Microsoft takes Agent 365 out of preview

AI agents are starting to look less like chatbots and more like software employees. They get credentials, touch company data, call tools, and act on their own. That is useful — but it also creates a new mess for IT and security teams. Microsoft’s move this week was to take Agent 365, its management layer for AI agents, out of preview and make it generally available on May 1, while also adding new preview tools for finding and governing “shadow AI” inside Microsoft 365 environments. ### What is Agent 365, exactly? Basically, Microsoft wants Agent 365 to be the control plane for agents. In plain English, that means one place to see which agents exist, what they are connected to, how they are being used, and what policies apply to them. Microsoft pitches it as a layer that works across Microsoft-built agents, partner agents, and other third-party agents, not just Copilot features. ### Why does “general availability” matter? Preview products are experiments with caveats. General availability means Microsoft is saying enterprises can treat this as a real production service. The May 1 launch also matters because it lines up with a broader push to make agent management part of the normal Microsoft 365 and security stack, instead of a side console for early adopters. ### What changed in this release? The big change is that Agent 365 now officially ships as a product, with expanded capabilities around observing, governing, and securing agents. Microsoft says the platform is also extending coverage to agents that operate with their own credentials and permissions — which is a big deal, because those are the agents that start behaving more like independent actors inside a company’s systems. ### What is “shadow AI” here? This is the part that makes the story more than a product launch. Microsoft uses “shadow AI” for unmanaged or unapproved agents showing up inside an organization — including local agents that employees can run outside the usual procurement and admin process. The new Shadow AI page in the Microsoft 365 admin center is still in preview, but it is designed to help admins discover those agents, monitor them, and take governance actions. ### Why are local agents such a problem? Because they break the old assumption that useful enterprise software arrives through a sanctioned app catalog. A local agent can be downloaded, pointed at company files, and given enough autonomy to do real work before IT even knows it exists. That is closer to shadow IT than to ordinary chatbot use — but with malicious agents such as OpenClaw. ### Is this mostly about security, or productivity? Both, but security is the forcing function. Microsoft’s own framing is that companies want the upside of autonomous agents without the “YOLO” version of deployment where anything can run anywhere. So Agent 365 sits in the middle — not to stop agents, but to make them legible enough to approve, audit, and contain. ### Why does this matter beyond Microsoft? Because it shows where enterprise AI is heading. The first phase was “let’s build agents.” The next phase is “who governs them?” Microsoft is turning that second question into a product category, and doing it inside the admin, identity, compliance, and security tools companies already use. That is a sign the market is moving from demos to operations. ### Bottom line? The news is not just that Agent 365 left preview. It is that Microsoft thinks rogue and unmanaged agents are now common enough to justify a permanent control layer. Turns out the hard part of agentic AI is not only making agents useful — it is keeping them visible, governed, and on a leash.