Microsoft turns Copilot into an agent platform with Agent 365
New IT console announced May 5 gives admins inventory, role-based controls, and policy hooks for autonomous agents — Microsoft's clearest signal yet that AI agents are the next operating layer for the productivity suite.
Microsoft on Tuesday unveiled Agent 365, a dedicated administrative console for IT teams to deploy, monitor, and secure autonomous AI agents at enterprise scale. The product reframes Copilot for the IT-leadership audience: less “AI in your sidebar,” more “platform for digital workers running across the business.”
The May 5 announcement formalizes a shift Microsoft has been telegraphing for two quarters. Agents — software that can reason over business processes, execute multi-step tasks, and collaborate with each other — are the next operating layer for work, in Microsoft’s framing. Agent 365 is the surface IT operates that layer through.
What Agent 365 actually does: it provides an inventory of every agent deployed across the organization. It gives administrators role-based controls for which agents can act on which data, with policy hooks for compliance. It supports lifecycle management — deployment, monitoring, retiring — for both Microsoft-shipped agents and customer-built ones. It is, in shape, what Microsoft Endpoint Manager was for managed devices in 2017: the IT-facing pane of glass that turns a scattered category into a managed surface.
The Frontier program — Microsoft’s tier of customers running the most aggressive Copilot deployments — gets reusable Cowork Skills (named, parameterized agent capabilities that any agent can call) and a new plugin model that lets enterprise teams build agent skills once and reuse them across multiple agents.
For enterprise software buyers, the procurement question is not whether to adopt Agent 365 in isolation. The question is whether the Microsoft agent stack — Copilot Cowork as the user-facing surface, Agent 365 as the IT-facing surface, Work IQ as the data layer — becomes the default agentic operating system inside Microsoft-shop enterprises. The deployment economics increasingly point in that direction. Once an organization has 78% of its knowledge workers using AI agents weekly, by the figure Microsoft cited in its accompanying Work Trend Index disclosure, the value of a unified control plane begins to dominate the value of any individual agent.
Salesforce, Anthropic, and Google have all pushed agent-first framings into enterprise pitches over the past several quarters. Microsoft’s May 5 release brings the same conceptual frame to the productivity install base — where the contest is between the platforms, not between the agents.