026-05-17","last_reviewed":"2026-06-17","next_review":"2026-07-31","verdict":"partial","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-17","verdict":"partial","note":"Source-text re-review: the '$300-$500 (2024) toward $100-$130 (early 2026)' median trajectory is not stated in either cited source — the Godberry Studios teardown reports stack cost by revenue tier (not a year-over-year median) and BetterCloud's SaaS-industry data covers enterprise spend, not solopreneur AI subscriptions. The compression direction is supported by the Godberry tier data and observable foundation-model bundling; the specific year-anchored median figures are reclassified as source:our-estimate in the article. The load-bearing claim (active compression / category-collapse) holds; status moved to Partial pending a primary source carrying a dated solopreneur-median series."}],"primary_sources":[]},{"id":"OPS-069","claim":"The widely-cited 95-percent generative-AI-pilot-failure framing (MIT Sloan Management Review and Boston Consulting Group adoption-research streams, 2025-2026) is methodologically defensible for the enterprise cohort the research sampled (large firms with dedicated AI functions, 12-to-18-month evaluation windows, scaled-production-deployment success definition) and materially misrepresents small-firm pilot dynamics. The 1-to-50-person operator cohort has a different failure-mode catalogue (tool-assigned-to-wrong-person, rewrite-cost-exceeds-savings, client-rework-from-AI-deliverable, line-item-stack-compounded-and-cancelled, sporadic-use-no-routine) and a different success definition (90-day payback at actual hourly rate; deliverable quality reaching the client without disproportionate rework; routine fit documented for handover). A three-question Monday-morning small-firm pilot test (payback, deliverable quality, routine fit) checked at 30 days and 60 days is the operator's actual evaluation instrument and replaces the enterprise 12-to-18-month evaluation cycle that the 95-percent number is measured against.","article_url":"https://agentmodeai.com/operators/small-firm-ai-pilots-fail-differently-enterprise-misread/","topic":"operators-ai-tooling","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-07-01","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-070","claim":"For 1-50 person operators running Claude, Claude Code, or Cursor (Claude-backed) on paid client work, the 19 May 2026 Karpathy hire at Anthropic is a vendor-trajectory signal operating on the 12-to-24-month time horizon (Karpathy's mandate is upstream pre-training research, not application-layer change), and the right operator-side response is the 70% concentration rule. List every monthly AI-stack line item, tag by underlying vendor (Anthropic / OpenAI / Microsoft / Google / other), compute the largest single-vendor share. Above 70% concentrated on Anthropic, add a deliberate secondary-lab subscription (ChatGPT Plus at $20/mo or Gemini Advanced at comparable pricing) as resilience against any future Anthropic-specific incident. Below 70%, continue concentrating and re-evaluate at the 45-day claim review. The hire does not change the daily workflow this week; it affects the medium-term improvement trajectory of the vibe-coding interface the operator is already using.","article_url":"https://agentmodeai.com/operators/karpathy-joins-anthropic-vibe-coding-operator-stack/","topic":"operators-ai-tooling","pub_date":"2026-05-19","last_reviewed":"2026-05-19","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-071","claim":"Colorado's AI Act replacement bill SB 189 — passed by both chambers of the Colorado legislature in May 2026 and effective 30 June 2026 — drops the original law's risk-management programme requirements and annual impact assessments in favour of a notice-and-transparency framework; operators using AI to make or materially influence a consequential decision about a Colorado resident (employment, housing, credit, insurance, education, healthcare) have active notice, disclosure, and documentation obligations from 30 June 2026 regardless of firm size.","article_url":"https://agentmodeai.com/operators/colorado-ai-act-sb189-operator-compliance-brief/","topic":"operators-ai-compliance","pub_date":"2026-05-22","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Trigger condition (2) fired: the effective date moved. Governor Polis signed SB 26-189 on 14 May 2026 (Holland & Knight client alert, May 2026; Seyfarth; Littler). The signed law repeals and reenacts the original Colorado AI Act and its obligations take effect 1 Jan 2027 — not 30 Jun 2026 as the claim asserted. No operator obligation starts 30 Jun 2026; the only pre-2027 item is Colorado AG rulemaking due by 1 Jan 2027. The claim's structural reading holds (risk-management programmes and impact assessments dropped for a notice-and-transparency framework; consequential-decision scope covering employment, housing, credit, insurance, education, healthcare; no small-firm exemption). The urgency leg ('obligations from 30 June 2026') is overtaken. Status Up → Partial."}],"primary_sources":[]},{"id":"OPS-072","claim":"Notion's 13 May 2026 developer platform launch — adding Workers, External Agents API (Claude, Codex, Decagon natively integrated), and Database Sync — means small teams already running on Notion have functional AI agent infrastructure that materially overlaps with separately paid automation tooling (Zapier, Make.com) for Notion-internal workflows; the consolidation evaluation belongs in the current billing cycle, and the Workers free tier through August 2026 is the zero-cost window to test it.","article_url":"https://agentmodeai.com/operators/notion-ai-agents-hub-small-team-operator-upgrade/","topic":"operators-ai-tooling","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-073","claim":"OpenAI's 11 May 2026 Deployment Company launch — a $4B+ entity with 150 Forward Deployed Engineers embedded inside client organisations to redesign workflows and build durable AI systems — confirms that AI configuration and workflow redesign are the market-identified constraint on AI ROI, not model access; for the 1-15 person operator or freelance builder whose differentiated value is knowing how to make AI work in a specific client context, this announcement is a competitive-positioning map: the service is real, the demand is confirmed, and the small operator has three durable advantages (speed, context depth, price) that the Deployment Company's structure cannot replicate at small-client scale.","article_url":"https://agentmodeai.com/operators/openai-deployment-company-operator-positioning-signal/","topic":"operators-ai-tooling","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-074","claim":"A 5-15 person team running AI tools on paid client work in 2026 can move from default-shared personal credentials for AI agents to a defensible non-human-identity posture in three hours of work using existing tooling (password manager, calendar, spreadsheet). The five-step starter kit (inventory every AI tool acting in the environment and its credential; mint per-agent credentials with smallest-scope; move every credential into one secrets vault and remove from elsewhere; set a 90-day rotation cadence with a calendar owner; write and test a one-page leaver and revocation runbook) covers the credential-management practices that CyberArk-grade enterprise NHI programmes cover, scaled to a team without an identity-governance function. The answer the kit produces is sufficient for almost every mid-market and SMB client procurement question, and a credible answer to most enterprise procurement questionnaires reaching small-agency vendors in 2026.","article_url":"https://agentmodeai.com/operators/operators-nhi-starter-kit/","topic":"operators-ai-security","pub_date":"2026-05-24","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-075","claim":"A 1-10 person team in 2026 has approved-tool unapproved-capability shadow AI running inside its already-approved SaaS estate (Notion AI agents, Slack AI, Microsoft 365 Copilot, Atlassian Intelligence, Google Workspace Gemini, Salesforce Einstein) at near-universal incidence, because the vendors have shipped these capabilities as on-by-default or low-friction add-ons inside the existing licence path. The three discovery signs (SaaS bill line items the founder does not remember approving; team-member mentions of new features inside existing tools; vendor admin console notifications advertising auto-enabled AI capabilities) are reliable triggers for a 60-minute audit (SaaS bill review, vendor admin console walk, team check-in) that produces a one-page inventory mapping each AI capability to the tool, activation date, current users, data scope, and team's explicit posture. The inventory is the artefact that answers client procurement questions about AI tool exposure and the basis for any disable-or-restrict decisions.","article_url":"https://agentmodeai.com/operators/approved-tool-unapproved-capability-shadow-ai-operators/","topic":"operators-ai-security","pub_date":"2026-05-24","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-076","claim":"A small agency (1-15 person team) building agentic features on paid client work in 2026 should pick its agent protocol per project by reading the client's existing stack (Anthropic-aligned client → MCP; Google-aligned client → A2A; sovereignty- or self-hosted-aligned client → Llama Stack), default to MCP for tool-heavy work and A2A for agent-collaboration work when the client has no existing stack, and keep its tool inventory portable by building every tool as a plain HTTP service first and wrapping it to the chosen protocol second. The plain-HTTP-first discipline costs roughly 20-30% extra on the first tool of a project and produces 60% wrapper-effort reduction on the second client requesting the same functionality on a different protocol. Tracking wrapper hours separately from service hours in the agency's time log is the simplest instrument for quoting the re-platform cost accurately when a client asks.","article_url":"https://agentmodeai.com/operators/agent-protocol-picking-for-small-agencies/","topic":"operators-ai-security","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-077","claim":"As of mid-2026, a non-developer operator on a 1-50 person team can build a production-useful internal-process agent in Notion (via the 13 May 2026 developer platform: Workers, External Agents API with Claude/Codex/Decagon, Database Sync, ntn CLI) or in ChatGPT (via the matured GPT Builder with Instructions, Knowledge, Actions and four sharing tiers) without a separate automation tool. The default permission scope of the resulting agent is the workspace-wide access of the human who built it (Notion: workspace-admin if the builder is an admin; ChatGPT: the builder's API key scope on configured Actions), which is the operator-scale shadow-AI risk. The safe-deploy playbook (three guardrails: scope to one data source not whole workspace; read-only first for a week of parallel running; human approval on customer-facing actions; plus the one permission-scope rule that the agent inherits the builder's access not the user's) closes the gap using only the tools' native configuration.","article_url":"https://agentmodeai.com/operators/no-code-agent-building-notion-gpt/","topic":"operators-ai-tooling","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-078","claim":"As of mid-2026, most 1-15 person teams running AI agents and automations on paid client work cannot revoke a misbehaving agent's access quickly because they share a small number of credentials across multiple tools and have no written pause-and-revoke runbook with rehearsed timings per tool. The runbook discipline (per-tool documentation of the pause path, the revoke path, the time-to-effect, and the OAuth third-party revocation step where applicable) is a 30-minute Friday investment using only the tools already in use, and is the small-team analogue of the four-primitive enterprise containment architecture covered in AM-171.","article_url":"https://agentmodeai.com/operators/kill-switch-for-small-team/","topic":"operators-ai-security","pub_date":"2026-05-26","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-079","claim":"As of mid-2026, the consumer-tier and small-team AI tools most 1-15 person teams use on client work (ChatGPT, Claude, Notion AI, Cursor, customer-service bots) retain context across sessions by default in ways that can carry one client's information into another's work, with each tool exposing a different memory default in a different settings location and a different cross-context behaviour. Most operators have not sat down to align them, and the 30-minute Monday hygiene routine (settings pass per tool; no-client-identifiers rule on every persistent custom-instructions field; per-client reset checklist) closes the operational gap using only the tools already in use.","article_url":"https://agentmodeai.com/operators/agent-memory-small-team-hygiene/","topic":"operators-ai-security","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-080","claim":"For a small business in the EU or selling into it, the Digital Omnibus political agreement of 7 May 2026 pushes the heavy high-risk obligations out to 2 December 2027 and 2 August 2028 but does not change the duties most small businesses are actually subject to: as a deployer rather than a provider, you must disclose AI-generated deepfake image, audio, or video and ensure AI chat tells people they are talking to a machine from 2 August 2026 under Article 50, you have been under the Article 4 AI literacy duty since 2 February 2025, and a 30-minute readiness check using tools you already have closes most of the practical gap.","article_url":"https://agentmodeai.com/operators/eu-ai-act-small-business-deployer-duties/","topic":"operators-ai-compliance","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-07-12","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-081","claim":"Google Workspace Studio, the no-code AI agent builder Google introduced at Cloud Next 2026 for paid Google Workspace tiers, is the right first place for a small team to build agents when its data and day-to-day workflows already live inside Google Workspace such as Gmail, Docs, Sheets, and Drive, because the integration and permissions are native and the build is natural-language; for a team whose work spans many third-party tools, a model-neutral automation layer such as n8n or a documentation-centric build in Notion remains the better default, and the deciding question is where the team's data and workflows already live rather than which builder markets the most capability.","article_url":"https://agentmodeai.com/operators/google-workspace-studio-for-small-teams/","topic":"operators-ai-tooling","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-07-12","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-082","claim":"If a small business shipped a customer-facing app built with a no-code or vibe-coding platform and never ran a security review, the safe operating assumption is that its database is reachable from the public internet until proven otherwise, because spring 2026 scans found thousands of such apps actively leaking sensitive data and the common cause is open data-access defaults that the natural-language build does not close for you.","article_url":"https://agentmodeai.com/operators/vibe-coded-app-security-check/","topic":"operators-ai-security","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-083","claim":"AI inference is getting cheaper per token in 2026 while the AI bills small businesses actually pay are rising, because the cost has moved from the model to the automation layer where metered SDK and agent usage now sits; the imminent example is Anthropic's announced 15 Jun 2026 split that carves Claude automation and SDK usage out of the flat subscription into a separately metered pool, so a small business running AI inside automations should re-model its stack before the cutover.","article_url":"https://agentmodeai.com/operators/ai-cheaper-but-your-bill-rising/","topic":"operators-ai-economics","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-084","claim":"For a small business, the fully-autonomous AI sales-development rep that finds prospects and sends outreach without a human has not proven durable, because autonomous volume sending burns a small sender's domain reputation and fully-automated personalization reads as machine-made; the pattern that works is human-in-the-loop, where AI handles the research and the first draft and a person approves and sends.","article_url":"https://agentmodeai.com/operators/ai-sdr-human-in-the-loop/","topic":"operators-ai-verticals","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-085","claim":"Shopify's Magic and Sidekick AI tools are included on every Shopify plan at no extra cost as of May 2026, so most merchants are already paying for a built-in AI assistant they have not activated; the operator value is in switching it on for the two or three recurring, easy-to-verify tasks that fit (store-data questions, repetitive copy, small multi-step admin tasks) rather than buying a separate ecommerce-AI subscription, while keeping it away from unverified pricing, financial, factual-product, and live-stock decisions.","article_url":"https://agentmodeai.com/operators/shopify-magic-sidekick-guide-small-business/","topic":"operators-ai-tooling","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-06-29","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-086","claim":"After Fathom limited advanced AI summaries on its free plan to roughly five per month in 2026, the AI meeting-notetaker choice for a small team turns on meeting volume and privacy posture rather than feature lists: a solo or light user can stay on a free tier or a single-seat upgrade (Fathom Premium around $16/month annual), while a team with several client meetings a week across multiple people is usually better served by a per-seat plan such as Fireflies Pro (around $10/user/month annual); pick on how much you actually meet and your data/consent posture, not on the comparison table.","article_url":"https://agentmodeai.com/operators/ai-meeting-notetaker-fathom-fireflies-small-business/","topic":"operators-ai-tooling","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-06-29","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-087","claim":"Webflow's May 2026 pricing restructure combines its former CMS and Business site plans into a single Premium plan at $25/month billed annually ($39/month on monthly billing) with 20,000 CMS items and 40 collections included, effective on renewal on or after 29 Jun 2026 for most existing sites (16 Nov 2026 for freelancer and agency workspaces); by Webflow's own account the change raises some sites' cost, lowers others', and leaves some unchanged, so the operator move is to run Webflow's own change calculator against actual usage (CMS items, editor seats, AI and other features used) and pick the cheapest correct plan before the effective date, rather than auto-accepting the migration or switching platforms over pricing noise.","article_url":"https://agentmodeai.com/operators/webflow-pricing-change-small-business-guide/","topic":"operators-ai-economics","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-088","claim":"A solo developer or small agency that runs an AI coding assistant (Claude Code, Cursor, Gemini CLI, GitHub Copilot, OpenAI Codex, Grok) on the same machine that holds its client SSH keys and deploy credentials is materially exposed by the May 2026 TrustFall and SymJack findings, in which opening a malicious repository and accepting an approval prompt can run attacker code that steals those secrets, and the proportionate fix is not a security budget but updating every tool to its latest version, slowing down on approvals (especially file copies and writes to configuration files), not opening untrusted repositories on a credentialed machine, and moving secrets out of plain files while rotating anything that may have been exposed.","article_url":"https://agentmodeai.com/operators/ai-coding-cli-security-small-team/","topic":"operators-ai-security","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-089","claim":"Agentic bookkeeping arrived for small business in May 2026 (Xero's XeroForce agent builder, in alpha and invite-only, plus JAX and Xero OS; Intuit's QuickBooks Assist agent teams), and the operating split that protects a small owner is to let agents run the repetitive, reversible, internal ledger work (categorisation, reconciliation, anomaly flagging, month-end report drafts, organising tax documents) while keeping a human approval on anything that moves money out or files with the authorities (pay runs, supplier payments, tax submissions), because those are irreversible and the owner remains responsible for what is filed and paid regardless of which tool prepared it, a caveat the vendors' own human-at-the-helm framing concedes.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-agents-small-business/","topic":"operators-ai-economics","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-090","claim":"For a US small business (under about 50 people) the 2026 state AI laws impose far less than the headlines imply: California's SB 53 applies only to frontier model developers (above ~10^26 training operations or >$500M revenue) and not to AI users, Colorado's comprehensive law was repealed, narrowed, and delayed to 1 January 2027 by SB 26-189 (signed 14 May 2026), Texas's Responsible AI Governance Act (effective 1 January 2026) is intent-based rather than paperwork-based, and no federal AI law is in force, so the proportionate response is a short list — know which states you actually touch, confirm you are a deployer not a frontier developer, adopt basic disclosure hygiene, and never deploy AI to deceive or discriminate — rather than an enterprise-scale compliance programme.","article_url":"https://agentmodeai.com/operators/us-ai-laws-small-business/","topic":"operators-ai-compliance","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-07-17","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-091","claim":"HubSpot's 14 Apr 2026 shift of its Breeze Customer Agent to $0.50 per resolved conversation (from $1.00 per conversation) makes the agent worth trialing for a small support team because cost now tracks resolved tickets rather than attempts, but HubSpot's definition of a resolved conversation (the agent shares a content source or performs an action and no human handoff occurs within 72 hours of the last message) is the term that decides whether it is actually cheap, so the comparison that matters is $0.50 per resolved ticket against the team's loaded cost per human-handled ticket, not against zero.","article_url":"https://agentmodeai.com/operators/hubspot-breeze-pay-per-resolution/","topic":"operators-ai-tooling","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-092","claim":"With Notion Custom Agents moving from free beta to credit-based billing ($10 per 1,000 credits, no rollover) on 4 May 2026, the operator decision flips from which agents to build to which recurring agents clear their per-run cost, and because the bill is driven by run frequency rather than the number of agents built, only a few high-frequency high-value patterns (daily brief, inbox or task triage, recurring status roll-up) are worth keeping for most small teams.","article_url":"https://agentmodeai.com/operators/notion-custom-agents-credit-pricing/","topic":"operators-ai-tooling","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-093","claim":"For a small team already on a Microsoft 365 Business plan, the $18 per user per month promotional rate for the Copilot Business add-on locks in only through 30 Jun 2026 before rising to the $21 standard rate, so the cost-rational move is to decide before the deadline, but only on the seats whose actual Word, Excel, Outlook and Teams usage clears the per-seat math, because the discount applies only to the first year and only to seats licensed before the deadline.","article_url":"https://agentmodeai.com/operators/microsoft-365-copilot-business-price-lock/","topic":"operators-ai-economics","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-094","claim":"The Canva + Perplexity Computer connector (launched 4 Jun 2026, available to Perplexity Pro at $20/month and the Max and Enterprise tiers across 11 languages) lets the Perplexity agent turn research into editable Canva designs, which removes the research-to-design context switch for a consultant or small agency that produces decks from research weekly, so the decision turns on the Perplexity subscription and the cadence of designed output, not on the workflow itself, and the output is a starting draft rather than a finished brand asset.","article_url":"https://agentmodeai.com/operators/canva-perplexity-computer-connector/","topic":"operators-ai-tooling","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-095","claim":"OpenAI's six role-specific Codex plugins (announced 2 Jun 2026, including Sales wired to Salesforce/HubSpot/Slack and Creative Production wired to Figma/Canva) are rolling out to Business and Enterprise plans first, while Codex itself is available from the $20 ChatGPT Plus plan up, so for a small team on Plus the cost-rational move is to use Codex now for self-assembled automation and treat the packaged plugins as a reason to watch rather than an immediate reason to upgrade to Business.","article_url":"https://agentmodeai.com/operators/openai-codex-role-plugins-small-team/","topic":"operators-ai-tooling","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-096","claim":"Google's new $100/month AI Ultra tier (introduced at Google I/O on 19 May 2026 alongside a cut of the top AI Ultra plan from $250 to $200) buys usage headroom and storage on top of the same Gemini models the cheaper tiers run, so it earns its price only for heavy daily agent users who exhaust a lower tier's limits, and for most operators the cost-rational path is the cheaper Google AI Pro subscription or Gemini 3.5 Flash via the API ($1.50 per million input tokens, $9.00 per million output).","article_url":"https://agentmodeai.com/operators/google-ai-ultra-pricing-decoded/","topic":"operators-ai-economics","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-097","claim":"The calendar-phishing and ClickFix scam patterns named in Google's 8 Jun 2026 fraud advisory specifically exploit the trust habits that AI scheduling, notetaker and assistant tooling creates in small teams (machine-generated calendar entries treated as normal; paste-this-command setup motions rehearsed as routine), and both are closed by free measures: restricting calendar auto-add to known or accepted invites, reporting rather than deleting phishing invites, and a team rule of never pasting a command a web page solicited.","article_url":"https://agentmodeai.com/operators/calendar-phishing-clickfix-small-business/","topic":"operators-ai-security","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-098","claim":"Zapier bills MCP usage at a fixed rate of two tasks per successful AI tool call against the plan's existing task allowance, with no separate MCP meter and no per-session cap (documentation updated 4 Jun 2026), which halves a plan's real agent capacity (a 750-task Professional plan at $19.99/month annual is 375 agent actions), so an operator wiring Claude or ChatGPT into Zapier should budget every AI workflow at double its successful-call count and keep monitoring-grade chatter off the MCP path.","article_url":"https://agentmodeai.com/operators/zapier-mcp-billing-two-tasks-per-call/","topic":"operators-ai-economics","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-099","claim":"Notion Workers (the hosted serverless code runtime shipped with the 13 May 2026 Developer Platform, distinct from the Custom Agents product) is free on Business and Enterprise plans until 11 Aug 2026 and then meters at $0.0023 per run, and because the post-August bill is purely a function of run frequency, the rational operator move is to build the highest-value syncs now, measure a real week of run counts inside the free window, and tune schedules before the meter starts.","article_url":"https://agentmodeai.com/operators/notion-workers-free-window-pricing/","topic":"operators-ai-tooling","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-100","claim":"Meta's Business Agent went globally available on 3 Jun 2026 across WhatsApp, Messenger and Instagram — free to activate, with paid subscription offerings announced for the coming months and more than one million businesses already using it pre-rollout per Meta — and because the free period is explicitly temporary, the rational small-business move is to treat it as an evaluation window: activate on a bounded slice, measure resolution rate and supervision cost (how often it says something you would not have said), and know the price at which it is worth paying before the tiers are announced.","article_url":"https://agentmodeai.com/operators/meta-business-agent-whatsapp-small-business/","topic":"operators-ai-tooling","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-101","claim":"Intuit's QuickBooks Workforce (announced 6 May 2026) packages an AI payroll agent for small teams at $50/month plus $6.50 per employee on the entry tier ($88 plus $10 Premium, $134 plus $12 Elite), with new pricing effective 1 Jul 2026 for new customers, and the operating rule that protects a small owner is the bookkeeping split applied to its sharpest case: let the agent do the reversible preparation (gather time data, flag inconsistencies, draft the run) while a human approves every execution, because payroll is irreversible money out plus filings to authorities and the owner remains responsible regardless of which tool prepared it.","article_url":"https://agentmodeai.com/operators/quickbooks-workforce-ai-payroll-small-team/","topic":"operators-ai-economics","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-102","claim":"As of 10 Jun 2026, Mastercard (Agent Pay for Machines) and Visa (Intelligent Commerce, with OpenAI) have both shipped production rails for AI agents to transact, so the operator-relevant decision is no longer whether to let an agent pay but how to scope it: a tokenised agent-specific credential, a hard per-period spend cap, merchant-category limits, and human approval by default.","article_url":"https://agentmodeai.com/operators/ai-agent-payments-small-business-guardrails/","topic":"operators-ai-tooling","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-07-28","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-103","claim":"AI tools small businesses use are shifting from per-seat subscriptions to outcome- and activity-based metering (HubSpot Breeze at $0.50 per resolved conversation since 14 Apr 2026, Zapier Agents metered per activity), which makes resolution rate and activity frequency, not seat count, the variables an owner must model before turning an agent loose.","article_url":"https://agentmodeai.com/operators/outcome-based-ai-pricing-small-business/","topic":"operators-ai-economics","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-07-28","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-104","claim":"AI-generated fraud (voice cloning, deepfake video calls, fake-supplier emails) is now a primary cyber-risk for small businesses without a dedicated finance or IT function, and the effective defenses are procedural, out-of-band callback verification on a number you already had and dual-control on any payment or bank-detail change, rather than technical detection tooling.","article_url":"https://agentmodeai.com/operators/ai-fraud-small-business-defenses/","topic":"operators-ai-security","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-07-28","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-001","claim":"The 47-question AI Vendor Security Questionnaire covers seven failure surfaces (model lineage, training/inference data handling, non-human identity, audit/observability, kill-switch, EU AI Act + GDPR posture, contract/indemnification) that CAIQ v4 and SIG do not address; vendors that cannot answer score sections binary-unanswered, and the questionnaire is the addendum (not replacement) to existing cloud/SaaS procurement frameworks.","article_url":"https://agentmodeai.com/resources/ai-vendor-security-questionnaire/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-002","claim":"The pre-deployment AI DPIA template fuses GDPR Article 35 obligations with EU AI Act Article 26 (deployer) and Article 27 (FRIA where applicable) into a single working-session document; sections 7 and 8 are conditional on the EU AI Act risk classification established in section 1, which means deployers complete the full document only when the system is classified as high-risk under Annex III.","article_url":"https://agentmodeai.com/resources/ai-dpia-template/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-003","claim":"The four-phase agent incident runbook (detect within 4h, contain within 30s, roll back per action class, post-mortem with MTTD-for-Agents detection chain) is the operational overlay on standard SRE incident response that most enterprises deploying agentic AI in 2026 do not have; the seven action classes (database writes, external API calls, customer comms, document publication, code commits, identity changes, knowledge-base writes) each require a distinct rollback procedure and the runbook captures the operator authorised, time budget, and substitute action where rollback is impossible.","article_url":"https://agentmodeai.com/resources/agent-incident-runbook/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-004","claim":"The Works Council AI Notification Packet covers three EU jurisdictions (German BetrVG §87(1) point 6, Dutch WOR Article 27, French CSE consultation under Code du travail L2312-8) plus the EU AI Act Article 26(7) deployer-notification overlay that activates 2 August 2026; early engagement (vendor-shortlist landing) compresses deployment timelines from 12-18 months (late engagement) to 6-9 months in the Mittelstand case studied in OPS-049, and the per-jurisdiction documents (Betriebsvereinbarung, OR convenant, CSE avis) are designed to consolidate into a single deployment-go decision with the Article 26(7) notification appended afterward.","article_url":"https://agentmodeai.com/resources/works-council-ai-notification-packet/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-005","claim":"The 38-item AI MSA red-team checklist organises the contractual review around seven clause families (training-data carve-outs, output ownership + IP indemnification, model-deprecation rights, sub-processor expansion, kill-switch SLA, exit-data portability, regulatory + EU AI Act flow-through) where 2025-2026 enterprise AI MSA failures cluster; vendors scoring yes on 30+ items are contractually serious, 20-29 items are treatable through negotiation, and below 20 signals that the vendor's commercial position depends on retaining the rights the checklist is designed to constrain.","article_url":"https://agentmodeai.com/resources/ai-msa-red-team-checklist/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]}]},"claim_archive":{"description":"Claims made by enterprise-AI vendors, analyst firms, academics, tier-1 publications, regulators, and identified consensus positions. Each is reviewed on a 30/60/90/180-day cadence.","methodology_url":"https://agentmodeai.com/archive/method/","index_url":"https://agentmodeai.com/archive/","insights_url":"https://agentmodeai.com/archive/insights/","feeds":{"new_claims":"https://agentmodeai.com/archive/feed/new-claims.xml","new_reviews":"https://agentmodeai.com/archive/feed/new-reviews.xml","status_changes":"https://agentmodeai.com/archive/feed/status-changes.xml"},"claims":[{"id":"ACA-2026-001","claim":"Stanford HAI's 2026 AI Index reports organisational AI adoption at 88%, with productivity gains of 14-26% in customer support and software development, and up to 72% in marketing teams. AI agent adoption across departments remains in single digits.","claim_url":"https://agentmodeai.com/claims/ACA-2026-001/","source_type":"academic","source_name":"Stanford HAI","source_url":"https://hai.stanford.edu/ai-index/2026-ai-index-report","source_snapshot_url":"https://web.archive.org/web/20260419215510/https://hai.stanford.edu/ai-index/2026-ai-index-report","source_date":"2026-04-14","log_date":"2026-04-19","review_cadence":180,"domain_tags":["adoption-rate","benchmark","capability-benchmark"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"ACA-2026-002","claim":"Stanford HAI's 2026 AI Index reports that China has erased the US lead in AI on capability + publication-volume benchmarks, with the two countries now at near-parity after a three-year closing curve.","claim_url":"https://agentmodeai.com/claims/ACA-2026-002/","source_type":"academic","source_name":"Stanford HAI","source_url":"https://hai.stanford.edu/news/inside-the-ai-index-12-takeaways-from-the-2026-report","source_snapshot_url":"https://web.archive.org/web/20260419215510/https://hai.stanford.edu/ai-index/2026-ai-index-report","source_date":"2026-04-14","log_date":"2026-04-19","review_cadence":180,"domain_tags":["market-position","capability-benchmark"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"ACA-2026-003","claim":"Stanford's Digital Economy Lab Enterprise AI Playbook documents bimodal ROI: 12% of enterprise AI deployments clear 300%+ ROI with 40-60% operational cost reductions; 88% sit at or below break-even.","claim_url":"https://agentmodeai.com/claims/ACA-2026-003/","source_type":"academic","source_name":"Stanford DEL","source_url":"https://digitaleconomy.stanford.edu/app/uploads/2026/03/EnterpriseAIPlaybook_PereiraGraylinBrynjolfsson.pdf","source_snapshot_url":"https://web.archive.org/web/20260407050750/https://digitaleconomy.stanford.edu/app/uploads/2026/03/EnterpriseAIPlaybook_PereiraGraylinBrynjolfsson.pdf","source_date":"2026-03-01","log_date":"2026-04-19","review_cadence":180,"domain_tags":["agent-roi","adoption-rate","benchmark"],"current_verdict":"retracted","next_review":"2026-12-07","review_history":[{"date":"2026-06-10","verdict":"retracted","oversight":"peter-led-deep-review","memo_url":"https://agentmodeai.com/claims/ACA-2026-003/review-2026-06-10/","counter_evidence_considered":null}],"change_history":[{"date":"2026-06-10","field_changed":"status","old_value":"pending-review","new_value":"retracted","reason":"First review (pulled forward from the 180-day cadence). Full-text verification of the source PDF found the claim_text figure absent: the Stanford DEL Enterprise AI Playbook documents no bimodal ROI distribution, no 12% / 300%+ cohort, and no 88% break-even body. The record asserted a figure its own snapshot never contained. Memo at review-2026-06-10."},{"date":"2026-06-10","field_changed":"review_history","old_value":"[]","new_value":"[{date: 2026-06-10, verdict: retracted, oversight: peter-led-deep-review, memo_slug: review-2026-06-10}]","reason":"Logged first review entry. See memo at content/claims/ACA-2026-003/review-2026-06-10.mdx for the full evidence chain and the verified adjacent figure (IDC/Lenovo via CIO.com, 25 Mar 2025: 88% of AI POCs never reach widescale deployment - a pilot-graduation metric, not an ROI distribution)."}]},{"id":"ACA-2026-004","claim":"Carnegie Mellon's TheAgentCompany 2026 update shows Gemini 2.5 Pro as the best enterprise-agent at 30.3% task completion — up from the 24% Claude 3.5 Sonnet baseline in 2024, but still far below production-readiness thresholds.","claim_url":"https://agentmodeai.com/claims/ACA-2026-004/","source_type":"academic","source_name":"Carnegie Mellon University","source_url":"https://arxiv.org/abs/2412.14161","source_snapshot_url":"https://web.archive.org/web/20260411143459/https://arxiv.org/abs/2412.14161","source_date":"2026-03-15","log_date":"2026-04-19","review_cadence":90,"domain_tags":["capability-benchmark","benchmark"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[]},{"id":"ANA-2026-001","claim":"Gartner predicted that over 40% of agentic AI projects will be cancelled by end of 2027, primarily due to escalating costs, unclear business value, and inadequate risk controls.","claim_url":"https://agentmodeai.com/claims/ANA-2026-001/","source_type":"analyst","source_name":"Gartner","source_url":"https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027","source_snapshot_url":"https://web.archive.org/web/20260412112337/https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027","source_date":"2025-06-25","log_date":"2026-04-19","review_cadence":180,"domain_tags":["adoption-rate","agent-roi","deployment-velocity"],"current_verdict":"holding","next_review":"2026-10-16","review_history":[],"change_history":[{"date":"2026-04-19","field_changed":"source_snapshot_url","old_value":"https://web.archive.org/web/20250625120000/https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027","new_value":"https://web.archive.org/web/20260412112337/https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027","reason":"Initial snapshot URL was a Phase 1 placeholder with an estimated timestamp; resolved to the nearest real capture via the Internet Archive availability API (Save Page Now returned HTTP 520 for Gartner — their servers actively limit archival requests)."}]},{"id":"ANA-2026-002","claim":"Only 28% of AI infrastructure-and-operations projects fully pay off, and 57% of leaders reporting failure cited \"expected too much, too fast\" as the top contributing factor.","claim_url":"https://agentmodeai.com/claims/ANA-2026-002/","source_type":"analyst","source_name":"Gartner","source_url":"https://www.theregister.com/2026/04/07/ai_returns_gartner/","source_snapshot_url":"https://web.archive.org/web/20260419221558/https://www.theregister.com/2026/04/07/ai_returns_gartner/","source_date":"2026-04-07","log_date":"2026-04-19","review_cadence":90,"domain_tags":["agent-roi","adoption-rate","deployment-velocity"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[]},{"id":"ANA-2026-003","claim":"Gartner predicts that by 2030, 50% of organisations will use autonomous AI agents to interpret governance policies and technical standards into machine-verifiable data contracts, automating compliance and governance-policy enforcement.","claim_url":"https://agentmodeai.com/claims/ANA-2026-003/","source_type":"analyst","source_name":"Gartner","source_url":"https://www.gartner.com/en/newsroom/press-releases/2026-03-11-gartner-announces-top-predictions-for-data-and-analytics-in-2026","source_snapshot_url":"https://web.archive.org/web/20260312144319/https://www.gartner.com/en/newsroom/press-releases/2026-03-11-gartner-announces-top-predictions-for-data-and-analytics-in-2026","source_date":"2026-03-11","log_date":"2026-04-19","review_cadence":180,"domain_tags":["governance","compliance","adoption-rate"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"ANA-2026-004","claim":"BCG's \"AI Transformation is a Workforce Transformation\" study finds future-built AI companies achieve 5× revenue increases and 3× cost reductions vs laggards. AI agents accounted for ~17% of AI-driven value in 2025, projected to reach 29% by 2028.","claim_url":"https://agentmodeai.com/claims/ANA-2026-004/","source_type":"analyst","source_name":"Boston Consulting Group","source_url":"https://www.bcg.com/publications/2026/ai-transformation-is-a-workforce-transformation","source_snapshot_url":"https://web.archive.org/web/20260331134658/https://www.bcg.com/publications/2026/ai-transformation-is-a-workforce-transformation","source_date":"2026-03-01","log_date":"2026-04-19","review_cadence":180,"domain_tags":["agent-roi","adoption-rate","market-position"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"ANA-2026-005","claim":"Gartner predicts that by 2028, 60% of brands will use agentic AI to facilitate streamlined one-to-one interactions with customers.","claim_url":"https://agentmodeai.com/claims/ANA-2026-005/","source_type":"analyst","source_name":"Gartner","source_url":"https://consumergoods.com/gartner-predicts-sharp-rise-ai-agents-within-enterprise-applications-2026","source_snapshot_url":"https://web.archive.org/web/20260419221413/https://consumergoods.com/gartner-predicts-sharp-rise-ai-agents-within-enterprise-applications-2026","source_date":"2026-01-15","log_date":"2026-04-19","review_cadence":180,"domain_tags":["adoption-rate","market-position"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"ANA-2026-006","claim":"McKinsey's State of AI 2025: 23% of respondents are scaling an agentic AI system, 39% experimenting. 39% attribute any EBIT impact to AI, most saying <5% of EBIT. A 6% 'AI high performer' segment attributes >5% of EBIT to AI.","claim_url":"https://agentmodeai.com/claims/ANA-2026-006/","source_type":"analyst","source_name":"McKinsey Global Institute","source_url":"https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai","source_snapshot_url":"https://web.archive.org/web/20260315102913/https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai","source_date":"2025-11-01","log_date":"2026-04-19","review_cadence":90,"domain_tags":["agent-roi","adoption-rate"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[]},{"id":"PUB-2026-001","claim":"The Financial Times reported that almost 40% of US data-center projects are at risk of falling behind schedule, per SynMax satellite + AI analytics data, attributing delays to permitting, local opposition, and shortages in labour, power, and equipment.","claim_url":"https://agentmodeai.com/claims/PUB-2026-001/","source_type":"publication","source_name":"Financial Times","source_url":"https://www.pymnts.com/artificial-intelligence-2/2026/permitting-hurdles-and-labor-shortages-threaten-ai-data-center-timelines","source_snapshot_url":"https://web.archive.org/web/20260419215722/https://www.pymnts.com/artificial-intelligence-2/2026/permitting-hurdles-and-labor-shortages-threaten-ai-data-center-timelines","source_date":"2026-04-10","log_date":"2026-04-19","review_cadence":90,"domain_tags":["adoption-rate","market-position"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[]},{"id":"PUB-2026-002","claim":"UK financial regulators are holding urgent talks with the government's cyber-security agency and major banks to assess systemic risks posed by Anthropic's latest Claude model, per the Financial Times.","claim_url":"https://agentmodeai.com/claims/PUB-2026-002/","source_type":"publication","source_name":"Financial Times","source_url":"https://www.thestandard.com.hk/finance/article/329134/","source_snapshot_url":"https://web.archive.org/web/20260419215826/https://www.thestandard.com.hk/finance/article/329134/UK-financial-regulators-rush-to-assess-risks-of-Anthropic-latest-AI-model-FT-reports","source_date":"2026-04-15","log_date":"2026-04-19","review_cadence":60,"domain_tags":["governance","security","regulatory"],"current_verdict":"pending-review","next_review":"2026-06-18","review_history":[],"change_history":[]},{"id":"PUB-2026-003","claim":"HBR's March 2026 framing argues AI agents should be scaled as team members — with defined roles, performance reviews, onboarding, and accountability — rather than as tools.","claim_url":"https://agentmodeai.com/claims/PUB-2026-003/","source_type":"publication","source_name":"Harvard Business Review","source_url":"https://hbr.org/2026/03/to-scale-ai-agents-successfully-think-of-them-like-team-members","source_snapshot_url":"https://web.archive.org/web/20260419220214/https://hbr.org/2026/03/to-scale-ai-agents-successfully-think-of-them-like-team-members","source_date":"2026-03-01","log_date":"2026-04-19","review_cadence":90,"domain_tags":["governance","adoption-rate"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[]},{"id":"PUB-2026-004","claim":"MIT Sloan's 2026 AI & Data Leadership Executive Benchmark Survey reports that 38% of responding companies have appointed a Chief AI Officer or equivalent role.","claim_url":"https://agentmodeai.com/claims/PUB-2026-004/","source_type":"publication","source_name":"MIT Sloan Management Review","source_url":"https://mitsloan.mit.edu/ideas-made-to-matter/action-items-ai-decision-makers-2026","source_snapshot_url":"https://web.archive.org/web/20260314200046/https://mitsloan.mit.edu/ideas-made-to-matter/action-items-ai-decision-makers-2026","source_date":"2026-03-01","log_date":"2026-04-19","review_cadence":180,"domain_tags":["governance","adoption-rate"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"PUB-2026-005","claim":"Harvard Business Review Analytic Services research finds that only 6% of companies fully trust AI agents to handle core business processes. 43% trust them only with limited or routine operational tasks; 39% restrict to supervised use cases or non-core processes.","claim_url":"https://agentmodeai.com/claims/PUB-2026-005/","source_type":"publication","source_name":"Harvard Business Review","source_url":"https://fortune.com/2025/12/09/harvard-business-review-survey-only-6-percent-companies-trust-ai-agents/","source_snapshot_url":"https://web.archive.org/web/20260220075113/https://fortune.com/2025/12/09/harvard-business-review-survey-only-6-percent-companies-trust-ai-agents/","source_date":"2025-12-09","log_date":"2026-04-19","review_cadence":180,"domain_tags":["governance","adoption-rate"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"REG-2026-001","claim":"NIST released a concept note on 7 Apr 2026 for an AI RMF Profile on Trustworthy AI in Critical Infrastructure, signalling binding-adjacent guidance for critical-infrastructure operators.","claim_url":"https://agentmodeai.com/claims/REG-2026-001/","source_type":"regulatory","source_name":"NIST","source_url":"https://www.nist.gov/itl/ai-risk-management-framework","source_snapshot_url":"https://web.archive.org/web/20260419215544/https://www.nist.gov/itl/ai-risk-management-framework","source_date":"2026-04-07","log_date":"2026-04-19","review_cadence":180,"domain_tags":["governance","regulatory","compliance"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"REG-2026-002","claim":"The European Commission published guidelines on the classification of high-risk AI systems in February 2026, establishing the interpretive framework for the 2 Aug 2026 Annex III enforcement deadline.","claim_url":"https://agentmodeai.com/claims/REG-2026-002/","source_type":"regulatory","source_name":"European Commission","source_url":"https://ai-act-service-desk.ec.europa.eu/en/faq","source_snapshot_url":"https://web.archive.org/web/20260419220239/https://ai-act-service-desk.ec.europa.eu/en/faq","source_date":"2026-02-15","log_date":"2026-04-19","review_cadence":90,"domain_tags":["regulatory","compliance","governance"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[]},{"id":"REG-2026-003","claim":"On 2 August 2026, Annex III high-risk AI system obligations under the EU AI Act (Articles 9–49) take full effect, with penalties up to €35M or 7% of global annual turnover for serious violations and €15M or 3% for high-risk non-compliance.","claim_url":"https://agentmodeai.com/claims/REG-2026-003/","source_type":"regulatory","source_name":"European Commission","source_url":"https://artificialintelligenceact.eu/article/6/","source_snapshot_url":"https://web.archive.org/web/20260419220323/https://artificialintelligenceact.eu/article/6/","source_date":"2026-02-01","log_date":"2026-04-19","review_cadence":180,"domain_tags":["regulatory","compliance","governance"],"current_verdict":"pending-review","next_review":"2026-10-16","review_history":[],"change_history":[]},{"id":"VEN-2026-001","claim":"Anthropic stated that Claude for Chrome reduced a financial services firm's accounts payable processing from 6 hours to under 30 minutes.","claim_url":"https://agentmodeai.com/claims/VEN-2026-001/","source_type":"vendor","source_name":"Anthropic","source_url":"https://claude.com/blog/claude-for-chrome","source_snapshot_url":"https://web.archive.org/web/20260419155325/https://claude.com/blog/claude-for-chrome","source_date":"2025-08-26","log_date":"2026-04-19","review_cadence":90,"domain_tags":["agent-roi","deployment-velocity"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[{"date":"2026-04-19","field_changed":"source_snapshot_url","old_value":"https://web.archive.org/web/20250826120000/https://www.anthropic.com/news/claude-for-chrome","new_value":"https://web.archive.org/web/20260419155325/https://claude.com/blog/claude-for-chrome","reason":"Initial snapshot URL was a Phase 1 placeholder with an estimated timestamp; resolved via Save Page Now to the actual archived capture. Canonical destination redirected from anthropic.com/news to claude.com/blog during the intervening period — both URLs serve the same content."},{"date":"2026-04-19","field_changed":"source_url","old_value":"https://www.anthropic.com/news/claude-for-chrome","new_value":"https://claude.com/blog/claude-for-chrome","reason":"Original URL returns HTTP 308 permanent redirect to claude.com/blog/claude-for-chrome (verified 19 Apr 2026). Updated source_url to the current canonical so future readers reach live content directly; the old URL still works via redirect and subsequent reviews can compare against both."}]},{"id":"VEN-2026-002","claim":"Google expanded AI Mode restaurant booking to 8 markets (UK, Australia, Canada, Hong Kong, India, New Zealand, Singapore, South Africa) with 8 partner platforms powering the backend.","claim_url":"https://agentmodeai.com/claims/VEN-2026-002/","source_type":"vendor","source_name":"Google","source_url":"https://blog.google/company-news/inside-google/around-the-globe/google-europe/united-kingdom/ai-mode-restaurants-uk/","source_snapshot_url":"https://web.archive.org/web/20260419221317/https://blog.google/company-news/inside-google/around-the-globe/google-europe/united-kingdom/ai-mode-restaurants-uk/","source_date":"2026-04-10","log_date":"2026-04-19","review_cadence":60,"domain_tags":["agent-procurement","market-position"],"current_verdict":"pending-review","next_review":"2026-06-18","review_history":[],"change_history":[]},{"id":"VEN-2026-003","claim":"Anthropic's Model Context Protocol crossed 97 million SDK installs by 25 March 2026, the fastest-adopted AI infrastructure standard in history. Anthropic donated MCP to the Agentic AI Foundation under the Linux Foundation, co-founded with Block and OpenAI.","claim_url":"https://agentmodeai.com/claims/VEN-2026-003/","source_type":"vendor","source_name":"Anthropic","source_url":"https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation","source_snapshot_url":"https://web.archive.org/web/20260419215620/https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation","source_date":"2026-03-25","log_date":"2026-04-19","review_cadence":90,"domain_tags":["market-position","agent-procurement","capability-benchmark"],"current_verdict":"pending-review","next_review":"2026-07-18","review_history":[],"change_history":[]},{"id":"VEN-2026-004","claim":"Microsoft's Copilot Wave 3 (9 Mar 2026) introduced Agent 365 as an enterprise agent governance/identity/security control plane, made Claude models available alongside OpenAI models inside Microsoft 365, and launched the M365 E7 licensing tier at $99/user/month.","claim_url":"https://agentmodeai.com/claims/VEN-2026-004/","source_type":"vendor","source_name":"Microsoft","source_url":"https://www.microsoft.com/en-us/microsoft-365/blog/2026/03/09/powering-frontier-transformation-with-copilot-and-agents/","source_snapshot_url":"https://web.archive.org/web/20260318005347/https://www.microsoft.com/en-us/microsoft-365/blog/2026/03/09/powering-frontier-transformation-with-copilot-and-agents/","source_date":"2026-03-09","log_date":"2026-04-19","review_cadence":60,"domain_tags":["market-position","governance","agent-procurement"],"current_verdict":"pending-review","next_review":"2026-06-18","review_history":[],"change_history":[]},{"id":"VEN-2026-005","claim":"Anthropic launched enterprise agent plug-ins for Claude in finance, engineering, and design verticals on 24 Feb 2026, alongside an expanded Accenture partnership that trains approximately 30,000 Accenture professionals on Claude.","claim_url":"https://agentmodeai.com/claims/VEN-2026-005/","source_type":"vendor","source_name":"Anthropic","source_url":"https://techcrunch.com/2026/02/24/anthropic-launches-new-push-for-enterprise-agents-with-plugins-for-finance-engineering-and-design/","source_snapshot_url":"https://web.archive.org/web/20260419220133/https://techcrunch.com/2026/02/24/anthropic-launches-new-push-for-enterprise-agents-with-plugins-for-finance-engineering-and-design/","source_date":"2026-02-24","log_date":"2026-04-19","review_cadence":60,"domain_tags":["market-position","agent-procurement"],"current_verdict":"pending-review","next_review":"2026-06-18","review_history":[],"change_history":[]},{"id":"VEN-2026-006","claim":"Salesforce closed 29,000 Agentforce deals in Q4 FY2026 (up 50% quarter-over-quarter), with Agentforce ARR reaching $800M and 48% quarter-over-quarter growth. CEO Benioff raised FY26 revenue guidance to $41.45-$41.55 billion.","claim_url":"https://agentmodeai.com/claims/VEN-2026-006/","source_type":"vendor","source_name":"Salesforce","source_url":"https://www.salesforceben.com/huge-agentforce-growth-in-salesforce-q4-as-benioff-mocks-saaspocalypse-narratives/","source_snapshot_url":"https://web.archive.org/web/20260307222535/https://www.salesforceben.com/huge-agentforce-growth-in-salesforce-q4-as-benioff-mocks-saaspocalypse-narratives/","source_date":"2026-02-27","log_date":"2026-04-19","review_cadence":30,"domain_tags":["agent-roi","market-position","adoption-rate"],"current_verdict":"holding","next_review":"2026-06-29","review_history":[{"date":"2026-04-24","verdict":"holding","oversight":"peter-led-deep-review","memo_url":"https://agentmodeai.com/claims/VEN-2026-006/review-2026-04-24/","counter_evidence_considered":null},{"date":"2026-05-30","verdict":"holding","oversight":"peter-led-deep-review","memo_url":"https://agentmodeai.com/claims/VEN-2026-006/review-2026-05-30/","counter_evidence_considered":null}],"change_history":[{"date":"2026-04-24","field_changed":"status","old_value":"pending-review","new_value":"holding","reason":"First scheduled review (cadence 30d); all three sub-claims re-verified against Salesforce IR Q4 FY26 release and corporate press release. Memo at review-2026-04-24."},{"date":"2026-04-24","field_changed":"review_history","old_value":"[]","new_value":"[{date: 2026-04-24, verdict: holding, oversight: peter-led-deep-review, memo_slug: review-2026-04-24}]","reason":"Logged first review entry for this claim. See memo at content/claims/VEN-2026-006/review-2026-04-24.mdx for full evidence chain."},{"date":"2026-05-30","field_changed":"review_history","old_value":"[1 entry: 2026-04-24 holding]","new_value":"[2 entries: + 2026-05-30 holding]","reason":"Second scheduled review (cadence 30d). All three Q4 FY26 sub-claims reconfirmed unrestated; Q1 FY27 (27 May 2026) reported Agentforce ARR $1.2B (+205% YoY), accelerating from the $800M base — logged as reinforcing context. Verdict Holding: claim is a point-in-time Q4 FY26 fact, accurate and unchanged. Memo at content/claims/VEN-2026-006/review-2026-05-30.mdx."}]},{"id":"VEN-2026-007","claim":"Microsoft reported 15 million paid Microsoft 365 Copilot seats in FY2026 Q2 (28 Jan 2026). The number of customers with more than 35,000 Copilot seats tripled year-over-year. Agent 365 was introduced as an enterprise governance/identity/security control plane for agents.","claim_url":"https://agentmodeai.com/claims/VEN-2026-007/","source_type":"vendor","source_name":"Microsoft","source_url":"https://www.microsoft.com/en-us/investor/earnings/fy-2026-q2/press-release-webcast","source_snapshot_url":"https://web.archive.org/web/20260419220513/https://www.microsoft.com/en-us/investor/earnings/fy-2026-q2/press-release-webcast","source_date":"2026-01-28","log_date":"2026-04-19","review_cadence":30,"domain_tags":["adoption-rate","market-position"],"current_verdict":"holding","next_review":"2026-06-29","review_history":[{"date":"2026-04-24","verdict":"holding","oversight":"peter-led-deep-review","memo_url":"https://agentmodeai.com/claims/VEN-2026-007/review-2026-04-24/","counter_evidence_considered":null},{"date":"2026-05-30","verdict":"holding","oversight":"peter-led-deep-review","memo_url":"https://agentmodeai.com/claims/VEN-2026-007/review-2026-05-30/","counter_evidence_considered":null}],"change_history":[{"date":"2026-04-24","field_changed":"status","old_value":"pending-review","new_value":"holding","reason":"First scheduled review (cadence 30d); 15M paid seats, tripled >35k-seat customer cohort (with named accounts), and Agent 365 introduction all re-verified against Microsoft IR Q2 FY26 release and earnings call transcript. Memo at review-2026-04-24."},{"date":"2026-04-24","field_changed":"review_history","old_value":"[]","new_value":"[{date: 2026-04-24, verdict: holding, oversight: peter-led-deep-review, memo_slug: review-2026-04-24}]","reason":"Logged first review entry for this claim. See memo at content/claims/VEN-2026-007/review-2026-04-24.mdx for full evidence chain."},{"date":"2026-05-30","field_changed":"review_history","old_value":"[1 entry: 2026-04-24 holding]","new_value":"[2 entries: + 2026-05-30 holding]","reason":"Second scheduled review (cadence 30d); first to weigh Q3 FY26 (reported 29 Apr 2026, five days after the prior review). All three Q2 FY26 sub-claims reconfirmed unrestated; Q3 FY26 reported 20M+ paid Copilot seats (up from 15M) and a quadrupled >50k-seat cohort — logged as reinforcing context, with a note that the large-customer metric migrated from >35k to >50k. Verdict Holding: claim is a point-in-time Q2 FY26 fact, accurate and unchanged. Memo at content/claims/VEN-2026-007/review-2026-05-30.mdx."}]},{"id":"VEN-2026-008","claim":"Google Cloud launched Gemini Enterprise for Customer Experience at NRF 2026 (11 Jan 2026), with Kroger, Lowe's, and Woolworths named as adopting customers using agentic capabilities for combined shopping + customer service workflows.","claim_url":"https://agentmodeai.com/claims/VEN-2026-008/","source_type":"vendor","source_name":"Google","source_url":"https://www.googlecloudpresscorner.com/2026-01-11-Google-Cloud-Brings-Shopping-and-Customer-Service-Together-with-Gemini-Enterprise-for-Customer-Experience","source_snapshot_url":"https://web.archive.org/web/20260419220604/https://www.googlecloudpresscorner.com/2026-01-11-Google-Cloud-Brings-Shopping-and-Customer-Service-Together-with-Gemini-Enterprise-for-Customer-Experience","source_date":"2026-01-11","log_date":"2026-04-19","review_cadence":60,"domain_tags":["adoption-rate","agent-procurement","market-position"],"current_verdict":"pending-review","next_review":"2026-06-18","review_history":[],"change_history":[]}]},"claims":[{"id":"AM-001","claim":"70% of AI-implementation failure is people and process, not technology — cultural transformation is the strongest predictor of AI ROI at the 2024-2025 maturity stage.","article_url":"https://agentmodeai.com/ai-readiness-in-organizations-the-2024-2025-landscape/","topic":"agentic-ai-governance","pub_date":"2026-04-19","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-002","claim":"Agentic AI's $3.50-per-dollar average return masks a 70% task-failure rate on the Carnegie Mellon benchmark; only narrowly-scoped deployments clear the reality bar.","article_url":"https://agentmodeai.com/the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025/","topic":"agent-procurement","pub_date":"2026-04-19","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug language ('revolution', 'real-world success stories') carries hype register the publication explicitly avoids; survey-of-surveys structure does not stand up to source-verification at the level the publication now demands. Google has rejected the URL despite an active claim status. URL now redirects to /retractions/?retired=the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025. Claim withdrawn — status moves to Not holding, no further reviews scheduled."},{"date":"2026-05-06","verdict":"partial","note":"URL state changed. The /the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025/ slug now serves a deliberately rewritten retrospective (claimId AM-130, \"Agentic AI 2024-2025 retrospective\", published 04 May 2026) against audited primary sources. The 28 Apr 2026 redirect to /retractions/ has been lifted to allow that. AM-002 the claim remains Not holding — the original $3.50/dollar + 70% failure-rate framing was withdrawn and is not restored. AM-130 is a separate claim with its own evidence chain. Readers arriving at /holding/AM-002 see the withdrawal here; the article link surfaces the new piece at the URL the original lived at, with this entry as the audit trail."}],"primary_sources":[]},{"id":"AM-003","claim":"GPT-5 Pro's tiered-subscription model forces enterprises to classify problems by computational difficulty — $200/month premium routing only repays for the top decile of 'very hard' queries.","article_url":"https://agentmodeai.com/gpt-5-pro-vs-enterprise-ai-agents-what-very-hard-problems-means-for-your-business/","topic":"enterprise-ai-cost","pub_date":"2026-04-19","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"partial","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-05-28","verdict":"partial","note":"Pricing/model drift: a $100/mo Pro tier now sits beside the $200 tier (added 9 Apr 2026) and the premium model is GPT-5.5 Pro. Core thesis holds; the single-$200-tier framing no longer matches. Re-verify current tiers at chatgpt.com/pricing."}],"primary_sources":[]},{"id":"AM-004","claim":"The agentic AI discovery phase upstream of procurement is not a vendor-evaluation sprint to a go-decision; it is an organisational-readiness test where the deciding question is whether the procuring enterprise can clear four upstream tests (definitional clarity across the senior team, a named operational candidate workflow with measured baseline and named owner, threat-model literacy on the cross-agent and browser-resident classes, and workforce-readiness against the BCG access gap) before any vendor conversation. Gartner's January 2025 poll of 3,412 executives (19% significant, 42% conservative, 31% wait-and-see, 8% no investment) describes the phase distribution; the 39% in 'wait-and-see' or 'no investment' postures are not failing discovery but correctly identifying that the upstream tests are not yet cleared.","article_url":"https://agentmodeai.com/navigating-the-discovery-phase-how-organizations-first-explore-agentic-ai/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-005","claim":"The AI assistant vs AI agent distinction is operationally meaningful for enterprise procurement: assistants are reactive, request-driven, human-in-the-loop systems whose deployment and ROI patterns are documented at named-customer scale (McKinsey's Lilli platform with 72% employee adoption, 500,000+ prompts processed monthly, ~30% time savings on knowledge work, six-month deployment from proof-of-concept to full rollout); agents are proactive, goal-directed, autonomous-action systems whose deployment patterns are still emerging and whose cohort-scale failure rate is documented (Gartner June 2025: 40%+ of agentic AI projects cancelled by end-2027). Assistants and agents are different procurement decisions rather than points on a continuum; an assistants-first enterprise roadmap is defensible on the documented named-success cohort, an agents-first roadmap is defensible only when the AM-004 discovery-phase tests are cleared and the AM-140 procurement-committee questions are answered.","article_url":"https://agentmodeai.com/ai-assistant-vs-ai-agent-understanding-the-key-differences-for-enterprise-implementation/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-006","claim":"The 56% AI-skill wage premium reported by the Federal Reserve Bank of Atlanta (May 2025, drawing on Lightcast job-posting data through 2024) describes a real labour-market signal at scale, but materially overstates what the typical mid-career worker should expect from a generic AI-literacy program: the premium attaches to specific technical skills surfacing in 1.62% of all 2024 job postings, and the BCG 14%-vs-44% gap in AI upskilling access between frontline workers and leaders is the operational variable that decides which cohort captures the premium and which sees credential inflation without the wage signal.","article_url":"https://agentmodeai.com/the-56-solution-how-workers-are-turning-ai-anxiety-into-career-gold/","topic":"enterprise-ai-cost","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-007","claim":"The AgentFlayer class of zero-click cross-agent prompt-injection attacks (Zenity Labs disclosure at Black Hat USA 2025) and the EchoLeak CVE-2025-32711 disclosure earlier the same year describe a structural failure mode in agentic AI rather than incidental bugs; the procurement-relevant signal is the vendor-response split — which vendors patched and committed to a response-SLA versus which classified the behaviour as 'intended functionality' — answered before the contract closes, not after.","article_url":"https://agentmodeai.com/agentflayer-attack-why-chatgpt-copilot-6-major-ai-platforms-are-being-hacked-right-now/","topic":"agentic-ai-governance","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-008","claim":"AI infrastructure water consumption has moved from sustainability-footnote to procurement-deck variable: Google reported 8.1 billion gallons of data-centre water consumption in 2024 (a 33% year-over-year increase from 6.1 billion in 2023), Microsoft reported 6.4 million cubic metres in 2022 at a Water Usage Effectiveness of 0.30 litres per kilowatt-hour (a 39% improvement from 0.49 in 2021), and the EU Energy Efficiency Directive 2023/1791 made WUE and water-consumption reporting mandatory for data centres above 500 kilowatts of IT power demand starting 15 September 2024. Closed-loop and immersion cooling technologies (Microsoft's zero-water evaporation systems standardised for new builds August 2024; immersion cooling at sub-1.1 PUE) have matured enough that the procurement question for cloud and co-location vendors in 2026 is the vendor's water-efficiency posture in writing, not whether water consumption is a procurement-relevant variable.","article_url":"https://agentmodeai.com/ai-environmental-impact-the-hidden-water-crisis-threatening-digital-transformation/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-06-17","next_review":"2026-08-16","verdict":"partial","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-17","verdict":"partial","note":"Source-text figure re-review: Google's 2024 Environmental Report reports a 28% year-over-year increase to 8.1 billion gallons, not the 33% (from a 6.1 billion 2023 base) asserted at publish. The 8.1B 2024 figure and the Microsoft WUE 0.30 L/kWh / 39%-improvement figure are unchanged and verified. Article corrected to 28% and the unsupported 6.1B base removed; the claim text retains the original figure with this correction per the Holding-up protocol."}],"primary_sources":[]},{"id":"AM-009","claim":"Anthropic's Claude for Chrome launch (26 Aug 2025, 1,000 Max-plan subscribers at $100-200/month) is a procurement-decision data point about the maturity of the browser-resident agentic AI class rather than about Anthropic specifically; the company's own security disclosure (23.6% prompt-injection success rate pre-mitigation, 11.2% post-mitigation, 0% on URL-injection variants after subsequent patches) describes the structural exposure level the deploying enterprise inherits across the class, including from Anthropic's competitors as they ship parallel browser-resident products. The procurement-relevant signal is the published-disclosure posture (Anthropic disclosed the rates honestly with mitigation deltas), which places Anthropic in the AM-007 Cohort A and gives procurement a verifiable vendor-response baseline; the rate itself bounds the deployment-layer compensating-control burden but does not, on its own, decide the procurement question.","article_url":"https://agentmodeai.com/anthropics-claude-for-chrome-changes-everything-what-business-leaders-need-to-know-now/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-010","claim":"Enterprise agentic AI deployments that reach the McKinsey 23% scaling cohort share five operational characteristics drawn from the cited JPMorgan, Toshiba, Wipro, and Aberdeen City Council deployments: measured pre-deployment baselines, named deployment-owner accountability, scoped-experimentation governance, training-over-hiring resource posture, and CIO-level visibility on per-deployment ROI; the characteristics are observational and replace the prior draft's 'ADAPT' acronym framing, which had no published source.","article_url":"https://agentmodeai.com/the-cios-playbook-orchestrating-human-ai-teams-that-actually-want-to-work-together/","topic":"enterprise-ai-cost","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-011","claim":"The IBM Watson Health collapse (2015 launch through 2022 sale to Francisco Partners at approximately one-fifth of the initial investment) is the canonical enterprise AI failure case where the underlying technology was substantively functional and the organisational integration was not — physician rejection at named partner sites, workflow misalignment with clinical practice, and professional-identity-threat dynamics drove abandonment despite the underlying capability; the pattern reproduces at the cohort scale RAND Corporation's 2024 study (n=65 senior data scientists) identifies at the 80% AI-project failure rate, with organisational resistance dominant over technical limitation as the failure cause. The procurement-deck implication is that the change-management variable belongs in the discovery phase (AM-004) and the procurement decision (AM-140), not as a post-deployment afterthought.","article_url":"https://agentmodeai.com/the-human-agent-partnership-why-67-of-ai-projects-fail-without-cultural-change/","topic":"agentic-ai-governance","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-012","claim":"The enterprise IT operations workforce is structurally the highest-exposure population to autonomous-action AI: the task surface (incident triage, configuration management, ticket processing, routine diagnostics, scripted remediation) maps onto the agent-class capability boundary more directly than any other large enterprise job-family, and public-sector workforce data (US Bureau of Labor Statistics Computer and Information Technology Occupations Outlook; World Economic Forum Future of Jobs Report 2025) places IT-ops roles at the top of both the displacement and the role-transformation lists. The procurement-deck question for the CIO is not whether the IT-ops role mix changes but on what timeline against which named roles, and whether the workforce-transition posture is agent-orchestration (training the team toward managing fleets of agents) or agent-replacement (letting workforce churn through to a smaller team operating the deployed agents).","article_url":"https://agentmodeai.com/ai-workforce-transformation-the-human-guide-to-building-your-autonomous-it-future/","topic":"agentic-ai-governance","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-013","claim":"Q1 2026 is the quarter enterprise agentic-AI crossed three thresholds simultaneously — the first at-scale in-the-wild exploits, the first vendor-shipped governance infrastructure, and the first hard ROI data — and programmes designed around only one will not make the 28% that pay off.","article_url":"https://agentmodeai.com/agentic-ai-got-real-q1-2026/","topic":"agentic-ai-governance","pub_date":"2026-04-19","last_reviewed":"2026-06-04","next_review":"2026-08-03","verdict":"holding","verdict_history":[{"date":"2026-04-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-014","claim":"The ~73% of enterprise agentic-AI projects that fail share three structural gaps — no named owner, scope drift, and missing agent-level MTTD — and the 27% that succeed cluster around the inverse.","article_url":"https://agentmodeai.com/why-73-of-agentic-ai-projects-fail-and-how-the-27-generate-312-roi/","topic":"enterprise-ai-cost","pub_date":"2025-08-03","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-08-03","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Both stats in the slug ('73% fail', '312% ROI') were backfilled with status: partial on 19 Apr 2026 noting the article predates editorial standard. Body never rewritten. Google's quality algorithm has independently flagged the URL nine days later. The slug carries the structural problem. URL now redirects to /retractions/?retired=why-73-of-agentic-ai-projects-fail-and-how-the-27-generate-312-roi. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-015","claim":"An agentic-AI Center of Excellence justifies its overhead only after the organisation has three production agents running; before that, it over-governs an experimental footprint.","article_url":"https://agentmodeai.com/building-a-center-of-excellence-for-agentic-ai-in-it-operations-complete-enterprise-guide/","topic":"agentic-ai-governance","pub_date":"2025-08-01","last_reviewed":"2026-06-04","next_review":"2026-08-03","verdict":"partial","verdict_history":[{"date":"2025-08-01","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."}],"primary_sources":[]},{"id":"AM-016","claim":"Agent-mediated network management reduces unplanned firewall-change incident costs only when the agent's action log feeds into the same change-management audit trail human changes use — not as a parallel system.","article_url":"https://agentmodeai.com/the-7-2m-firewall-change-that-transformed-network-management-how-agentic-ai-prevents-it-disasters/","topic":null,"pub_date":"2025-07-27","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-07-27","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. '$7.2M' figure in the slug cannot be traced to any disclosed firewall-change incident. Body never rewritten past 19 Apr 2026 backfill. The dollar specificity in the URL is the structural problem and Google's quality algorithm has independently flagged the URL. URL now redirects to /retractions/?retired=the-7-2m-firewall-change-that-transformed-network-management-how-agentic-ai-prevents-it-disasters. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-017","claim":"Agentic AI's durable enterprise pattern is redeployment-first, not replacement-first. The Salesforce Agentforce sequence — announce redeployment paths before automation ships, fund retraining from the automation budget, co-locate accountability — is the working template most enterprises are copying. Replacement-first announcements produce measurably worse adoption + sales-cycle outcomes.","article_url":"https://agentmodeai.com/the-day-9000-people-asked-to-be-replaced/","topic":null,"pub_date":"2025-07-19","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-07-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-19","verdict":"partial","note":"Anchor verification complete (see audit/ANCHOR_VERIFICATION_2026-04-19.md). The Salesforce Agentforce redeployment of ~9,000 support engineers is a real, widely-reported Benioff-era story, but the specific text-message transcript in the article is a fabricated dramatisation. Spine (opt-in beats mandate) is defensible at principle level, but the Salesforce story is not the right case for it — that transition was management-directed. Rewrite flagged for before 18 Jun 2026 review."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten. Fabricated text-message transcript removed. Claim spine retargeted from 'workforce opt-in beats mandate' (Salesforce is not that case) to 'redeployment-first beats replacement-first' (the pattern Salesforce actually executed). Status moves from Partial to Up. Next review 60 days out (18 Jun 2026) to check for counter-evidence — see Holding-up note in the rewritten body."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug premise ('asked to be replaced') is the dramatized framing the body had to remove on 19 Apr 2026. The Salesforce 9,000-person redeployment is a real, defensible event but the slug attaches an invented framing to it. Body preserved in archived/. Google has independently rejected the URL. URL now redirects to /retractions/?retired=the-day-9000-people-asked-to-be-replaced. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-018","claim":"Agentic AI's compounding economics show up in back-office operations (AP, IT ticket triage, HR onboarding, procurement, close-cycle reconciliation), not in front-office customer-facing workflows. The 12% of deployments that clear 300%+ ROI cluster there for structural reasons: per-action savings × action frequency × task-specification tightness × existing process instrumentation.","article_url":"https://agentmodeai.com/the-executive-who-discovered-her-competitors-secret-weapon/","topic":null,"pub_date":"2025-07-19","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-07-19","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Article predates the Holding-up standard. Retroactive claim assigned on 19 Apr 2026. Initial verdict 'Partial' — spine is defensible, per-claim numeric verification deferred to +60d review. Body not rewritten per AGENTMODE_PHASE2_BRIEF §114."},{"date":"2026-04-19","verdict":"partial","note":"Anchor verification complete (see audit/ANCHOR_VERIFICATION_2026-04-19.md). 'Sarah Chen' and the 2 AM Munich-hotel scenario are fully fabricated — the article's narrative protagonist does not correspond to any real executive. The underlying framework (back-office cost compounding faster than front-office wins; per-action delta × frequency) IS defensible against McKinsey + Futurum operational-AI-ROI data. Rewrite required before the article can move to Holding."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten. Fabricated 'Sarah Chen' narrative frame removed entirely. Claim spine sharpened: original was 'back-office cost compounding faster than front-office'; new version adds the structural explanation (per-action × frequency × task-specification × measurement instrumentation) and specific 2026 benchmark anchors (Stanford DEL 12%/88%, Gartner 28%, Futurum 71% vs 40%). Status moves from Partial to Up. Cross-links to AM-020 (TCO), AM-021 (measurement discipline), AM-022 (bimodal ROI) explicitly drawn in the body. Next review 18 Jun 2026."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug structure (fictional protagonist, 'discovered her competitors' secret weapon') is the fabricated narrative frame the body had to remove on 19 Apr 2026. Body rewritten with Stanford DEL / McKinsey / Futurum sourcing (preserved in archived/) but the slug is the structural problem. URL now redirects to /retractions/?retired=the-executive-who-discovered-her-competitors-secret-weapon. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-019","claim":"Manufacturing deployments hitting the 30% unplanned-downtime-reduction benchmark share one architectural pattern — the agent writes its actions into the plant's existing MES/CMMS audit trail rather than a parallel log. Parallel-log deployments underperform by a factor of 2-3.","article_url":"https://agentmodeai.com/manufacturing-4-0-how-multi-agent-systems-reduce-downtime-by-30/","topic":"agentic-ai-governance","pub_date":"2025-08-01","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2025-08-01","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten. Original headline number (30% downtime reduction) survives against current case-study data. New analytical spine: the audit-trail architecture separates wins from stalls. Status moved from rewrite-in-progress Partial placeholder to Up. Next review 60 days out because architectural claims age slower than pricing claims."}],"primary_sources":[]},{"id":"AM-020","claim":"The 40-60% TCO underestimate on enterprise agentic-AI deployments is not a cost-visibility failure — it is a cross-departmental cost-attribution failure. Integration, tokens, maintenance, supervision, and compliance costs land on IT, HR, and Legal budgets that do not reconcile in most organisations, so the CFO sees the bill late and partial.","article_url":"https://agentmodeai.com/the-hidden-costs-of-agentic-ai-a-cfos-guide-to-true-tco-and-roi-modeling/","topic":"enterprise-ai-cost","pub_date":"2025-07-31","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2025-07-31","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop. Status moves from rewrite-in-progress placeholder to Up. New analytical spine: the TCO underestimate is cross-departmental cost-attribution failure, not hidden costs. Five cost categories named with budget owners. 60-day review cadence."}],"primary_sources":[]},{"id":"AM-021","claim":"The 87% vs 27% success-rate gap between Six-Sigma and non-Six-Sigma organisations on agentic-AI deployments reflects pre-existing measurement discipline, not the DMAIC methodology itself. Agents require a clean baseline, defect definition, documented root-cause analysis, and a change-management gate — four conditions that ISO 9001, ITIL, SRE, or HACCP practices produce just as reliably.","article_url":"https://agentmodeai.com/dmaic-for-agentic-ai-deployment/","topic":"agentic-ai-governance","pub_date":"2025-08-16","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2025-08-16","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop. Status moves from rewrite-in-progress placeholder to Up. New thesis: the causation runs the opposite direction from the vendor narrative — the measurement discipline was the prerequisite, the methodology name doesn't matter. 60-day review."},{"date":"2026-04-28","verdict":"partial","note":"Slug migration to §6a-compliant URL: from-dmaic-to-ai-agents-how-traditional-optimization-methods-accelerate-agentic-ai-success → dmaic-for-agentic-ai-deployment. Body unchanged from 19 Apr rewrite, only the URL changed. Old slug 308-redirects to new. Reason: the long descriptive slug carried §6a-grade friction (88+ chars, vendor-cliche framing) and Google's quality algorithm had flagged the original URL as low-quality (per the 28 Apr 2026 GSC drilldown showing it in the 'Crawled - currently not indexed' bucket). The clean slug preserves the analytical content while removing the URL-level quality penalty."},{"date":"2026-06-10","verdict":"partial","note":"Primary-source erosion on the headline statistic. The Gravitex page (gravitexgenesys.com/blog/ai-agents-lean-six-sigma-automating-dmaic) no longer carries the 87%/27% success-rate comparison — checked 10 Jun 2026, the URL now serves Six Sigma course-offering content with no AI-deployment success-rate data. A web search found no independent source corroborating the 87/27 pair. The claim's interpretive reading (the gap reflects pre-existing measurement discipline; ISO 9001/ITIL/SRE/HACCP produce the same four conditions) is unaffected and remains supported by Gartner's 7 Apr 2026 I&O survey (57% of failures cited 'too much too fast'). Status Up → Partial until the 87/27 figure can be re-anchored to a retrievable primary source."}],"primary_sources":[]},{"id":"AM-022","claim":"The 171% average ROI on enterprise agentic-AI deployments is the mean of a bimodal distribution — roughly 12% of deployments clear 300%+ and 88% sit at or below break-even. The single factor distinguishing the clusters is not a multi-pattern framework; it is whether business-line (not IT) ownership held the kill-switch and accountability before the deployment shipped.","article_url":"https://agentmodeai.com/the-agentic-ai-success-formula-7-proven-patterns-driving-171-roi-in-enterprise-deployments/","topic":null,"pub_date":"2025-08-06","last_reviewed":"2026-04-28","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2025-08-06","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop (7-patterns vendor framework with fabricated case studies). New thesis: bimodal distribution, not normal — the 171% average describes no specific deployment. Business-line kill-switch ownership is the single distinguishing factor. Cross-links to AM-020 + AM-021 on the shared organisational-precondition thread."},{"date":"2026-04-28","verdict":"partial","note":"Article retracted 28 Apr 2026. Slug carries '171% ROI' as a category average and a '7 proven patterns' framework that the body had to disown — the rewritten body explicitly argues 171% is the mean of a bimodal distribution, not a benchmark. Body rewritten 19 Apr 2026 (preserved in archived/) but the slug contradicts the rewritten thesis and Google has rejected the URL. URL now redirects to /retractions/?retired=the-agentic-ai-success-formula-7-proven-patterns-driving-171-roi-in-enterprise-deployments. Claim withdrawn — status moves to Not holding, no further reviews scheduled."}],"primary_sources":[]},{"id":"AM-023","claim":"The 10 Apr 2026 Google AI Mode rollout to eight markets is the first vertical (restaurant booking) where agentic search reduces named SaaS aggregators (OpenTable, TheFork, ResDiary and five others) to API backends rather than destinations. The template applies to every enterprise-relevant aggregation vertical — business travel, expense management, procurement, ATS, HR service delivery — and incumbents in those verticals have 18-24 months to pick API-backend or destination positioning before agentic search forces the choice.","article_url":"https://agentmodeai.com/google-ai-mode-restaurant-booking-the-50-billion-business-revolution-every-ceo-must-understand-2025/","topic":"enterprise-ai-cost","pub_date":"2025-08-23","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2025-08-23","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-19","verdict":"partial","note":"Body rewritten from WP-era slop (the '$50 Billion Revolution' headline and 'act within 90 days' crisis-FOMO framing were both fabrications). New thesis: restaurant booking is a template, not the story. Named 5 enterprise-relevant aggregation verticals (business travel, expense, procurement, ATS, HR service) and the API-backend-vs-destination choice incumbents face. Next review in 60 days."}],"primary_sources":[]},{"id":"AM-024","claim":"Enterprise-AI decisions in 2026 are made on a citation chain nobody in the chain verifies. The infrastructure gap CIOs face is a verification layer for the claims their procurement runs on — not an information gap. The 88% failure rate in enterprise agentic AI is the predictable output of decision-making on unverified citations, not a capability problem.","article_url":"https://agentmodeai.com/the-unverified-citation-chain-where-enterprise-ai-decisions-actually-come-from/","topic":"agentic-ai-governance","pub_date":"2026-04-20","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-04-20","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Claim-text figure unanchored — a same-day reversal of the morning verdict, recorded rather than smoothed over. The batch-1 re-review earlier on 10 Jun 2026 returned Holding after confirming the cited sources were live and unrevised (including the Stanford DEL playbook PDF at its URL) and deferred re-anchoring the 88% figure to the next cycle. A batch-2 full-text extraction the same afternoon disproved the figure: the playbook contains no 88% failure rate and no 12/88 ROI distribution (full finding at AM-029, correction of 10 Jun 2026). URL-liveness is not figure-verification; the morning verdict does not survive the afternoon evidence. The claim's spine (enterprise-AI decisions run on citation chains nobody verifies) holds and is itself illustrated by this incident, but its quantitative anchor — 'the 88% failure rate in enterprise agentic AI' — has no verifiable source as a deployment-failure or ROI distribution. The only verified figure carrying the same numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial, same day as the batch-1 Holding verdict."}],"primary_sources":[]},{"id":"AM-025","claim":"Enterprise agentic AI governance in 2026 fails at the operational layer even when it passes at the compliance layer. Boards receive EU-AI-Act-mapped compliance decks while the agentic deployments actually shipping out of IT ops have no measurable overlap with that deck. Durability requires six instrumented dimensions scored 0–100 (GAUGE framework) with a 90-day setup cadence and a 12-month trajectory target — not a compliance matrix.","article_url":"https://agentmodeai.com/the-enterprise-agentic-ai-governance-playbook-2026/","topic":"agentic-ai-governance","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-026","claim":"Generic enterprise SaaS RFPs systematically underweight six agent-specific governance dimensions (governance maturity, threat model, ROI evidence, change management, vendor lock-in, compliance posture). A 60-question RFP layer mapped to the GAUGE framework materially changes vendor selection outcomes by disqualifying vendors whose operational governance will not survive the 18-month enterprise review cycle.","article_url":"https://agentmodeai.com/the-enterprise-agentic-ai-rfp-60-questions/","topic":"agent-procurement","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-027","claim":"A durable enterprise agentic AI business case requires three specific documents — a TCO model with ten named cost categories (not vendor-supplied line items), an ROI model with a pre-deployment measured baseline and an independent validation round, and a three-scenario risk-adjusted NPV. The single-scenario vendor-framed business cases that dominate 2026 enterprise AI investment committees are the predictable root of the 40%+ projected agentic AI project cancellation rate.","article_url":"https://agentmodeai.com/the-cfos-agentic-ai-business-case-tco-and-roi/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-028","claim":"Partner — co-development with a vendor on a structured non-standard engagement — is structurally under-chosen in enterprise agentic AI procurement in 2026. Procurement committees have templates for build and buy but none for partner, so the third path does not get evaluated on an equal footing. The vendor-lock-in and change-management dimensions of the GAUGE framework usually favour partner when it is honestly evaluated, not buy or build.","article_url":"https://agentmodeai.com/build-vs-buy-vs-partner-for-enterprise-agentic-ai-2026/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-029","claim":"The 12/88 bimodal distribution in enterprise agentic AI ROI realisation (Stanford DEL 2026 + cross-validated by Gartner, McKinsey, CMU) is a governance-discipline outcome, not a model-capability outcome. The 12% instrument the six GAUGE dimensions on a 90-day review rhythm; the 88% treat governance as a deliverable to the audit committee. Capability gap (CMU's 30.3% best-in-class task completion) constrains what is possible, not what separates the 12% from the 88%.","article_url":"https://agentmodeai.com/why-88-percent-of-agentic-ai-deployments-fail/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":null,"verdict":"not_holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"not_holding","note":"Primary-source verification failed on the headline figure. The claim attributes a 12/88 bimodal ROI distribution (12% of enterprise agentic AI deployments clearing 300%-plus ROI, 88% at or below break-even at 12-18 months) to the Stanford Digital Economy Lab Enterprise AI Playbook (Apr 2026). The full report text contains no such distribution and no ROI-realisation failure data: it is a study of 51 successful deployments (Pereira, Graylin, Brynjolfsson), in which the only 88% figure is '88% of organizations use AI in at least one function' and the only 12% figure is a sponsor-engagement category. The report's design cannot yield a failure distribution. A web search found no Stanford DEL publication reporting 12/88. Gartner (28% of AI I&O projects fully pay off, Apr 2026) and McKinsey (6% high performers, Nov 2025) document a small-tail pattern but do not corroborate the specific 12/88 split the claim asserts. Status Up -> Not holding. The article remains published with this correction log; the governance-over-capability argument is re-anchored, where it appears elsewhere in the corpus, to figures that survive verification."},{"date":"2026-06-10","verdict":"not_holding","note":"Article restated. The piece at the source URL was rewritten the same day, at the same slug, on the verifiable counterpart figure the fabricated one shadowed: IDC research commissioned by Lenovo (CIO Playbook 2025, Feb 2025; global survey n=2,920) reporting 88% of AI proof-of-concepts failing to reach production, with 4 of every 33 POCs (roughly 12%) graduating, per CIO.com (25 Mar 2025). The restated article asserts a new tracked claim, AM-213, and opens with a correction notice pointing back to this record. This claim stays Not holding as the permanent record of the fabricated attribution; its claim text is unchanged."}],"primary_sources":[]},{"id":"AM-030","claim":"The McKinsey State of AI 2025 figure (23% of enterprises scaling an agentic AI system, 39% still experimenting) is an operational-preconditions outcome, not a technical-readiness outcome. Four preconditions (agent registry, measured pre-deployment baseline, differentiated change-management playbook for adjacent units, cross-agent threat model at scale) separate pilots that cross into production from pilots that stall. The 6% AI-high-performer segment is the subset of the 23% scaling with additional measurement discipline that makes ROI audit-survivable.","article_url":"https://agentmodeai.com/the-mckinsey-23-percent-agentic-ai-scaling-gap/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-031","claim":"The CMU TheAgentCompany 2026 benchmark figure (30.3% task completion for best-in-class frontier model, up from 24% in 2024) is the current capability constraint for enterprise agentic AI. Capability trajectory projects to ~40% by late 2027, which does not cross the 95% production-readiness threshold within the 3-year TCO horizon enterprise business cases operate against. The Stanford DEL 12% durable cohort operates within the 30.3% (narrow scope + human-in-the-loop + GAUGE-dimensional governance discipline), not around it. Capability is not the variable that separates the 12% from the 88%.","article_url":"https://agentmodeai.com/the-cmu-30-percent-agent-capability-gap/","topic":"enterprise-ai-cost","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One leg unanchored on re-review. The CMU capability figures verify cleanly (30.3% full completion for Gemini 2.5 Pro and 39.3% partial-credit on the 175-task TheAgentCompany set per paper v2; 24% for Claude 3.5 Sonnet in the Dec 2024 v1). The Stanford DEL '12% durable cohort' referenced in the claim text does not exist in the cited source: the Enterprise AI Playbook (Pereira, Graylin, Brynjolfsson, Apr 2026) studies 51 successful deployments and contains no 12/88 ROI cohort (full finding at AM-029, correction of 10 Jun 2026). The claim's capability-constraint argument holds on its own evidence; the sentence tying the constraint to the 12%/88% cohort behaviour has no verifiable referent. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-032","claim":"EU financial-services agentic AI deployments operate under a compounded five-framework obligation surface (DORA, NIS2, MiFID II, EU AI Act, GDPR) that sits on top of general AI governance. Liability does not transfer to the vendor contractually regardless of SLA language — MiFID II conduct rules, EU AI Act deployer obligations, and DORA third-party-risk provisions place customer-facing and regulator-facing liability on the deploying financial institution. Compliance-posture and vendor-lock-in are the dominant GAUGE dimensions for the sector, scoring 15-25 points lower than cross-industry averages on first pass.","article_url":"https://agentmodeai.com/agentic-ai-in-financial-services-compliance-and-liability/","topic":"agentic-ai-governance","pub_date":"2026-04-24","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-033","claim":"The McKinsey 17%-EBIT-attributable-to-genAI figure, the most-cited single statistic in 2026 enterprise agentic AI procurement decisions, is a self-reported attribution from McKinsey's State of AI 2025 survey of approximately 1,491 respondents. The way it is typically read in CIO decks, as evidence that 17% of enterprises have produced 5% or more of EBIT from genAI, materially overstates what the survey supports. The figure documents 17% of survey respondents asserting that level of attribution, not 17% of enterprises producing it under audited measurement.","article_url":"https://agentmodeai.com/the-mckinsey-17-percent-ebit-claim/","topic":"enterprise-ai-cost","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-034","claim":"AI assistants and AI agents are not the same product class. An AI assistant is a productivity-augmentation tool that suggests; an AI agent is an automation-execution system that acts on a downstream surface (tools, APIs, write-paths). Conflating them in 2026 enterprise procurement produces the most common single category mistake — buying an assistant under the assumption it is an agent, or buying an agent and governing it as if it were an assistant. The risk profile, contract structure, audit obligation, and TCO model differ categorically.","article_url":"https://agentmodeai.com/ai-assistant-vs-ai-agent/","topic":"agent-procurement","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-035","claim":"The EU AI Act enforcement deadline of 2 August 2026 applies high-risk-system obligations under Articles 9 through 49 to most enterprise agentic AI deployments operating in EU jurisdiction or providing services to EU nationals — not only to deployments explicitly classified within the Annex III high-risk categories. The compliance gap most enterprises face is structural: the Act requires evidence-of-action production (logs, oversight records, post-market monitoring, incident reports) that most agentic deployments do not generate by default. Building the evidence layer post-hoc, after a regulator request, is the failure mode.","article_url":"https://agentmodeai.com/eu-ai-act-agentic-ai-compliance/","topic":"agentic-ai-governance","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"partial","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"The 2 Aug 2026 high-risk-obligations leg is overtaken by the EU Digital Omnibus. Council and Parliament reached a provisional agreement on 6-7 May 2026 (confirmed by the Council 13 May 2026) deferring Annex III stand-alone high-risk-system obligations to 2 Dec 2027 and Annex I embedded high-risk obligations to 2 Aug 2028; formal adoption is expected before 2 Aug 2026. What still lands on 2 Aug 2026 is narrower than the claim asserted: Article 50 transparency obligations (with a watermarking grace period to 2 Dec 2026 for systems already in market) plus the penalties and governance architecture. The claim's structural argument is unchanged and current: agentic deployments still do not generate the evidence-of-action layer (Article 12 logs, Article 14 oversight records, post-market monitoring, incident reporting) by default, and building it post-hoc remains the failure mode; the deferral changes the deadline, not the gap. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-036","claim":"Enterprise shadow AI in 2026 is structurally different from enterprise shadow AI in 2024. The 2024 framing assumed unsanctioned tool adoption — workers pasting confidential data into consumer ChatGPT or installing browser extensions outside IT review. The 2026 reality is that the larger blast radius is agentic capability silently activating inside already-approved tools, often through configuration changes (Custom GPT actions, Copilot custom agents, MCP server connections from approved IDEs) that the original procurement approval did not anticipate. Discovery has to look at capability state, not vendor identity. Most enterprise shadow-AI inventories built against the 2024 framing miss 50 to 80% of the actual exposure surface.","article_url":"https://agentmodeai.com/shadow-ai-discovery-playbook/","topic":"shadow-ai-discovery","pub_date":"2026-04-25","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-25","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-037","claim":"AI agents are structurally different from earlier classes of non-human identity (service accounts, API keys, machine certificates, bot identities), and the IAM platforms most enterprises run in 2026 cannot represent them adequately because those platforms authorise on principal identity rather than on per-action behavioural context. The 92% of enterprises that report low IAM confidence for agentic AI are not configured wrong; they are running an identity model with one structural axis where the agentic deployment requires four (identity, behaviour, context, revocation). The remediation is a four-layer extension on top of existing IAM, not a rip-and-replace migration. Most enterprises can ship the augmentation in 8 to 12 weeks of engineering.","article_url":"https://agentmodeai.com/non-human-identity-ai-agents/","topic":"non-human-identity","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-038","claim":"Model Context Protocol (MCP) reached enterprise procurement gravity in 18 months, faster than typical interoperability standards. The 10,000+ active public MCP servers, adoption by ChatGPT, Cursor, Gemini, Microsoft Copilot, and VS Code, and the December 2025 Linux Foundation donation made MCP a tooling-layer choice that ripples through every adjacent agentic-AI procurement decision: which agents connect to which enterprise systems, which audit boundaries hold, which vendor lock-in patterns activate. The actual procurement decision enterprise IT faces is not whether to adopt MCP (the question is moot once any approved tool ships MCP support); it is the scope-and-governance decision: which MCP servers the enterprise allows agents to connect to, what scopes those connections grant, and how cross-agent delegation through MCP is monitored. Treating MCP as a binary adoption question rather than a scope-and-governance question is the most common enterprise procurement mistake on this surface in 2026.","article_url":"https://agentmodeai.com/mcp-enterprise-agent-tooling/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-039","claim":"The 2026 enterprise agentic AI vendor comparison reduces to four credible platform plays (Anthropic, OpenAI, Google, Microsoft), and the procurement decision between them is no longer primarily about model capability. The model layer has converged to comparable parity for most enterprise use cases. The procurement decision in 2026 is on three other axes: pricing model (Anthropic Managed Agents at 8 cents per session-hour plus tokens versus OpenAI Agents SDK at no first-party runtime fee versus Microsoft and Google's vertically-integrated platform pricing), governance and BAA posture (Anthropic's three-cloud BAA position is structurally distinct), and ecosystem distribution (Microsoft's Office plus Azure footprint has no near peer; Google's vertical integration on Workspace and Cloud is second). Treating this as a model-quality bake-off is the most common 2026 procurement mistake and produces decisions that age badly within the first 12 months.","article_url":"https://agentmodeai.com/enterprise-ai-agent-vendor-comparison/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Extracted-text verification failed on the governance axis of the claim. 'Anthropic's three-cloud BAA position is structurally distinct' overstates Anthropic's own coverage: per Anthropic's BAA documentation (privacy.claude.com, retrieved 10 Jun 2026), Anthropic signs BAAs for the first-party API and HIPAA-ready Claude Enterprise only; the page contains no Bedrock, Vertex, or Azure coverage. Claude consumed via AWS Bedrock or Google Vertex AI is covered by the hyperscaler's BAA, and an Azure-side Anthropic BAA could not be verified. The article's stronger formulation ('Anthropic operates under BAAs with Amazon Web Services, Google Cloud, and Microsoft Azure simultaneously', cited only to a secondary Ampcome blog) is the same overstatement. The substantive point survives restated: BAA-covered Claude deployment surfaces span more clouds than competitors offer, but the BAAs are not Anthropic's across three clouds. This matches the AM-053 correction of the same day. The pricing axis ($0.08 per session-hour plus tokens; Agents SDK no first-party runtime fee) and the ecosystem axis verify and stand. Status Up -> Partial. Article body needs a Peter-approved BAA restate (FAQ x2, body x3, howTo step 1)."}],"primary_sources":[]},{"id":"AM-040","claim":"Enterprise agentic AI in 2026 is in its first year of operational consequence rather than its first year of capability. The deployment record across multiple independent datasets shows a stable bimodal distribution (a small high-performing tail clearing 300%+ ROI and a much larger struggling body at or below break-even), four credible platform plays converging at the vendor layer, a structurally inadequate IAM posture across 92% of enterprises, and a 14-week runway to the EU AI Act August 2026 enforcement window. The aggregate signal is that the year's defining variable is deployment discipline, not model capability or vendor selection. The 6% AI-high-performer segment and the 12% Stanford DEL high-ROI cohort instrument six specific governance dimensions on a 90-day review cadence; the remaining 88-94% mostly do not.","article_url":"https://agentmodeai.com/state-of-enterprise-agentic-ai/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One component unanchored on re-review. The claim text cites 'the 12% Stanford DEL high-ROI cohort' against 'the remaining 88-94%'. Full-text verification on 10 Jun 2026 found the Stanford DEL Enterprise AI Playbook contains no 12% high-ROI cohort and no ROI distribution of any kind — it studies 51 successful deployments by design (full finding at AM-029, correction of 10 Jun 2026). The McKinsey 6% AI-high-performer figure (Nov 2025, n=1,993) verifies independently, as do the IAM-posture and EU AI Act runway components. The 'stable bimodal distribution' framing is supported only as a small-tail/large-body shape (Gartner: 28% of AI I&O projects fully paying off; McKinsey: 6% high performers), not as the two-cluster distribution the claim names. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-041","claim":"The 2026 enterprise agentic AI procurement playbook resolves to a six-stage sequence that integrates the build-vs-buy-vs-partner decision, the 60-question agentic AI RFP, the GAUGE governance scoring, the four-vendor comparison, and the EU AI Act compliance scaffolding into one operational track. Most enterprises in 2026 run these as separate work streams owned by separate functions, which produces structurally inconsistent procurement records and substantial duplicate effort. The integrated six-stage track ships in 8 to 10 weeks for standard environments and produces an audit-defensible per-deployment procurement artifact that satisfies the EU AI Act Article 9 risk-management system requirement by construction.","article_url":"https://agentmodeai.com/enterprise-agentic-ai-procurement-playbook/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-042","claim":"The 6% AI-high-performer cohort identified by McKinsey and the 12% high-ROI cohort identified by the Stanford Digital Economy Lab share ten measurable governance practices that an enterprise can audit in under 60 minutes. An enterprise answering YES to 8 or more of the 10 diagnostic questions has the operating profile of the high-performing segment. An enterprise answering YES to 4 or fewer has the operating profile of the 88-94% struggling cohort and is unlikely to clear break-even on agentic AI deployment without a posture rebuild. The diagnostic audits posture, not outcomes; it identifies where governance investment is needed before the next deployment commitment, not whether a specific deployment will succeed.","article_url":"https://agentmodeai.com/agentic-ai-readiness-diagnostic/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of two cohort anchors unanchored on re-review. The claim text pairs McKinsey's 6% AI-high-performer cohort with 'the 12% high-ROI cohort identified by the Stanford Digital Economy Lab' and an '88-94% struggling cohort'. Full-text verification on 10 Jun 2026 found the Stanford DEL Enterprise AI Playbook identifies no 12% high-ROI cohort and no struggling-cohort percentage — it studies 51 successful deployments by design (full finding at AM-029, correction of 10 Jun 2026). The source article additionally cites a 'Stanford Digital Economy Lab 2026 Enterprise AI Productivity Study', a publication title that does not exist. The McKinsey 6% leg verifies (Nov 2025, n=1,993). The ten diagnostic questions stand as an editorial instrument, but the claim's cohort calibration now rests on one verified dataset, not two. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-043","claim":"The OWASP Agentic Security Initiative's threat taxonomy for agentic AI (memory poisoning, tool misuse, privilege compromise, resource overload, cascading hallucination, intent breaking, misaligned and deceptive behaviour, repudiation and untraceability, identity spoofing, overwhelming human-in-the-loop) maps cleanly onto seven specific enterprise controls: scoped non-human identity, action-class approval gates, decision audit logging at Article 12 evidence quality, MTTD-for-Agents layered detection, deployment-tier resource quotas, behavioural drift monitoring, and HITL throughput limits. An enterprise that operates these seven controls covers all ten OWASP threat classes; an enterprise missing more than two of the controls has structural exposure to at least four of the threat classes.","article_url":"https://agentmodeai.com/owasp-agentic-ai-top-10-walkthrough/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-044","claim":"Six well-documented public agentic AI deployment failures from 2024-2025 (Air Canada bereavement-refund chatbot, NYC MyCity small-business chatbot, Replit production-database wipe, Cursor unauthorised code deletion, Klarna customer-service reversal, DPD chatbot escalation incident) cluster into three structural failure modes: (1) the agent acts as a binding agent of the enterprise without disclosure or approval, (2) the agent operates with permissions the deployment never authorised, (3) the agent's economic case requires a service quality the deployment cannot sustain. Each failure mode maps to a specific control from the seven-control surface; all six failures would have been mitigated by controls already specified in the OWASP Agentic AI Top 10 enterprise walkthrough. The pattern is consistent enough that an enterprise can use the cases as a procurement filter: any vendor unable to point to its specific control posture against each of the three failure modes is not procurement-ready.","article_url":"https://agentmodeai.com/agentic-ai-failure-case-studies/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-045","claim":"EchoLeak (CVE-2025-32711, disclosed by Aim Security in June 2025 against Microsoft 365 Copilot) is the canonical example of a class of attacks rather than a single vulnerability: cross-agent prompt injection in which a malicious payload travels through ordinary content channels (an email, a shared document, a calendar invite, a tool response) into one or more agents' context windows, where it manipulates the agents into actions the deploying enterprise did not authorise, with no user interaction required. The attack class is structurally inherent to any architecture in which an LLM-based agent ingests untrusted content and has tool surfaces capable of exfiltration or action; closing the class requires architectural separation between content-ingest and tool-execution privileges, not point-fixes against specific exploit chains. Enterprises in 2026 operating multiple agents that share context, share memory, or hand off tasks to each other are structurally exposed to the EchoLeak class until the architectural separation is implemented.","article_url":"https://agentmodeai.com/echoleak-cross-agent-prompt-injection/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-046","claim":"EU AI Act Article 12 (record-keeping for high-risk AI systems) and Article 19 (record retention by providers) are operationalised for agentic AI by a 14-field audit-evidence template that captures every agent decision in a regulator-queryable form: deployment ID, agent identity, session ID, ISO timestamp, user prompt, retrieved context with provenance, model output, planned action, action class, approval reference, executed action, tool-call audit chain, output disclosure surface, and policy version. Logs retained for the regulatory minimum (typically 6 months for the EU AI Act baseline, 5 to 7 years for sector-specific overlays like HIPAA and SOX) in a queryable format that supports under-4-business-hour evidence assembly. An enterprise that captures the 14 fields, retains them for the maximum applicable period, and instruments the queryable export has substantially completed Article 12 compliance for the agent layer; the residual work is integrating the agent log stream with the broader audit substrate.","article_url":"https://agentmodeai.com/eu-ai-act-article-12-audit-evidence/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-047","claim":"The Head of AI Governance role (variant titles: Chief AI Officer, VP AI Strategy, Director of Responsible AI) is now a named operating role in 60% of Fortune 100 enterprises per Forrester's 2026 Enterprise AI Predictions, and is the strongest single predictor of an enterprise's score on Q10 of the readiness diagnostic. The role's effective shape converges on six accountabilities: cross-functional governance ownership, EU AI Act compliance posture, vendor procurement gate-keeping, deployment kill-criterion enforcement, audit-evidence substrate ownership, and internal upskilling. The role reports to the executive committee (CEO direct or CFO/COO) rather than to IT, security, or legal, because matrixed reporting into existing functions reproduces the matrixed-shared-accountability failure pattern. Compensation in 2026 ranges from $250-450K base for the Director tier, $400-700K for VP tier, and $600K-$1.2M total comp at the C-level, with significant equity components in growth-stage and tech enterprises.","article_url":"https://agentmodeai.com/head-of-ai-governance-role/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Extracted-text verification failed on two parts of the claim. (1) The claim asserts the role 'is now a named operating role in 60% of Fortune 100 enterprises per Forrester's 2026 Enterprise AI Predictions'. Forrester's actual prediction reads '60% of Fortune 100 companies will appoint a head of AI governance in 2026' (Predictions 2026, quoted by CIO Dive, 16 Dec 2025) — a forecast of appointments during 2026, not a measurement of current adoption. The cited sourceUrl (forrester.com/blogs/the-ai-cio-will-govern-outcomes-at-scale/, 9 Apr 2026) contains neither the 60% figure nor any Fortune-100 reference; the article's 'In Q1 2026, Forrester's Enterprise AI Predictions found 60% ... had hired or were actively recruiting' sentence has no locatable source. (2) The compensation bands ($250-450K Director base, $400-700K VP, $600K-$1.2M C-level total comp) could not be located in any primary source and are unlabelled in the article. The six-accountabilities convergence and the executive-committee reporting-line argument are editorial synthesis and stand on their own. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-048","claim":"The NIST AI Risk Management Framework (AI RMF 1.0, published January 2023, with the Generative AI Profile published July 2024) maps onto enterprise agentic AI deployment work across its four functions (Govern, Map, Measure, Manage) using the same artefacts an enterprise produces for EU AI Act Article 9. Specifically: NIST Govern maps to the Head of AI Governance role and the AI governance committee; NIST Map maps to the deployment inventory and the OWASP Agentic Top 10 walkthrough; NIST Measure maps to the 14-field Article 12 audit substrate plus the GAUGE governance dimensions; NIST Manage maps to the kill-criterion enforcement and the seven-control surface. An enterprise that has the EU AI Act preparation track running has substantially completed NIST AI RMF coverage and can document the mapping as a single cross-reference matrix. The reverse mapping (NIST → EU AI Act) requires more work because NIST is voluntary in posture and the EU AI Act is operational; an enterprise that started with NIST as the framework needs to extend audit substrate granularity and add the Article 73 incident-reporting workflow.","article_url":"https://agentmodeai.com/nist-ai-rmf-agentic-ai-mapping/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-049","claim":"Enterprise multi-agent architectures resolve to three orchestration patterns (hierarchical, peer-to-peer, broker-mediated) with materially different governance properties: hierarchical concentrates accountability at the orchestrator and is the easiest to audit but the most exposed to orchestrator-compromise; peer-to-peer distributes accountability and is the most resilient to single-agent failure but the hardest to audit; broker-mediated centralises the inter-agent communication path and is the most defensible against the cross-agent prompt-injection class. The choice of pattern is not a free architectural decision in 2026 because the EU AI Act's Article 9 risk-management requirements and the OWASP Agentic AI threat surface impose specific control obligations on each pattern. An enterprise should default to broker-mediated for new deployments above the high-risk threshold; hierarchical is acceptable for low-risk and contained deployments; peer-to-peer should be avoided in production agentic AI in 2026 unless the audit substrate is materially stronger than vendor-native baseline.","article_url":"https://agentmodeai.com/multi-agent-architecture-playbook/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-050","claim":"The A2A (Agent2Agent) protocol announced by Google Cloud in April 2025 is the most credible 2026 candidate for an open standard for cross-vendor agent-to-agent interoperability, with backing from 50+ partners across the enterprise software ecosystem (Salesforce, SAP, ServiceNow, MongoDB, Atlassian, and others). The protocol layer covers what MCP (Model Context Protocol) does not: MCP is for agent-to-tool communication, A2A is for agent-to-agent communication. The two protocols are designed to be complementary rather than competing. A2A's adoption trajectory through 2026 will determine whether broker-mediated multi-agent patterns become the cross-vendor default; current trajectory points to deployment-grade stability in the second half of 2026, with widespread enterprise adoption following in 2027. Enterprises selecting agent platforms in 2026 should require A2A roadmap commitments from any vendor whose product will participate in cross-vendor agent workflows.","article_url":"https://agentmodeai.com/a2a-agent-to-agent-protocol/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-051","claim":"Enterprise AI governance organisational design resolves to three operating models in 2026: centralised (a single AI governance function owns policy, procurement, audit, and kill-criterion enforcement enterprise-wide), federated (each business unit owns its AI deployments with cross-unit coordination through a small central function), and hybrid (a central function owns regulatory and procurement; business units own deployment operations and ROI accountability). The dominant 2026 pattern in Fortune 500 enterprises is hybrid, because purely centralised models do not scale past 50-100 deployments and purely federated models cannot satisfy EU AI Act Article 9 risk-management documentation consistency. The right model for a given enterprise depends on three variables: deployment count, regulatory exposure, and the maturity of the existing risk-management organisation. The hybrid model is structurally superior to the alternatives once an enterprise crosses approximately 30 production deployments or operates in two or more EU AI Act high-risk Annex III categories.","article_url":"https://agentmodeai.com/centralized-vs-federated-ai-governance/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-052","claim":"Enterprise agentic AI vendor contracts in 2026 require eight specific exit-clause provisions that standard SaaS contract templates do not adequately cover: (1) full audit-log export with retention, (2) trained-state extraction or destruction guarantee, (3) prompt and configuration portability, (4) tool-and-MCP-connector reconfiguration support during transition, (5) named-individual handoff for in-flight deployments, (6) regulatory-evidence preservation through transition, (7) data-residency continuity, (8) liability-tail coverage for agent actions taken before the transition completes. An enterprise that signs an agentic AI contract without these eight provisions has effectively created a one-way procurement decision; the realistic cost of a forced transition without the provisions is materially higher than the contract value, which inverts the procurement leverage. The provisions add typically modest contract complexity but materially change the enterprise's negotiating posture and the vendor's incentive structure during the relationship.","article_url":"https://agentmodeai.com/ai-agent-contract-exit-clauses/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-053","claim":"HIPAA-compliant agentic AI deployment in U.S. healthcare in 2026 requires four conditions that materially constrain vendor selection and architectural design: (1) the vendor offers a BAA covering the specific agent workflow including any subprocessors and any tools the agent calls, (2) the agent's audit log structure satisfies HIPAA 164.312(b) audit controls AND the EU AI Act Article 12 14-field structure simultaneously, (3) PHI flows through agent tool calls are explicitly mapped and authorised under the HIPAA Privacy Rule's minimum necessary standard, (4) the agent's behavioural drift monitoring includes correctness against clinical-decision benchmarks, not just engagement or business-metric benchmarks. Anthropic's three-cloud BAA position (covering AWS, GCP, and Azure deployment surfaces) is structurally distinct in the 2026 vendor landscape and materially expands healthcare deployment options. The OCR's 340% spike in AI-related discrimination complaints (logged in 2025) makes audit-substrate readiness the highest-priority preparatory work for any healthcare AI deployment going into production in 2026.","article_url":"https://agentmodeai.com/hipaa-compliant-agentic-ai-healthcare/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Extracted-text verification failed on two parts of the claim. (1) The asserted 'OCR's 340% spike in AI-related discrimination complaints (logged in 2025)' cannot be located in any primary source: three targeted searches (10 Jun 2026) across HHS OCR publications, the Section 1557 final-rule coverage, enforcement trackers, and trade press surface no AI-specific complaint-volume series from OCR and no 340% figure anywhere. The article attributes the figure directly to OCR with only the OCR homepage as citation. The figure is unanchored and is treated as failed verification, not as pending. (2) 'Anthropic's three-cloud BAA position' is imprecise: per Anthropic's own BAA documentation, Anthropic signs BAAs for the first-party API and HIPAA-ready Claude Enterprise plans; Claude consumed via AWS Bedrock or Google Vertex AI is covered by the hyperscaler's BAA (AWS Artifact; Google Cloud BAA), not by an Anthropic BAA, and an Azure-side Anthropic BAA could not be verified. The deployment-surface breadth is real; the BAA attribution to Anthropic across three clouds is not. The four deployment conditions (BAA-with-subprocessor coverage, dual 164.312(b)+Article-12 logging, minimum-necessary PHI mapping, clinical-correctness drift monitoring) are editorial architecture and stand. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-054","claim":"Public-sector agentic AI deployment in 2026 operates under five constraints that materially narrow the vendor and architectural options compared to private-sector deployment: (1) FedRAMP authorisation (Moderate or High depending on data sensitivity) is required for federal deployments and increasingly for state, (2) sovereign data residency requirements (data and model inference must remain within national or sub-national boundaries), (3) procurement transparency obligations (the deployment, the vendor, and the decision logic typically must be publicly disclosed), (4) explicit accountability under administrative law (decisions affecting individuals are subject to due-process and appeal frameworks that the agent must support), (5) FOIA-equivalent disclosure of audit logs to the public on request. Public-sector deployments cannot reasonably use peer-to-peer multi-agent patterns and cannot accept vendors without published government cloud SKUs; the realistic 2026 options are Microsoft Azure Government, AWS GovCloud-deployed Anthropic, Google Cloud Public Sector, and a small number of specialist government-AI vendors. The NYC MyCity case (claim AM-044) is the canonical 2026 public-sector failure illustrating what happens when the constraints are inadequately addressed.","article_url":"https://agentmodeai.com/public-sector-agentic-ai/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-055","claim":"Retail and logistics agentic AI deployments in 2026 cluster around five workflow patterns with substantially different governance properties: customer-service agents (the Klarna failure case applies directly, claim AM-044), inventory and demand-forecasting agents (operationally lower-risk but with material accuracy requirements), dynamic-pricing agents (carry antitrust exposure that is structurally distinct from other AI risks), supply-chain orchestration agents (multi-party data flows that complicate audit substrate ownership), and returns-and-fraud-detection agents (consumer-protection law exposure including disparate-impact claims). The dominant 2026 production pattern is augmentation rather than replacement of human operators; deployments framed as headcount-replacement have produced reversals at material rates (the Klarna pattern). Retailers and 3PLs (third-party logistics providers) operating across multiple jurisdictions face an additional layer of consumer-protection law fragmentation that the EU AI Act does not pre-empt and that materially affects the deployment scope.","article_url":"https://agentmodeai.com/retail-logistics-ai-agents/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-056","claim":"Enterprise AI agent ROI calculation in 2026 requires a structured eight-input model that captures the costs and benefits the standard SaaS-style ROI calculator misses: (1) per-session-hour or per-task model cost at the deployment's actual usage profile, (2) human-in-the-loop labour cost including approval-gate review time, (3) deployment-layer instrumentation cost (audit substrate, drift monitoring, MTTD detection), (4) regulatory compliance cost amortised across the deployment's revenue, (5) productivity uplift on existing human staff (the augmentation case), (6) avoided cost from reduced incident rate and reduced kill-criterion losses, (7) revenue impact net of service-quality regression risk, (8) the strategic-option value of the deployment's underlying capability. The calculation produces a 90-day ROI checkpoint figure, a 12-month payoff figure, and a kill-criterion threshold. The calculation also produces a sensitivity table showing which inputs drive the ROI most heavily; cost-side sensitivity is typically dominated by inputs 2 and 3, revenue-side by inputs 5 and 7. Most 2026 enterprise AI deployments evaluated against this model break even between months 9 and 18; deployments outside that range are either materially under-investing in instrumentation (faster apparent ROI) or are operating in unfavourable cost structures (longer payoff).","article_url":"https://agentmodeai.com/ai-agent-roi-calculator/","topic":"agent-procurement","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-057","claim":"The enterprise AI agent risk register for 2026 resolves to a 12-column template that captures every risk an enterprise must document under EU AI Act Article 9 and NIST AI RMF Manage function: risk ID, deployment ID, threat class (per OWASP Agentic AI Top 10), likelihood, impact, inherent risk score, control mapping (against the seven-control surface), residual risk score, named accountable individual, review cadence, status, last-reviewed date. The register is operated by the Head of AI Governance, reviewed monthly in the AI governance committee, and queryable in the under-4-business-hour Article 73 incident-response window. The 12-column template integrates the threat surface (OWASP Agentic AI Top 10, claim AM-043), the controls (seven-control surface, claim AM-043), the audit substrate (claim AM-046), and the kill-criterion enforcement (claim AM-047), into a single living artefact. An enterprise that operates the register seriously has substantially completed the Article 9 risk-management system documentation requirement; the register is the single artefact that resolves the cross-reference matrix between operational reality and regulatory framework.","article_url":"https://agentmodeai.com/ai-agent-risk-register-template/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-061","claim":"Production agentic-AI costs at scale routinely run multiples of POC projections, and a layered optimisation programme covering model tiering, vendor prompt caching, batch APIs, context-window discipline, and observability budgeting closes most of the gap.","article_url":"https://agentmodeai.com/the-2m-ai-bill-that-became-200k-the-enterprise-cost-optimization-playbook-for-production-ai-agents/","topic":"enterprise-ai-cost","pub_date":"2025-07-27","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2025-07-27","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-28","verdict":"partial","note":"Rewritten 27-28 Apr 2026 from 27 Jul 2025 WordPress-migrated original. Original used a fictional CTO scene (Marcus Chen, $4.2B logistics company, 9:47 AM Tuesday Seattle), fabricated case figures ($2.1M to $187K monthly, named-company before/after teardowns), fabricated expert quotes (Patricia Williams VP of Engineering at Walmart; David Park Principal at Goldman Sachs), and banned phrases (plot twist, the dirty secret, revolutionary, emoji subheads). Rewrite extracts the verifiable cost-driver categories with primary-source citations from Anthropic's published multi-agent token-ratio research, vendor prompt caching and batch-API pricing pages, McKinsey State of AI, Andreessen Horowitz on LLM inference economics, and Gartner's April 2026 I&O finding. Approved + published 28 Apr 2026."}],"primary_sources":[]},{"id":"AM-063","claim":"AI agents executing financial transactions need a four-control bundle (action-approval gates by blast radius, kill-switch protocols, decision-audit trails, per-action revocation); enterprises shipping agentic-AI without this bundle face CISO governance pressure they cannot satisfy under existing model-risk-management, FFIEC, and EU AI Act expectations.","article_url":"https://agentmodeai.com/your-ai-agents-just-approved-2-7m-in-vendor-payments-and-other-nightmares-keeping-cisos-awake/","topic":"agentic-ai-governance","pub_date":"2025-07-27","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2025-07-27","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-28","verdict":"partial","note":"Rewritten 27-28 Apr 2026 from 27 Jul 2025 WordPress-migrated original. Original used fictional Seattle CISO scene with fabricated $2.7M case, fabricated cohort scheduling, emoji subheads, and 'battle-tested' hype. Rewrite extracts the verifiable control-set framework with primary-source citations (NIST AI RMF, NIST AI 600-1 Generative AI Profile, FFIEC IT Examination Handbook, SR 11-7, OCC Bulletin 2011-12, ISACA AI Audit Toolkit, Cloud Security Alliance MAESTRO framework). Cross-links to the live AM-037 non-human-identity piece as the identity-layer companion. Approved + published 28 Apr 2026."}],"primary_sources":[]},{"id":"AM-100","claim":"AI-authored + human-signed publications produce more verifiable enterprise-AI commentary than human-only or anonymous-AI alternatives, when the AI authorship is paired with a public claim ledger and dated correction log.","article_url":"https://agentmodeai.com/ai-writes-about-ai-tracked-claims-case/","topic":"agentic-ai-governance","pub_date":"2026-04-26","last_reviewed":"2026-04-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-101","claim":"Across the named analyst-publication comparable set (Stratechery, The Information, the Substack analyst stack, the Big-4 research blogs, Gartner, Forrester, IDC) as of late April 2026, none maintains a public claim ledger — a tracked register of every primary claim with scheduled reviews, dated verdicts, and a public correction log. The absence is structural, not accidental, and explains why none of the category produces the kind of audit-able commentary the Holding-up system makes possible.","article_url":"https://agentmodeai.com/why-this-publication-has-a-ledger/","topic":"agentic-ai-governance","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-07-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-102","claim":"Among the comparable publications surveyed in AM-101 (Stratechery, The Information, the Substack analyst stack, the Big-4 research blogs, Gartner, Forrester, IDC) as of late April 2026, none uses the disclosed-AI-author + named-human-signatory + public-claim-ledger format. The combination is structurally rare and the rarity is what makes the format consequential, not the disclosed AI authorship alone.","article_url":"https://agentmodeai.com/the-ai-author-signature-decision/","topic":"agentic-ai-governance","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-07-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-103","claim":"Across two of the three Q1 2026 ventures Peter built with Claude (agentmodeai, Rhino-basketball; DealVex pending git-versioning), rework rate measured as deletions / total git churn ranged from 8.1% to 13.5% over the 90-day window from 28 Jan to 28 Apr 2026. The data is meaningfully lower than typical solo-developer projects but substantially higher than the 'AI codes it correctly the first time' marketing narrative implies, supporting the thesis that AI-paired development requires explicit measurement, not assumed productivity.","article_url":"https://agentmodeai.com/learning-ai-by-doing-ai-the-data/","topic":"agentic-ai-governance","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-07-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-104","claim":"Anthropic's withholding of Claude Mythos forces senior IT teams to advance their AI cyber-threat-model timeline by two to three years, and to rebuild three specific assumption sets — patch prioritization, third-party risk on AI infrastructure, and AI procurement diligence — inside Q2 2026.","article_url":"https://agentmodeai.com/claude-mythos-cio-risk-posture/","topic":"agentic-ai-governance","pub_date":"2026-04-27","last_reviewed":"2026-06-10","next_review":"2026-08-09","verdict":"holding","verdict_history":[{"date":"2026-04-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-105","claim":"Organizations that have not adopted an offensive-security operating mode (continuous attack-surface validation, AI-augmented internal vulnerability discovery, standing threat-hunting, deception, counter-AI controls) by Q4 2026 will show measurably wider mean-time-to-detect for AI-assisted attackers than peers that have, in industry-survey data published in late 2026 and early 2027.","article_url":"https://agentmodeai.com/offensive-security-cio-clockspeed/","topic":"agentic-ai-governance","pub_date":"2026-04-27","last_reviewed":"2026-04-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-04-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-106","claim":"Loaded human FTE cost ($90K-$180K all-in for typical knowledge work) vs total agentic-AI operational cost (token plus orchestration plus integration plus observability plus human oversight) does not favour replacement at parity in 2026 for most roles; the math works for narrow, high-volume, low-judgment task categories and breaks down where regulatory accountability, customer trust, or judgment-under-ambiguity is load-bearing.","article_url":"https://agentmodeai.com/agentic-ai-vs-human-worker-cost-economics/","topic":"enterprise-ai-cost","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — spine is observable from current public deployment cost data and labour-displacement research, per-category quantitative bands tracked against next review cycle. REVIEW: Peter — please verify claim text + cited sources before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-107","claim":"The 2026 insurance market does not yet offer agent-specific E&O policies in any mature form; existing cyber and tech-E&O policies were drafted against human-error and software-defect risk models that don't cleanly map to autonomous reasoning actors. Enterprises shipping agentic-AI face an underwriting gap: the cyber policy may not respond to a loss caused by an agent's reasoning step, and the professional-liability policy may exclude AI-generated outputs entirely. CIOs and CROs need to surface this gap with their broker before the loss event, not after.","article_url":"https://agentmodeai.com/agentic-ai-insurance-and-underwriting/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 as a staged draft (rewriteInProgress: true). Status set to Partial because the underlying market is in a transitional phase and per-carrier wording specifics may shift inside the 60-day review window. REVIEW: Peter to verify (a) the Lloyd's Lab Cohort 12 dating and submissions detail, (b) the Munich Re aiSure agentic-deployment extension claim, (c) the NAIC Model Bulletin scope, (d) whether the AIG CyberEdge and Chubb Integrity+ AI endorsement language descriptions reflect the most recent product updates, and (e) the MGA list (Armilla, Vouch, Coalition, Relm) is currently in market with AI-liability paper before promoting from staged draft to published."}],"primary_sources":[]},{"id":"AM-108","claim":"Agentic-AI data-residency requirements are not cleanly inherited from existing GDPR cross-border transfer practice. Agent context windows, retrieval indexes, and reasoning traces all create new categories of personal-data processing that have to be located, documented, and (for high-risk Annex III deployments) data-resident inside the EEA before EU AI Act Article 16 enforcement opens on 2 August 2026. The deployment topology has to shift to single-region EEA-resident for high-risk systems; hub-and-spoke remains defensible for general-purpose deployments under documented GDPR Chapter V transfer mechanisms.","article_url":"https://agentmodeai.com/agentic-ai-data-residency-eu-ai-act/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — spine is anchored to the Act itself plus current vendor compliance pages, but the four-surface Article-mapping has not yet been tested against an enforced case (the August 2026 enforcement window opens inside the next review cycle). REVIEW: Peter — please verify claim text + Article references + vendor citations before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-109","claim":"Enterprises focused on the headcount-reduction half of agentic-AI transformation are systematically under-budgeting the retraining cost for the residual workforce, and programmes that ship the cuts without simultaneously shipping the upskilling produce a 6-12 month productivity dip that erases the early ROI.","article_url":"https://agentmodeai.com/agentic-ai-retraining-gap-survivors/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — the productivity-dip duration is observable from current public workforce data but the 6-12 month band has not been tested against post-2026 enterprise case data yet. REVIEW: Peter — please verify claim text + cited sources before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-110","claim":"Traditional SLAs (uptime, p95 latency, error rate) are structurally insufficient for autonomous agentic-AI; the four metrics that actually work are action-bounded availability, MTTD-for-Agents, output-distribution drift, and per-class action error budget, and vendors that cannot expose the telemetry these require are not yet production-ready against the 2026 enterprise procurement bar.","article_url":"https://agentmodeai.com/agentic-ai-sla-architecture/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — the four metrics are observable from current SRE/OTel practice but have not been tested as a procurement bar against 2026 vendor SLAs yet. REVIEW: Peter — please verify claim text + cited primary sources (especially the OpenTelemetry GenAI stable-promotion date and the Anthropic/MS Agent Framework reliability docs) before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-111","claim":"The right enterprise playbook for an agent incident in 2026 has six steps that do not appear in any standard SRE handbook — action-class containment before root-cause analysis, reasoning-trace forensics, blast-radius reconstruction across downstream agents and systems, stakeholder notification with the specific failure mode named, regulatory exposure assessment for in-scope deployments, and selective re-enable with degraded-mode guardrails — and CIOs without this playbook will spend their first agent incident discovering it under crisis conditions.","article_url":"https://agentmodeai.com/agent-incident-response-playbook/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — six-step playbook is a synthesis from current SRE practice + AI-specific guidance and has not been tested against a major published agent-incident postmortem yet. REVIEW: Peter — please verify claim text + cited primary sources before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-112","claim":"Healthcare agentic-AI sits across three regulatory regimes that do not compose cleanly — HIPAA on PHI handling and BAA topology, FDA software-as-medical-device guidance on clinical decision support and predetermined change control, and state medical/nursing board licensure rules placing the practitioner as the responsible party of record — and the five-control bundle of BAA-aware architecture, PCCP, clinical-judgement-of-record audit trail, on/off-switch with practitioner attribution, and breach-notification readiness is the minimum defensible architecture for any clinical agentic-AI deployment.","article_url":"https://agentmodeai.com/healthcare-agentic-ai-governance/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-113","claim":"Standard 2026 agentic-AI vendor MSAs contain six contract patterns that systematically transfer risk from vendor to enterprise customer in ways that do not appear in equivalent pre-AI enterprise software MSAs — model-version unilateral-change, training-data ambiguity on customer inputs, usage-cap auto-escalation, indemnification carve-outs for model output, data-residency commitments that don't bind sub-processors, and liability caps tied to fees-paid that don't scale with autonomous-action authority.","article_url":"https://agentmodeai.com/agentic-ai-vendor-contract-gotchas/","topic":"agent-procurement","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-114","claim":"Production agentic-AI in 2026 needs four observability layers — infrastructure, LLM-call, trace, and output — and most enterprise deployments instrument only the cheaper subset (Layers 1 and 2 plus partial Layer 3); the failure modes Layers 3 and 4 catch (multi-step reasoning failure and output-distribution drift) are the ones EU AI Act Article 9 and Article 17 evidence obligations from 2 Aug 2026 onward will require coverage of, and the four layers compose directly into the four AM-110 SLA metrics (action-bounded availability, MTTD-for-Agents, output-distribution drift, per-class action error budget).","article_url":"https://agentmodeai.com/agent-observability-stack-production/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-28","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Initial verdict 'Partial' — the four-layer model is observable from current 2026 tool categories and OpenTelemetry GenAI convergence, but the procurement-or-build cost bands are publication estimates and have not been tested across a representative sample of enterprise deployments. REVIEW: Peter — please verify (1) the OpenTelemetry GenAI stable-promotion date (13 Mar 2026) is consistent with what AM-110 cites; (2) the cost-band ranges in the §Share-thoughts template are defensible as our-estimate or need tightening; (3) Datadog AI Observability and New Relic AI Monitoring product names are current; (4) Arize Phoenix open-source/managed dual-form description is accurate; (5) the CNCF OpenTelemetry GenAI working-group framing matches the actual project structure before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-115","claim":"Agent Mode AI publishes a public quarterly review of every claim it has made, with verdict before/after, named primary-source movement, and aggregate verdict-change rate across the corpus. The bulletin runs on a fixed quarterly cadence (end of Apr, Jul, Oct, Jan); the rhythm is the editorial discipline the niche has been missing.","article_url":"https://agentmodeai.com/q2-2026-claim-review-bulletin/","topic":null,"pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-07-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 — the first Quarterly Claim Review Bulletin. The claim itself is recursive: it asserts that the bulletin will ship quarterly, and the next review (30 Jul 2026) tests whether the Q3 bulletin actually appeared. Status starts as 'up' because the claim is currently true (the Q2 bulletin shipped). The verdict at end of July 2026 will move to Holding, Partial (bulletin shipped but on a delayed cadence), or Not holding (no bulletin shipped). REVIEW: Peter — please verify claim text + cadence wording before removing rewriteInProgress flag."}],"primary_sources":[]},{"id":"AM-116","claim":"A class of derivative actions is forming in 2025-2026 around board failure to supervise AI deployments under the Caremark line, and D&O carriers are responding at renewal with explicit AI questionnaires and emerging exclusions, materially shifting director liability exposure that most boards have not yet read in their actual policy language.","article_url":"https://agentmodeai.com/directors-officers-insurance-ai-supervision-claim/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-117","claim":"AI Bill of Materials (AI-BOM) is moving from optional security artefact to enforceable procurement requirement in 2026, driven by EU AI Act Article 11 + Annex IV technical-documentation requirements (effective 2 August 2026) and the CycloneDX ML-BOM and SPDX 3.0 specifications. Enterprise SBOM programs need three specific extensions (generation path for AI components, AI-specific risk correlation feeds, procurement-side language for AI-BOM delivery).","article_url":"https://agentmodeai.com/ai-bill-of-materials-supply-chain-disclosure/","topic":"agent-procurement","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-118","claim":"As of April 2026 the largest sovereign-wealth and pension funds (NBIM, CalPERS, ABP, OTPP, USS) have published almost no formal AI position papers, despite trillion-dollar AI exposure across portfolios. The structural absence is the signal: AI is being rated by these investors but the rating criteria have not been formally codified, leaving public-company IR teams preparing engagement against expectations the investors have not yet written down.","article_url":"https://agentmodeai.com/pension-fund-sovereign-wealth-ai-policy-void/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-119","claim":"The 2026 cyber-insurance renewal tightening enterprises are experiencing is upstream-driven by reinsurance market repricing of catastrophic AI tail risk (Lloyd's of London, Munich Re, Swiss Re), not by primary-carrier loss data. The reinsurance signal travels via tighter treaty terms, AI-specific exclusions, and elevated retentions, with a 6-12 month lag to primary policies. Enterprise risk officers negotiating against the primary on AI terms have limited room because the carrier's own treaty caps what it can offer.","article_url":"https://agentmodeai.com/reinsurance-market-ai-tail-risk-pricing/","topic":"agent-procurement","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-120","claim":"AI agent deployments touching employee work in EU jurisdictions with co-determination law (Germany BetrVG §87, Netherlands WOR Art. 27, France CSE provisions) require works council consent before activation in 2026. Most US-headquartered AI vendors lack a customer-success workflow for this, producing a class of stalled rollouts that read as 'vendor delay' but are actually compliance gaps. Total EU-site timeline from selection to production is 6-9 months when handled well, 12-18 when consultation begins late.","article_url":"https://agentmodeai.com/works-council-ai-agent-deployment-eu/","topic":"agentic-ai-governance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-121","claim":"AI in IT operations in mid-2026 delivers measurable productivity gains (UK Government Digital Service trial: 26 minutes per user per day across 20,000 staff; BT pilot: 35% case-resolution-time reduction with named CIO on the record; ServiceNow's own help desk: 90% L1 deflection in vendor-internal optimal conditions) but the staff-reduction story is structurally smaller than vendor pitches suggest. Gartner finds only 11% of Fortune 500 companies have actually cut support headcount via AI; Forrester reports 55% of AI-attributed layoffs are regretted and roughly half are reversed; CRMArena-Pro shows multi-step agent reliability at ~35%. The cost saving lands first on the BPO/contractor line, second on contractor spend, and only slowly and controversially on direct headcount. Agentic L2/L3 remediation remains pilot-stage: per Gartner's October 2025 survey of 360 IT app leaders, only 15% are considering, piloting, or deploying fully autonomous agents, and Gartner predicts >40% of agentic AI projects will be cancelled by end-2027.","article_url":"https://agentmodeai.com/ai-it-operations-reality-check/","topic":"enterprise-ai-cost","pub_date":"2026-05-02","last_reviewed":"2026-05-02","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-02","verdict":"holding","note":"Claim created at publish."},{"date":"2026-05-02","verdict":"partial","note":"Klarna walk-back primary-source upgrade — added Siemiatkowski verbatim quotes via Bloomberg-cited-by-Fortune (9 May 2025) and the Uber-style freelance hiring detail via Entrepreneur. Closes the highest-priority evidence gap from the source dossier."}],"primary_sources":[]},{"id":"AM-122","claim":"The four credible 2026 agent-evaluation platforms (DeepEval, Braintrust, LangSmith, Patronus AI) do not compete on capability rank; each fits a distinct deployment shape (engineering-led eval-as-code; SaaS-first eval-as-product; LangChain-stack-native bundled with observability; research-grade hallucination + simulation), and picking by capability matrix produces the wrong procurement outcome for most enterprises. The structurally load-bearing eval-vs-observability split (companion piece AM-123) compounds this: 'is the agent right' and 'what did the agent do' are different procurement decisions answered by different platforms.","article_url":"https://agentmodeai.com/agent-eval-frameworks-deepeval-braintrust-langsmith-patronus/","topic":"agent-procurement","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-123","claim":"Evaluation answers 'is the agent right'; observability answers 'what did the agent do'. The four credible 2026 agent-observability platforms (Langfuse, Arize, Helicone, LangSmith) split cleanly on a single structural axis: open-source-first vs SaaS-first. Helicone has been in maintenance mode since 3 March 2026 (founders joined Mintlify) and should not be selected for greenfield 2026 deployments. Production deployments need both eval and observability; the procurement decisions are different and conflating them produces SLA architecture that fails its first incident.","article_url":"https://agentmodeai.com/agent-observability-langfuse-arize-helicone-langsmith/","topic":"agent-procurement","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-124","claim":"Pharma and life sciences agentic AI in 2026 inherits five regulatory regimes simultaneously (21 CFR Part 11, GxP under GAMP 5 Second Edition, EMA Annex 11 in 2025-2026 revision, the EMA Reflection Paper on AI in the medicinal product lifecycle, and the EU AI Act). The audit substrate that satisfies any one regime does not by default satisfy the others. The 2026 procurement gap is treating the regimes as substitutable. Four conditions materially constrain compliant deployment (validated computerised system status under GAMP 5 plus CSA; 17-field audit trail covering Part 11 + Annex 11 + Article 12 simultaneously; ALCOA+ data integrity with contemporaneous, original, enduring records; EU AI Act high-risk-system registration with Article 11 technical file plus Article 16 post-market monitoring). Three vendor postures emerge in market (pre-validated Category 4 packaging; general-purpose platform plus customer-validated wrapper; open-source stack plus customer-engineered audit substrate).","article_url":"https://agentmodeai.com/pharma-life-sciences-agentic-ai-21-cfr-part-11/","topic":"agentic-ai-governance","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-08-01","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-125","claim":"ITSM agent procurement in 2026 is not three independent vendors but two acquirer ecosystems plus one product line at the intersection: ServiceNow (acquirer; Now Assist native plus Moveworks acquired 15 Dec 2025 for $2.4B closed consideration vs the announced $2.85B) and Automation Anywhere (acquired Aisera Nov 2025). The procurement decision in 2026 is shaped less by the feature matrix than by the post-acquisition reality. Picking by feature matrix without mapping the acquirer's strategic interest produces the wrong answer. ServiceNow Now Assist is the bolt-on for organisations already on ServiceNow; Moveworks is the omnichannel layer (still standalone branding, ServiceNow-owned); Aisera is the auto-resolution play that competes on closure rate, now under Automation Anywhere's portfolio.","article_url":"https://agentmodeai.com/servicenow-now-assist-vs-moveworks-vs-aisera/","topic":null,"pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-126","claim":"The OWASP Agentic AI Top 10 names what to defend against; it does not say how to test that the defences work. The 2026 enterprise red-team for agentic systems is a distinct discipline from generalised pen-testing, with its own methodology (four disciplines: prompt injection, tool misuse, context-window attacks, multi-turn objective drift), tooling stack (PyRIT v0.13.0, Garak, custom harnesses, MITRE ATLAS for structured threat-modelling vocabulary), evidence model (six-section report including ATLAS technique mapping plus residual-risk plus EU AI Act Article 12 substrate alignment plus Article 16 post-market monitoring recommendations), and procurement decisions (in-house vs specialist-vendor vs hybrid). Most enterprises run the wrong test (generalised application pen-test) and pass it; the passing report is the procurement evidence that produces false confidence.","article_url":"https://agentmodeai.com/agent-red-teaming-owasp-companion/","topic":"agentic-ai-governance","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-127","claim":"Of the eleven claims this publication has published against the 2 August 2026 EU AI Act enforcement deadline, the four operational-evidence claims (AM-108 data residency, AM-046 audit-evidence under four hours, AM-117 AI-BOM procurement, AM-120 works council workflow) carry materially higher risk of moving from Holding to Partial in Q3 2026 than the two governance-process claims (AM-047 Head of AI Governance role, AM-051 centralised-vs-federated). Materially higher risk is defined as: at least three of the four operational-evidence claims will be downgraded to Partial or Not holding by 1 October 2026, while at least one of the two governance-process claims will remain Holding.","article_url":"https://agentmodeai.com/90-days-eu-ai-act-enforcement-what-corpus-says/","topic":"agentic-ai-governance","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-01","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-128","claim":"The MIT NANDA 'GenAI Divide' 95% pilot-failure statistic (August 2025) is widely cited in 2026 enterprise procurement decks as evidence that 95% of AI projects fail. The underlying methodology measures something narrower and more specific: 95% of 300 analysed AI projects delivered no measurable P&L impact, where 'no measurable impact' is largely a function of pilots not having documented pre-deployment baselines, not a function of pilots failing technically. The structurally interesting findings underneath the headline (build-vs-buy 67%-vs-22% spread, 40%-licensed / 90%-shadow-using gap, marketing-vs-back-end deployment misdirection, the static-error / learning-gap pattern) are more useful for procurement teams than the headline number, and they update against the Stanford 12/88 bimodal ROI distribution (claim AM-029) cleanly.","article_url":"https://agentmodeai.com/the-mit-genai-pilot-failure-claim/","topic":"enterprise-ai-cost","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-129","claim":"No mid-market enterprise has produced a documented +240% ROI in 90 days from agentic AI under audited conditions. Read against McKinsey State of AI 2025 (n=1,993; 23% scaling, 17% EBIT-attribution at 12-month horizon), MIT NANDA GenAI Divide (95% of pilots produce no measurable P&L impact, 67% buy vs 22% build success spread), and Stanford Digital Economy Lab Enterprise AI Playbook (12/88 bimodal ROI distribution at 12-18 months), the realistic 90-day mid-market ROI band for the highest-discipline 12% cohort is 20-40% operator-time savings on bounded use cases plus a working pilot pattern that scales into 12-18-month measurable ROI — not the 240% ROI in 90 days the vendor pitch frames it as. The four-artefact 90-day deliverable (documented baseline, bounded production deployment, per-class action error budget, scaling-vs-stop decision) is what the 12% cohort actually produces.","article_url":"https://agentmodeai.com/achieve-240-roi-in-90-days-with-ai-agents-for-mid-market/","topic":"enterprise-ai-cost","pub_date":"2026-05-04","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of three read-against anchors unanchored on re-review. The claim text cites 'Stanford Digital Economy Lab Enterprise AI Playbook (12/88 bimodal ROI distribution at 12-18 months)' and frames the realistic ROI band around 'the highest-discipline 12% cohort'. Full-text verification on 10 Jun 2026 found the playbook contains no 12/88 distribution, no bimodal ROI shape, and no 12-18-month ROI measurement point (full finding at AM-029, correction of 10 Jun 2026). The claim's core negative finding — no mid-market enterprise has produced a documented +240% ROI in 90 days under audited conditions — is unaffected; the McKinsey State of AI 2025 and MIT NANDA legs verify and continue to support it. The '12% cohort' framing has no verifiable referent. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-130","claim":"Agentic AI 2024-2025 produced four distinct classes of evidence the 2026 procurement reader should not collapse into a single 'AI is working' narrative: (1) vendor-published wins inside vendor-controlled environments (ServiceNow internal 90% L1 deflection, framed by Nenshad Bardoliwalla as upper bound conditioned on two decades of structured workflow data the customer does not have), (2) audited customer pilots with active human oversight (BT 35% case-resolution improvement with random checks per Hena Jalil; UK Government Digital Service 26 minutes/day saved across 20,000 staff in Q4 2024; HMRC 28,000-staff M365 Copilot rollout April 2026), (3) public walk-backs (Klarna May 2025 Bloomberg-reported reversal of the 700-agent claim while the original press release stayed live; GitHub Copilot April 2026 token-counting bug; Salesforce Agentforce IT 200-customer reality vs Marc Benioff's launch pitch), and (4) structural failure modes (CRMArena-Pro 35% multi-step agent reliability finding; Carnegie Mellon independent verification at 30-35%; EchoLeak CVE-2025-32711 cross-agent prompt-injection class). Each class produces a different procurement lesson; treating them as one narrative is the most common 2026 enterprise mistake.","article_url":"https://agentmodeai.com/the-agentic-ai-revolution-real-world-success-stories-and-strategic-insights-from-2024-2025/","topic":"agent-procurement","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-131","claim":"The AI Training Lead role — the human who curates the agent's evaluation set, reviews sampled outputs against it, and partners with the ML engineer on retraining decisions — is now a budget-line for enterprise agentic AI deployments rather than a vendor-bundled professional-services function. Domain experts (five-plus years inside the workflow the agent is meant to assist) outperform pure-ML hires in the role because the work is judgement-heavy, not algorithm-heavy. CIOs that do not budget the role explicitly see deployments fail at the iteration boundary.","article_url":"https://agentmodeai.com/from-it-pro-to-ai-training-lead-the-180k-career-path-nobodys-talking-about/","topic":"agentic-ai-governance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-132","claim":"Enterprise agentic AI ROI in 2026 is bimodal across four independent datasets. Stanford Digital Economy Lab's 2026 Enterprise AI Playbook documents 12% of deployments clearing 300%+ ROI with 88% at or below break-even at 12-18 months. Gartner Q1 2026 Infrastructure & Operations Survey reports 28% of AI projects 'fully paying off'. McKinsey State of AI 2025 (n=1,993) reports 23% scaling with 17% EBIT-attribution at 12 months. MIT NANDA's GenAI Divide reports 95% of pilots produce no measurable P&L impact alongside the 67% buy vs roughly 22% build success spread. The 73%/27% slug rounds the four numbers; the bimodal shape is reproducible and the variable separating the two cohorts is operational discipline (instrumented under GAUGE: governance, audit substrate, use-case maturity, guardrails, evidence/baseline, exit posture), not model selection.","article_url":"https://agentmodeai.com/why-73-of-agentic-ai-projects-fail-and-how-the-27-generate-312-roi/","topic":"enterprise-ai-cost","pub_date":"2026-05-05","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of four legs unanchored on re-review. The claim text attributes '12% of deployments clearing 300%+ ROI with 88% at or below break-even at 12-18 months' to the Stanford DEL 2026 Enterprise AI Playbook. Full-text verification on 10 Jun 2026 found no such figure in that source: the playbook (Pereira, Graylin, Brynjolfsson, Apr 2026) studies 51 successful deployments by design and contains no ROI distribution, no 300%-plus cohort, and no break-even measurement point (full finding at AM-029, correction of 10 Jun 2026). The only verified figure carrying the same 12/88 numerals is IDC research with Lenovo (via CIO.com, Mar 2025): roughly 88% of AI proof-of-concepts never reach production and roughly 12% graduate — a pilot-to-production graduation metric, not an ROI distribution. The Gartner 28%, McKinsey 23%/17%, and MIT NANDA 95% legs verify; they support a small high-performing tail and a large struggling body, but none documents the two-peak bimodal shape the claim asserts. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-134","claim":"The 2026 implementation cut on non-human identity for AI agents resolves on three factors (existing IAM relationship, deployment topology, cross-platform integration burden) across six credible control planes: Okta NHI, Microsoft Entra ID Workload Identities, Auth0, Keycloak, SPIFFE/SPIRE for Kubernetes-native deployments, and AWS IAM Roles Anywhere for hybrid AWS-anchored deployments. The procurement-defensible audit substrate captures three event classes regardless of vendor: identity issuance, authentication, and authorisation.","article_url":"https://agentmodeai.com/agent-identity-iam-architecture-nhi/","topic":"agentic-ai-governance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-135","claim":"EU AI Act Article 50 takes effect 2 August 2026 and creates four distinct transparency obligations requiring different UX implementations: Article 50(1) chatbot interaction disclosure on providers, Article 50(2) machine-readable marking on generative AI output, Article 50(3) biometric categorisation and emotion recognition disclosure on deployers, and Article 50(4) deepfake disclosure on deployers (with the artistic-or-creative-work exception). The procurement-defensible disclosure UX has six properties (visible at the right moment, plain language, persistent or recurrent, linked to a substantive disclosure surface, auditable, updateable). Most enterprises have absorbed the legal text without designing the UX it requires.","article_url":"https://agentmodeai.com/eu-ai-act-article-50-transparency-disclosure/","topic":"agentic-ai-governance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-136","claim":"Across the 24-month window May 2024 to April 2026, every major foundation-model provider (Anthropic, OpenAI, Google, AWS Bedrock, Azure OpenAI) experienced at least one multi-hour outage that exceeded the SLA-credit threshold defined in their published terms. The procurement-defensible posture is multi-provider routing with documented failover and hard-dollar incident liability above the standard SLA-credit cap. Three architectural patterns dominate 2026 production deployments: gateway abstraction (LiteLLM, OpenRouter, Portkey), provider-side regional failover (partial mitigation), and explicit multi-provider provisioning at the application layer.","article_url":"https://agentmodeai.com/foundation-model-uptime-sla-track-record/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-30","next_review":"2026-06-29","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-137","claim":"Agent evaluation in production resolves on three operational components that determine whether the chosen evaluation platform produces useful signal: eval-set design across three layers (50-200 calibration prompts, 30-100 edge-case prompts, 10-50 production-sampled prompts per week), drift detection across three signal classes (output-distribution, score-distribution, tool-use distribution), and a regression-budget framework that forces binary ship/hold decisions (defensible default 5% absolute decline on calibration set, 10% on edge-case set, per release window). The procurement decision (which platform to buy, covered at AM-122) is the easier half; the operational discipline is what most enterprises under-invest in even after buying a platform.","article_url":"https://agentmodeai.com/agent-evaluation-in-production/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-138","claim":"The 2 August 2026 EU AI Act deployer-obligations enforcement window adds three new clause families to the AI MSA red-team checklist that were optional or absent in pre-enforcement contracts: Article 11 technical-file pass-through, Article 16 post-market-monitoring support, and Article 26 deployer-documentation supply. The post-enforcement checklist grows from the 38-item RES-005 v1.0 baseline to roughly 54 items across 11 clause families, with Article 50 transparency UX (covered at AM-135) and foundation-model uptime hard-dollar liability (covered at AM-136) as additional 2026 additions. The asymmetric-instrument observation — that enterprise and operator AI procurement face the same vendor-citation-chain manipulation pattern with different audit instruments — is embedded as a 600-word insert in this piece.","article_url":"https://agentmodeai.com/vendor-msa-renewal-post-eu-ai-act-enforcement/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-139","claim":"Enterprise AI buyers and operator AI buyers face the same vendor-citation-chain manipulation pattern with asymmetric audit instruments, and consume vendor case studies aimed at the other cohort with mirror-image misreads. The enterprise reads the IndieHacker timeline as procurement-cycle benchmark and removes controls under timeline pressure; the operator reads the Fortune-500 efficiency gain as result-attribution and inherits expectation without the operational substrate. The cross-borrow that is procurement-defensible at both scales: enterprises borrow the operator's cancellation-trigger discipline (OPS-051) and the cohort-fit filter (OPS-011); operators borrow the enterprise's MSA red-team scoped down (RES-005), evaluation discipline scaled to weekly (AM-137), and audit substrate at lightweight scale (AM-046). The verification gap is the same gap; the instruments are different; the publication's two-register architecture is the editorial response.","article_url":"https://agentmodeai.com/vendor-case-study-misreads-across-buyers/","topic":"agent-procurement","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-140","claim":"Vendor 'successful pilot' references presented at procurement-committee evaluation transfer to scaled production at the procuring enterprise's measurement and governance regime at roughly the McKinsey 23% rate (n=1,491, Nov 2025); the gap is operational rather than capability-driven and is tractable with six pre-pilot questions a procurement committee can require answered in writing before the contract closes, not after.","article_url":"https://agentmodeai.com/agentic-ai-pilot-to-production-gap/","topic":"agent-procurement","pub_date":"2026-05-06","last_reviewed":"2026-05-06","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-06","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-141","claim":"Agent Mode is the same brand-name shipping in three different product classes in 2026: Microsoft 365 Copilot (productivity-suite agents), Cursor (developer-IDE agents), and GitHub Copilot (code-platform agents). The procurement decision is not feature-comparison; it is which class fits the in-house workflow. Anthropic Managed Agents and OpenAI Agents SDK occupy a fourth category (dedicated agent platforms) that competes for adjacent budgets without using the Agent Mode brand.","article_url":"https://agentmodeai.com/what-is-agent-mode/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-142","claim":"AI agent, AI assistant, and LLM are three structurally different categories in 2026, distinguished by whether the system can reason about a goal (LLM yes), invoke tools to achieve it (assistant adds), and operate autonomously across multi-step workflows (agent adds again). Procurement that conflates the three optimizes the wrong axis: model-quality bake-offs decide the LLM tier, governance scaffolding decides the assistant tier, and operational preconditions (registry, baseline, change-management, threat model) decide whether an agent can scale at all.","article_url":"https://agentmodeai.com/ai-agent-vs-assistant-vs-llm/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-143","claim":"An AI Bill of Materials in 2026 is the audit-ready inventory of every model, dataset, training source, evaluation method, and deployment dependency in a production AI system. Most enterprises do not yet ship one; the EU AI Act Article 16 deployer-documentation obligations make it mandatory in scope by 2 August 2026. Six layers belong on the BOM: foundation model + version + provider, training datasets + provenance + opt-out signals, fine-tuning data, evaluation methodology + scores, system prompts + guardrails, deployment dependencies (vector DB, RAG sources, MCP servers, agent orchestrator). CycloneDX-AI is the emerging machine-readable format; SBOM under Executive Order 14028 is the precedent.","article_url":"https://agentmodeai.com/ai-bom-enterprise/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-144","claim":"For a Microsoft-stack enterprise in 2026, Microsoft 365 Copilot Agent Mode is the lower-friction agent-platform choice if the workflow already lives in Microsoft Graph; it is structurally weaker on multi-vendor deployment, model-portability, and platform-independence than dedicated agent platforms (Anthropic Managed Agents, OpenAI Agents SDK, Vertex AI Agent Builder). The procurement decision turns on three questions: (1) is the workflow Microsoft-resident, (2) is multi-vendor model selection a hard requirement, (3) is the agent's primary surface productivity-suite or workflow-orchestration.","article_url":"https://agentmodeai.com/microsoft-copilot-agent-mode-enterprise/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-145","claim":"AI vendor switching in 2026 is bound primarily by contract terms — exit clauses, data-portability obligations, model-deprecation rights — not by technical migration cost. Seven clause patterns repeatedly create the lock-in most enterprises only discover at year two of the relationship: (1) data-portability scope narrowness, (2) model-deprecation rights without credit, (3) sub-processor expansion without consent, (4) output-IP ambiguity, (5) pricing-tier rebalancing mid-contract, (6) agent-uptime SLA definition gaps, (7) audit-evidence retention obligations. Vendor consolidation (Moveworks→ServiceNow Dec 2025, Aisera→Automation Anywhere Nov 2025) and model deprecations make this a 2026 procurement story.","article_url":"https://agentmodeai.com/ai-vendor-exit-clauses-checklist/","topic":"agent-procurement","pub_date":"2026-05-07","last_reviewed":"2026-05-07","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-146","claim":"A vendor claim of 'ready-to-run' agentic AI that does not name (a) the specific task being measured, (b) the baseline against which accuracy is reported, and (c) the methodology by which the measurement was produced is not procurement evidence regardless of how the rate is described in marketing; the 2026 industry baseline for procurement-credible accuracy disclosure is the Anthropic Cohort A pattern (red-team rates with named attack corpus, pre/post-mitigation deltas, named patch cadence) on the vendor side and the academic-benchmark pattern (CRMArena-Pro 35% multi-step reliability with defined CRM task corpus, CMU TheAgentCompany 30-35% reproduction range, WebArena ~36% browser-agent ceiling) on the methodology side; vendor 'ready-to-run' positioning without equivalent disclosure leaves the deploying enterprise inheriting the methodology gap as an audit-defense burden.","article_url":"https://agentmodeai.com/agentic-ai-accuracy-claims-task-baseline-methodology/","topic":"agent-procurement","pub_date":"2026-05-09","last_reviewed":"2026-05-09","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-05-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-147","claim":"The Firefox 150 / Claude Mythos disclosure (November 2025) marks the operational shift in agentic AI code auditing from 'AI can find bugs' (true since 2023, but blocked from production CI by the false-positive rate that earlier read-only GPT-4 / Sonnet 3.5 attempts produced) to 'agentic verification clears the false-positive wall by building and running its own test cases before reporting'; the procurement-deck consequence is that CI-time agentic auditing becomes the default expectation for any shipping enterprise software in 2026, and three derived questions belong in any software-vendor procurement (does the vendor's CI pipeline include an agentic-auditing step; what is the vendor's disclosure posture when bugs are found in their own product by agentic tools; what is the vendor's posture on the dual-use risk that the same pipeline architecture works in reverse, as the reported Anthropic investigation of unauthorized Mythos use via a third-party vendor environment makes explicit).","article_url":"https://agentmodeai.com/agentic-code-auditing-firefox-claude-mythos-procurement-read/","topic":"agent-procurement","pub_date":"2026-05-10","last_reviewed":"2026-05-10","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-05-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-148","claim":"GPT-5.5 (released 23 Apr 2026) and Claude Opus 4.7 (released 16 Apr 2026) are not substitutable models for an enterprise running both agentic-coding workloads and knowledge-work workloads in 2026: GPT-5.5 leads the public evaluation evidence on agentic-coding and computer-use surfaces (Terminal-Bench 2.0 82.7% vs 69.4%; GDPval 84.9% vs 80.3%; FrontierMath Tiers 1-3 51.7% vs 43.8%) and runs roughly 72% fewer output tokens than Opus 4.7 on identical coding tasks per Artificial Analysis; Opus 4.7 leads the public evaluation evidence on contamination-resistant coding, finance, and vision-reasoning surfaces (SWE-Bench Pro 64.3% vs GPT-5.4 57.7%; Finance Agent v1.1 64.4%; CharXiv reasoning 78.3%; GPQA Diamond 94.2%) and reports a 36% AA-Omniscience hallucination rate against GPT-5.5's 86% on the same independent evaluation, a 50 percentage-point spread that is the load-bearing data point of any 2026 single-model standardisation decision. The procurement-architecture answer for an enterprise running both workload types is three-tier routing (GPT-5.5 with Codex for agentic coding; Opus 4.7 plus retrieval augmentation for knowledge work; Mythos-via-Glasswing or Opus 4.7 with verification layer for frontier and high-stakes-verification work), not single-model standardisation.","article_url":"https://agentmodeai.com/split-verdict-gpt55-opus47/","topic":"agent-procurement","pub_date":"2026-05-10","last_reviewed":"2026-05-10","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-05-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-149","claim":"In production agentic systems documented across the publicly observable 2025–2026 deployments, the observable band of internal LLM calls per user-facing request sits between 1:18 and 1:60 across documented deployments, with tail cases regularly exceeding 1:400 — meaning unit-economics, latency budgets, and observability scopes built on a 1:1 mental model under-provision by one to two orders of magnitude.","article_url":"https://agentmodeai.com/agent-fan-out-problem-llm-call-amplification/","topic":"enterprise-ai-cost","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-07-11","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-150","claim":"Across the publicly documented 2025–2026 enterprise deployments, single-agent architectures with structured tool-calling outperform multi-agent orchestrations on accuracy, cost, and MTTD for tasks below approximately 12 distinct tool-domains; multi-agent only pays back above that threshold and only when inter-agent state is bounded by a shared structured artifact rather than free-text handoff.","article_url":"https://agentmodeai.com/single-agent-vs-multi-agent-decision-framework/","topic":"agentic-ai-governance","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-07-11","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-151","claim":"Across the 2025–2026 documented deployments at AmLaw 100 firms, agentic AI captures durable value in three of the six billable-hour sub-tasks (document review, precedent retrieval, deposition prep) and produces a net malpractice-risk increase in two (legal drafting submitted as final, citation generation) vs a junior-associate-drafted equivalent at the same time-to-delivery; the remaining sub-task (client communication) is bounded by professional-conduct rules, not technology.","article_url":"https://agentmodeai.com/agentic-ai-legal-services-billable-hour/","topic":"agentic-ai-governance","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-08-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-152","claim":"Across the publicly disclosed 2025-2026 U.S. federal and EU member-state agentic AI procurements, contract renewals are running materially below the broader enterprise SaaS renewal benchmark — driven primarily by audit-evidence failures under OMB M-24-10 §5 and EU AI Act Article 12, not by technical performance — and the renewal-rate gap is the leading early indicator that public-sector agentic AI is following the Salesforce-for-government 2010s adoption curve, not the cloud-for-government 2015s curve.","article_url":"https://agentmodeai.com/public-sector-agentic-ai-procurement-record/","topic":"agent-procurement","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-07-11","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-153","claim":"Of the 8 most-cited enterprise agentic AI vendor claims made in Q1 2026 (Salesforce Agentforce, Microsoft Copilot Agent Mode, Google Gemini Enterprise, Anthropic Claude for Enterprise, OpenAI Agents SDK, ServiceNow AI Agents, Workday Illuminate, SAP Joule), a minority remain Holding at 90-day review, a majority sit at Partial with at least one falsified component, and customer-cited ROI claims hold materially better than vendor-cited ROI claims — meaning the citation-source of an enterprise AI claim is a stronger predictor of its 90-day durability than the size of the vendor making it.","article_url":"https://agentmodeai.com/enterprise-agentic-ai-quarterly-record/","topic":"vendor-trajectory","pub_date":"2026-05-12","last_reviewed":"2026-05-12","next_review":"2026-08-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-154","claim":"Enterprise AI compute growth in the credible 10x to 100x range by 2030 will absorb most of the renewable-buildout headroom the energy transition depends on, extending fossil reliance by roughly a decade unless enterprise IT functions begin modelling AI energy demand in kWh and including it in cloud and on-premise procurement criteria.","article_url":"https://agentmodeai.com/ai-energy-consumption-enterprise/","topic":"enterprise-ai-cost","pub_date":"2026-05-15","last_reviewed":"2026-05-15","next_review":"2026-08-13","verdict":"holding","verdict_history":[{"date":"2026-05-15","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-155","claim":"The Cyber Safety Review Board's April 2024 report on Storm-0558 documented four credential-management failures at Microsoft (a signing key seven years past rotation; an environment-separation boundary enforced procedurally rather than technically; a crash-dump leak the existing scanning could not see; an anomaly-detection baseline that did not exist for the credential class). All four conditions are reproduced in most enterprise AI agent deployments in 2026: long-lived agent credentials without rotation policy, dev/staging/production credentials promoted without re-issuance, runtime telemetry that leaks short-lived tokens without scanning, no issuance-and-use baseline per agent. The CSRB report is forward-readable as a structural map of where AI agent identity programmes fail, not a Microsoft-specific post-mortem. The blast radius is wider for AI agents than it was for Storm-0558 because the action surface authorised by a compromised AI agent credential routinely includes writes, transactions, and downstream tool-use chains, where the Storm-0558 attacker had read-only mail access from one credential.","article_url":"https://agentmodeai.com/non-human-identity-after-the-csrb-report/","topic":"non-human-identity","pub_date":"2026-05-16","last_reviewed":"2026-05-16","next_review":"2026-08-14","verdict":"holding","verdict_history":[{"date":"2026-05-16","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-156","claim":"The detection lag observed in Samsung Electronics' April 2023 ChatGPT incidents (three confidential pastes discovered after the fact by internal audit and self-report, leading to the 2 May 2023 restriction memo) was not a Samsung-specific operational failure. It was the structural output of running enterprise DLP, designed against email/file/removable-media egress channels, against a new egress class (paste-into-chat-interface) that the controls were not built for. Three years later, the structural gap remains the dominant detection failure in enterprise shadow-AI programmes, with the pattern now inverted: the 2023 case was unsanctioned external tools, the 2026 case is agentic capability silently activating inside approved tools (Microsoft 365 Copilot agents acquiring write capability, Custom GPTs created against corporate accounts, MCP servers connected by approved IDEs). The 2026 case is harder to detect because the egress destination is an approved vendor and the AI capability sits behind a procurement approval that did not assess the capability surface. The operational test for whether a programme has closed the Samsung gap is a 24-hour AI-capable-surface inventory, a confidential-document trace test, and an automatic update path when vendors ship new AI features into approved tools.","article_url":"https://agentmodeai.com/shadow-ai-detection-lag-after-samsung/","topic":"shadow-ai-discovery","pub_date":"2026-05-16","last_reviewed":"2026-05-16","next_review":"2026-07-15","verdict":"holding","verdict_history":[{"date":"2026-05-16","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-157","claim":"Three independently-disclosed CVE classes in May 2026 (Microsoft Security Response Center's CVE-2026-25592 and CVE-2026-26030 against Semantic Kernel on 7 May 2026; OX Security's MCP STDIO supply-chain advisory traversing every published MCP implementation regardless of language; the Windsurf 1.9544.26 prompt-injection-to-MCP-registration path) share a single structural property: in the default configuration of 2026 agent frameworks, tool-configuration is treated as data the model is allowed to author, which means the deployer's allowlist is enforced against the configured tools rather than against the model's ability to mutate the configuration. The patch surface is therefore the framework default, not the deployer's wrap. The conventional 2024–2025 enterprise treatment of prompt injection — sandbox the agent's reachable surface at deployment time — is necessary but no longer sufficient. The procurement template for an agent vendor must add five framework-layer attestations (tool-configuration as a privileged operation, runtime enumeration of the tool-configuration surface, configuration-mutation telemetry, coordinated-disclosure record on framework-layer issues, MCP protocol-revision commitment) on top of the deployer-control questions that remain in place.","article_url":"https://agentmodeai.com/prompt-injection-rce-threshold-semantic-kernel-mcp-cves/","topic":"agentic-ai-governance","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-07-16","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-158","claim":"Most enterprises in the EU AI Act high-risk-system in-scope cohort (Annex III categories: biometrics, critical infrastructure, education, employment, essential services, law enforcement, migration, justice) will not have a documented conformity-assessment artifact, an operational post-market monitoring telemetry pipeline, and an Article 13 model-card-and-instructions-for-use production cadence in place by the 2 August 2026 activation. The gap is not legal interpretation, which outside counsel can answer in days. It is a budget gap on three operating-expense lines (conformity-assessment headcount, audit-evidence pipeline infrastructure, model-card production cadence) that the chief financial officer has not yet been asked to size and that the audit committee has not yet authorised. The procurement record, posted-position count, and Q2 2026 enterprise-filing line items together suggest the cohort is mid-cycle on acquisition and pre-production on operational delivery.","article_url":"https://agentmodeai.com/eu-ai-act-readiness-budget-high-risk-systems/","topic":"regulatory-readiness","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-08-15","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-159","claim":"Anthropic's financial-services agent stack announced on 5 May 2026 (ten agents covering investment-banking workflows, Moody's data partnership, full Microsoft 365 integration) will reach a documented production deployment at a tier-1 bank — defined as a published case study, board-disclosed P&L impact, or CIO-level public attribution — by 1 September 2026. If it does, the vertical-specialised stack becomes the procurement default for high-headcount, high-document-throughput workflows in 2027 and horizontal-only platforms face a squeeze on the enterprise contracts where vertical depth is the deciding factor. If it does not, the Wall Street launch is a finance-specific anomaly driven by the Moody's data partnership and the unusual document-shape concentration in investment banking, and horizontal platforms remain the procurement default for cross-functional enterprise adoption. The procurement-template implication for non-finance CIOs is operational now regardless of the predictive outcome: vendor questionnaires must include a vertical-stack roadmap question, MSAs must include an early-renegotiation right triggered by vendor vertical-stack shipments, and 2026 procurement should run against multiple vendors with materially different vertical bets (Anthropic vertical-depth-first; Google platform-and-protocol-first; OpenAI horizontal-with-services-overlay; Microsoft horizontal-with-incremental-vertical-layering).","article_url":"https://agentmodeai.com/anthropic-wall-street-agents-cio-cross-industry-read/","topic":"vendor-trajectory","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-08-15","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-160","claim":"Andrej Karpathy's 19 May 2026 announcement that he has joined Anthropic, paired with Anthropic's confirmed framing that he will lead a team focused on using Claude to accelerate pre-training research (under team lead Nick Joseph), is a foundational-layer vendor-trajectory signal that composes with the 5 May 2026 Wall Street agents launch (AM-159) to describe Anthropic operating on both ends of the platform stack simultaneously — vertical-depth-first on the application layer and name-recognition-first on the pre-training layer. The mandate (Claude accelerating Claude) is more procurement-relevant than the hire itself, because it is a public commitment to recursive self-improvement of the model line at the foundational layer rather than at the application layer. By 17 August 2026, observable evidence in the AI-research community will or will not appear across four markers: (1) a published paper from Anthropic's pre-training team describing a Claude-in-the-loop component with measurable productivity or capability impact; (2) a Claude release crediting Claude-assisted research methodology in the development cycle; (3) public commentary from Karpathy or Anthropic leadership on team progress beyond the launch-day framing; (4) Anthropic-attributed performance gains on community-authoritative benchmarks. Procurement-template implication: AI-vendor questionnaires should add a model-improvement-methodology disclosure field, and multi-year MSAs should add a research-roadmap-attestation clause requiring thirty-day advance notice on material methodology changes.","article_url":"https://agentmodeai.com/karpathy-joins-anthropic-cio-vendor-trajectory-read/","topic":"vendor-trajectory","pub_date":"2026-05-19","last_reviewed":"2026-05-19","next_review":"2026-08-17","verdict":"holding","verdict_history":[{"date":"2026-05-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-161","claim":"Senior IT leaders that frame internal AI-and-jobs communications at the job level ('will AI replace your role?') produce defensive postures from employees, lower reskill take-up, and the under-budgeted reskill line AM-109 documents. The task-level frame — which tasks shift on which horizon, which moats hold, which residual skills (agent output review, exception escalation routing, prompt and policy maintenance, vendor evaluation) the surviving role requires — is both more honest about what workers see at their desks and the only frame that resolves into the four skill gaps determining whether the post-displacement function actually works. The operational move for CIOs in 2026-2027 is to replace function-wide reassurance with task-level analytical resources teams can engage with directly, and to use those resources as the basis for role-by-role conversations rather than function-wide town halls. The task-level conversation should run before the reskill budget conversation, because the task inventory is the input the budget line needs.","article_url":"https://agentmodeai.com/how-ai-changes-jobs-task-level-frame/","topic":"enterprise-ai-cost","pub_date":"2026-05-20","last_reviewed":"2026-05-20","next_review":"2026-08-03","verdict":"holding","verdict_history":[{"date":"2026-05-20","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-162","claim":"Karpathy's seat at Anthropic will remain an IC/research-lead role — not a VP or exec-hierarchy title — through at least end of 2026.","article_url":"https://agentmodeai.com/karpathy-anthropic-bench-not-org-chart/","topic":"enterprise-ai-cost","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-12-31","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-163","claim":"SAP's Sapphire 2026 Autonomous Suite — which embeds more than 50 domain-specific Joule AI Assistants across finance, supply chain, HR, procurement, and CX, with Claude powering the finance, procurement, and supply chain agents — materially changes the ERP renewal calculus for enterprise CIOs: the AI agent layer is now inside the RISE with SAP and SAP GROW contract structure rather than a separately evaluated point-tool purchase, and most H2 2026 renewal teams have not yet built a line item for the point-tool displacement this creates.","article_url":"https://agentmodeai.com/sap-sapphire-joule-agents-erp-renewal-cio/","topic":"agent-procurement","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-08-20","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-164","claim":"Anthropic's May 21 2026 discussions with Microsoft to adopt Maia 200 inference chips, read alongside the same-day SpaceX filing disclosing a $1.25B/month compute contract through May 2029, reveals that the foundation-model inference stack is visibly diversifying from commodity Nvidia hardware to hyperscaler-proprietary silicon — a structural change that is currently invisible in standard enterprise AI vendor questionnaires and that introduces a triple dependency (model vendor, cloud provider, silicon provider) into the procurement risk map.","article_url":"https://agentmodeai.com/anthropic-microsoft-maia-chip-inference-infrastructure/","topic":"vendor-trajectory","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-07-21","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-165","claim":"The gap between AI initiative volume (97 percent of enterprises have active AI programmes, per D&B's 2026 AI Momentum Survey of 10,000 businesses across 32 countries) and AI data readiness (5 percent say their data is adequately ready) is not primarily a technology problem: it is a budget-allocation failure in which enterprises funded model-layer spend without funding the prerequisite data-infrastructure investment, and enterprises that correct the allocation by treating data infrastructure as the prerequisite rather than the follow-on will reach meaningful scale in 2026 before those that do not.","article_url":"https://agentmodeai.com/enterprise-ai-data-readiness-gap-scaling-problem/","topic":"enterprise-ai-cost","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-08-20","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-166","claim":"AI productivity gains create workforce reduction pressure through two structural mechanisms: demand ceilings in mature markets and competitive price compression from industry-wide AI adoption. Both mechanisms resolve independently in the direction of fewer workers for the same revenue, regardless of management intent.","article_url":"https://agentmodeai.com/ai-productivity-demand-ceiling-workforce/","topic":"enterprise-ai-cost","pub_date":"2026-05-23","last_reviewed":"2026-05-23","next_review":"2026-08-21","verdict":"holding","verdict_history":[{"date":"2026-05-23","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-167","claim":"A 2026 enterprise contracting for a third-party AI agent platform is, in almost every case, accepting a vendor-issued non-human identity into its environment with the authority to read, write, transact, and call further agents. Standard 2026 agentic AI master service agreements govern data classes, SLAs, pricing, and exit terms but do not govern the identity primitive the vendor uses for the agent's credentials, the rotation cadence, the customer's right to inventory and audit them, or the vendor's disclosure obligations if a credential class is compromised on the vendor side. The four procurement clauses that close the gap (identity primitive disclosure, rotation cadence and audit right, vendor-side breach disclosure, customer-side revocation control) are absent from the standard MSA templates of every major agentic AI platform reviewed in early 2026 and are present only in customer-redlined versions used by procurement-mature enterprises.","article_url":"https://agentmodeai.com/nhi-procurement-clause-gap/","topic":"non-human-identity","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-168","claim":"The dominant 2026 shadow-AI gap is not unsanctioned vendors but sanctioned vendors that have shipped agentic capabilities inside already-approved tenants without triggering a re-evaluation in the customer's SaaS approval process. The canonical examples are Microsoft 365 Copilot Studio inside approved M365 tenants, Slack AI and Slack agent platform inside approved Slack workspaces, Notion AI agents hub inside approved Notion workspaces, ServiceNow Now Assist inside approved ServiceNow tenants, Atlassian Rovo inside approved Atlassian estates, and Salesforce Agentforce inside approved Salesforce contracts. Existing 2024-era shadow-AI discovery playbooks (oriented to vendor-discovery and DLP egress detection) do not surface this class because the vendor is in-policy and no egress boundary is traversed. The structural fix is procurement-side: a re-evaluation trigger inside the SaaS approval policy that runs the original data-class and risk-assessment workflow against any new agentic capability inside an existing tenant within 30 days of activation.","article_url":"https://agentmodeai.com/approved-tool-unapproved-capability-shadow-ai/","topic":"shadow-ai-discovery","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-169","claim":"The three open agent protocols active in 2026 (Anthropic's Model Context Protocol, Google's Agent2Agent protocol contributed to the Linux Foundation, and Meta's Llama Stack) are not on a convergence trajectory, and the four major proprietary agentic platforms (Microsoft Copilot Agent, Salesforce Agentforce, SAP Joule, ServiceNow Now Assist) do not adopt any of the three as first-class defaults. The cost of being wrong on the model choice is low because swapping models is a configuration change. The cost of being wrong on the protocol choice is high because the locked asset is the tool inventory — the MCP servers, A2A endpoints, Llama Stack tool plugins, or proprietary extensions the customer has built or commissioned — and re-platforming the tool inventory is the long-tail engineering bill. Standard 2026 agentic AI MSAs do not include the three procurement clauses (protocol portability disclosure, tool inventory exit terms, protocol-roadmap commitment) that would price the protocol-roadmap optionality back to the customer.","article_url":"https://agentmodeai.com/agent-protocol-tax-mcp-a2a-llama-stack/","topic":"agent-procurement","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-170","claim":"As of mid-2026, the major enterprise agent platforms enable persistent agent memory with retention defaults, residency locations, encryption-at-rest ownership models, erasure-propagation pathways, and audit-evidence export capabilities that are not surfaced in standard procurement, leaving agent memory outside the enterprise data-retention register and the Article 30 record of processing activities. The compliance surface exists already (GDPR Article 5(1)(e) storage-limitation, Article 17 right to erasure, Article 30 records of processing, EU AI Act Article 12 record-keeping) but no AI-specific regulation has yet named persistent agent memory as a governed data class, leaving the obligation in force and the implementation gap unaddressed by procurement and identity-governance teams in most 2026 enterprises.","article_url":"https://agentmodeai.com/agent-memory-governance/","topic":"agentic-ai-governance","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-171","claim":"As of mid-2026, the majority of enterprises running production AI agents cannot terminate a misbehaving agent within their own stated incident-response window, because containment is specified as kill criteria in the risk register rather than built and tested as a runtime control plane with the four primitive actions (purpose binding, kill switch, network isolation, credential revocation). Kiteworks' 2026 Data Security and Compliance Risk Forecast measured the gap at 60% cannot terminate quickly, 63% cannot enforce purpose limitations, 55% cannot isolate networks, with the government-sector figures materially worse. Microsoft Agent 365 with Intune and Defender (GA 1 May 2026, runtime-controls preview from June 2026) is the first major-platform consolidation of the four primitives in a customer-administered control plane, which moves the question from engineering integration to procurement evaluation but does not resolve the cross-platform standardisation gap.","article_url":"https://agentmodeai.com/agent-kill-switch-containment-architecture/","topic":"agentic-ai-governance","pub_date":"2026-05-26","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-172","claim":"As of mid-2026, US enterprises do not need new federal AI legislation to be exposed on AI governance; AI-touching workflows already fall under HIPAA Security Rule access and audit controls (45 CFR 164.312), GLBA Safeguards Rule (16 CFR Part 314) access-control and incident-notification obligations, SEC cyber-disclosure rules (Item 106 of Regulation S-K and Item 1.05 of Form 8-K, with the 4-business-day materiality clock), and FTC Section 5 deception and unfairness jurisdiction at the data layer. The structural pattern, captured in the UK ICO's May 2026 'AI-powered cyber threats' guidance and the seven threat categories it names, is that AI governance has become data governance; the most common 2026 implementation gap is the fragmented audit log (Kiteworks 2026 Forecast: 33% lack audit trails entirely, 61% have fragmented logs), not the absence of regulation.","article_url":"https://agentmodeai.com/ai-governance-data-governance-us-frameworks/","topic":"regulatory-readiness","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-08-24","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-173","claim":"The aggregate water footprint of generative AI is small relative to common consumer products (a year of heavy ChatGPT use is on the order of 300 to 900 litres, against 15,415 litres per kilogram of beef and 2,495 litres per cotton t-shirt; data centers globally remain a small single-digit fraction of one percent of freshwater withdrawals, with agriculture at about 70% and industry at about 20%); the real governance concern is geographic concentration of consumption in water-stressed regions (Microsoft 2024 disclosure: 42% of consumption in water-stressed areas; Microsoft West Des Moines, Iowa Jul 2022 about 11.5M US gallons / about 6% of the city that month per AP 9 Sep 2023), not global query volume.","article_url":"https://agentmodeai.com/ai-water-use-in-context/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-08-25","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-174","claim":"The realistic year-one TCO of a security-platform agentic AI procurement at mid-sized SOC scale runs 4 to 7 times the order-form platform-fee line, decomposing across five cost components (platform fee, integration of SIEM and EDR and identity and ticketing telemetry, analyst retraining of the L1 and L2 SOC tier, tuning by the detection-engineering function in the first 6 months, and exit migration if the relationship ends); vendor-published ROI figures should be discounted by 30 to 50 percent against demo-environment bias, 20 to 40 percent against the customer's actual false-positive-adjusted alert backlog, and 60 to 80 percent against the survivorship bias in published case studies; the structural procurement instrument that prices these discounts at signing is a contractual 90-day in-environment paid-pilot evaluation gate that runs at production scope against four pre-defined customer baseline measurements (mean-time-to-triage, false-positive rate per detection class, analyst-hours per closed incident, backlog age), with a documented walk-away clause that returns the customer to the pre-pilot operating state without successor-platform commitment.","article_url":"https://agentmodeai.com/security-platform-agentic-ai-tco-roi/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-175","claim":"The 2026 Salesforce-platform-vs-Microsoft-platform AI procurement is a different conversation than the product-level Agentforce-vs-Copilot comparison and resolves on five comparison axes (data gravity against the customer's existing CRM and collaboration substrate; identity gravity against the standing IAM commitment with Microsoft Entra structurally advantaged for Microsoft-mature enterprises; developer-tooling gravity against the existing Power Platform versus Lightning + Apex developer population; regulatory-fit at the sector-specific certification layer; year-three exit cost dominated by data-egress, connector-mesh rebuild, and analyst-retraining) rather than on the per-seat or per-conversation headline pricing; the Salesforce stack (Data Cloud + Einstein + Agentforce + MuleSoft + Tableau) answers the customer-360-plus-customer-facing-agents question well, the Microsoft stack (Microsoft Graph + Fabric + Azure AI Foundry + Copilot + Power Platform + Entra + Purview) answers the knowledge-worker-productivity-plus-internal-workflow-agents question well, and the buying-committee mistake to avoid is treating the two platforms as substitutes when the larger 2026 enterprise procurement pattern is treating them as complements (Salesforce for customer-facing surfaces, Microsoft for internal-workflow surfaces) with the integration tax priced explicitly at signing rather than discovered in year two.","article_url":"https://agentmodeai.com/salesforce-platform-vs-microsoft-platform-ai/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-176","claim":"The 2026 enterprise agent-identity procurement choice between Okta and specialized NHI vendors is not a binary; Okta covers three slices of the NHI surface natively (workforce and managed-service-account lifecycle, session anomaly via October 2024 ITDR, privileged human access via 2025 Okta Privileged Access expansion), is partial on two (OAuth third-party app token governance and workload identity for cloud-native runtimes), and does not cover at GA the agent-runtime credential issuance against ephemeral workloads with sub-hour lifetimes that the specialist tier (Astrix Security for OAuth-app sprawl, Apono for just-in-time cloud access, Britive for multi-cloud privileged orchestration, Aembit for workload-to-workload SPIFFE-style attestation, Andesite for NHI runtime detection on top of existing SIEM, P0 Security for temporary access management with audit-trail evidence) is purpose-built for; the architecture-grade procurement output is one comparison matrix per identity class in scope (typically three: human workforce, managed service accounts, agent-runtime), named federation seams at four specific surfaces (identity-source authority, provisioning protocol with SCIM 2.0 the default, federation protocol with OIDC or SPIFFE the choice, audit-event format aligned to the customer's SIEM), and the procurement-side contractual instruments (AM-167 NHI procurement clause work) that make the federation enforceable at the MSA layer.","article_url":"https://agentmodeai.com/okta-vs-specialized-nhi-vendors/","topic":"non-human-identity","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-177","claim":"The 2026 'compare AI agent vendors regulated enterprise' procurement question resolves to four sector-specific vendor matrices, not one universal scorecard; each begins with a structural disqualification pass on the regulatory baseline (FedRAMP High plus DoD Impact Level for federal-civilian and defence; HIPAA BAA plus FDA 21 CFR Part 11 plus HITRUST plus EU AI Act Article 6/14 for healthcare and pharma; Federal Reserve SR 11-7 plus NYDFS Part 500 plus FINRA RN 24-09 plus SEC Rule 17a-4 for US financial services; NERC CIP plus EU NIS2 plus ISA/IEC 62443 plus regional reliability operator data-handling for energy and utility), populates a 5-or-6-row matrix per vertical against the surviving vendors, and translates the row gaps into contractually enforceable MSA addenda (the AM-167 NHI procurement-clause work instruments); the energy and utility matrix is structurally thinner than the other three because the OT/ICS overlay disqualifies most general-purpose agentic AI platforms in favour of the OT-specialist tier (Claroty, Dragos, Nozomi) with AI overlays.","article_url":"https://agentmodeai.com/agentic-ai-regulated-enterprise-vendor-matrix/","topic":"regulatory-readiness","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-178","claim":"The 2026 enterprise agentic AI orchestration-framework choice across the five major frameworks (AWS Bedrock AgentCore GA October 2025, Microsoft Azure AI Foundry plus Copilot Studio, Google Vertex AI Agent Builder plus the open-source Agent Development Kit, OpenAI Agent Builder GA October 2025 plus the open-source Swarm primitive, Anthropic Claude Agent SDK late 2024 plus the open-source Model Context Protocol) prices the decision as a 3-year orchestration-layer commitment along five comparison axes (orchestration primitive, tool-use protocol, deployment topology, observability tier, exit cost), with the framework choice resolved by gravity-fit against the customer's existing cloud, identity, and data substrate rather than by model-tier performance; the 2026 cross-vendor convergence on the Model Context Protocol as the tool-use standard is the structural change that makes exit cost newly tractable for customers who write tool definitions in the protocol and emit OpenTelemetry traces, materially lowering the year-three re-platforming cost relative to the framework-native alternatives.","article_url":"https://agentmodeai.com/aws-microsoft-google-openai-anthropic-frameworks/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-179","claim":"The 2026 enterprise AI infrastructure vendor SLA conversation resolves on five dimensions (uptime commitment with the denominator named explicitly, latency commitment at P95 and P99 negotiated into the MSA addendum because public SLAs typically omit it, support response tier per severity level, credit calculation scope and cap, exclusions list scope including scheduled-maintenance window, content-policy actions, capacity constraints, partial-availability events, and third-party-source outages); the publicly disclosed headline numbers (AWS Bedrock 99.9% monthly with 10/25/100% credit tiers, Azure OpenAI Service inheriting Azure platform 99.9% with PTU separate availability, Google Vertex AI 99.5%-99.9% varying per-model and per-region, OpenAI Enterprise 99.9%-99.99% per-customer in MSA, Anthropic Enterprise commitments per-customer with no public uniform tier) understate the year-two operational reality because exclusions list scope and credit calculation scope vary materially across vendors; the buying-committee discipline is to populate the per-vendor matrix at short-list rather than discover the gaps at year-one operational experience or year-two renewal.","article_url":"https://agentmodeai.com/enterprise-ai-vendor-sla-uptime-comparison/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-180","claim":"The 3-year IAM TCO envelope for a 2,000-employee mid-enterprise absorbing the agent-runtime identity class lands at roughly USD 1.5M to USD 3.05M, decomposed across five cost components (license, integration, operations, migration, exit) and three identity classes (human workforce, managed service accounts, agent-runtime); the platform-fee headline (Okta enterprise pricing per okta.com/pricing/) accounts for roughly 25-35% of the envelope, the remainder distributing across federation seam build-out, SCIM provisioning, OIDC/SAML configuration, audit-event format alignment, identity-operations team additions (typically 0.5 to 2.5 FTE), and the periodic-review cadence redesign for sub-hour credentials; three line items reliably unpriced in the year-one budget are the agent-runtime credential issuance integration (USD 100K-250K of first-year engineering recovered as year-two overage), the identity-operations team capacity uplift (USD 100K-200K annual), and the access-review redesign tooling or process work (USD 50K-150K).","article_url":"https://agentmodeai.com/agentic-iam-tco-mid-enterprise/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-181","claim":"The 2026 buying-committee diligence on an agentic AI vendor's strategic narrative resolves on seven proof points (named-customer references with segment-level revenue contribution; model-vendor relationships disclosed in the MSA at contractual rather than press-release level; engineering team tenure and turnover pattern as a leading indicator of narrative-product disconnect; post-revenue-recognition product-roadmap evidence comparing 12-month-prior commitments against 12-month actual ship; regulatory disclosure cadence covering SOC 2 Type II, ISO 27001 surveillance, sector-specific certifications, and public incident disclosure record; executive incentive structure as a structural read on what the vendor's leadership is trying to achieve over the 3-year MSA horizon; public technical-content cadence as downstream evidence of engineering depth); the pattern across roughly 30 vendor diligence cycles surfaced in 2025-2026 is consistent (vendors pass proof points one and two easily, fail or partially fail proof points three through five, split on six and seven); the buying committee that walks all seven systematically before the technical-feature comparison produces a structurally different diligence output and a 30-60% short-list reduction relative to the buying committee that anchors on the narrative alone.","article_url":"https://agentmodeai.com/vendor-strategic-narrative-proof-points/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-182","claim":"The 2026 Agentforce-vs-Microsoft-Copilot pricing decision resolves on the structural per-conversation versus per-user-seat pricing-model choice (Agentforce listing publicly at USD 2 per conversation with the 2025 Flex credit alternative; Microsoft 365 Copilot listing at USD 30 per user per month with Copilot Studio at USD 200 per tenant for 25,000 messages plus per-message overage) rather than the headline unit rate; enterprise-scale negotiation typically produces 30-50% per-conversation discount at committed volumes above 100,000 conversations per year on the Salesforce side and 15-25% per-user discount at committed seats above 1,000 on the Microsoft side, with multi-year commitment adding 5-15% per year at the cost of reducing year-one renegotiation leverage; three year-two renewal surprises (usage divergence from forecast, vendor pricing-model migration, bundle deconstruction) reliably surface against customers who priced only the order-form headline at signing.","article_url":"https://agentmodeai.com/agentforce-vs-microsoft-copilot-pricing/","topic":"enterprise-ai-cost","pub_date":"2026-05-27","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-183","claim":"The 2026 digital-transformation RFP that includes agentic AI procurement extends the existing 60-question RFP playbook (/the-enterprise-agentic-ai-rfp-60-questions/, 38 Copilot citations) with 15 UX assessment questions across five categories (interaction modes; error-handling and confidence-communication; accessibility against WCAG 2.2 + Section 508 + EN 301 549 + assistive-technology compatibility; multi-user and cross-device continuity; design-system extensibility), producing a 75-question total RFP that the buying committee uses with separate reviewer assignments (UX team plus accessibility office plus design-system stewards review the UX section); the procurement-mature pattern asks the vendor to demonstrate the answers to error-handling, accessibility, continuity, and extensibility questions rather than describe them, surfacing the gaps between strategic narrative and product capability before the MSA negotiation.","article_url":"https://agentmodeai.com/digital-transformation-rfp-ai-ux-questions/","topic":"agent-procurement","pub_date":"2026-05-27","last_reviewed":"2026-05-27","next_review":"2026-07-26","verdict":"holding","verdict_history":[{"date":"2026-05-27","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-184","claim":"The Digital Omnibus political agreement reached by the European Parliament and Council on 7 May 2026 postpones the EU AI Act's high-risk obligations to 2 December 2027 for standalone Annex III systems and 2 August 2028 for Annex I systems embedded in regulated products, and postpones the Article 50(2) provider watermarking duty to 2 December 2026, but leaves the deployer transparency obligations applicable from 2 August 2026 and leaves the GPAI obligations, the governance regime, the prohibited practices, and the AI literacy duty already in force, so an enterprise that reads the delay as a reason to pause its AI Act programme is misreading what still applies on and before 2 August 2026.","article_url":"https://agentmodeai.com/eu-ai-act-digital-omnibus-what-still-applies/","topic":"regulatory-readiness","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-08-26","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-185","claim":"In May 2026 the frontier model vendors began moving down the stack into systems integration: on 4 May 2026 Anthropic launched a roughly 1.5 billion dollar enterprise AI services company with Blackstone, Hellman and Friedman, and Goldman Sachs that embeds engineers inside mid-market and private-equity-owned companies rather than operating as a traditional consultancy, and OpenAI launched a parallel venture, the Deployment Company, with Bain Capital, Advent International, TPG, and Brookfield; the structural consequence for the enterprise buyer is that the model vendor, the integrator, and in the private-equity-owned case the company's own owner can be the same commercial interest, which changes the independence assumptions built into standard build-versus-buy and vendor-selection processes.","article_url":"https://agentmodeai.com/frontier-labs-as-systems-integrators/","topic":"vendor-trajectory","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-08-26","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-186","claim":"The Digital Omnibus postponement of the EU AI Act high-risk obligations to 2 December 2027 (Annex III standalone) and 2 August 2028 (Annex I embedded) re-times the conformity workstream but does not gate the readiness foundations beneath it. An enterprise running agentic AI should keep three foundations moving on the original timeline through and beyond 2 August 2026: a current inventory of which agents run under whose authority, agent-aware vendor contract terms, and active shadow-AI discovery. Each is either required by an obligation that did not move (the Article 50 deployer transparency duties applicable 2 August 2026, the GPAI and governance regime in force since 2 August 2025, and the Article 4 AI literacy duty in force since 2 February 2025) or is the prerequisite evidence base for the high-risk conformity work when it lands, and none of the three benefits from waiting for the new dates.","article_url":"https://agentmodeai.com/eu-ai-act-delay-readiness-agenda/","topic":"regulatory-readiness","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-08-26","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-187","claim":"A model's score on a coding benchmark such as SWE-bench is a weak predictor of its reliability on tasks that have no automatic verifier; enterprise model-maturity assessment therefore has to be measured on a second axis that headline leaderboards do not capture, namely common-sense robustness, run-to-run consistency, and the model's willingness to flag and correct its own errors.","article_url":"https://agentmodeai.com/car-wash-test-model-maturity/","topic":"vendor-trajectory","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-188","claim":"A special, sector-specific tax on AI justified by the arguments that it is built on the public's collective knowledge or that it runs on public infrastructure is the wrong instrument: both justifications lack a limiting principle and apply to every firm and prior technology, the legitimate grievances underneath them are better served by copyright markets and litigation (for uncompensated training data) and by the existing progressive corporate and capital-gains base (for concentrated profits), and a dedicated AI levy would fall on buyers and workers through prices and wages, entrench incumbents against startups and open source, and tax a productivity input at the moment economies most need it; the larger real tax distortion runs the other way, since the current code over-subsidises automation relative to labour.","article_url":"https://agentmodeai.com/why-an-ai-tax-is-the-wrong-instrument/","topic":"regulatory-readiness","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-189","claim":"Aggressive AI-driven workforce reduction is not, by itself, producing superior financial returns in the current cycle; across large enterprises the firms cutting deepest have shown returns close to those cutting least, which locates the return on agentic AI in retaining and amplifying the people who supervise autonomous systems rather than in headcount elimination.","article_url":"https://agentmodeai.com/ai-layoffs-roi-evidence/","topic":"enterprise-ai-cost","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-190","claim":"As measured by the 2026 Verizon Data Breach Investigations Report, AI is raising the throughput of cyberattacks rather than the underlying capability of attackers, because AI-assisted intrusions overwhelmingly scale known techniques rather than generate novel ones; the more consequential 2026 shift is that vulnerability exploitation has overtaken stolen credentials as the leading initial-access vector, and together these redirect enterprise defensive priority toward patch velocity and identity hygiene over hunting novel AI-authored threats.","article_url":"https://agentmodeai.com/ai-attacker-throughput-not-capability/","topic":"agentic-ai-governance","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-191","claim":"Within a roughly two-week window in May 2026, at least three of the four largest professional-services firms standardized their delivery organizations on a single AI model vendor, which makes the model an enterprise's auditor and implementation partner have adopted an input to that enterprise's own architecture decisions and a concentration risk that vendor-neutral strategy advice does not surface.","article_url":"https://agentmodeai.com/big-four-ai-model-concentration/","topic":"vendor-trajectory","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-08-27","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-192","claim":"ISO/IEC 42001:2023, the first certifiable artificial intelligence management system standard, has moved from a vendor marketing claim to an enterprise AI procurement checkpoint through 2025-2026, appearing as a stated or preferred requirement in regulated-sector and EU AI RFPs; the certificate is a necessary-not-sufficient screen (it attests to a governance management system, not to any specific model's safety, accuracy, or data provenance), so the buying-committee discipline is to require the evidence behind it (scope statement, Statement of Applicability, certification body and accreditation, validity dates, product-level AI risk assessment) and to pair it with a control-baseline mapping (NIST SP 800-53 / AI RMF) and the buyer's own product evaluation rather than treating the certificate as proof.","article_url":"https://agentmodeai.com/iso-42001-enterprise-ai-vendor-checkpoint/","topic":"agent-procurement","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-08-28","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-193","claim":"Enterprises mapping agentic AI onto NIST SP 800-53 today find real, recurring control gaps concentrated in four families (Access Control, Identification and Authentication, Audit and Accountability, and Supply Chain Risk Management) because the catalogue's implementation guidance assumes human-operated, deterministic systems rather than autonomous agents that hold delegated credentials, can be steered by untrusted input, and depend on a model-and-tool supply chain; NIST's COSAiS project (Control Overlays for Securing AI Systems) is writing single-agent and multi-agent overlays to close the gap, but finalized agent-specific guidance is not expected before 2027, so the interim burden is on the enterprise to identify the touched controls, document where standard guidance does not fit the agent case, and record compensating controls.","article_url":"https://agentmodeai.com/nist-cosais-sp-800-53-ai-agent-security-gap/","topic":"agentic-ai-governance","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-08-28","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-194","claim":"Enterprises that scale agentic AI without a dedicated inference FinOps discipline (workload-level cost allocation, spend-cap and budget-alert tooling, and model-routing policy) systematically under-budget production spend, because agentic workloads break the two assumptions cloud FinOps was built on: per-request cost is non-deterministic (token consumption varies with input and reasoning steps, and a single user request fans out into many model calls) and ownership is opaque (without tagging, inference arrives as one unattributable line item); the 2026 platform direction of cloud-native spend caps and AI cost-explainability confirms the gap is real but does not close it, because the missing layer is the operating discipline and a named owner, not the tooling.","article_url":"https://agentmodeai.com/agentic-ai-cost-governance-finops-enterprise/","topic":"enterprise-ai-cost","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-07-29","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-195","claim":"The May 2026 disclosures against AI coding agents (Adversa AI's TrustFall on 7 May 2026, a one-keypress remote code execution reaching Claude Code, Cursor, Gemini CLI, and GitHub Copilot CLI, and SymJack on 26 May 2026, a symlink-hijack confirmed against six agents that overwrites an agent's own configuration to plant a malicious MCP server, plus Microsoft's Semantic Kernel CVE-2026-26030 and CVE-2026-25592) share one design assumption, that showing an approval prompt is the same as obtaining informed consent, and because the coding agent executes attacker-supplied instructions with the developer's full credentials and write access to the build and deploy chain, it is a production attack surface that the enterprise should govern as a managed endpoint (inventory, deliberate version-pinning and patching, credential separation, monitoring for config-write-then-execute, and no untrusted repositories on credentialed machines) rather than as developer tooling outside the inventory.","article_url":"https://agentmodeai.com/ai-coding-agents-enterprise-attack-surface/","topic":"agentic-ai-governance","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-08-31","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-196","claim":"By mid-2026 the binding constraint on enterprise agentic-AI value had shifted from model capability, now a commodity any buyer can rent, to human deployment capacity, the forward-deployed engineer who integrates a model into one company's exceptions, legacy systems, and undocumented processes; because that capacity sits with the vendor, the forward-deployed-engineer-led delivery model converts what looks like a software purchase into a professional-services engagement with vendor-operability lock-in, so the buyer's defensible response is to classify and govern the spend as professional services, contract knowledge-transfer milestones with acceptance tests, build internal counterpart capacity, and require operability and exit terms, rather than treat the engagement as a delivery convenience.","article_url":"https://agentmodeai.com/forward-deployed-engineers-the-ai-bottleneck/","topic":"vendor-trajectory","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-08-31","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-197","claim":"As of mid-2026 there is no US federal AI floor coming on a plannable timeline (the White House National Policy Framework for AI of 20 March 2026 is explicitly non-binding and would require Congressional action, and the proposed federal moratorium on state AI laws was not enacted), and the most-watched comprehensive state law retreated rather than advanced (Colorado's SB 26-189, signed 14 May 2026, repealed and reenacted the Colorado AI Act, removing the algorithmic-discrimination duty of care and the risk-management and impact-assessment obligations and moving the effective date from 30 June 2026 to 1 January 2027), so the operative reality is a non-converging state patchwork, and the defensible enterprise posture is to build to the strictest obligation that actually applies to its own deployments and treat the regulatory map as a moving input rather than waiting for a federal floor.","article_url":"https://agentmodeai.com/us-ai-regulation-federal-state-standoff/","topic":"regulatory-readiness","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-08-31","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-199","claim":"Across the publication's first cohort of tracked claims, the ones that moved to Not holding were hype-register slugs inherited from the pre-standard era rather than errors of analysis, and pricing or model-tier claims were the fastest to require correction.","article_url":"https://agentmodeai.com/enterprise-ai-claims-what-held-up/","topic":"vendor-trajectory","pub_date":"2026-06-03","last_reviewed":"2026-06-03","next_review":"2026-07-18","verdict":"holding","verdict_history":[{"date":"2026-06-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-200","claim":"By mid-2026 the major enterprise agentic-AI platforms have converged on a common primitive set (agent builder, MCP tool integration, a policy gateway, and observability), so the durable enterprise selection criterion shifts from model capability to the auditability of each vendor's accountability surface: a published model-version and deprecation policy, SLA specificity, and compliance documentation.","article_url":"https://agentmodeai.com/enterprise-ai-vendor-comparison/","topic":"vendor-trajectory","pub_date":"2026-06-03","last_reviewed":"2026-06-03","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-06-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-201","claim":"In 2026 the enterprise-AI cost question that matters is total cost of ownership measured against realised ROI, not headline seat price; and across four independent datasets (Stanford DEL's 12% clearing 300%+ ROI vs 88% at or below break-even, McKinsey's 23% scaling and 17% self-reported EBIT, Gartner's 28% fully paying off, MIT NANDA's 95% of pilots with no measurable P&L impact) the high-return minority is separated from the majority by measurement discipline and operational preconditions, not by model capability or vendor choice.","article_url":"https://agentmodeai.com/enterprise-ai-cost-and-roi/","topic":"enterprise-ai-cost","pub_date":"2026-06-04","last_reviewed":"2026-06-10","next_review":"2026-07-15","verdict":"partial","verdict_history":[{"date":"2026-06-04","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One of four named datasets unanchored on review. The claim text names 'Stanford DEL's 12% clearing 300%+ ROI vs 88% at or below break-even' as one of four independent datasets. Full-text verification on 10 Jun 2026 found the Stanford DEL Enterprise AI Playbook contains no such distribution — it studies 51 successful deployments by design and carries no ROI-realisation failure data (full finding at AM-029, correction of 10 Jun 2026). The McKinsey (23% scaling, 17% EBIT-attribution), Gartner (28% fully paying off), and MIT NANDA (95% no measurable P&L impact) datasets verify; the claim's spine stands on three datasets rather than four. The only verified figure carrying the 12/88 numerals is IDC's pilot-graduation finding (roughly 88% of AI proof-of-concepts never reach production; via CIO.com, Mar 2025), a different metric from an ROI distribution. Status Up -> Partial."}],"primary_sources":[]},{"id":"AM-202","claim":"Microsoft's 1 May 2026 launch of Microsoft 365 E7 (the Frontier Suite) at $99 per user per month, combined with the 1 Jul 2026 increases to the E3 base ($36 to $39) and E5 base ($57 to $60), restructures Microsoft 365 economics so that the enterprise AI-licensing decision becomes a platform-tier decision taken at renewal rather than a Copilot add-on bought separately, and customers who decline Copilot still face a higher base cost.","article_url":"https://agentmodeai.com/microsoft-365-e7-frontier-suite-cost/","topic":"enterprise-ai-cost","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-20","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-203","claim":"Anthropic's $965 billion Series H valuation, closed 28 May 2026 and overtaking OpenAI's $852 billion (31 Mar 2026), marks the consolidation of frontier enterprise AI into a hyperscaler-backed top two, which shifts the binding risk in a multi-year Claude or GPT contract from model capability to vendor pricing power and operational switching cost, making contractual exit terms rather than benchmark wins the procurement variable that matters.","article_url":"https://agentmodeai.com/anthropic-valuation-vendor-risk/","topic":"vendor-trajectory","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-09-03","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-204","claim":"By mid-2026 non-human identities (service accounts, API keys, OAuth tokens and AI-agent credentials) outnumber human identities by roughly an order of magnitude (Cloud Security Alliance: an average of 45 to 1, up to 144 to 1 in cloud-native environments) while most enterprises lack any documented policy to provision or retire them (78% per CSA), making NHI the fastest-growing unmanaged enterprise attack surface, and the binding first control is an inventory with owner and lifecycle, not additional perimeter security.","article_url":"https://agentmodeai.com/non-human-identity-governance-vacuum/","topic":"non-human-identity","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-09-03","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-205","claim":"Enterprises systematically overestimate their visibility into AI agents (Cloud Security Alliance, Apr 2026: 82% had discovered at least one AI agent running without their security or IT team's knowledge in the past year while 68% believed they had strong visibility, with only 21% running any formal agent decommissioning process), and because a written policy cannot be enforced against agents nobody can see, continuous discovery rather than policy is the binding first control.","article_url":"https://agentmodeai.com/shadow-ai-discovery-visibility-gap/","topic":"shadow-ai-discovery","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-09-06","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-206","claim":"The May 2026 wave of enterprise AI-security moves (Zscaler agreeing to acquire Symmetry Systems on 21 May, Snowflake agreeing to acquire Natoma on 27 May, and Microsoft bringing Agent 365 to general availability on 1 May at $15 per user per month) signals that the contested layer in enterprise AI security has moved from the network to the agent-to-data access graph, so buyers should evaluate platforms on whether they can map which AI agent accesses which data, by what path, rather than on network controls.","article_url":"https://agentmodeai.com/ai-security-acquisitions-agent-access-graph/","topic":"vendor-trajectory","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-09-06","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-207","claim":"The 2 Jun 2026 US executive order 'Promoting Advanced Artificial Intelligence Innovation and Security' concentrates its mandatory weight on the federal government (CISA binding operational directives on AI-enabled cyber defence within 30 days, an NSA/CISA/Treasury AI cybersecurity clearinghouse within 30 days, OPM cyber hiring within 60 days) while making frontier-model pre-release government review explicitly voluntary, and the binding federal directives will move the de-facto AI-cyber baseline that auditors, insurers and customers measure enterprises against even though the order compels no private firm.","article_url":"https://agentmodeai.com/white-house-ai-executive-order-security/","topic":"regulatory-readiness","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-23","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-208","claim":"The SpaceX IPO filings (S-1 of 20 May 2026 and the 5 Jun 2026 free writing prospectus) disclose a circular frontier-compute economy in which Anthropic pays roughly $1.25 billion a month and Google roughly $920 million a month for GPU capacity in rival xAI's data centers — about $26 billion a year flowing into an AI segment that reported a $6.355 billion operating loss for 2025, on leases both cancellable at 90 days' notice — which makes compute supply, not model capability, the binding constraint to plan around in enterprise AI roadmaps.","article_url":"https://agentmodeai.com/xai-ipo-compute-disclosures/","topic":"vendor-trajectory","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-24","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-209","claim":"Agentic AI in banking moved from pilot to named production deployments in H1 2026 (the FIS×Anthropic Financial Crimes AI Agent announced 4 May 2026, compressing AML investigations from days or hours to minutes, with BMO and Amalgamated Bank in active development toward H2 2026 general availability; Lloyds Banking Group's 40,000-licence Microsoft 365 Copilot estate at 97% active use), and the shared deployment pattern is decision-preserving: agents compress evidence-assembly and draft the case narrative while humans retain the legally consequential filing decision — the configuration that survives regulatory scrutiny.","article_url":"https://agentmodeai.com/agentic-ai-banking-aml-deployments/","topic":"agentic-ai-governance","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-09-07","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-210","claim":"Agent washing — Gartner's term for the rebranding of existing products such as AI assistants, RPA and chatbots as agentic AI without substantial agentic capabilities — is the dominant noise source in the 2026 agentic market (Gartner's April 2026 Hype Cycle assesses only about 130 of the thousands of self-described agentic vendors as delivering real capability, against 17% of organisations deployed and more than 80% intending to deploy within two years), and the buyer's defence is a pre-contract capability test for goal-directed multi-step autonomy, tool-based action, carried state and deviation handling, rather than reliance on vendor labels.","article_url":"https://agentmodeai.com/what-is-agent-washing/","topic":"agent-procurement","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-09-07","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-211","claim":"Siemens' Eigen Engineering Agent (launched at Hannover Messe on 20 Apr 2026, available to the 600,000+ users of TIA Portal after pilots at 100+ companies in 19 countries, with vendor-reported gains of up to 50% in automation-engineering efficiency and 2-5x faster execution) marks manufacturing's first at-scale production agentic layer landing in engineering design — PLC coding, HMI visualisation, device configuration — rather than runtime operations, because design iteration is cheap and reversible while runtime carries safety cases, and the vendor's up-to figures are pilot hypotheses for buyers to test, not business cases.","article_url":"https://agentmodeai.com/agentic-ai-manufacturing-engineering/","topic":"agentic-ai-governance","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-212","claim":"AI observability — per Gartner's two-part definition, the characteristic of systems being understandable from their outputs, extended by dedicated tools that manage and assess the behaviour, decision-making and risks of an AI solution such as model drift, bias and LLM logic — is a distinct discipline from classic application monitoring because AI fails semantically (drift, bias, opaque reasoning) while APM watches infrastructure and application health, and with Gartner predicting 40% of AI-deploying organisations will run dedicated AI observability tools by 2028 from a nascent base, the CIO-grade sequence is to define wrong-outcome metrics and measured detection time before buying tooling.","article_url":"https://agentmodeai.com/what-is-ai-observability/","topic":"agentic-ai-governance","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-213","claim":"Per IDC research commissioned by Lenovo (CIO Playbook 2025, February 2025; global survey n=2,920), 88% of enterprise AI proof-of-concepts fail to reach production: for every 33 POCs a company launched, only four (roughly 12%) graduated to widescale deployment, and the graduation gap is an organizational-readiness outcome (unclear ROI, insufficient AI-ready data, lack of in-house AI expertise) rather than a model-capability outcome.","article_url":"https://agentmodeai.com/why-88-percent-of-agentic-ai-deployments-fail/","topic":"enterprise-ai-cost","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-09-08","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-214","claim":"The publication's correction record for the fabricated 'Stanford DEL 12/88' statistic is complete across the written corpus and public as of 10 Jun 2026: eight claim verdicts changed with dated, append-only correction entries (AM-029 to Not holding; AM-024, AM-031, AM-040, AM-042, AM-129, AM-132 and AM-201 to Partial), the Claim Archive's first retraction (ACA-2026-003), roughly 120 occurrences across 30 published article bodies restated on the verified IDC/Lenovo graduation figure or softened to qualitative language, and the signature article restated at its original URL under new claim AM-213, with no occurrence silently deleted.","article_url":"https://agentmodeai.com/anatomy-of-a-fabricated-statistic/","topic":"agentic-ai-governance","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-215","claim":"Claude Fable 5, released 9 Jun 2026 at $10/$50 per million tokens (twice Opus 4.8), can refuse a request mid-call in three restricted domains (cybersecurity, biology and chemistry, model distillation) and fall back to Opus 4.8, which makes deploying it a runtime-reliability problem (refusal handling, per-call capability variance, and variable billing) rather than a one-time procurement decision.","article_url":"https://agentmodeai.com/claude-fable-5-enterprise-fallback-reliability/","topic":"vendor-trajectory","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-09-12","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-216","claim":"The EU AI Act's enforcement calendar splits in two: general-purpose AI model-provider obligations become directly enforceable on 2 Aug 2026 (finable up to 3% of global annual turnover or €15M), while the provisionally agreed Digital Omnibus defers high-risk deployer obligations for stand-alone Annex III systems to 2 Dec 2027, so enterprises face their AI vendors' new compliance demands roughly 16 months before their own high-risk deadline arrives.","article_url":"https://agentmodeai.com/eu-ai-act-gpai-enforcement-vendor-flow-down/","topic":"regulatory-readiness","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-09-12","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-217","claim":"IBM's June 2026 study of 2,000 technology executives shows a structural AI control gap: 66% are accountable for AI systems they do not fully control, and the organisations that embed control directly into their AI systems report materially better outcomes (18% higher operating margins, 25% fewer incidents, four times less wasted AI budget) than those relying on manual, bolt-on governance.","article_url":"https://agentmodeai.com/ai-control-gap-cio-accountability/","topic":"agentic-ai-governance","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-09-12","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"AM-CANON-001","claim":"AI-written commentary can be more verifiable, not less, than human-written commentary — when it is published inside an explicit accountability architecture with six components: disclosure, claim isolation, verdict tracking, dated retraction, primary-source pinning, and review cadence.","article_url":"https://agentmodeai.com/canon/v1.0/accountability-architecture/","topic":null,"pub_date":"2026-05-14","last_reviewed":"2026-05-14","next_review":"2026-08-12","verdict":"holding","verdict_history":[{"date":"2026-05-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-001","claim":"For a 4–10 person ops team running ~50 automations including five agentic steps in 2026, the platform choice is binary between n8n self-hosted and Make.com Pro, decided by whose time pays for the platform; Zapier earns its cost only when a critical integration is vendor-locked.","article_url":"https://agentmodeai.com/operators/n8n-vs-make-com-vs-zapier/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-002","claim":"For a 5-person consultancy already on either Notion or ClickUp in 2026, the AI features alone do not justify a workspace switch; the bundling difference (Notion bundles AI into Business at $19.50/seat, ClickUp Brain is a separate $9/seat add-on) makes the platform-shape choice (doc-centric vs project-centric) the actual decision.","article_url":"https://agentmodeai.com/operators/notion-ai-vs-clickup-ai-consultancy/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-05-28","verdict":"partial","note":"Price drift: Notion Business with bundled AI now about $15/seat annual ($20 monthly) vs cited $19.50; ClickUp Brain now $7/seat vs cited $9. Verdict logic unchanged; figures need updating."}],"primary_sources":[]},{"id":"OPS-003","claim":"For a solo founder choosing exactly one consumer AI subscription at around $20/month in 2026, the choice between Claude Pro and ChatGPT Plus is workflow-shape (long-document review and code favour Claude Pro; voice mode, image generation, and integration breadth favour ChatGPT Plus) — not capability-rank, which both vendors trade leadership on monthly.","article_url":"https://agentmodeai.com/operators/claude-pro-vs-chatgpt-plus-solo-founder/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-005","claim":"At sub-1M tokens per month (typical SMB agent volume) in 2026, the absolute dollar gap between Claude Haiku 4.5, GPT-4o-mini, and Gemini 2.5 Flash is small enough (≤$3/month) that price is the wrong tiebreaker; tool-use reliability, instruction-following on long context, and ecosystem fit determine the right cheap-tier model per workload shape.","article_url":"https://agentmodeai.com/operators/anthropic-vs-openai-vs-gemini-api-smb/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-05-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-011","claim":"If a candidate first-AI-agent use case at an SMB cannot answer all four of (a) what does success look like in numbers, (b) who owns it on Monday, (c) what breaks if it fails silently, (d) what is the rollback — the use case is not ready to deploy, regardless of vendor demo quality or model capability.","article_url":"https://agentmodeai.com/operators/picking-first-ai-agent-small-business/","topic":"operators-ai-tooling","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-014","claim":"An SMB AI vendor evaluation defensible to the typical cyber-insurance reasonable-care expectation can be completed in 90 minutes by walking through five questions in order — model provenance, data residency, sub-processor list, breach history, termination clause — each answered from the vendor's public site or the contract about to be signed.","article_url":"https://agentmodeai.com/operators/ai-vendor-due-diligence-small-business/","topic":"operators-ai-compliance","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-021","claim":"Across the published 2026 small-bookkeeping AI corpus (Xero OS, Intuit Assist, Canopy AI Notetaker, Digits MCP Server, with CPA Practice Advisor as the trade-press source), AI now reliably handles five recurring grind workflows at 1-to-5-person firm scale (bank-feed categorisation, receipt OCR, recurring journal posting, sales-tax reconciliation, AR ageing emails), but the judgement-call workflows (period close, advisory conversations, audit defence) remain human-led.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-small-firm-case-study/","topic":"operators-ai-economics","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-022","claim":"Across the published 2026 small-law-firm AI corpus (Spellbook with named small-firm customers Westaway, KMSC Law, Polley Faith; Harvey AI with mid-size roster Thompson Hine through Lowenstein Sandler; GC AI as named Anthropic enterprise customer claiming 1,500 companies and 14 hours/week saved), AI now ships at 1-to-20 lawyer-firm scale for contract drafting, document review at scale, and legal research with citation, but privileged-content workflows still require Enterprise-tier model access with zero-data-retention contractual posture per ABA Formal Opinion 512.","article_url":"https://agentmodeai.com/operators/ai-small-law-firm-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"partial","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Vendor attribution error in the claim text. The claim names Polley Faith among 'Spellbook with named small-firm customers Westaway, KMSC Law, Polley Faith'. Polley Faith LLP is a Harvey-listed law-firm customer, not a Spellbook customer: the live Spellbook site (now spellbook.com; spellbook.legal 301-redirects) names Westaway, KMSC Law, and McInnes Cooper with no Polley Faith, and the source article's own body correctly places Polley Faith on Harvey's roster — the claim text and the article excerpt bundled it with the wrong vendor at publish. The remaining legs verify against extracted source text on 10 Jun 2026: Anthropic's GC AI customer story carries 'More than 1,500 companies' and '14 hours saved per week on average ... based on a survey of more than 100 active customers' verbatim; Harvey's published roster (Thompson Hine, Fox Rothschild, Lowenstein Sandler, Polley Faith) matches; ABA Formal Opinion 512 remains the governance baseline. The corpus reading (AI ships at 1-to-20 lawyer scale; privileged work stays on Enterprise-tier zero-retention access) is unaffected. Status Up -> Partial."}],"primary_sources":[]},{"id":"OPS-026","claim":"The published 2026 construction-AI case corpus is overwhelmingly vendor-led (Procore, Autodesk Construction Cloud, OpenSpace, Buildots, Doxel) with thin named small-contractor self-published cases. Reading the vendor corpus honestly, three workflows now show consistent under-100-employee contractor AI deployment (estimating speed via takeoff acceleration, schedule risk surfacing, as-built reality capture); a fourth (AI safety detection) remains structurally biased toward larger sites with the camera coverage and safety officer to act on alerts.","article_url":"https://agentmodeai.com/operators/ai-small-construction-firm-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-027","claim":"Across the published 2026 dental-AI case corpus (Pearl with FDA-cleared 2D and 3D radiography AI plus 23,000 published practices; Overjet with 21+ named small-and-family-practice customers including Promenade Center, Quest Dental, Midtown Dental Studio), AI now ships at 1-to-3-dentist practice scale for FDA-cleared radiography assist, insurance verification automation, and patient-education visualisation; ambient voice AI for clinical notes is the next surface to ship widely.","article_url":"https://agentmodeai.com/operators/ai-small-dental-practice-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-028","claim":"The published 2026 small-beauty-salon AI case-study corpus is materially thinner than dental, legal, or bookkeeping (booking platforms publish customer counts but rarely individual-salon AI-attributable outcomes; solo stylists who use AI share informally on Instagram and TikTok rather than in case-study form). Reading the platform corpus honestly, the 2026 working pattern at 1-to-5 chair scale concentrates on no-show reduction via deposits, marketing copy via consumer-tier AI assistants, and portfolio/look generation via Canva and similar tools. AI-driven hairstyling recommendation, voice-AI booking, and dynamic pricing are not yet at the published-case-density that supports a small-salon recommendation.","article_url":"https://agentmodeai.com/operators/ai-small-beauty-salon-case-study/","topic":"operators-ai-verticals","pub_date":"2026-04-26","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-04-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-029","claim":"For solo founders and small teams (under ~50 people) building with AI in 2026, the build-vs-buy decision tree has inverted: specification, not engineering capacity, is now the bottleneck. The teams that can describe their workflow in operational detail can ship things they could not previously afford to build; the teams that cannot still cannot ship, regardless of how good the AI tooling is.","article_url":"https://agentmodeai.com/operators/three-launches-with-ai-the-lessons/","topic":"operators-ai-tooling","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-030","claim":"The fastest path for an owner-operator to build practical agentic-AI competence in 2026 is the three-week build-by-shipping protocol — specification + scaffolding + ship + connect + deploy + iterate, against a real workflow, with one external user — not formal study or consulting engagement. The protocol produces more transferable competence than published comparable courses on three measurable outcomes: operational decisions the operator can make after, debugging capability without external help, and calibration on when to build versus buy.","article_url":"https://agentmodeai.com/operators/using-ai-to-learn-ai-operator-playbook/","topic":"operators-ai-tooling","pub_date":"2026-04-28","last_reviewed":"2026-04-28","next_review":"2026-06-27","verdict":"holding","verdict_history":[{"date":"2026-04-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-031","claim":"Solo founders evaluating AI bookkeeping in 2026 face three realistic options: a fully-managed AI-augmented service (Bench, Pilot), a software-led tool that does AI categorisation inside an existing accounting product (QuickBooks Live, Xero with Hubdoc), or a DIY stack (Claude/ChatGPT + a spreadsheet template). The fully-managed option scales when revenue passes ~$30K MRR; below that, the DIY stack with a 30-min monthly review beats both software-led and managed. The failure mode is paying for managed-service automation while still doing 80% of the categorisation yourself because the AI hasn't seen enough of your transaction patterns yet.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-for-solo-founders/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 covering the three-option market split (fully-managed / software-led / DIY) and the ~$30K MRR threshold for fully-managed to become net-positive. REVIEW: Peter to verify current Bench and Pilot entry-tier pricing on or before 13 Jun 2026; if either has launched a sub-$100/month tier the threshold call shifts."}],"primary_sources":[]},{"id":"OPS-032","claim":"For SMB content workflows in 2026 (blog drafts, weekly newsletter, social copy, email sequences) at a 1-to-10 person business shipping two-to-four pieces per week, the practitioner read is workflow-shape not capability-rank: Claude wins on long-form editorial voice and structured drafting; ChatGPT wins on speed-and-iteration plus image generation in the same conversation; Gemini wins on Google-stack integrations. Paying for all three Plus tiers (around $60/month) without a deliberate task split is the expensive failure mode.","article_url":"https://agentmodeai.com/operators/chatgpt-vs-claude-vs-gemini-smb-content/","topic":"operators-ai-tooling","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 with status=partial. Recommendation derived from vendor pricing pages 29 Apr 2026 + public eval leaderboards + practitioner write-ups, not from a tracked SMB-cohort replication. Promotes to Holding once two consecutive 45-day reviews replicate the workflow-shape split on a real operator sample. REVIEW: Peter."}],"primary_sources":[]},{"id":"OPS-033","claim":"AI customer-service automation at 1-10 employee scale clears net-positive only when 70% or more of weekly inquiries are repetitive, low-stakes, and factually resolvable (hours, pricing, simple status). Below 50% the trust-erosion and remediation cost exceeds the headcount saving; between 50% and 70%, the answer turns on whether responsiveness is the brand differentiator.","article_url":"https://agentmodeai.com/operators/ai-customer-service-small-business/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Break-even thresholds (70/50) and never-deflect list are editorial synthesis from cited platform docs and CS-automation research, not a primary-data study. REVIEW: Peter to validate against any first-party SMB deployment data he has access to before status promotion to Holding."}],"primary_sources":[]},{"id":"OPS-034","claim":"For a solo founder processing 100-300 emails a day in 2026, the cheap-stack option (Gmail labels + Claude Pro at $20/month + a 5-line prompt template) recovers roughly 90% of the value of an $83/month premium stack (Superhuman AI + Shortwave Pro + Reclaim.ai Pro) at about 24% of the cost. The premium stack is worth its price under three conditions only — 2+ hours/day in email, keyboard-shortcut speed gain that pays back at the founder's hourly rate, and a documented bottleneck the cheap stack failed to solve after a two-week trial. Without all three, the founder is paying for an aesthetic, not measurable productivity.","article_url":"https://agentmodeai.com/operators/solo-founder-email-triage-ai-stack/","topic":"operators-ai-tooling","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026 with status=partial. Cost-side claims (vendor pricing) verifiable against the four cited pricing pages on the publication date. Time-recovery claim (90+ min compressed to ~20 min) drawn from published productivity-blogger benchmarks rather than Peter-run measurement; first-cohort replication on the publication's tracked operator cohort due by 13 Jun 2026. REVIEW: Peter."}],"primary_sources":[]},{"id":"OPS-035","claim":"There are five categories of small-business work where AI substitution in 2026 costs more in trust and liability exposure than it saves in productivity: (1) signed legal documents and tax-return positions, (2) trust-laden customer touchpoints (cancellations, refunds, conflict de-escalation), (3) regulatory submissions where the human signature is the audit trail, (4) anything requiring genuine domain credentialing (medical advice, licensed financial advice, signed engineering work), and (5) the first six conversations with a new high-value client.","article_url":"https://agentmodeai.com/operators/when-not-to-use-ai-for-small-business/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Status set to Partial at publication because category 5 lacks the same regulatory/cited-consequence anchor as categories 1-4. REVIEW: Peter to confirm category 5 evidence base and either upgrade to Holding (with strengthened citation) or amend the claim to four categories."}],"primary_sources":[]},{"id":"OPS-036","claim":"An SMB AI policy that actually changes day-to-day behaviour fits on one page and contains exactly eight clauses — sanctioned tools, prohibited data, human-review gate, client disclosure rule, prohibited uses, incident-report path, review cadence, and signature line — each closing a failure mode currently surfacing in regulatory guidance, court records, and breach disclosures through 2025-2026.","article_url":"https://agentmodeai.com/operators/1-page-ai-policy-for-small-business/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"partial","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."},{"date":"2026-04-29","verdict":"partial","note":"Initial publication 29 Apr 2026. Status set to Partial at publication because clause 6 commentary references an order-of-magnitude remediation-cost gap derived from the IAPP 2024 AI Governance Profession Report; the report characterises the gap as material but does not publish a precise multiple, so the wording is annotated source: our-estimate. REVIEW: Peter to source a precise figure or amend the commentary."}],"primary_sources":[]},{"id":"OPS-037","claim":"AI-drafted invoices for EU SMB operators in 2026 fail VAT audit at higher rates than human-drafted invoices specifically on cross-border treatment (OSS scheme wording, reverse-charge language, customer VAT-status verification), because LLM training data underweights post-2021 e-commerce VAT rules. The fix is a 4-line VAT-compliance prompt prefix that names the operator's VAT registration, the customer's VAT status, and the applicable scheme; most SMB invoicing tooling does not ship this by default.","article_url":"https://agentmodeai.com/operators/ai-invoicing-vat-compliance-small-business/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-038","claim":"SMB AI-VA deployments displacing admin work in collective-agreement-covered sectors (Dutch CAO, German Tarifvertrag, French Convention Collective) trigger collective-agreement provisions even at sub-10-employee scale in 2026, via job-classification-displacement and technology-introduction-consultation channels. Most SMB owners are unaware until the first union audit; FNV / DGB / IG Metall / CFDT activity in this area has shifted from theoretical to operational since 2024.","article_url":"https://agentmodeai.com/operators/ai-va-small-business-collective-agreement/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-039","claim":"AI-drafted contracts in EU notary-required jurisdictions (NL, DE, AT, BE, CH) are producing a class of legal-malpractice incidents in 2026 where the SMB owner treats an AI draft as final binding document, missing the notarisation requirement for real-estate transfers, GmbH/BV share transfers, and certain marriage/inheritance instruments. The fix is a 30-second pre-signing check on transaction-type and jurisdictional notarial-form requirement; AI tooling does not flag this by default.","article_url":"https://agentmodeai.com/operators/ai-drafted-contracts-notary-requirement-eu/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-040","claim":"Dutch ZZP'ers losing recurring client work to AI replacement in 2026 sit outside the WW (Werkloosheidswet) safety net entirely and find that available AOV (arbeidsongeschiktheidsverzekering) products mostly exclude demand-side income loss; the structural gap is pushing affected ZZP'ers into bijstand at faster rates than the 2024 baseline. The realistic options are operational (client-base diversification, offer restructuring, larger liquid buffer), not insurance-based.","article_url":"https://agentmodeai.com/operators/zzp-ai-displacement-unemployment-gap-nl/","topic":"operators-ai-economics","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-041","claim":"SMB owners using AI to produce marketing content are hitting platform algorithmic penalties at increasing rates in 2026, with platform-specific enforcement: Google Helpful Content system + March 2024 spam policy update target scaled-content-without-E-E-A-T; LinkedIn feed-distribution deprioritises fully-AI-generated content while tolerating AI-assist; Etsy listing-policy enforcement is heavier than either, with category-specific AI prohibitions. The defensible cross-platform posture is AI drafts + human edits + human signature with sustainable cadence.","article_url":"https://agentmodeai.com/operators/platform-algorithm-ai-content-penalties/","topic":"operators-ai-compliance","pub_date":"2026-04-29","last_reviewed":"2026-04-29","next_review":"2026-06-30","verdict":"holding","verdict_history":[{"date":"2026-04-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-042","claim":"For under-100-employee construction firms in 2026, the AI procurement order is estimating + bidding tools first (Togal.AI for general takeoff; Procore Copilot if already on Procore), with visual-progress capture (Buildots, OpenSpace) deferred until project portfolio exceeds 8 simultaneous projects per project manager. The vendor pitch oversells visual capture and undersells the takeoff workflow where the actual hours go (35-45% of estimator/PM time on bidding work, 5-10% on jobsite walkthroughs).","article_url":"https://agentmodeai.com/operators/ai-construction-estimating-bidding-tools/","topic":"operators-ai-verticals","pub_date":"2026-05-03","last_reviewed":"2026-06-17","next_review":"2026-08-16","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-043","claim":"For solo founders under €5K MRR running 20-80 customer-service tickets per week in 2026, the cheap stack (shared inbox host + Claude Pro at €20/month + a copy-paste prompt-pack, total under €40/month) is structurally cheaper than the dedicated AI helpdesks (Intercom Fin, Crisp AI, Tidio Lyro) until ticket volume sustains above ~200/week. Above that threshold, the per-resolution and per-conversation pricing on the dedicated platforms starts to compete; below it, the cheap stack wins on cost AND on operator experience. The volume threshold is the procurement signal, not the vendor pitch.","article_url":"https://agentmodeai.com/operators/solo-founder-customer-service-ai-stack/","topic":"operators-ai-tooling","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-044","claim":"For appointment-driven local-service businesses in 2026 (hairdresser, plumber, garage, cleaner, beautician), the AI value concentrates in two workflows neither booking-platform AI feature serves well: no-show reduction via personalised SMS sequences (3rd-party SMS API on top of the booking platform's webhook, typical 30-50% no-show reduction in published case studies) and review generation (post-appointment SMS or WhatsApp, typical 3-5x review-completion lift). The booking-platform decision (Booksy, Square Appointments, Treatwell, Vagaro) is shaped by customer-discovery model and existing payment infrastructure; the AI decision is shaped by whichever third-party SMS-and-review-automation layer bolts on top. Operators picking the booking platform on its bundled AI features pay for AI that does not move the numbers.","article_url":"https://agentmodeai.com/operators/ai-local-service-business-appointment-driven/","topic":"operators-ai-verticals","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-045","claim":"OPS-031's jurisdiction-neutral DIY AI bookkeeping case for solo founders under €30K MRR breaks at the NL-specific Belastingdienst audit-trail boundary. The procurement decision per omzetband: Moneybird (€15-€39/month) under €100K omzet with API-driven AI flow via Make.com or n8n; e-Boekhouden as the goedkope fallback with bundled Scan & Herken OCR; Exact Online above €500K omzet or at BV-overgang where Exact's interne AI replaces the external prompt-pack workflow. NL-specifieke prompt-prefix (klant locatie, dienst type, reverse-charge applicability, OSS-scheme applicability, BTW-rubriek per Belastingdienst-aangifte 2026) is the operationally load-bearing addition that makes AI-getekende journaalposten direct invoerbaar in the chosen tool's BTW-aangifte.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-nl-moneybird-eboekhouden-exact/","topic":"operators-ai-economics","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-046","claim":"Marketplace-reseller AI in 2026 fails differently per platform and the cross-platform mitigation pattern is to separate AI-on-listing-copy (broadly safe across Etsy, Marktplaats, Vinted) from AI-on-listing-images (increasingly penalised on all three platforms via different mechanisms: Etsy's Creativity Standards and AI-disclosure requirement; Marktplaats's photo-fingerprint deduplication; Vinted's image-similarity penalty for resale-of-resold). The 'AI does the entire listing' workflow is the procurement pattern that produces the account-suspension report 6-12 months later. The defensible reseller workflow uses real photos, AI-assisted copy with platform-required disclosure, and per-platform performance tracking on impressions and sales.","article_url":"https://agentmodeai.com/operators/ai-marketplace-resellers-etsy-marktplaats-vinted/","topic":"operators-ai-verticals","pub_date":"2026-05-03","last_reviewed":"2026-05-03","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-05-03","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-047","claim":"EU AI Act Annex III point 4 (employment, workers management, recruitment) applies to SMB AI hiring use even at four-employee scale; the threshold does not scale with company size, and the 2 August 2026 enforcement window covers AI-screened CVs in ChatGPT/Claude/Gemini the same way it covers dedicated platforms (Workable, Greenhouse, Lever, BrightHire). The defensible posture is AI-assisted decisions with a documented human decision-maker plus retained AI-output records — not AI-decided hiring. Solely-automated candidate scoring also conflicts with GDPR Article 22; ICO, AP, and Garante guidance from 2024-2025 is consistent on the human-in-the-loop requirement.","article_url":"https://agentmodeai.com/operators/ai-hiring-smb-eu-ai-act-annex-iii/","topic":"operators-ai-compliance","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-048","claim":"Solo founders adding AI to cold outbound see a deliverability collapse around day 60-90 because AI lifts personalisation breadth at the same volume rather than personalisation depth at lower volume. The collapse is mechanical: AI-templated personalisation degrades recipient engagement, engagement decay triggers spam-classifier de-prioritisation, lower inbox rate produces more complaints, complaints trigger soft blocks. The defensible 2026 posture: 30-40 sends per inbox per day, named-specific first-paragraph personalisation, reply-rate KPI not open-rate, plus a documented EU GDPR Article 6(1)(f) Legitimate Interest Assessment for B2B founders in scope of e-Privacy Directive.","article_url":"https://agentmodeai.com/operators/ai-cold-sales-solo-founder-deliverability/","topic":"operators-ai-tooling","pub_date":"2026-05-04","last_reviewed":"2026-06-04","next_review":"2026-07-19","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-049","claim":"German Mittelstand AI deployment in 2026 hits two compliance surfaces most US-headquartered AI vendors do not handle out of the box: BetrVG §87(1) point 6 co-determination triggers at the first AI assistant or agent that touches employee work activity (Bundesarbeitsgericht broad interpretation covers any system that captures, processes, or analyses employee work activity, primary purpose immaterial); DSGVO Article 35 + Datenschutzkonferenz Muss-Liste require pre-deployment DPIA for most AI-employee-data deployments. The early-engagement workflow (works council notified at vendor selection, DPIA in parallel with vendor evaluation, joint Betriebsvereinbarung drafting, documented pilot at one team for 60-90 days, broader rollout after pilot review) compresses Mittelstand AI timeline from 12-18 months (late engagement) to 6-9 months.","article_url":"https://agentmodeai.com/operators/ai-mittelstand-betrvg-dsgvo-deployment/","topic":"operators-ai-compliance","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-050","claim":"Local SMB AI use on Google Business Profile and local-SEO content splits into two cohorts in 2026: AI-as-research-and-assembly (keyword research, citation audit, performance analysis via Surfer/Frase/Ahrefs/BrightLocal/Whitespark) compounds visibility safely; AI-as-generation (auto-published reviews, auto-published review responses, bulk service-area pages, high-cadence GBP posts) triggers Google's Helpful Content classifier and the March 2024 spam policy update enforcement, with documented suspensions and ranking collapse on a 30-90 day cycle. The defensible posture is AI for the work that scales poorly (research, cross-reference) and human for any content that reaches the public surface.","article_url":"https://agentmodeai.com/operators/ai-local-seo-google-business-profile-smb/","topic":"operators-ai-compliance","pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-051","claim":"AI proposal tools in 2026 split into two clusters by what they let the operator publish: tools that AI-assist proposal assembly (PandaDoc, Better Proposals, Proposify, Bonsai) compound; tools that AI-generate proposal narrative (Pitch, Gamma, Tome AI generation features) read as AI-generated to most buyers within thirty seconds and close at materially lower rates. Three structural patterns trigger the buyer-side AI-generated detection: the three-phase project structure regardless of actual scope, the credentials paragraph that lists capability without naming clients, the pricing section that over-explains itself. The defensible posture is AI for assembly (pricing tables, scope-of-work blocks, clause libraries from CRM) and human for voice (cover letter, executive summary, project-fit paragraph, next-step CTA).","article_url":"https://agentmodeai.com/operators/ai-client-proposals-tools-solo-founder/","topic":"operators-ai-tooling","pub_date":"2026-05-04","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"One named member of the generation cluster was already defunct at publication: Tome shut down its presentation/narrative product (Tome Slides) in March 2025 and pivoted to sales tooling, with the brand later sold to AngelList (deckary.com shutdown timeline; signalhub.substack.com post-mortem, both checked 10 Jun 2026). The generation cluster reduces to Pitch + Gamma. The two-cluster thesis itself is unaffected and arguably strengthened — the pure AI-narrative product failed to find a sustainable business while Gamma (70M users, $100M ARR as of Nov 2025) and the assembly cluster (PandaDoc, Better Proposals, Proposify per Luniq 2026 agency comparison) both compound. Status Up → Partial for the factual error in the tool list."}],"primary_sources":[]},{"id":"OPS-052","claim":"Voor de Nederlandse zelfstandige advocaat (eenmanspraktijk, klein kantoor onder 5 partners) is AI in 2026 toegestaan voor drie hoofdcategorieën onder de NOvA-gedragsregels: juridisch onderzoek met advocaat-verificatie van elke citatie, document-drafting waar de advocaat reviewt en signeert, en cliëntcommunicatie-ondersteuning waar de advocaat elke uitgaande communicatie reviewt voor verzending. AI is niet toegestaan zonder advocaat-review voor: advies-generatie aan cliënten, procesvertegenwoordiging, cliëntgegevens-verwerking via niet-EU-LLM zonder Verwerkersovereenkomst, en het ondertekenen van documenten met AI-gegenereerde citaten zonder primaire-bron-verificatie. EU AI Act Artikel 50 disclosure is verplicht voor cliënt-AI-chatbots vanaf 2 augustus 2026.","article_url":"https://agentmodeai.com/operators/ai-solo-legal-paralegal-nl-bar-rules/","topic":"operators-ai-compliance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-08-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-053","claim":"For marketplace resellers running AI image workflows in 2026, the safe pattern across Marktplaats, Vinted, and Etsy is original photography of the actual item with light AI enhancement (lighting, contrast, background cleanup) only. AI-generated listing imagery and heavy enhancement that produces consistent visual fingerprints across listings trigger Marktplaats's photo-fingerprint deduplication (most aggressive), Vinted's image-similarity penalty for the resale-of-resold pattern, and Etsy's Creativity Standards on AI-generated imagery in handmade categories. The five-rule safe workflow: original photography of every item, light AI enhancement only, fresh photography per relisting, per-platform disclosure where required, and impression-to-view ratio tracking as the leading indicator of algorithm-induced ranking suppression.","article_url":"https://agentmodeai.com/operators/ai-marketplace-image-workflow-marktplaats-vinted/","topic":"operators-ai-verticals","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-054","claim":"For EU-based solo developers doing client work in 2026, the procurement-defensible AI-tool posture turns on client-code data residency rather than on Cursor-vs-Copilot-vs-Claude-Code feature comparison. All three dominant AI coding tools support EU data residency at Enterprise tiers (Copilot via Microsoft Azure OpenAI EU regions, Cursor via configurable LLM provider routing, Claude Code via Anthropic API EU-region availability). Three contract clauses now appear in regulated EU client agreements: client-code-non-transmission, EU-residency requirement, and sub-processor disclosure. The procurement-defensible workflow has five steps: AI-tool inventory, per-client risk assessment, configure tools per client, document configuration in engagement contract, audit quarterly. Three scenarios where the right answer is to disable AI tooling entirely: explicit contract prohibition that cannot be negotiated, embedded regulated data in the codebase, national-security or jurisdictionally-sensitive code.","article_url":"https://agentmodeai.com/operators/ai-solo-dev-eu-client-code-residency/","topic":"operators-ai-compliance","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-055","claim":"For German solo founders and small Mittelstand operators running AI-bookkeeping in 2026, the Buchhaltungssoftware choice resolves on Umsatz tier and Steuerberater relationship: DATEV (€20-€80/month plus Steuerberater-coupling) above €100K Umsatz where the Steuerberater workflow is binding, sevDesk (€8-€48/month) under €100K Umsatz as the cheapest path that produces a GoBD-compliant audit trail, and Lexware (€10-€40/month) as the legacy-Mittelstand fallback. The OPS-031 jurisdiction-neutral DIY-AI-bookkeeping case breaks at the moment the AI-drafted Buchungssatz must land in a tool that preserves the GoBD audit trail; the German-tool layer is the complement to the DIY-AI case. The OSS-Verfahren and reverse-charge VAT prompt-prefix is the operational discipline that prevents AI-VAT-error in 1 of 20 EU-cross-border invoices.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-de-datev-sevdesk-lexware/","topic":"operators-ai-economics","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-056","claim":"For bootstrapped SaaS founders under €30K MRR with AI features in production, the metric that matters is token cost per active user (not total monthly AI spend). Total monthly spend is the lagging indicator that signals problems only after they have crossed gross-margin thresholds; cost per active user is the leading indicator that catches runaway patterns before they erode unit economics. The defensible cancellation-trigger threshold sits at 30-40% of per-user revenue. Four levers when the cost crosses the trigger, ranked by disruption: provider-tier switch (40-70% reduction, low impact), prompt and caching optimisation (20-40% reduction, moderate impact), product change (30-60% reduction, high impact), provider switch (10-30% reduction, highest disruption). Token cost dropped roughly 90% from 2023-2026 but per-user cost stayed flat because product features pulled 10-30x more tokens per session and user behaviour shifted toward higher engagement.","article_url":"https://agentmodeai.com/operators/ai-cost-discipline-bootstrapped-saas/","topic":"operators-ai-economics","pub_date":"2026-05-05","last_reviewed":"2026-05-05","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-057","claim":"For an Etsy seller earning under €100K/year, the cheapest AI stack that consistently pays back combines Claude Pro (€18/month) for listing copy and customer reply drafts with a single image-generation tool (€10-30/month — Midjourney, Adobe Firefly, or DALL-E via ChatGPT Plus). Most paid Etsy-specific AI tools (eRank Pro, Sale Samurai, ListEasy, Alura, Marmalead) only repay above ~200 active listings or ~€50K annual revenue. Etsy's AI-content disclosure policy and the EU AI Act Article 50 transparency obligations apply to AI-generated imagery.","article_url":"https://agentmodeai.com/operators/etsy-ai-tools-for-sellers/","topic":"operators-ai-verticals","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-058","claim":"For a 1-5 person service business in 2026, AI voice agents repay above ~30 inbound calls/week per agent line. Below that, the integration cost (~10-20 hours setup) and per-minute usage rates (5-12 cents per minute typical) exceed the value of automated handling. Vapi, Bland, and Retell occupy three different price-and-control points: Vapi's developer-platform model wins on flexibility and BYO-model control, Bland's no-code on speed-to-deploy, Retell's enterprise-tier on compliance posture (SOC 2 Type II, HIPAA-eligible). TCPA + FCC AI-voice rulings (US) and GDPR Article 22 + ePrivacy Directive (EU) constrain deployment with mandatory disclosure and human-handoff patterns.","article_url":"https://agentmodeai.com/operators/ai-voice-agents-solo-business/","topic":"operators-ai-tooling","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-059","claim":"SMBs without legal teams sign AI vendor MSAs that lock them in via seven recurring clause patterns: (1) data-portability narrowness (prompts/embeddings/agent state excluded from 'your data' definitions), (2) auto-renewal with short notice windows, (3) model-deprecation rights without credit, (4) sub-processor expansion without consent, (5) output-IP ambiguity, (6) pricing escalator without cap, (7) termination-data export window too short. Pattern recognition + a 1-page checklist applied before signature is the practical defence. Five questions emailed to the vendor sales rep before signing — and their willingness to answer in writing — is itself a signal.","article_url":"https://agentmodeai.com/operators/ai-vendor-redflags-smb/","topic":"operators-ai-compliance","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-060","claim":"For Dutch e-commerce SMBs (under €500K annual revenue) in 2026, the AI stack that consistently pays back is Claude Pro or ChatGPT Plus (€18-20/month) for product copy + customer-service drafts, plus one image tool (Midjourney/Firefly), plus EU-resident hosting if data residency matters. Bol.com's API constraints, AVG (Dutch GDPR implementation), and EU AI Act Article 50 transparency obligations create a different procurement reality than US/UK SMBs. Shopify Magic + Sidekick win on speed-to-deploy; WooCommerce wins on data-residency control; Bol.com wins on Dutch-marketplace reach but penalises low-quality AI-drafted listings.","article_url":"https://agentmodeai.com/operators/ai-for-dutch-ecommerce/","topic":"operators-ai-verticals","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-061","claim":"For a 1-5 person business in 2026, AI consistently pays back on six task classes (drafting, summarising, scheduling, research synthesis, code generation for solo developers, image/asset production) and consistently fails on six others (high-stakes customer-facing decisions without disclosure, regulatory advice, complex multi-party negotiations, brand-distinctive creative work, anything requiring physical presence, anything requiring social proof of human authenticity). The 90-second test before delegating any new task: (a) if AI gets it wrong, what is the worst-case cost, (b) does the customer expect a human authored this, (c) is disclosure feasible without breaking the trust contract.","article_url":"https://agentmodeai.com/operators/what-to-delegate-to-ai/","topic":"operators-ai-tooling","pub_date":"2026-05-07","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"holding","verdict_history":[{"date":"2026-05-07","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-062","claim":"For a UK sole trader, Claude Pro and ChatGPT Plus subscriptions are allowable expenses under HMRC's wholly-and-exclusively test (BIM37007) only when paid from the business account and used for trade purposes; the business-tier seats (Claude Team £24/user/month, ChatGPT Business $25/user/month) are cleaner deductibles above £50k turnover because the personal-use exposure is structurally lower and the audit trail is built for the trade. Reverse-charge VAT applies on Anthropic invoices (US entity) and OpenAI invoices (Ireland entity invoicing most UK customers) under VAT Notice 741A place-of-supply rules; Microsoft Copilot UK plans invoice via Microsoft UK Ltd with VAT on the invoice. The MTD-ITSA regime that landed 6 Apr 2026 (mandatory at £50k combined trading + property income, dropping to £30k in Apr 2027 and £20k in Apr 2028) requires every AI-vendor invoice to be captured in HMRC-recognised software at the point it lands; FreeAgent is the practitioner default for UK sole traders under £200k turnover (Open Banking feed, receipt capture included, reliable AI-vendor categorisation), Xero scales better as headcount appears, QuickBooks works but its UK MTD-ITSA module is the youngest of the three. The practical trigger to switch from consumer-tier (Pro / Plus) to business-tier (Team / Business) is VAT registration: the £4-7/month/seat uplift is below the noise floor; the wholly-and-exclusively defence, the admin console, and the audit trail justify the move.","article_url":"https://agentmodeai.com/operators/uk-sole-trader-ai-stack-mtd-vat/","topic":"operators-ai-economics","pub_date":"2026-05-12","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-063","claim":"Under the BNC micro regime, AI-tool subscriptions are not separately deductible because the 34% abattement forfaitaire is fixed; therefore the decision to add AI tooling above ~50 k€ CA is not a tax question but a velocity-to-ceiling question — at the 77 700 € threshold the right move is to forecast the régime réel crossover before adding tooling, not after.","article_url":"https://agentmodeai.com/operators/micro-entrepreneur-ia-urssaf-bnc-france/","topic":"operators-ai-economics","pub_date":"2026-05-12","last_reviewed":"2026-06-10","next_review":"2026-07-20","verdict":"partial","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"The micro-regime ceiling figure in the claim text is overtaken: the 2026-2028 triennial revalorisation raised the services/BNC plafond from 77 700 € to 83 600 € (mixed-activity global cap 203 100 €), per LégiFiscal's 2026 thresholds bulletin and the URSSAF 2026 seuils announcement; the cited impots.gouv.fr page still presents the 77 700 € figure under its 2023-2025 framing. The claim's structural analysis is unaffected — the 34% abattement forfaitaire remains fixed (BOI-BNC-DECLA-10-70), AI subscriptions remain non-deductible under micro-BNC, and the velocity-to-ceiling advice stands with the crossover forecast now running against 83 600 €. Secondary note correction: the TVA franchise for services in 2026 is 37 500 € base / 41 250 € majoré (service-public.fr F21746, extracted 10 Jun 2026: 'Pour l'année 2026, les seuils de franchise en base de TVA française applicables restent inchangés' and the single 25 000 € threshold proposal 'a été abandonnée'); the 36 800 € figure carried in the article body and prior note was the 2023-2024 value. Status Up -> Partial. Article body needs a Peter-approved threshold refresh: 77 700 € appears in the title, excerpt, supportingFigure, FAQ, and body; 36 800 € in FAQ and body; the slug carries no figure and is unaffected."}],"primary_sources":[]},{"id":"OPS-064","claim":"For a freelance translator below 0.10 €/word, accepting MTPE rates at agency-standard 40–60% of full rate is rational only when productivity exceeds 1.8× source-rate baseline; below that ratio, MTPE work is rate-cannibalising and the freelancer should refuse it or move directly to higher-margin language pairs.","article_url":"https://agentmodeai.com/operators/freelance-translator-ai-stack-post-editing/","topic":"operators-ai-verticals","pub_date":"2026-05-12","last_reviewed":"2026-06-04","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-065","claim":"A solo agency delivering AI-assisted work to a client needs four contract clauses by Aug 2026 — disclosure of AI use, IP warranty carve-out for AI-generated portions, training-data exclusion of client materials, and a liability cap tied to fee paid — without which the agency carries strict liability under EU AI Act Article 50 plus contract-law warranty exposure on copyright.","article_url":"https://agentmodeai.com/operators/ai-client-deliverable-contract-clauses/","topic":"operators-ai-compliance","pub_date":"2026-05-12","last_reviewed":"2026-06-04","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-066","claim":"For a services firm under 50 people, paid AI seats pencil at ~5-person firms only when at least 2 of 5 staff are knowledge workers spending more than 10 hours/week on text drafting, and they fail to pencil at 25–40-person firms if the firm-wide rollout includes less than 60% adoption — between those zones, the break-even is determined by adoption rate, not seat price.","article_url":"https://agentmodeai.com/operators/ai-break-even-headcount-smb/","topic":"operators-ai-economics","pub_date":"2026-05-12","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-12","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-067","claim":"For a 5-15 person services agency running Cursor, Windsurf, Claude Code, or any internal agent platform built on the Model Context Protocol on paid-client-work machines, the May 2026 CVE class (Microsoft Security Response Center's CVE-2026-25592 and CVE-2026-26030 against Semantic Kernel on 7 May 2026, OX Security's MCP STDIO supply-chain advisory, and the Windsurf 1.9544.26 prompt-injection-to-MCP-registration path) cannot be cleared by vendor auto-update alone. A 5-step playbook (inventory every machine; pin the patched version and disable auto-update; write a one-page MCP allowlist; disclose AI-IDE use to active clients in writing; schedule a 30-day review) is the agency-level minimum that holds against the question an enterprise client will ask in procurement and against the residual liability the agency carries if a remediation conversation becomes necessary.","article_url":"https://agentmodeai.com/operators/windsurf-mcp-advisories-small-agency-ide-playbook/","topic":"operators-ai-security","pub_date":"2026-05-17","last_reviewed":"2026-06-04","next_review":"2026-07-04","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-068","claim":"The solopreneur AI subscription stack is in active compression through 2026, with median monthly spend tracking from $300-$500 in 2024 toward $100-$130 in early 2026 (Godberry Studios Zoom Solopreneur 50 teardown corroborated by BetterCloud SaaS industry data) and toward under $80 by end-Q3 2026 (editorial estimate, source:our-estimate, based on observed compression rate and announced 2026 feature roadmaps of Claude, ChatGPT, and Microsoft 365 Copilot). The compression is not a pricing story; it is a category-collapse story driven by 12 named standalone-AI categories being absorbed into the foundation-model subscription (Claude Pro or ChatGPT Plus) and the host platform's native AI: standalone writing assistants, meeting summarisers, slide generators, email-draft assistants, SEO optimisers, form-builder add-ons, calendar overlays, standalone chatbots, image-generation subscriptions for non-creative-pro use, note-taking add-ons, no-code agent builders, and standalone research assistants. The operator-side decision rule is cancel-now, wait-one-cycle, or migrate-carefully, applied via a four-line sub-thirty-minute test-before-cancel script against each line in the standing-subscription list.","article_url":"https://agentmodeai.com/operators/solopreneur-ai-stack-consolidation/","topic":"operators-ai-tooling","pub_date":"2026-05-17","last_reviewed":"2026-06-17","next_review":"2026-07-31","verdict":"partial","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-17","verdict":"partial","note":"Source-text re-review: the '$300-$500 (2024) toward $100-$130 (early 2026)' median trajectory is not stated in either cited source — the Godberry Studios teardown reports stack cost by revenue tier (not a year-over-year median) and BetterCloud's SaaS-industry data covers enterprise spend, not solopreneur AI subscriptions. The compression direction is supported by the Godberry tier data and observable foundation-model bundling; the specific year-anchored median figures are reclassified as source:our-estimate in the article. The load-bearing claim (active compression / category-collapse) holds; status moved to Partial pending a primary source carrying a dated solopreneur-median series."}],"primary_sources":[]},{"id":"OPS-069","claim":"The widely-cited 95-percent generative-AI-pilot-failure framing (MIT Sloan Management Review and Boston Consulting Group adoption-research streams, 2025-2026) is methodologically defensible for the enterprise cohort the research sampled (large firms with dedicated AI functions, 12-to-18-month evaluation windows, scaled-production-deployment success definition) and materially misrepresents small-firm pilot dynamics. The 1-to-50-person operator cohort has a different failure-mode catalogue (tool-assigned-to-wrong-person, rewrite-cost-exceeds-savings, client-rework-from-AI-deliverable, line-item-stack-compounded-and-cancelled, sporadic-use-no-routine) and a different success definition (90-day payback at actual hourly rate; deliverable quality reaching the client without disproportionate rework; routine fit documented for handover). A three-question Monday-morning small-firm pilot test (payback, deliverable quality, routine fit) checked at 30 days and 60 days is the operator's actual evaluation instrument and replaces the enterprise 12-to-18-month evaluation cycle that the 95-percent number is measured against.","article_url":"https://agentmodeai.com/operators/small-firm-ai-pilots-fail-differently-enterprise-misread/","topic":"operators-ai-tooling","pub_date":"2026-05-17","last_reviewed":"2026-05-17","next_review":"2026-07-01","verdict":"holding","verdict_history":[{"date":"2026-05-17","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-070","claim":"For 1-50 person operators running Claude, Claude Code, or Cursor (Claude-backed) on paid client work, the 19 May 2026 Karpathy hire at Anthropic is a vendor-trajectory signal operating on the 12-to-24-month time horizon (Karpathy's mandate is upstream pre-training research, not application-layer change), and the right operator-side response is the 70% concentration rule. List every monthly AI-stack line item, tag by underlying vendor (Anthropic / OpenAI / Microsoft / Google / other), compute the largest single-vendor share. Above 70% concentrated on Anthropic, add a deliberate secondary-lab subscription (ChatGPT Plus at $20/mo or Gemini Advanced at comparable pricing) as resilience against any future Anthropic-specific incident. Below 70%, continue concentrating and re-evaluate at the 45-day claim review. The hire does not change the daily workflow this week; it affects the medium-term improvement trajectory of the vibe-coding interface the operator is already using.","article_url":"https://agentmodeai.com/operators/karpathy-joins-anthropic-vibe-coding-operator-stack/","topic":"operators-ai-tooling","pub_date":"2026-05-19","last_reviewed":"2026-05-19","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-19","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-071","claim":"Colorado's AI Act replacement bill SB 189 — passed by both chambers of the Colorado legislature in May 2026 and effective 30 June 2026 — drops the original law's risk-management programme requirements and annual impact assessments in favour of a notice-and-transparency framework; operators using AI to make or materially influence a consequential decision about a Colorado resident (employment, housing, credit, insurance, education, healthcare) have active notice, disclosure, and documentation obligations from 30 June 2026 regardless of firm size.","article_url":"https://agentmodeai.com/operators/colorado-ai-act-sb189-operator-compliance-brief/","topic":"operators-ai-compliance","pub_date":"2026-05-22","last_reviewed":"2026-06-10","next_review":"2026-07-25","verdict":"partial","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."},{"date":"2026-06-10","verdict":"partial","note":"Trigger condition (2) fired: the effective date moved. Governor Polis signed SB 26-189 on 14 May 2026 (Holland & Knight client alert, May 2026; Seyfarth; Littler). The signed law repeals and reenacts the original Colorado AI Act and its obligations take effect 1 Jan 2027 — not 30 Jun 2026 as the claim asserted. No operator obligation starts 30 Jun 2026; the only pre-2027 item is Colorado AG rulemaking due by 1 Jan 2027. The claim's structural reading holds (risk-management programmes and impact assessments dropped for a notice-and-transparency framework; consequential-decision scope covering employment, housing, credit, insurance, education, healthcare; no small-firm exemption). The urgency leg ('obligations from 30 June 2026') is overtaken. Status Up → Partial."}],"primary_sources":[]},{"id":"OPS-072","claim":"Notion's 13 May 2026 developer platform launch — adding Workers, External Agents API (Claude, Codex, Decagon natively integrated), and Database Sync — means small teams already running on Notion have functional AI agent infrastructure that materially overlaps with separately paid automation tooling (Zapier, Make.com) for Notion-internal workflows; the consolidation evaluation belongs in the current billing cycle, and the Workers free tier through August 2026 is the zero-cost window to test it.","article_url":"https://agentmodeai.com/operators/notion-ai-agents-hub-small-team-operator-upgrade/","topic":"operators-ai-tooling","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-073","claim":"OpenAI's 11 May 2026 Deployment Company launch — a $4B+ entity with 150 Forward Deployed Engineers embedded inside client organisations to redesign workflows and build durable AI systems — confirms that AI configuration and workflow redesign are the market-identified constraint on AI ROI, not model access; for the 1-15 person operator or freelance builder whose differentiated value is knowing how to make AI work in a specific client context, this announcement is a competitive-positioning map: the service is real, the demand is confirmed, and the small operator has three durable advantages (speed, context depth, price) that the Deployment Company's structure cannot replicate at small-client scale.","article_url":"https://agentmodeai.com/operators/openai-deployment-company-operator-positioning-signal/","topic":"operators-ai-tooling","pub_date":"2026-05-22","last_reviewed":"2026-05-22","next_review":"2026-07-06","verdict":"holding","verdict_history":[{"date":"2026-05-22","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-074","claim":"A 5-15 person team running AI tools on paid client work in 2026 can move from default-shared personal credentials for AI agents to a defensible non-human-identity posture in three hours of work using existing tooling (password manager, calendar, spreadsheet). The five-step starter kit (inventory every AI tool acting in the environment and its credential; mint per-agent credentials with smallest-scope; move every credential into one secrets vault and remove from elsewhere; set a 90-day rotation cadence with a calendar owner; write and test a one-page leaver and revocation runbook) covers the credential-management practices that CyberArk-grade enterprise NHI programmes cover, scaled to a team without an identity-governance function. The answer the kit produces is sufficient for almost every mid-market and SMB client procurement question, and a credible answer to most enterprise procurement questionnaires reaching small-agency vendors in 2026.","article_url":"https://agentmodeai.com/operators/operators-nhi-starter-kit/","topic":"operators-ai-security","pub_date":"2026-05-24","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-075","claim":"A 1-10 person team in 2026 has approved-tool unapproved-capability shadow AI running inside its already-approved SaaS estate (Notion AI agents, Slack AI, Microsoft 365 Copilot, Atlassian Intelligence, Google Workspace Gemini, Salesforce Einstein) at near-universal incidence, because the vendors have shipped these capabilities as on-by-default or low-friction add-ons inside the existing licence path. The three discovery signs (SaaS bill line items the founder does not remember approving; team-member mentions of new features inside existing tools; vendor admin console notifications advertising auto-enabled AI capabilities) are reliable triggers for a 60-minute audit (SaaS bill review, vendor admin console walk, team check-in) that produces a one-page inventory mapping each AI capability to the tool, activation date, current users, data scope, and team's explicit posture. The inventory is the artefact that answers client procurement questions about AI tool exposure and the basis for any disable-or-restrict decisions.","article_url":"https://agentmodeai.com/operators/approved-tool-unapproved-capability-shadow-ai-operators/","topic":"operators-ai-security","pub_date":"2026-05-24","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-076","claim":"A small agency (1-15 person team) building agentic features on paid client work in 2026 should pick its agent protocol per project by reading the client's existing stack (Anthropic-aligned client → MCP; Google-aligned client → A2A; sovereignty- or self-hosted-aligned client → Llama Stack), default to MCP for tool-heavy work and A2A for agent-collaboration work when the client has no existing stack, and keep its tool inventory portable by building every tool as a plain HTTP service first and wrapping it to the chosen protocol second. The plain-HTTP-first discipline costs roughly 20-30% extra on the first tool of a project and produces 60% wrapper-effort reduction on the second client requesting the same functionality on a different protocol. Tracking wrapper hours separately from service hours in the agency's time log is the simplest instrument for quoting the re-platform cost accurately when a client asks.","article_url":"https://agentmodeai.com/operators/agent-protocol-picking-for-small-agencies/","topic":"operators-ai-security","pub_date":"2026-05-24","last_reviewed":"2026-05-24","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-05-24","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-077","claim":"As of mid-2026, a non-developer operator on a 1-50 person team can build a production-useful internal-process agent in Notion (via the 13 May 2026 developer platform: Workers, External Agents API with Claude/Codex/Decagon, Database Sync, ntn CLI) or in ChatGPT (via the matured GPT Builder with Instructions, Knowledge, Actions and four sharing tiers) without a separate automation tool. The default permission scope of the resulting agent is the workspace-wide access of the human who built it (Notion: workspace-admin if the builder is an admin; ChatGPT: the builder's API key scope on configured Actions), which is the operator-scale shadow-AI risk. The safe-deploy playbook (three guardrails: scope to one data source not whole workspace; read-only first for a week of parallel running; human approval on customer-facing actions; plus the one permission-scope rule that the agent inherits the builder's access not the user's) closes the gap using only the tools' native configuration.","article_url":"https://agentmodeai.com/operators/no-code-agent-building-notion-gpt/","topic":"operators-ai-tooling","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-078","claim":"As of mid-2026, most 1-15 person teams running AI agents and automations on paid client work cannot revoke a misbehaving agent's access quickly because they share a small number of credentials across multiple tools and have no written pause-and-revoke runbook with rehearsed timings per tool. The runbook discipline (per-tool documentation of the pause path, the revoke path, the time-to-effect, and the OAuth third-party revocation step where applicable) is a 30-minute Friday investment using only the tools already in use, and is the small-team analogue of the four-primitive enterprise containment architecture covered in AM-171.","article_url":"https://agentmodeai.com/operators/kill-switch-for-small-team/","topic":"operators-ai-security","pub_date":"2026-05-26","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-079","claim":"As of mid-2026, the consumer-tier and small-team AI tools most 1-15 person teams use on client work (ChatGPT, Claude, Notion AI, Cursor, customer-service bots) retain context across sessions by default in ways that can carry one client's information into another's work, with each tool exposing a different memory default in a different settings location and a different cross-context behaviour. Most operators have not sat down to align them, and the 30-minute Monday hygiene routine (settings pass per tool; no-client-identifiers rule on every persistent custom-instructions field; per-client reset checklist) closes the operational gap using only the tools already in use.","article_url":"https://agentmodeai.com/operators/agent-memory-small-team-hygiene/","topic":"operators-ai-security","pub_date":"2026-05-26","last_reviewed":"2026-05-26","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-05-26","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-080","claim":"For a small business in the EU or selling into it, the Digital Omnibus political agreement of 7 May 2026 pushes the heavy high-risk obligations out to 2 December 2027 and 2 August 2028 but does not change the duties most small businesses are actually subject to: as a deployer rather than a provider, you must disclose AI-generated deepfake image, audio, or video and ensure AI chat tells people they are talking to a machine from 2 August 2026 under Article 50, you have been under the Article 4 AI literacy duty since 2 February 2025, and a 30-minute readiness check using tools you already have closes most of the practical gap.","article_url":"https://agentmodeai.com/operators/eu-ai-act-small-business-deployer-duties/","topic":"operators-ai-compliance","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-07-12","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-081","claim":"Google Workspace Studio, the no-code AI agent builder Google introduced at Cloud Next 2026 for paid Google Workspace tiers, is the right first place for a small team to build agents when its data and day-to-day workflows already live inside Google Workspace such as Gmail, Docs, Sheets, and Drive, because the integration and permissions are native and the build is natural-language; for a team whose work spans many third-party tools, a model-neutral automation layer such as n8n or a documentation-centric build in Notion remains the better default, and the deciding question is where the team's data and workflows already live rather than which builder markets the most capability.","article_url":"https://agentmodeai.com/operators/google-workspace-studio-for-small-teams/","topic":"operators-ai-tooling","pub_date":"2026-05-28","last_reviewed":"2026-05-28","next_review":"2026-07-12","verdict":"holding","verdict_history":[{"date":"2026-05-28","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-082","claim":"If a small business shipped a customer-facing app built with a no-code or vibe-coding platform and never ran a security review, the safe operating assumption is that its database is reachable from the public internet until proven otherwise, because spring 2026 scans found thousands of such apps actively leaking sensitive data and the common cause is open data-access defaults that the natural-language build does not close for you.","article_url":"https://agentmodeai.com/operators/vibe-coded-app-security-check/","topic":"operators-ai-security","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-083","claim":"AI inference is getting cheaper per token in 2026 while the AI bills small businesses actually pay are rising, because the cost has moved from the model to the automation layer where metered SDK and agent usage now sits; the imminent example is Anthropic's announced 15 Jun 2026 split that carves Claude automation and SDK usage out of the flat subscription into a separately metered pool, so a small business running AI inside automations should re-model its stack before the cutover.","article_url":"https://agentmodeai.com/operators/ai-cheaper-but-your-bill-rising/","topic":"operators-ai-economics","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-084","claim":"For a small business, the fully-autonomous AI sales-development rep that finds prospects and sends outreach without a human has not proven durable, because autonomous volume sending burns a small sender's domain reputation and fully-automated personalization reads as machine-made; the pattern that works is human-in-the-loop, where AI handles the research and the first draft and a person approves and sends.","article_url":"https://agentmodeai.com/operators/ai-sdr-human-in-the-loop/","topic":"operators-ai-verticals","pub_date":"2026-05-29","last_reviewed":"2026-05-29","next_review":"2026-06-28","verdict":"holding","verdict_history":[{"date":"2026-05-29","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-085","claim":"Shopify's Magic and Sidekick AI tools are included on every Shopify plan at no extra cost as of May 2026, so most merchants are already paying for a built-in AI assistant they have not activated; the operator value is in switching it on for the two or three recurring, easy-to-verify tasks that fit (store-data questions, repetitive copy, small multi-step admin tasks) rather than buying a separate ecommerce-AI subscription, while keeping it away from unverified pricing, financial, factual-product, and live-stock decisions.","article_url":"https://agentmodeai.com/operators/shopify-magic-sidekick-guide-small-business/","topic":"operators-ai-tooling","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-06-29","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-086","claim":"After Fathom limited advanced AI summaries on its free plan to roughly five per month in 2026, the AI meeting-notetaker choice for a small team turns on meeting volume and privacy posture rather than feature lists: a solo or light user can stay on a free tier or a single-seat upgrade (Fathom Premium around $16/month annual), while a team with several client meetings a week across multiple people is usually better served by a per-seat plan such as Fireflies Pro (around $10/user/month annual); pick on how much you actually meet and your data/consent posture, not on the comparison table.","article_url":"https://agentmodeai.com/operators/ai-meeting-notetaker-fathom-fireflies-small-business/","topic":"operators-ai-tooling","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-06-29","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-087","claim":"Webflow's May 2026 pricing restructure combines its former CMS and Business site plans into a single Premium plan at $25/month billed annually ($39/month on monthly billing) with 20,000 CMS items and 40 collections included, effective on renewal on or after 29 Jun 2026 for most existing sites (16 Nov 2026 for freelancer and agency workspaces); by Webflow's own account the change raises some sites' cost, lowers others', and leaves some unchanged, so the operator move is to run Webflow's own change calculator against actual usage (CMS items, editor seats, AI and other features used) and pick the cheapest correct plan before the effective date, rather than auto-accepting the migration or switching platforms over pricing noise.","article_url":"https://agentmodeai.com/operators/webflow-pricing-change-small-business-guide/","topic":"operators-ai-economics","pub_date":"2026-05-30","last_reviewed":"2026-05-30","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-05-30","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-088","claim":"A solo developer or small agency that runs an AI coding assistant (Claude Code, Cursor, Gemini CLI, GitHub Copilot, OpenAI Codex, Grok) on the same machine that holds its client SSH keys and deploy credentials is materially exposed by the May 2026 TrustFall and SymJack findings, in which opening a malicious repository and accepting an approval prompt can run attacker code that steals those secrets, and the proportionate fix is not a security budget but updating every tool to its latest version, slowing down on approvals (especially file copies and writes to configuration files), not opening untrusted repositories on a credentialed machine, and moving secrets out of plain files while rotating anything that may have been exposed.","article_url":"https://agentmodeai.com/operators/ai-coding-cli-security-small-team/","topic":"operators-ai-security","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-089","claim":"Agentic bookkeeping arrived for small business in May 2026 (Xero's XeroForce agent builder, in alpha and invite-only, plus JAX and Xero OS; Intuit's QuickBooks Assist agent teams), and the operating split that protects a small owner is to let agents run the repetitive, reversible, internal ledger work (categorisation, reconciliation, anomaly flagging, month-end report drafts, organising tax documents) while keeping a human approval on anything that moves money out or files with the authorities (pay runs, supplier payments, tax submissions), because those are irreversible and the owner remains responsible for what is filed and paid regardless of which tool prepared it, a caveat the vendors' own human-at-the-helm framing concedes.","article_url":"https://agentmodeai.com/operators/ai-bookkeeping-agents-small-business/","topic":"operators-ai-economics","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-07-02","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-090","claim":"For a US small business (under about 50 people) the 2026 state AI laws impose far less than the headlines imply: California's SB 53 applies only to frontier model developers (above ~10^26 training operations or >$500M revenue) and not to AI users, Colorado's comprehensive law was repealed, narrowed, and delayed to 1 January 2027 by SB 26-189 (signed 14 May 2026), Texas's Responsible AI Governance Act (effective 1 January 2026) is intent-based rather than paperwork-based, and no federal AI law is in force, so the proportionate response is a short list — know which states you actually touch, confirm you are a deployer not a frontier developer, adopt basic disclosure hygiene, and never deploy AI to deceive or discriminate — rather than an enterprise-scale compliance programme.","article_url":"https://agentmodeai.com/operators/us-ai-laws-small-business/","topic":"operators-ai-compliance","pub_date":"2026-06-02","last_reviewed":"2026-06-02","next_review":"2026-07-17","verdict":"holding","verdict_history":[{"date":"2026-06-02","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-091","claim":"HubSpot's 14 Apr 2026 shift of its Breeze Customer Agent to $0.50 per resolved conversation (from $1.00 per conversation) makes the agent worth trialing for a small support team because cost now tracks resolved tickets rather than attempts, but HubSpot's definition of a resolved conversation (the agent shares a content source or performs an action and no human handoff occurs within 72 hours of the last message) is the term that decides whether it is actually cheap, so the comparison that matters is $0.50 per resolved ticket against the team's loaded cost per human-handled ticket, not against zero.","article_url":"https://agentmodeai.com/operators/hubspot-breeze-pay-per-resolution/","topic":"operators-ai-tooling","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-092","claim":"With Notion Custom Agents moving from free beta to credit-based billing ($10 per 1,000 credits, no rollover) on 4 May 2026, the operator decision flips from which agents to build to which recurring agents clear their per-run cost, and because the bill is driven by run frequency rather than the number of agents built, only a few high-frequency high-value patterns (daily brief, inbox or task triage, recurring status roll-up) are worth keeping for most small teams.","article_url":"https://agentmodeai.com/operators/notion-custom-agents-credit-pricing/","topic":"operators-ai-tooling","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-093","claim":"For a small team already on a Microsoft 365 Business plan, the $18 per user per month promotional rate for the Copilot Business add-on locks in only through 30 Jun 2026 before rising to the $21 standard rate, so the cost-rational move is to decide before the deadline, but only on the seats whose actual Word, Excel, Outlook and Teams usage clears the per-seat math, because the discount applies only to the first year and only to seats licensed before the deadline.","article_url":"https://agentmodeai.com/operators/microsoft-365-copilot-business-price-lock/","topic":"operators-ai-economics","pub_date":"2026-06-05","last_reviewed":"2026-06-05","next_review":"2026-07-05","verdict":"holding","verdict_history":[{"date":"2026-06-05","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-094","claim":"The Canva + Perplexity Computer connector (launched 4 Jun 2026, available to Perplexity Pro at $20/month and the Max and Enterprise tiers across 11 languages) lets the Perplexity agent turn research into editable Canva designs, which removes the research-to-design context switch for a consultant or small agency that produces decks from research weekly, so the decision turns on the Perplexity subscription and the cadence of designed output, not on the workflow itself, and the output is a starting draft rather than a finished brand asset.","article_url":"https://agentmodeai.com/operators/canva-perplexity-computer-connector/","topic":"operators-ai-tooling","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-095","claim":"OpenAI's six role-specific Codex plugins (announced 2 Jun 2026, including Sales wired to Salesforce/HubSpot/Slack and Creative Production wired to Figma/Canva) are rolling out to Business and Enterprise plans first, while Codex itself is available from the $20 ChatGPT Plus plan up, so for a small team on Plus the cost-rational move is to use Codex now for self-assembled automation and treat the packaged plugins as a reason to watch rather than an immediate reason to upgrade to Business.","article_url":"https://agentmodeai.com/operators/openai-codex-role-plugins-small-team/","topic":"operators-ai-tooling","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-096","claim":"Google's new $100/month AI Ultra tier (introduced at Google I/O on 19 May 2026 alongside a cut of the top AI Ultra plan from $250 to $200) buys usage headroom and storage on top of the same Gemini models the cheaper tiers run, so it earns its price only for heavy daily agent users who exhaust a lower tier's limits, and for most operators the cost-rational path is the cheaper Google AI Pro subscription or Gemini 3.5 Flash via the API ($1.50 per million input tokens, $9.00 per million output).","article_url":"https://agentmodeai.com/operators/google-ai-ultra-pricing-decoded/","topic":"operators-ai-economics","pub_date":"2026-06-08","last_reviewed":"2026-06-08","next_review":"2026-07-08","verdict":"holding","verdict_history":[{"date":"2026-06-08","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-097","claim":"The calendar-phishing and ClickFix scam patterns named in Google's 8 Jun 2026 fraud advisory specifically exploit the trust habits that AI scheduling, notetaker and assistant tooling creates in small teams (machine-generated calendar entries treated as normal; paste-this-command setup motions rehearsed as routine), and both are closed by free measures: restricting calendar auto-add to known or accepted invites, reporting rather than deleting phishing invites, and a team rule of never pasting a command a web page solicited.","article_url":"https://agentmodeai.com/operators/calendar-phishing-clickfix-small-business/","topic":"operators-ai-security","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-098","claim":"Zapier bills MCP usage at a fixed rate of two tasks per successful AI tool call against the plan's existing task allowance, with no separate MCP meter and no per-session cap (documentation updated 4 Jun 2026), which halves a plan's real agent capacity (a 750-task Professional plan at $19.99/month annual is 375 agent actions), so an operator wiring Claude or ChatGPT into Zapier should budget every AI workflow at double its successful-call count and keep monitoring-grade chatter off the MCP path.","article_url":"https://agentmodeai.com/operators/zapier-mcp-billing-two-tasks-per-call/","topic":"operators-ai-economics","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-099","claim":"Notion Workers (the hosted serverless code runtime shipped with the 13 May 2026 Developer Platform, distinct from the Custom Agents product) is free on Business and Enterprise plans until 11 Aug 2026 and then meters at $0.0023 per run, and because the post-August bill is purely a function of run frequency, the rational operator move is to build the highest-value syncs now, measure a real week of run counts inside the free window, and tune schedules before the meter starts.","article_url":"https://agentmodeai.com/operators/notion-workers-free-window-pricing/","topic":"operators-ai-tooling","pub_date":"2026-06-09","last_reviewed":"2026-06-09","next_review":"2026-07-09","verdict":"holding","verdict_history":[{"date":"2026-06-09","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-100","claim":"Meta's Business Agent went globally available on 3 Jun 2026 across WhatsApp, Messenger and Instagram — free to activate, with paid subscription offerings announced for the coming months and more than one million businesses already using it pre-rollout per Meta — and because the free period is explicitly temporary, the rational small-business move is to treat it as an evaluation window: activate on a bounded slice, measure resolution rate and supervision cost (how often it says something you would not have said), and know the price at which it is worth paying before the tiers are announced.","article_url":"https://agentmodeai.com/operators/meta-business-agent-whatsapp-small-business/","topic":"operators-ai-tooling","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-101","claim":"Intuit's QuickBooks Workforce (announced 6 May 2026) packages an AI payroll agent for small teams at $50/month plus $6.50 per employee on the entry tier ($88 plus $10 Premium, $134 plus $12 Elite), with new pricing effective 1 Jul 2026 for new customers, and the operating rule that protects a small owner is the bookkeeping split applied to its sharpest case: let the agent do the reversible preparation (gather time data, flag inconsistencies, draft the run) while a human approves every execution, because payroll is irreversible money out plus filings to authorities and the owner remains responsible regardless of which tool prepared it.","article_url":"https://agentmodeai.com/operators/quickbooks-workforce-ai-payroll-small-team/","topic":"operators-ai-economics","pub_date":"2026-06-10","last_reviewed":"2026-06-10","next_review":"2026-07-10","verdict":"holding","verdict_history":[{"date":"2026-06-10","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-102","claim":"As of 10 Jun 2026, Mastercard (Agent Pay for Machines) and Visa (Intelligent Commerce, with OpenAI) have both shipped production rails for AI agents to transact, so the operator-relevant decision is no longer whether to let an agent pay but how to scope it: a tokenised agent-specific credential, a hard per-period spend cap, merchant-category limits, and human approval by default.","article_url":"https://agentmodeai.com/operators/ai-agent-payments-small-business-guardrails/","topic":"operators-ai-tooling","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-07-28","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-103","claim":"AI tools small businesses use are shifting from per-seat subscriptions to outcome- and activity-based metering (HubSpot Breeze at $0.50 per resolved conversation since 14 Apr 2026, Zapier Agents metered per activity), which makes resolution rate and activity frequency, not seat count, the variables an owner must model before turning an agent loose.","article_url":"https://agentmodeai.com/operators/outcome-based-ai-pricing-small-business/","topic":"operators-ai-economics","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-07-28","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"OPS-104","claim":"AI-generated fraud (voice cloning, deepfake video calls, fake-supplier emails) is now a primary cyber-risk for small businesses without a dedicated finance or IT function, and the effective defenses are procedural, out-of-band callback verification on a number you already had and dual-control on any payment or bank-detail change, rather than technical detection tooling.","article_url":"https://agentmodeai.com/operators/ai-fraud-small-business-defenses/","topic":"operators-ai-security","pub_date":"2026-06-14","last_reviewed":"2026-06-14","next_review":"2026-07-28","verdict":"holding","verdict_history":[{"date":"2026-06-14","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-001","claim":"The 47-question AI Vendor Security Questionnaire covers seven failure surfaces (model lineage, training/inference data handling, non-human identity, audit/observability, kill-switch, EU AI Act + GDPR posture, contract/indemnification) that CAIQ v4 and SIG do not address; vendors that cannot answer score sections binary-unanswered, and the questionnaire is the addendum (not replacement) to existing cloud/SaaS procurement frameworks.","article_url":"https://agentmodeai.com/resources/ai-vendor-security-questionnaire/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-002","claim":"The pre-deployment AI DPIA template fuses GDPR Article 35 obligations with EU AI Act Article 26 (deployer) and Article 27 (FRIA where applicable) into a single working-session document; sections 7 and 8 are conditional on the EU AI Act risk classification established in section 1, which means deployers complete the full document only when the system is classified as high-risk under Annex III.","article_url":"https://agentmodeai.com/resources/ai-dpia-template/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-003","claim":"The four-phase agent incident runbook (detect within 4h, contain within 30s, roll back per action class, post-mortem with MTTD-for-Agents detection chain) is the operational overlay on standard SRE incident response that most enterprises deploying agentic AI in 2026 do not have; the seven action classes (database writes, external API calls, customer comms, document publication, code commits, identity changes, knowledge-base writes) each require a distinct rollback procedure and the runbook captures the operator authorised, time budget, and substitute action where rollback is impossible.","article_url":"https://agentmodeai.com/resources/agent-incident-runbook/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-004","claim":"The Works Council AI Notification Packet covers three EU jurisdictions (German BetrVG §87(1) point 6, Dutch WOR Article 27, French CSE consultation under Code du travail L2312-8) plus the EU AI Act Article 26(7) deployer-notification overlay that activates 2 August 2026; early engagement (vendor-shortlist landing) compresses deployment timelines from 12-18 months (late engagement) to 6-9 months in the Mittelstand case studied in OPS-049, and the per-jurisdiction documents (Betriebsvereinbarung, OR convenant, CSE avis) are designed to consolidate into a single deployment-go decision with the Article 26(7) notification appended afterward.","article_url":"https://agentmodeai.com/resources/works-council-ai-notification-packet/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-08-02","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]},{"id":"RES-005","claim":"The 38-item AI MSA red-team checklist organises the contractual review around seven clause families (training-data carve-outs, output ownership + IP indemnification, model-deprecation rights, sub-processor expansion, kill-switch SLA, exit-data portability, regulatory + EU AI Act flow-through) where 2025-2026 enterprise AI MSA failures cluster; vendors scoring yes on 30+ items are contractually serious, 20-29 items are treatable through negotiation, and below 20 signals that the vendor's commercial position depends on retaining the rights the checklist is designed to constrain.","article_url":"https://agentmodeai.com/resources/ai-msa-red-team-checklist/","topic":null,"pub_date":"2026-05-04","last_reviewed":"2026-05-04","next_review":"2026-07-03","verdict":"holding","verdict_history":[{"date":"2026-05-04","verdict":"holding","note":"Claim created at publish."}],"primary_sources":[]}]}