Microsoft used its 21 April 2026 partner update to sharpen the enterprise case for Agent 365: a unified control plane for building, deploying and governing AI agents across the Microsoft cloud. The announcement arrived alongside an IDC forecast that there will be 1.3 billion AI agents operating in enterprises by 2028.
Put those two numbers together and the picture is clear. Agents are about to proliferate faster than most organisations can catalog them. A control plane is not a nice-to-have. It is infrastructure.
What Agent 365 is supposed to solve
Agent 365 aims to bring order to a fragmented landscape. Today, agents can be built in Copilot Studio, Azure AI, Power Automate, custom code and an expanding list of third-party tools. Each has its own identity model, permissions, logging and lifecycle. The result is predictable: duplicate capabilities, orphaned agents, inconsistent access policies and no clear owner when something goes wrong.
A unified control plane promises to centralise:
- Identity and access. Every agent has a defined identity, scoped permissions and a clear owner.
- Policy enforcement. Guardrails apply across build tools, not just the ones IT remembers to check.
- Lifecycle management. Agents can be versioned, tested and retired rather than left running indefinitely.
- Observability. Logs and audit trails are aggregated in one place.
In principle, this is exactly what a CIO needs. In practice, the value depends on how comprehensively the control plane covers the agent population — including the ones built outside Microsoft’s walls.
The sprawl problem is bigger than Microsoft
IDC’s forecast of 1.3 billion agents by 2028 is useful as a directional signal. Even if the precise number is debatable, the scale is not. Agents will outnumber applications in many organisations. Some will be built by IT. More will be built by business users with low-code tools. Many will be bought as features inside SaaS products.
Agent 365 can govern the Microsoft-native portion of that estate. It cannot govern an agent built in a separate platform that connects to the same data via API. That means enterprise governance still needs an inventory layer above any single vendor’s control plane.
Three governance moves to make now
Create an agent register. Before you can govern agents, you need to know they exist. The register should record purpose, owner, data sources, permissions, build platform and retirement date. Treat it as a living list, not a one-time audit.
Define agent classes. Not every agent needs the same level of oversight. A draft-generating assistant is different from an agent that approves expenses or triggers payments. Classify agents by decision authority and data sensitivity, then apply controls proportionally.
Plan for cross-platform policy. Even if Microsoft is your primary stack, assume agents will be built elsewhere. Write policies in terms of outcomes — what an agent may access, what decisions it may make, what logging it must produce — rather than in terms of specific tools. That way the policy survives the next vendor announcement.
The realistic read
Agent 365 is a credible response to a real problem. It will help enterprises that are already Microsoft-centric and willing to invest in the governance discipline. It will not, by itself, prevent agent sprawl. No single control plane can.
The firms that navigate this well will be the ones that treat agent governance as a cross-functional capability, not a Microsoft feature. Legal, compliance, security and operations all have a stake. The technology provides leverage, but the rules still have to come from the business.
With billions of agents on the horizon, the time to build that capability is now — while the estate is still small enough to inventory.