We only publish what we can defend in a vendor meeting. Every claim carries an ID, a review date, and a verdict you can check.
- Our ledger180
- Holding168
- Partial06
- Not holding06
- Industry claims tracked26
- Last reviewtoday
Quiet — no verdict transitions in the last 30 days. See the ledger →
Agent Mode AI — claim-tracked agentic AI analysis
Karpathy joins Anthropic's pre-training team: what the May 19 hire signals for CIO vendor-trajectory models
Andrej Karpathy announced on Tuesday 19 May 2026 that he has joined Anthropic. Anthropic confirmed he will lead a team focused on using Claude to accelerate pre-training research, working under Nick Joseph on the pre-training team. The trade-press framing is the hiring coup. The CIO framing is different. Karpathy's specific mandate — applying Claude to the work of building the next Claude — is the load-bearing signal. It indicates Anthropic is betting on recursive self-improvement of its model line at the foundational layer, not just at the application layer. For enterprises sizing multi-year platform commitments, that materially changes the vendor-trajectory model on which the commitment rests.
27 years enterprise IT operations. Global organisation. Major incidents. Editorially independent.
- 108pieces
- 180tracked claims
- 14public retractions
The Enterprise Agentic Governance Benchmark. Six dimensions, scored 0–100. Free 5-minute web diagnostic; 30–45 minute Excel for governance groups.
Recently reviewed
Three claims most recently re-tested against their primary sources. Status changes log to the corrections page; nothing quietly vanishes.
- AM-133HoldingQ3 2026 Claim Review Bulletin: which claims moved, which held, and what the EU AI Act enforcement window did to the corpusReviewed 30 Jul 2026Read article →
- AM-161HoldingAI and jobs: why the task-level frame is the one CIOs needReviewed 20 May 2026Read article →
- OPS-070HoldingKarpathy joined Anthropic on 19 May 2026: what the vibe-coding inventor's move means for the 1-50p operator stackReviewed 19 May 2026Read article →
Why this publication has a ledger
Most AI commentary gets paid for being loud about what's new. Almost none gets measured on whether what it said last quarter still holds this one. That is the gap this publication exists to close. Every published argument carries an ID, a review date, and one of three verdicts — Holding, Partial, or Not holding — that updates over time as evidence accumulates. The verdict log is the product.
When a claim stops holding, the page says so. The original sentence stays visible. The correction is dated and appended. Nothing is quietly removed. You do not need to trust the author to trust the verdicts — the receipts are public, on a 30–90 day review rhythm, and the corrections record is permanent.
Two registers
Same Holding-up disciplineMid-market and large enterprise. Procurement, governance, EU AI Act, multi-vendor agentic stacks. 30–90 day claim review cadence.
No IT department. Practitioner-advisory voice; faster 30–45 day cadence. Tools, vendor red flags, hours-per-week evaluation budgets.
Topic pillars
Five clusters- 2 articlesNon-human identity
How enterprise IT manages AI agents as first-class identities — lifecycle, credentials, procurement clauses, audit.
- 29 articlesAgent procurement
The contracts, SLAs, and evaluation criteria that distinguish agentic-AI procurement from SaaS procurement.
- 2 articlesShadow AI discovery
Detecting unauthorised agentic-AI deployments inside the enterprise — telemetry patterns, inventory methods, policy response.
- 48 articlesAgentic AI governance
Governance frameworks, oversight patterns, and compliance postures for enterprise agentic-AI deployment.
- 18 articlesEnterprise AI cost
Verifying, tracking, and challenging the ROI claims vendors and analysts make about enterprise agentic AI.
Editor's picks
One per topic cluster- Governance90 days to EU AI Act enforcement: what the corpus says enterprises still haven't done
- Cost economicsThe hidden costs of agentic AI: a CFO's guide to true TCO and ROI modeling
- SecurityClaude Mythos: what 'too dangerous to release' means for your risk appetite and cyber posture
- ArchitectureNon-human identity for AI agents: the 2026 IAM playbook
- StrategyWhy 88% of agentic AI deployments fail
Latest pieces
Full archive →The Samsung lesson for shadow AI: detection lag is structural, not procedural
Samsung Electronics restricted ChatGPT and other generative AI on company devices in May 2023, after three separate internal incidents in April where employees pasted confidential source code, meeting transcripts, and yield-test code into the public ChatGPT interface. The detail in the public reporting is the load-bearing point. Samsung found the leaks after the fact, by audit, not by detection at the moment the paste happened. The detection lag was not a Samsung-specific operational failure. It was the predictable output of running enterprise data-loss prevention against a category of egress channel the controls were not built for. Three years on, most enterprise shadow-AI programmes still have the same gap.
Storm-0558 and the structural risk in AI agent credentials
The Cyber Safety Review Board's April 2024 report on the Storm-0558 intrusion catalogued the credential-management practices that produced the breach: a four-year-old signing key past its rotation policy, an environment boundary that did not enforce its own separation, a crash-dump leak that the existing detection tooling could not see, and a corporate account compromise that completed the chain. Read it forward, not backward: those same four practices describe how most enterprises are storing AI agent credentials in 2026. Storm-0558 was a forward indicator for the structural risk in non-human identity, not a one-off Microsoft incident.
The Energy Bill Nobody Budgeted For
Nvidia says agentic AI may need up to a thousand times the compute of a chatbot. The credible enterprise range is 10x to 100x by 2030. Even the floor of that range absorbs the renewable headroom the energy transition depends on, and almost no enterprise AI roadmap is pricing it.
Single-agent or multi-agent: what the 2026 deployment record actually says
The 2025–2026 deployment record shows single-agent architectures win on accuracy, cost, and MTTD below roughly 12 tool-domains. Multi-agent only pays back above that threshold, and only when inter-agent state is bounded by a shared structured artifact.
Public-sector agentic AI procurement: what the GSA and EU records show
Federal and EU member-state agentic AI contract records show renewals running materially below the enterprise SaaS benchmark. The driver is not technical performance but audit-evidence completeness under OMB M-24-10 §5 and EU AI Act Article 12. The procurement implication is structural.
Enterprise agentic AI in Q2 2026: what shipped, what slipped, what held
Of 8 major enterprise agentic AI vendor claims from Q1 2026, a minority are Holding at 90-day review. The pattern that predicts durability is not vendor size. It is whether the ROI evidence came from a customer or from the vendor itself.
Agentic AI in legal services: what survives the billable-hour decomposition
Three of the six billable-hour sub-tasks capture durable value with agentic AI. Two increase malpractice risk vs a junior-associate equivalent at the same time-to-delivery. One is bounded by conduct rules, not technology. The evidence from AmLaw 100 deployments now allows a clear-eyed breakdown.
The agent fan-out problem: when one prompt becomes 400 LLM calls
Production agentic systems amplify a single user request into dozens or hundreds of internal LLM calls. Most enterprise unit-economics, latency budgets, and observability setups are still priced for 1:1.
Browse by topic pillar
Five strategic pillarsComing next
Peter's editorial calendar — honest dates, bumped-with-notes if missed.- Week 1726 Apr 2026Non-human identity — the first procurement question CIOs aren't asking yet
Every enterprise agent deployment passes through a credential. Most teams still hand the agent a human's credential. Naming the NHI gap is the next Q2 procurement conversation.
- Week 1803 May 2026Shadow agent sprawl — what telemetry catches and what it misses
The browser-as-agent-runtime pattern creates a detection gap that MDM/CASB don't see. What the first wave of shadow-AI discovery tools actually find, and the three categories they miss.
- Week 1910 May 2026The AI agent MSA — four clauses every enterprise contract needs by August
EU AI Act enforcement activates 2 Aug 2026. The clauses that survive legal review in the next quarter will be the ones that don't pretend the agent is conventional SaaS.