Microsoft presented a broad change to Copilot's structure on September 25, with three new work areas: Home, Code and Autopilot. Alongside the interface change, the company is changing how some advanced capabilities are billed, separating ordinary Copilot use from agentic tasks performing sustained work for the user.
The most significant part is Autopilot, a persistent agent intended to continue working after users define and approve a task. Meanwhile, Code is intended for building apps from natural-language instructions, and Home unifies Chat and Cowork in one workspace.
What changed in Copilot
According to Microsoft, the new Copilot has three main destinations.
Home brings together Chat and Cowork. The goal is to let users begin with an ordinary conversation and move to more complex tasks without switching to a separate product.
Code is intended for building apps from natural-language descriptions. Microsoft presents a workspace where Copilot can help plan, build and run apps, not just write code snippets inside an editor.
Autopilot is the most notable change. It is a persistent agent, previously called Scout, with its own identity and memory. According to the company, it is intended to work over time after users define the goal and approve the action.
The announcement also includes Copilot Managed Runtime, a managed execution layer for apps and agents created in the new environment.
The billing model is changing too
Microsoft is not moving all Copilot use to usage-based billing.
According to the announcement, Chat and Copilot capabilities within Office apps remain part of fixed per-user licensing. By contrast, Cowork, Code, Autopilot and certain frontier model uses move to usage-based billing.
Microsoft names Astra and Fable among the available models in this context.
The change has practical implications for IT managers. Agent-based uses can create variable costs, so the company presents spending limits and usage controls through Agent 365.
Exact prices and the complete structure of new billing plans remain unclear across all scenarios.
Autopilot is intended to work even when users are not sitting in front of it
Autopilot represents a shift from Copilot answering users to Copilot performing tasks over time.
According to Microsoft, the agent can retain context and memory, continue work after human approval and operate with its own identity within the organizational environment.
This is significant for permissions and management. As an agent gains sustained working ability, questions grow about which data it can see, which actions it may perform and how the organization tracks its costs and activity.
Microsoft emphasizes Agent 365's management and oversight role, but Autopilot's practical ability in real environments will need testing as access expands.
Fabric IQ enters the picture
The announcement also connects Copilot to Fabric IQ.
According to Microsoft, Fabric IQ can ground Copilot in more than 20 million Power BI semantic models. This means organizations using Power BI and Fabric can give Copilot more structured business context rather than relying only on documents or free-form conversation.
This may be particularly significant for tasks requiring an agent to understand metrics, business structures and organizational data already defined within Microsoft.
When it is available
Availability is not uniform across components.
Copilot Managed Runtime enters public preview on September 25.
Autopilot is expected to enter private preview at the end of September.
Code is expected to reach Frontier at the end of September, with broader access in subsequent weeks.
Microsoft also notes a preview later in 2026 for Microsoft 365 Premium and Pro users.
It is therefore inaccurate to describe the entire new Copilot as already available to everyone.
Why this is an important change
Until now, many knew Copilot mainly as an assistant appearing inside Word, Excel, Teams or Windows.
The new announcement seeks to turn it into a platform where users can converse, build apps and operate persistent agents from the same workspace.
Meanwhile, moving some capabilities to usage-based billing changes the cost question. An organization can no longer look only at a fixed monthly license price. It will also need to monitor agentic use and understand which tasks create additional consumption.
What remains unknown
Microsoft did not provide a complete pricing picture for every use scenario in the announcement.
Autopilot and Code General Availability dates are also not final, and the company did not present broad usage data allowing Autopilot's reliability on long tasks to be assessed.
At this stage, the direction is clear: Copilot is turning from a chat and add-on environment into a broader system of agents and apps. Actual cost and performance will be easier to assess as the preview expands.