More and more organizations are running AI agents without a clear picture of what they have or who owns them. This is a practical look at Microsoft Agent 365: what it does, what it does not do, how the licenses compare, and who actually needs one.
Walk into most Microsoft environments today and ask three simple questions. How many AI agents are running in your tenant? Who owns them? What can they touch? You will rarely get a straight answer.
Copilot Studio made building agents almost trivial. A business analyst spins one up on a Friday afternoon. Someone in finance connects one to a SharePoint site. A vendor drops a partner agent into the mix. Six months later you have dozens, sometimes hundreds, and no single list of what actually exists.
This is not hypothetical. It is the same movie we watched with Power Platform. Makers built Power Automate flows and Power Apps faster than IT could track them, and governance showed up years too late. Agents are that story again, only faster and with far more access.
Agent 365 is Microsoft's answer to that gap.
Agent 365 is Microsoft's central management and governance layer for AI agents. Microsoft calls it a control plane. In plain language, it is where you register, see, own, monitor, secure and retire the agents in your organization.
It became generally available on 1 May 2026 and plugs into the tools your admins probably already use: the Microsoft 365 admin center, Entra ID, Defender, Purview and Intune. So not a whole new console to learn from scratch.
What it covers:
In practical terms, Agent 365 hands IT and security a set of capabilities that map onto problems teams already have:
Agent 365 is not for the end user chatting with an agent. It is for the people who have to keep the whole thing under control.
In other words, if you are accountable for the AI landscape as a whole rather than for using one agent, Agent 365 is aimed at you.
Agent 365 is licensed per user, not per agent. One user license covers all the agents that person uses, owns, sponsors or manages. The agents themselves do not need a license. That is a deliberate design choice: cost follows the humans responsible for agents, not the count of agents, so growing your fleet does not multiply your governance bill.
Note: A "sponsor" is the human responsible for an AI agent. They provide accountability for the agent’s purpose, access, lifecycle, and governance, ensuring every agent has a clear business owner and does not become an unmanaged or orphaned agent.
The free foundational plan gives every Microsoft Cloud customer agent identity, an inventory and basic insights. For a small organization experimenting with a few agents, that visibility can genuinely be enough to start. You get to answer the "what do we have" question before you spend anything.
The paid plan is what you reach for once agents touch sensitive data or run at scale. It adds the advanced governance, security posture, threat detection and data controls that a security or compliance team will ask for the moment agents move beyond a pilot. If you operate in a regulated sector, you will land here quickly.
Two things are easy to overlook here. The first is that the standalone license assumes you already have a qualifying Microsoft 365 base. In practice that means one of the following:
The second is that the license covers governance and security only. Building and running the agents is separate consumption through Copilot Studio or Foundry, billed on your Azure invoice.
Microsoft's longer-term vision goes further than governance. Its documentation describes "agent users", a full identity type where an agent gets its own mailbox, its own OneDrive and a place on the org chart, with a name such as agent@yourtenant.onmicrosoft.com.
Agent 365 will not do that thinking for you. It gives you the registry, the identities, the policies and the monitoring, but you still need an ownership model, an approval process and someone accountable when an agent does something it should not. Treat it the way the mature organizations eventually treated Power Platform.