A weekly read on what the AI noise is hiding.
SAP LeanIX asked 228 enterprise architects and IT decision-makers about AI agents: 98% of their companies have deployed or plan to deploy them, and 49% report even partial visibility into the ones in use or into their applications’ AI features. Europe has given companies until 2 December 2027 on its high-risk rules. Our read: use the time to write the list.
A vendor survey from SAP LeanIX, 228 enterprise architects and IT decision-makers fielded March 2026, finds 98% of companies have deployed or plan to deploy AI agents and 49% report visibility, even if only partial, into the agents in use or their applications’ AI features; the EU’s AI Omnibus moves Annex III high-risk obligations to 2 December 2027. Every stat sourced. The hype named as hype. One number worth acting on. Read the full brief below, free, no email required.
One survey asks how much of their AI companies can see. One regulator moved the date it will start asking. The deadline moved — the list still has to be written.
SAP LeanIX published its Agentic AI Survey 2026 from fieldwork in March 2026. The sample is 228 enterprise architects and IT decision-makers, 70% of them working for European companies and 18% for firms in the US or Canada. The subject is AI agents (SI agents, in the term the White House adopted for the executive branch on 29 September): software that takes actions inside a business, rather than software that answers questions about it.
98% of companies surveyed have deployed or plan to deploy AI agents. In the LeanIX sample, 44% are already deploying them, 40% are experimenting and another 14% plan to.
49% report visibility, even if only partial, into the AI agents currently in use or into the AI features offered by their existing applications. LeanIX states this one as a share of the enterprise architects surveyed, and the question bundles agents with application features, so the figure does not isolate agents. Read the other way, about half did not report even partial visibility into either.
Ownership is thin in the same sample. 48% have no clear roles or responsibilities for AI agents.
Population and screen. A vendor survey. SAP LeanIX sells enterprise architecture software, including tooling for keeping an inventory of AI agents, so the company measuring the gap sells the remedy. 228 respondents, fielded March 2026, mostly European. The page describes the sample as balanced in terms of company size, with 37% of respondents working in organizations of more than 10,000 employees; it does not break results out for companies in the $10M-$250M range. “Visibility, even if only partial” is the respondents’ own reading, not an audit of their systems.
The EU’s Digital Omnibus on AI was published in the Official Journal on 24 July 2026, after Parliament adopted it on 16 June and the Council on 29 June. It moves the date the AI Act’s high-risk obligations apply to standalone Annex III systems from 2 August 2026 to 2 December 2027. High-risk AI embedded in products already under EU product-safety law (Annex I) moves from 2 August 2027 to 2 August 2028.
Annex III is the list a mid-market operator is most likely to touch without knowing it — it includes AI used in employment, alongside education, credit, biometrics, critical infrastructure, law enforcement, justice and migration. A screening tool in a hiring workflow sits on that list.
The deferral changes the date. The Future of Privacy Forum’s reading of the text is that “the core obligations and legislative philosophy of the AI Act were not touched.”
Population and screen. A legal timeline, read through a policy analysis. Dates are taken from the Future of Privacy Forum’s post of 28 July 2026; the Official Journal text was not opened for this issue. Under Article 2, the AI Act reaches providers placing AI systems on the EU market wherever they are based, deployers established in the EU, and providers and deployers outside the EU where their system’s output is used in the EU. A US company with none of those links is outside it, and this issue does not say otherwise.
Last week the question was who owns the hours agents free up. This week’s question comes before it: which agents are running, and where does a human have to say yes.
Singapore already wrote one answer down. Its Model AI Governance Framework for Agentic AI, announced by Minister Josephine Teo at the World Economic Forum on 22 January 2026, builds on the Model AI Governance Framework Singapore introduced in 2020. It is voluntary guidance, not law. Its second dimension is “making humans meaningfully accountable for agents by defining significant checkpoints at which human approval is required.”
Put the three side by side. A vendor survey says about half of its respondents report no visibility, even partial, into their agents or their software’s AI features. Europe gave companies sixteen more months before its high-risk rules apply. A voluntary government framework says the first job is to name the points where a person approves. None of the three needs the others to be true.
On our reading, the extra sixteen months are the time to do that work. In AJ Maxwell’s terms, execution is the quality of a decision times its acceptance — E = Q x A. An agent nobody has listed is a decision nobody has accepted, because nobody knows it was made.
The takeaway for a CEO this week: ask for one page. Every agent running in the company, what it is allowed to do, and the one checkpoint where a named person approves its work. If the page takes longer than a week to produce, that is your finding.
Humans decide. Agents coordinate. Strategy enforces itself.
Find out where your company stands: the Align-ify readiness assessment takes about five minutes and shows where coordination is costing you today. readiness.align-ify.com
From SAP LeanIX, these figures run: 228, 98% with its 44%, 40% and 14% split, 49% and 48%, plus the 70% Europe and 18% US/Canada sample split and the 37% large-company share. Refused, and described here without their numbers: the page’s share of respondents lacking insight into how deployed agents support business capabilities, because its numeral is reserved on this publication’s list for a McKinsey subgroup share; the agent-performance visibility figure, which measures a different question from the 49% and invites a subtraction this publication does not make; and the two enterprise-architect shares, which are a profession’s view of its own role.
The 37% in this issue describes the LeanIX sample’s company size. It is unrelated to the McKinsey full-sample 37% on this publication’s list.
Not run: a claim circulating in our own research notes that “fewer than half hold a complete inventory of what the agents do.” The LeanIX page does not say that. It says 49% report visibility, even if only partial, into the agents in use or the AI features of existing applications, and that is the sentence printed here.
Singapore is not this week’s news. The framework launched in January 2026 and runs here as the standing answer to a question this week’s two items raise, not as a new development.
Named gap. No source in this issue reports results for companies in the $10M-$250M range.
Tooling limit, stated once: the fetch tool returns rendered content with no status line and no byte count, so neither appears below.
SAP LeanIX, Agentic AI Survey 2026. Fetched 9 November 2026, content returned; re-fetched by the claim pass the same day. Spans: callout “98 % of companies have deployed or plan to deploy AI agents”; “Forty-four percent of the companies surveyed are already deploying AI agents”; “40 percent are currently experimenting with AI agents”, “14 percent plan to do so”; “Slightly less than half (49%) of the enterprise architects surveyed report visibility, even if only partial, into the AI agents currently in use or into the availability of AI features offered by existing applications” (tail clause restored, R14); callout “48 % have no clear roles or responsibilities for AI agents”. Sample: “In March 2026, SAP LeanIX conducted an online survey of 228 enterprise architects and IT decision-makers”; “Seventy percent work for European companies, while another 18 percent are employed by firms in the USA or Canada”; “The sample is balanced in terms of company size”; 37 percent “in organizations with more than 10,000 employees”. Product disclosure (R7): the same page promotes “SAP AI Agent Hub: Discover and govern your AI portfolio”; product page https://www.leanix.net/en/ai-agent-hub found in search, not fetched. Negative claim re-asked 9 November (R31): no “complete inventory” figure, no $10M-$250M breakout. Carries §02, signals 1 and 2.
Future of Privacy Forum, “The AI Act Implementation Timeline: What Changes Under the AI Omnibus?”, Andreea Șerban and Bianca-Ioana Marcu, 28 July 2026. Fetched 9 November 2026, content returned. Dates read: political agreement 7 May 2026; Parliament 16 June 2026; Council 29 June 2026; Official Journal 24 July 2026; Annex III 2 August 2026 to 2 December 2027; Annex I 2 August 2027 to 2 August 2028. Span on the page itself (claim pass, 9 November): “the core obligations and legislative philosophy of the AI Act were not touched.” Official Journal text: NOT READ. Carries §03, signal 3.
AI Act, Article 2 (scope) and Annex III, Regulation (EU) 2024/1689 as reproduced by artificialintelligenceact.eu (Future of Life Institute). Fetched 9 November 2026, content returned. Art. 2(1)(a) providers placing on the EU market “irrespective of whether those providers are established or located within the Union or in a third country”; (b) deployers established or located in the Union; (c) third-country providers and deployers “where the output produced by the AI system is used in the Union”. Annex III areas: biometrics; critical infrastructure; education; employment; essential private and public services (5(b) creditworthiness and credit score); law enforcement; migration; administration of justice. Mirror of the 2024 text, not an Omnibus-consolidated version; EUR-Lex: NOT READ (R34, SUPPORTED). Carries §03 scope sentence and Annex III list.
MDDI Singapore, “Singapore Launches New Model AI Governance Framework for Agentic AI”. Fetched 9 November 2026, content returned. Release date 22 January 2026; Minister Josephine Teo; WEF Davos; span “Making humans meaningfully accountable for agents by defining significant checkpoints at which human approval is required”; “builds upon the governance foundations of MGF for AI, which was introduced in 2020”. The MDDI release does not use the words “voluntary”, “binding” or “law” (checked 9 November); voluntary status is carried by Rödl & Partner, 23 February 2026: “Although the framework is voluntary and does not create legally binding obligations”. The framework PDF itself: NOT READ. Carries §04.
SI gloss: Executive Order 14434, “Inaugurating the Era of Super Intelligence”, signed 29 September 2026, as cited and verified in Issue 17; existence and date re-confirmed by search 9 November (whitehouse.gov, UCSB Presidency Project). FETCHED AND VERIFIED 9 November 2026 (whitehouse.gov, HTTP 200): EO 14434, “Inaugurating the Era of Super Intelligence”, signed 29 September 2026. Note for any future use: the order is terminology only (directs “Super Intelligence”/“SI” in place of “AI” in non-statutory documents); it says nothing about agents, autonomy, or oversight — it may never be cited for an agent-governance point.
readiness.align-ify.com, rendered via r.jina.ai 9 November 2026 by the claim pass: “Find your number-one bottleneck in five minutes.” Carries the §04 CTA (R37).
One weekly read on what the AI noise is hiding. Every stat sourced, the hype named as hype, one number worth acting on. No spam, unsubscribe anytime.
Start by measuring the coordination tax you’re paying right now. Free, a few minutes.
Next Tuesday: another scan, same rule. Cut the hype. Show you what the noise is hiding.