Skip to content

Microsoft gives Copilot a persistent Autopilot agent and splits its billing in two

· by Pondero Newsdesk

The short version

Microsoft rebuilt Copilot around three modes on September 25, 2026: Home, Code, and a tenant-resident Autopilot agent with its own identity and memory, paired with a new split between flat per-user licenses and usage-based billing for agentic work.

Microsoft gives Copilot a persistent Autopilot agent and splits its billing in two

Microsoft rebuilt Copilot on September 25, 2026 around an agent that has its own email address and keeps working after an employee logs off, then drew a hard line through the product's pricing to separate that kind of agentic work from everyday chat.

What happened

Microsoft's Copilot app now organizes around three tabs, per the company's announcement: Home folds Chat and the existing Cowork mode together and now embeds full Word, Excel, and PowerPoint editing directly in the interface. Code lets any employee describe an app, dashboard, or automation in plain language and get a working solution back, running on what Microsoft describes as the same underlying technology as GitHub Copilot. Autopilot is the new piece: a persistent, proactive agent that lives inside the customer's Microsoft 365 tenant with its own identity, memory, computer, and workspace, built on a context layer Microsoft calls Microsoft IQ. A user gives it a name, a role, and a goal, and it keeps running unattended, checking channels, following up on threads, and picking work back up days later without a new prompt.

Autopilot is not entirely new under the hood. Per GeekWire's reporting, it grew out of Microsoft's OpenClaw personal-agent project, which the company called Scout as recently as June 2026. Home and Code begin rolling out through Microsoft's Frontier early-access program in the coming weeks; Autopilot moves to private preview at the end of September.

Pricing changed alongside the product. Microsoft's blog post splits Copilot spend into two tracks: a flat user subscription license still covers Chat and the Office apps, while a new usage-based billing track covers Cowork, Code, Autopilot, and calls to frontier models. That agentic track already includes OpenAI's GPT-6 Astra and Anthropic's Claude Fable 5.1, both billed at provider list pricing rather than folded into the flat seat price.

IT budgets get a new line item for agent work

The billing split is the part enterprise buyers have to plan around first. A Copilot seat used only for chat, drafting, and Office work stays a predictable per-user cost. The moment an organization turns on Cowork, Code, or an Autopilot agent, spend becomes metered and variable, tied to how much work the agent actually does rather than how many people have a license. Microsoft frames this as matching cost to value; for a finance or IT admin, it means the agentic rollout needs its own budget line and its own usage monitoring, which is why the same announcement introduces expanded cost-management tooling inside Agent 365 for tracking Code and Copilot Managed Runtime spend, with Copilot Studio agent cost tracking promised for October.

The tenant-resident design matters more than the pricing mechanics for governance teams. An Autopilot agent with its own identity and memory is closer to provisioning a new employee account than assigning a software license: it needs permissions, an audit trail, and boundaries set at creation, not bolted on after the fact. Microsoft's blog says permissions, audit, and governance controls apply, but the model itself, an agent that acts on threads and reaches out to stakeholders without a live prompt, is a genuinely different deployment shape than the request-response chat session Copilot shipped for its first three years.

Context and reactions

Microsoft is drawing an explicit contrast with how rivals run agentic work. Jacob Andreou, Copilot's EVP, told GeekWire that "sending your most sensitive files down to a local machine into a VM was a complete nonstarter," a direct reference to Anthropic's Claude Cowork running client-side in a virtual machine. Satya Nadella pitched the opposite argument for keeping multiple frontier models inside one product, telling GeekWire that customers need "a private eval that you can hill climb on and substitute a model," meaning Copilot now offers a menu that includes OpenAI's and Anthropic's models alongside an Auto mode that routes each request by accuracy, speed, and cost.

The stakes for Microsoft are sizable but still early. GeekWire reported that Microsoft counted 30 million paid Copilot seats as of July 2026, roughly 7 percent of the more than 450 million commercial Microsoft 365 seats the company reports. That gap between the installed base and paid adoption is the pressure behind rebuilding the product around agentic work rather than a chat window: Microsoft needs a reason for the other 93 percent to convert, and a persistent agent doing unattended work is a different pitch than a smarter autocomplete for email.

The rebrand from Scout to Autopilot also lands months after Microsoft first previewed the underlying agent concept, suggesting the company iterated on the product before committing to a public identity and pricing model for it, rather than shipping the June preview as a finished feature.

What to watch next

Two rollouts land within weeks: Autopilot's private preview opens at the end of September, and a related "Today" proactive dashboard for mail, calendar, and Teams threads enters private preview in October. Watch whether usage-based billing on Cowork, Code, and Autopilot draws the same pushback GitHub Copilot saw when it moved developer seats to per-request credits earlier in 2026, since IT buyers who budgeted for flat per-seat Copilot costs now have to forecast a metered line item on top of it.

Sources