Microsoft's New Copilot Bills Agents by Usage. What an Always-On Agent Needs Before You Switch It On

Microsoft's new Copilot app includes an agent that keeps working when you are away. It also changes how part of Copilot is paid for: by usage.

Those two facts belong together. An agent that runs on its own and a bill that follows what it does is a different kind of purchase from a per-seat license.

Here is what Microsoft has said publicly, and a short list of things I would settle before turning an always-on agent loose.

What Microsoft announced

On September 25, Microsoft unveiled a redesigned Copilot app. The Verge describes three tabs: Home, which combines Copilot Chat and Cowork, Code, and Autopilot. The Next Web reports the same structure.

According to those reports, Autopilot is the renamed Scout. Users give it a name, a role, and a goal. It runs in the cloud with its own identity, memory, and workspace, and keeps working when its user is away. People can mention it in Teams and Outlook like a colleague. The Verge quotes Microsoft saying it lives in the customer's tenant with permissions, audit, and governance behind it.

Code lets people build apps, trackers, dashboards, and automations by describing them. Both outlets say it runs in a sandbox and can be hosted in the company's own tenant.

On availability, The Verge reports that Home and Code roll out to Microsoft's Frontier early-access program in the coming weeks, and Autopilot goes into private preview. Dates and access may have moved since September 25, so check the current status for your tenant.

How it is paid for

This is the part that matters for planning.

The Verge reports that the regular per-user license covers Copilot in Chat, Word, Excel, PowerPoint, Outlook, and Teams. Cowork, Code, and Autopilot are billed by usage. The Next Web adds that the newest frontier models, such as Astra and Fable, are also billed by usage.

The Verge says IT admins will need Microsoft's new FinOps for AI to manage AI spend and keep agentic usage from getting out of control. I did not find pricing details for these usage meters in the sources I reviewed, so I am not quoting any rates.

A flat license has a ceiling. A usage meter does not, unless someone sets one. For an agent that runs without a person watching, "someone sets one" is the first task, not an afterthought.

What the Microsoft executive said, and who paid for the interview

In an October 11 newsletter, The Rundown AI published a Q&A with Jacob Andreou, Microsoft's EVP of Copilot. The newsletter says it was produced in partnership with Microsoft, so read it as a company-aligned interview, not independent reporting.

Two points from it are worth noting as his statements, not as findings. He said Microsoft removed Copilot from several Windows apps after finding entry points that brought traffic but little value, and that fewer entry points led to more usage per user. And he described his own experience with a consumer agent: it felt magical for a few weeks, then like eating too much ice cream, and for harder tasks it felt shallow.

The Next Web quotes him at the launch event, via GeekWire, saying the autonomy that makes a long-running agent good is the same thing that makes it frightening to an IT admin. That is the right framing for the checklist below.

Settle these before an agent runs unattended

None of this is Microsoft-specific. It applies to any agent that runs on its own and costs money per use.

An owner. Name one person responsible for the agent's goal, its spend, and its output. An agent with its own identity and no owner is an orphan by default.

A spend ceiling. Set a cap per agent and per month, and decide what happens at the cap. Pause is safer than continue. Find out whether your tooling alerts before the cap or only at it.

A narrow goal. "Handle follow-ups for this one project" is testable. "Help the team" is not. Microsoft describes the agent as taking a role and a goal, so write both down.

Scoped access. Give the agent access to the sources and channels the goal needs. Check what its identity can read and write, and review that list whenever the goal changes.

An action boundary. Decide which actions it can take alone, such as drafting, and which need approval, such as sending to people outside the team or changing records.

A review rhythm. Read the agent's activity log on a schedule. Look at what it did, what it cost, and whether the result was worth it.

A stop rule. Decide in advance when you will switch it off: cost per result above a threshold, or repeated corrections from a person.

A small pilot

If Autopilot reaches your tenant, start with one agent and one narrow job.

1. Pick a task that already has a human owner and a way to check the result.

2. Set a low spend cap before the first run.

3. Run it for two weeks, with a person reviewing every output.

4. Record cost per completed task, corrections needed, and time saved by the person.

5. Compare against doing the job by hand. Keep it only if the numbers justify it.

This is a plan, not a result. I have not used Autopilot, and I am not reporting any measurements.

What I would not conclude

Do not read the usage model as a price increase or decrease. I found no rates.

Do not assume enterprise governance removes the need for limits. Microsoft says audit and governance sit behind the agent, but a log of what an agent did is not the same as a control on what it may do.

And do not treat a rollout announcement as a status report. The preview terms in the September coverage may already have changed.

Frequently asked questions

What is Copilot Autopilot?

According to The Verge and The Next Web, it is Microsoft's always-on agent, formerly called Scout, which runs in the cloud with its own identity, memory, and workspace.

What is billed by usage?

Reports say Cowork, Code, Autopilot, and the newest frontier models are billed by usage, while the regular per-user license covers Copilot in Chat and the Office apps.

How much does it cost?

I did not find published rates in the sources reviewed. Check Microsoft's current pricing for your plan.

Is it generally available?

The September 25 coverage described early access and private preview stages. Verify current availability for your tenant.

How do admins control spend?

The Verge says admins will use Microsoft's new FinOps for AI. I did not review its features in detail.

What should I set before starting an agent?

An owner, a spend cap, a narrow goal, scoped access, an action boundary, a review rhythm, and a stop rule.

Is the Microsoft executive interview independent?

No. The newsletter states it was produced in partnership with Microsoft.

Sources

• The Verge, Microsoft's new Copilot super app: Home, Code, Autopilot, usage-based billing, FinOps for AI, rollout status.
• The Next Web: Autopilot description, billing split, executive quote via GeekWire, and launch details.
• The Rundown AI: Q&A with Jacob Andreou, produced in partnership with Microsoft.