Microsoft Agent 365 Just Went GA — and Quietly Made Claude Code, Copilot CLI, and OpenClaw All Subjects of the Same Identity System

If you spent the last twelve months arguing in standup that "agents are just like service accounts, but worse," congratulations — on May 1, 2026, Microsoft made you right at planetary scale. Agent 365 hit general availability across commercial Microsoft 365 tenants, and the design choice the industry is still digesting is this: every AI agent now has its own first-class identity in Entra ID, exactly like a human employee, with onboarding, group membership, conditional access, and offboarding. The era of an agent borrowing a service principal's creds is over for any organization in the M365 stack.

1. What "Control Plane" Actually Means

Agent 365 sits above the agents themselves. It does not run them; it observes, governs, and secures them. At GA, that means three things in practice:

Observe — A unified inventory of every agent across Windows endpoints, Azure, and (newly in public preview) AWS Bedrock and Google Cloud Gemini Enterprise Agent Platform. You finally have a CMDB for non-humans.

Govern — Purview data labels follow the agent across tools, so a confidential SharePoint document referenced by an agent prompt remains confidential when the agent feeds it to a downstream model. Defender for Cloud Apps applies runtime policy: an agent that suddenly tries to enumerate /etc/shadow gets blocked the same way a user would.

Secure — Intune now manages local agents on Windows. Day-one supported runtimes include the open-source OpenClaw, GitHub Copilot CLI, and Anthropic's Claude Code. That last one is striking: a Microsoft control plane is shipping first-class management for a competitor's agent.

2. The Pricing Story Nobody Wants to Talk About

Agent 365 is $15 per user per month standalone, or bundled into Microsoft 365 E7 — the new "Frontier Suite" SKU — at $99 per user per month. E7 also includes Copilot Wave 3, Cowork, and the upgraded Defender XDR with agent-aware detections. For an enterprise already on E5 ($57), the upgrade math is roughly "am I willing to pay $42 more per seat to stop having a dozen agent tools that nobody can audit?" My read of three large IT roadmaps this week: yes, but only after security review.

3. Why This Hits PHP and Web Teams Sooner Than You Think

You might assume an Entra/Microsoft control plane is irrelevant if your stack is LAMP and your CI is GitHub Actions. It isn't. The moment your customers' IT departments mandate that any agent reading their tenant data must be Agent 365-registered, your SaaS app's embedded agents (the chatbot, the auto-doc tool, the AI search box) need to support OAuth flows that mint Entra agent identities. Vendors who don't will be quietly removed from approved-tools lists in Q3.

4. The Real Story: Identity Is the New Network

Ten years ago, network segmentation was the primary defensive surface. Five years ago, it was endpoint. Today, with agents acting on behalf of users across a dozen SaaS tenants per minute, identity has become the only durable boundary. Agent 365 is Microsoft's bet that the company that owns identity for humans owns it for agents, and the integrations with AWS Bedrock and Google's Gemini Enterprise platform on day one suggest the bet is paying off — everyone is happier delegating governance than rebuilding it.

My Take

The most underrated detail of this launch is that Microsoft is willingly managing Claude Code — an Anthropic product — from inside its own control plane. That decision tells you the platform layer of agentic AI is being settled the same way the SaaS identity layer was settled around Okta and Entra: not by who has the best agent, but by who controls the seat the agent sits in. If your roadmap still treats "our AI agent" as a feature, you're fighting yesterday's war. The fight that matters in Q3 2026 is who governs the agent, not who built it.