Copilot Autopilot ships — and the agent tier is metered

Microsoft shipped the Copilot super app today, and the Copilot Autopilot tier answers the question we flagged in August: no, your existing seat does not cover it. In the announcement, Jared Spataro laid out three destinations — Home, Code and Autopilot — and, underneath them, a billing split that turns background agent work into a variable line item. If you run Microsoft 365, that split is the whole story.

What actually happened

Copilot now opens into Home (Chat and Cowork together, with Word, Excel and PowerPoint generating real editable documents in-line), Code (describe a widget, dashboard or app in plain language and get a working one, built on GitHub Copilot technology and sandboxed inside your tenant), and Autopilot — the agent formerly called Scout. You give an Autopilot a name, a role and a goal. It watches channels, chases threads, runs recurring work, and picks a project back up days later without a prompt, with its own identity and its own workspace.

Microsoft also opened Copilot Managed Runtime in public preview: Microsoft-hosted execution for the code Copilot writes, running inside the Microsoft 365 tenant boundary under IT governance, with an SDK so third-party developers can build against the same host.

Then the pricing model, stated plainly in Microsoft's own post. A User Subscription License covers Chat and the Office integrations at a fixed cost, including "Auto" model routing. Usage-based billing covers Cowork, Code, Autopilot and frontier models. Agent 365's FinOps controls extend to Code and the managed runtime — admins set spending policies and restrict model availability by group; end users see their credit balance inside Copilot.

Rollout: Home and Code go to Frontier participants over the coming weeks, Autopilot expands to private preview at month end, and Code reaches Microsoft 365 Premium and Pro subscribers later in 2026. VentureBeat notes that Microsoft did not publish per-unit rates for Autopilot, Code or the runtime at announcement — against published baselines like $30 per user per month for Microsoft 365 Copilot. So the meter exists; the rate card does not yet.

Why usage-based agent billing matters for your business

A seat price is a budget. A meter is a risk. You can forecast 40 seats at a known rate. You cannot forecast what an always-on agent costs when it decides a thread needs following for six days. Before anyone in your company names an Autopilot, decide the monthly ceiling and set it in Agent 365 FinOps. A spending policy you configure on day one is cheap. The same policy after a surprise invoice is a meeting.

"Its own identity and its own workspace" is a permissions question wearing a product name. An agent that watches channels and picks up projects has standing access to whatever you granted it, indefinitely, with no human in the loop to notice drift. Write down — actually write down — which mailboxes, channels and SharePoint sites each Autopilot can reach, and who reviews that list quarterly. Nadella's own framing at the Ignite presentation was that every agent needs an identity and everything it does needs to be observed. Observed by whom, on your side, is your call to make.

Code plus a managed runtime means non-developers now ship software. Someone in ops will describe a dashboard, get a real app hosted inside your tenant, and share it. That is genuinely useful and it is also a new surface: undocumented tools holding live company data, with no owner when that person changes roles. Decide now whether generated apps get an owner field and a review date, or whether you find out about them during an audit.

Meter one workflow yourself. Usage billing cuts both ways — you can run a real pilot without buying a tier for everyone. Put one Autopilot on one boring recurring task, track credits consumed against hours returned for a month, and let that number decide the rollout. Not the demo.

Sources: Microsoft, VentureBeat.