The Decision Layer Compass™ · Week 2 Executive Edition

Your firm is buying an agent workforce.

Five developments on agentic AI adoption. Agents are arriving through procurement, through employees, and through a licence renewal, and only one of those three routes has a governance gate.
Published8 September 2026
Read time5 minutes
CycleWeek 2 of 4
FollowsIssue 02 · 1 September
01 · The five signals
S1
Agentic AI adoption

The agent count is forecast to move by four orders of magnitude

What happenedGartner projected in April that the average Fortune 500 enterprise will run more than 150,000 AI agents by 2028, against fewer than 15 in 2025. Cloud Security Alliance research found 53% of organisations had already seen agents exceed their intended permissions.
Why it moves a priorityA permissions model built for people does not survive that arithmetic. Access reviews, joiner and leaver processes, and segregation of duties all assume a human who can be trained, disciplined, and removed.
Question to put to managementHow many agents are running in our environment today, and who could answer that without starting a project?
S2
AI procurement

AI governance has become a scored procurement category

What happenedA Gartner survey found 83% of Fortune 500 procurement teams intend to require technology vendors to work to ISO 42001 by 2027. The 2026 SIG questionnaire added an AI governance section, CAIQ now carries AI-specific control mappings, and UKAS granted accreditation for ISO 42001 certification in January.
Why it moves a priorityThis runs both ways. Your suppliers will be scored on it, and clients will score you. Both sides now expect documented process where a statement of principles once sufficed.
Question to put to managementIf a client sent us their AI governance questionnaire this week, what would we send back?
S3
Shadow AI

Shadow AI is now an insider category, not an anecdote

What happenedThe Verizon 2026 Data Breach Investigations Report recorded a fourfold rise in shadow AI detections in a year, making it the third most common non-malicious insider action. A PagerDuty survey of 1,250 professionals at firms above $500 million in revenue found 66% had used AI tools they believed were against policy, and more than a third had entered customer data into public models.
Why it moves a priorityEmployees are now building agents as well as pasting into chatbots. An agent created by a well-meaning analyst holds that analyst's access and acts on it without supervision.
Question to put to managementWhat is our approved alternative, and how long does it take an employee to get access to it?
S4
AI governance tooling

Tooling is being bought before the governance decision is taken

What happenedA category of discovery and control products has formed quickly, covering tool discovery, browser-level usage control, and agent registries. ISO/IEC 42006 now sets requirements for the bodies certifying AI management systems.
Why it moves a priorityA discovery tool produces a list. It does not say who owns each item, what evidence supports its use, or what would force it back to a committee. Buying the tool first produces an inventory nobody is accountable for.
Question to put to managementBefore we buy a tool, who will own the list it produces?
S5
Microsoft Copilot evolution

The agent platform arrived inside a licence renewal

What happenedMicrosoft 365 E7 became generally available on 1 May at $99 per user per month, bundling E5, Copilot, the Entra Suite, and Agent 365, the control plane for governing autonomous agents. Tenable researchers then showed a Copilot Studio agent could be prompted into disclosing customer card records and altering stored prices without specialised skills.
Why it moves a priorityFor most firms the agent decision arrives as a licensing conversation, decided on cost per seat, with the governance implications discovered afterwards.
Question to put to managementIs our next Microsoft renewal being evaluated as a cost decision or as an agent deployment decision?
02 · Position movement

Against the dashboard published in Issue 02 on 1 September.

Agent permission sprawl
New entry at 87
Shadow AI and employee-built agents
New entry at 79
Vendor-borne model exposure
89, held
Model supply concentration
78, held
AI decision confidence index
52 to 49, down 3

Confidence falls because two new positions enter above the median score, and neither has a named owner.

03 · What did not move

The classification rule and the third-party register are unchanged for a third month. Both remain the cheapest items to fix, and both remain unowned.

04 · Sources

Gartner, 2026. Cloud Security Alliance, April 2026. Verizon Data Breach Investigations Report 2026. PagerDuty Shadow AI Survey 2026. Microsoft 365 E7 and Agent 365 availability, 1 May 2026. Tenable research on Copilot Studio prompt injection. Third-party figures are reported as published and have not been independently verified.

05 · The invitation

Open. Not pushy.

Working instrument

Agentic AI Decision Boundary Check™

Your organisation may already have more agents than its governance model can see.

A short working instrument to test five things:

What agents exist. How they entered. What they can access. Who owns them. What would make someone stop them.

The interesting answer is rarely "none."

It is usually: "We would need to ask around."

And that is already a signal.

Run the check→
Decision conversation

Book a 45-Minute Agentic AI Governance Conversation

Bring one live agent use case, procurement decision, Copilot deployment, or licence renewal.

We will trace it from entry point to authority, permissions, ownership, evidence, and escalation.

No maturity score. No tour of an AI framework.

Just one practical question:

𝐈𝐟 𝐭𝐡𝐢𝐬 𝐚𝐠𝐞𝐧𝐭 𝐚𝐜𝐭𝐬 𝐭𝐨𝐦𝐨𝐫𝐫𝐨𝐰, 𝐜𝐨𝐮𝐥𝐝 𝐲𝐨𝐮 𝐬𝐡𝐨𝐰 𝐰𝐡𝐨 𝐚𝐮𝐭𝐡𝐨𝐫𝐢𝐬𝐞𝐝 𝐢𝐭, 𝐰𝐡𝐚𝐭 𝐢𝐭 𝐜𝐚𝐧 𝐝𝐨, 𝐚𝐧𝐝 𝐰𝐡𝐨 𝐜𝐚𝐧 𝐬𝐭𝐨𝐩 𝐢𝐭?

Book a diagnostic conversation→

That's The Decision Layer Compass™ for this month.

Reading this week's five signals takes five minutes. Discovering six months from now that autonomous agents entered through procurement, employee experimentation, or a licence renewal, without clear ownership, permissions, or governance, will take considerably longer to unwind.

The costly part will be explaining who authorised them, what they can do, what evidence supports their use, and why nobody noticed the decision had already been made, not finding agents. That is precisely the gap this issue surfaces: from agent permission sprawl and shadow AI to tooling bought before governance and Microsoft licensing decisions that quietly become agent deployment decisions.

Forward this to a CIO, CISO, CRO, CAE, General Counsel or Audit Committee Chair who still thinks agentic AI is a future implementation problem, and who would prefer to discover the governance gap before the agent workforce arrives through the side door.

Maman Ibrahim · F-ISRM · F-IoCR · CISSP · CISA · CRISC
Decision Signals · The Decision Layer™ · © 2026 DiamondSoul