Agent inventory blind spot: Microsoft Agent 365 and the register you owe the board · Enterprise Agentic AI Insights
Enterprises will run 1,600+ AI agents by year end but fewer than 1 in 5 keep a live inventory. Use Agent 365 and Agent Fabric to stand up the register before the EU AI Act Article 50 deadline.
You can open your HR system and name every person on payroll. Try the same with the AI agents now acting inside Microsoft 365, Salesforce, ServiceNow and your cloud accounts. Most CIOs cannot produce that list, and the gap is widening fast. The numbers are blunt. IBM's Think 2026 data points to 1,600+ agents running inside the average large enterprise by year end, with 70% of leaders saying they cannot govern the ones they already have. OutSystems, surveying 1,900 IT leaders, found 96% adoption and 94% already worried about sprawl. Gravitee's State of AI Agent Security 2026 reports 88% of organizations had a confirmed or suspected agent incident in the past year, while 48% of production agents run unmonitored and only 12% have a single console to manage them. That is the operational fire on the CIO and CDO desk this quarter: agents spun up in a dozen tools, no shared register, no single owner. It shows up as a mystery integration touching a customer record, a spend line no one forecast, and a "who approved this agent" question with no clean answer. The platform vendors have started handing you the register. Microsoft Agent 365 is now generally available as a control plane: one view of agent inventory, permissions, behavior and activity, with discovered local agents surfacing in Defender and Intune and registry sync that inventories agents across AWS Bedrock and Google Cloud. Salesforce answered with Agent Fabric and Agent Script in Summer '26, adding guided determinism so an agent's actions are constrained rather than open-ended. The tooling to see and box in your agents exists. The discipline to use it is the missing piece. Here is the board-level turn. The EU AI Act Omnibus, finalized 16 June 2026, delayed high-risk obligations (standalone Annex III systems move to 2 December 2027, product-embedded Annex I systems to 2 August 2028). That delay is not a reprieve on transparency. Most Article 50 obligations still land on 2 August 2026, eleven days out: EU-facing chatbots and generative systems must disclose they are AI, and synthetic media must be labeled. An agent you cannot name is an agent you cannot prove is compliant, and for the Chief Risk Officer that is an audit-evidence gap that exists today, delay or not. The move is simple and it is not a rip-and-replace. Stand up the register this quarter using the control plane your platform already gives you, assign every agent a named owner and a purpose, and mark which agents touch EU users so the Article 50 disclosure question has an answer before the deadline. That is the difference between "we run agents" and "we can account for every agent," which is the sentence a regulator, a board, and a customer all want to hear. Sources: IBM Think 2026; OutSystems 2026 State of AI Development; Gravitee State of AI Agent Security 2026; Microsoft Agent 365 GA; Salesforce Agent Fabric; EU AI Act Omnibus (Gibson Dunn, Jones Walker).