What Is Microsoft Agent 365?

Microsoft Agent 365 is the company’s new control plane for AI agents inside Microsoft 365: a single place for IT and security teams to see what agents are running, who is allowed to use them and what they are allowed to do. It went generally available on 1 May 2026 at $15 per user per month, and is also bundled into a new Microsoft 365 E7 “Frontier Suite” tier at $99 per user per month.

Agent 365 is not a chatbot, and it is not something most employees will ever open. It is a back-office control layer aimed squarely at IT, security and compliance teams. The pitch from Microsoft is that AI agents are now turning up everywhere inside large organisations, built by Microsoft, by software vendors, by internal developers, and increasingly installed on people’s laptops without anyone in IT knowing about it. Agent 365 is the place where those agents are supposed to be catalogued, named, given an identity, scoped to a set of permissions and watched. If an auditor or regulator asks what AI is running on the company’s data, Agent 365 is meant to be the place where the answer lives.

The headline launch was on 9 March 2026, when Judson Althoff, CEO of Microsoft Commercial Business, set out a 1 May 2026 general availability date in a post on the Official Microsoft Blog. That date has now passed: on 1 May 2026, Corporate Vice President Nirav Shah and colleagues confirmed Agent 365 was generally available for commercial customers, and used the GA post to announce a series of new previews layered on top. So the basic product is live; several of the more interesting capabilities sit in public preview rather than general availability, which we will come back to.

The core building block is the Agent 365 Registry, plus a piece of identity infrastructure called Microsoft Entra Agent ID. Together these treat every AI agent like an employee: each agent has an identity, is logged somewhere central, can be granted or revoked access to specific data and tools, and can be observed when it does something the security team does not like. Microsoft Defender, Microsoft Entra, Microsoft Intune and Microsoft Purview all plug into the same control plane, which means an organisation that has already invested in Microsoft’s security tooling gets to reuse it for agents rather than buy a parallel governance stack. For a business that has already standardised on Microsoft 365, this is the most coherent part of the proposition.

One of the more interesting design choices is that the Registry is open to non-Microsoft agents. Microsoft says Agent 365 is meant to cover agents built on its own platforms, agents from named software development partners (the May GA blog lists Genspark, Zensai, Egnyte, Zendesk, Kasisto, Kore and n8n), and agents built on rival agent platforms such as AWS Bedrock and Google’s Gemini Enterprise Agent Platform via a new public preview of “registry sync”. On paper, that turns Agent 365 into something closer to a multi-vendor inventory than a Microsoft-only walled garden. Whether it works in practice depends on whether third-party vendors actually plug in, whether internal developers bother to register the agents they build, and what happens when they do not. Microsoft has not published a clear enforcement story for unregistered or non-compliant agents in the materials reviewed for this article. That gap is worth keeping an eye on as adoption grows.

The other half of the announcement is the new Microsoft 365 E7 “Frontier Suite” tier. E7 bundles Microsoft 365 E5, Microsoft 365 Copilot, Agent 365, the Entra Suite and a set of advanced Defender, Intune and Purview capabilities into one $99-per-user-per-month licence, available with and without Teams. Microsoft says this works out cheaper than buying the components separately, though it has not published a like-for-like price comparison and the individual a la carte prices are not all public, so the “cheaper than the sum of the parts” claim is a vendor claim rather than something an outside reader can verify. For customers already on E5 plus Copilot plus a chunk of Entra and Purview add-ons, E7 may simplify the bill. For anyone further down the Microsoft 365 stack, the jump is sizeable, and the comparison worth doing on a spreadsheet rather than on a sales call.

Agent 365 was announced alongside, but is distinct from, the next round of changes to Microsoft 365 Copilot itself, branded Wave 3. Wave 3 brings what Microsoft calls “agentic experiences” into Word, Excel, PowerPoint and Outlook: rather than just answering a prompt, Copilot is meant to carry out multi-step tasks inside the document a person is already working on, and employees can build their own small agents inside the same application canvases. An intelligence layer called Work IQ uses signals from across an organisation’s Microsoft 365 data to make those experiences context-aware. There is also a new enhanced chat experience that can produce and refine longer-form artefacts directly in conversation. All of this is positioned as Microsoft 365 Copilot evolving, not as a separate product.

The model strategy inside Copilot is now openly multi-vendor. Microsoft 365 Copilot can route to both OpenAI’s latest models and to Anthropic’s Claude. Claude is available in mainline Copilot chat for customers who join Microsoft’s “Frontier program”, a programme Microsoft describes but has not fully priced in the public materials seen here, so whether Frontier access carries an additional cost on top of a standard Copilot or E7 licence is not clear from the announcement. A second Anthropic collaboration, Copilot Cowork, was described as a research preview at the March announcement, and is built on the same long-horizon technology that powers Anthropic’s Claude Cowork. Microsoft has not committed to a general availability date for Copilot Cowork, and its scope may change.

A few of the GA-day previews are worth flagging because they hint at where this is heading. Windows 365 for Agents, a separate but related product, is now in public preview in the United States only and provides a managed Cloud PC environment where agents can run with the same identity and security controls as employee devices. Microsoft Defender and Intune are gaining the ability to discover and block “shadow” AI agents installed on Windows machines, starting with OpenClaw and expanding to GitHub Copilot CLI and Claude Code, with policy-based runtime blocking and asset context mapping in preview from June 2026. None of those features are at GA yet. Anyone reading the announcement and assuming the full shadow-AI story is already live should slow down and check what is and is not generally available before sizing a project around it.

For a small business or a non-technical manager, the most important thing to take away from Agent 365 is what it is not. It is not an out-of-the-box AI assistant, and it is not designed to be useful without somebody who can set up identity, access and security policies. It is also priced for the enterprise: at $15 per user per month on top of whatever Microsoft 365 licences are already in place, or $99 per user per month for the full E7 bundle, it is unlikely to land on the desks of most ten-person businesses any time soon. The questions worth asking before signing up are practical: who in the organisation will own the Agent 365 Registry day to day, what policy will govern which agents are allowed to access which data, how will employees be told what they can and cannot install on their work devices, and how will the company react when an agent does something unexpected? The answers to those questions matter more than any feature in the product.

It is also worth being clear-eyed about who is verifying which claims. The adoption metrics Microsoft has published, more than 160% year-on-year growth in Copilot paid seats, ten times more daily active use, 90% of the Fortune 500 using Microsoft 365 Copilot, tens of millions of agents registered during the Agent 365 preview, and more than 500,000 agents inside Microsoft itself with around 65,000 agent responses per day, all come from Microsoft itself and have not been independently audited. The named enterprise customers Microsoft has highlighted, including Mercedes-Benz, NASA, Fiserv, ING, the University of Manchester, the US Department of the Interior and Westpac, are real and on the record, but most are described as Copilot customers rather than Agent 365 customers specifically. The IDC forecast Microsoft cites, of 1.3 billion agents in circulation by 2028, is exactly that: a forecast.

The wider context to keep in mind is that Microsoft is no longer locked into a single model partner. Its exclusivity arrangement with OpenAI shifted during April 2026, opening the door for Microsoft to offer a wider range of models across its cloud and for OpenAI to sell directly through other clouds. The model-diverse design of Copilot and Agent 365 fits that new arrangement: Microsoft is positioning itself as the platform you run AI on, not as the supplier of any one underlying model. For customers, the practical effect is that the choice of which AI sits behind Copilot or behind an agent is becoming a procurement decision in its own right, with Microsoft now happy to take payment on whichever model wins inside a given customer.

So, what is Microsoft Agent 365 actually for? It is for the IT, security and compliance teams who already know they are losing track of AI agents inside their organisation and want a single place to start getting that back under control. For those teams, the most useful starting move is probably to look at what is already running before deciding whether the E7 bundle is worth it: an Agent 365 Registry full of agents nobody can account for is itself a useful, if uncomfortable, piece of information. To go deeper on the trade-offs and risks of letting AI agents act on your systems, see our explainers on what AI agents can actually do today, what can go wrong when AI agents act on your behalf, and how to create a simple AI policy for a small business.

Sources: Microsoft, “Introducing the First Frontier Suite built on Intelligence + Trust” (9 March 2026); Microsoft Security Blog, “Microsoft Agent 365, now generally available, expands capabilities and integrations” (1 May 2026). Image: Microsoft Agent 365 launch material, used for editorial reporting purposes.