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.
Claim created at publish; review on 60-day cadence (the model-capability and agentic-framework release tempo means the fan-out ratio band can shift within a single quarter). The 1:18 floor is derived from constrained, well-instrumented production deployments documented in LangSmith multi-step agent case studies and the Anthropic Building Effective Agents post (December 2024); the 1:60 upper band reflects research and orchestration agents documented in the Microsoft Magentic-One paper (arXiv 2411.04468) and Helicone cost-per-task analysis; the 1:400 tail reflects unbounded ReAct agents documented in OpenAI Agents SDK warnings and LangSmith trace data. The band is asserted as the observable range across documented case studies, not as a measured median across a systematic sample — the systematic cross-deployment study does not yet exist in the public literature. Three triggers to revisit before next cadence: (a) a frontier model that internalises multi-step reasoning within a single API call, collapsing the billing-layer fan-out ratio, which would require reframing the cost section while preserving the latency and observability sections; (b) a major observability platform (LangSmith, Helicone, Datadog AI) publishing a cross-deployment fan-out ratio study that either validates or revises the 1:18–1:60 band with systematic data; (c) provider-level task-completion pricing replacing per-call pricing, which changes the unit-economics framing without changing the detection or latency implications. Sister claims: AM-013 (Q1 2026 agentic AI thresholds — the detection-time and observability themes are directly related), AM-148 (GPT-5.5 vs Opus 4.7 cost-per-task framing — the fan-out ratio is a multiplier on the per-call cost comparison). The MTTD-for-Agents tool-use-frequency Z-score tripwire described in this article is the primary detection mechanism; full methodology at /mttd/.
/holding/AM-149/Embed this claimiframe + oEmbed
The card auto-updates when the claim's status, last-reviewed date, or correction log changes. Embedders never need to refresh — the card is rendered live from the canonical record.
Email-me when AM-149's status, next review date, or correction log changes. One email per change. No newsletter subscription, no other mail.
The 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.
About this register
The Reporting register tracks claims published from articles addressed to senior enterprise IT leaders — CIOs, IT directors, heads of platform. Claims are reviewed on a 30–90 day cadence; each review either reaffirms the claim, marks one substantive part as Partial, or marks it Not holding once the underlying evidence has been overtaken.
Recent corrections in Reporting
- AM-008 · Partial · 17 Jun 2026
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.
- AM-132 · Partial · 10 Jun 2026
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.
- AM-129 · Partial · 10 Jun 2026
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.
Reviews coming up in Reporting
- AM-063 · Holding · next +9d (27 Jun 2026)
AI agents executing financial transactions need a four-control bundle (action-approval gates by blast radius, kill-swit…
- AM-061 · Holding · next +9d (27 Jun 2026)
Production agentic-AI costs at scale routinely run multiples of POC projections, and a layered optimisation programme c…
- AM-003 · Partial · next +9d (27 Jun 2026)
GPT-5 Pro's tiered-subscription model forces enterprises to classify problems by computational difficulty — $200/month…