What Microsoft Actually Shipped
Microsoft unveiled its redesigned Copilot app on September 25, folding the previously separate consumer Copilot and Microsoft 365 Copilot into one product organized around three tabs. Home merges chat with Cowork, a task-delegation feature, and now embeds full working versions of Word, Excel and PowerPoint. Code is a natural-language app and dashboard builder running in a sandboxed, tenant-hosted environment, built on GitHub Copilot technology. Autopilot, a rebrand of the Scout assistant Microsoft introduced at Build in June, is a named digital teammate with an assigned role that can act on tasks without anyone prompting it in the moment.
Corporate VP Annie Pearl said Copilot draws primarily on models from Anthropic and OpenAI, with added access to xAI's Grok, and an Auto mode that routes each request by accuracy, speed and cost.
One App, Two Different Meters
The part that will show up on an invoice, not a press release, is the billing split Microsoft built into the same launch.
| Bucket | What's in it | Billing model | What that means |
|---|---|---|---|
| Everyday AI | Chat, Office apps (Word, Excel, PowerPoint), standard Auto routing | User Subscription License - fixed seat cost | Predictable, budgets the way a normal software seat does |
| Agentic work | Cowork, Code, Autopilot, and any premium model access | Usage-Based Billing - metered by consumption | Cost scales with how much the agents actually do, not with headcount |
Microsoft is bundling "FinOps for AI" into the same release - cost management, spending policies and model-selection controls - the same toolkit enterprises already use to keep cloud-compute bills from running away. Shipping that governance layer on day one is itself a tell: Microsoft is not guessing that agentic billing needs a leash, it is building the leash before general availability.
The Part No Launch Headline Says Out Loud
A per-seat license has a hard ceiling: one person, one monthly fee, capped by headcount. A usage meter attached to a persistent agent has no equivalent ceiling, because Autopilot is explicitly built to keep working while you are offline - handling recurring tasks, supplier reviews and stakeholder follow-ups on its own schedule, not yours.
That is a different risk shape than the seat-based SaaS most finance teams know how to budget. You cannot pause an agent's spend by logging off, because logging off is exactly when it is designed to keep running. For any organization that turns Autopilot on broadly, the FinOps controls Microsoft ships alongside it are the only thing standing between a digital teammate's initiative and an unbounded bill.
How to Budget a Copilot Rollout Without Guessing
Treat Home and Office chat as the predictable line item it is designed to be - a normal seat cost, easy to forecast against headcount. Treat Cowork, Code and Autopilot as a metered utility from day one, not an add-on you will tune later.
Before turning Autopilot on for a team, set the spending policy and per-agent cost cap first, not after the first invoice arrives. Pilot it on one named role with one clear task, watch what a full month of unattended operation actually costs, and only then decide whether to widen it. Microsoft is racing OpenAI's own super app from July and Anthropic's enterprise push with this launch. That competitive pressure belongs to Microsoft; let your rollout pace match your own cost visibility instead of their release calendar.
Read next: Microsoft Trained Its Sellers to Call Claude Slower | OpenAI Priced Its Model at Half of Anthropic's
