Three modes for a broader Copilot
Microsoft has presented a redesigned Copilot organised around Home, Code and Autopilot. Its Microsoft announcement, published on 25 September, describes a shift from a chat assistant towards a place to ask questions, delegate work, build tools and run longer-lived agents. The names represent different kinds of work, and their release dates are not identical.
Home brings Chat and Cowork together with Office inside Copilot. Chat is for immediate questions and drafts; Cowork is for work handed off to produce a result. Microsoft says Word, Excel and PowerPoint can create or update real editable files from the same experience, keeping changes in sync with the familiar Office applications. This is a more consequential promise than generating a file-shaped answer in chat.
A buyer should still test what “in sync” means for templates, permissions and concurrent editing in its own tenant. The vendor’s examples include a launch brief, budget model and presentation. A useful pilot would examine whether a colleague’s later edits remain visible and whether the assistant preserves the organisation’s formatting and source evidence.
Code turns prompts into small solutions
Code is intended to let non-developers describe an app, dashboard, tracker, widget or automation and have Copilot build it. Microsoft says it uses underlying technology shared with GitHub Copilot, runs in a sandbox and can host applications within a company tenant. This is different from asking Copilot to write a snippet that someone must deploy manually.
The associated Copilot Managed Runtime is in preview. Microsoft describes it as infrastructure for code produced in Cowork, Code and Copilot Studio, with access planned for third-party and professional developers too. IT governance remains important: an app may be easy to create, but its data connections, retention, sharing and ongoing ownership still need explicit decisions.
Code is beginning a Frontier rollout at month-end, with broader availability following in coming weeks; Microsoft also describes a later preview for some consumer subscription tiers. These stages should not be collapsed into a single general-availability claim. Teams planning a workflow should confirm eligibility in their actual tenant before committing to an internal launch.
Autopilot works between prompts
Autopilot, formerly called Scout, is described as a persistent, proactive personal agent. A user can give it a role and goal, then have it follow channels, run recurring work and resume a project later. Microsoft says it has its own identity, memory, computer and workspace inside the tenant and can appear in Teams, Outlook and other work surfaces.
That persistence is useful only with boundaries. A supplier review agent might prepare a schedule and follow-ups, but its authority to contact stakeholders, alter records or spend money should be defined separately. Administrators need an understandable way to inspect actions, revoke access and stop a task that is no longer appropriate.
Microsoft says Autopilot expands to private preview at the end of September. It is therefore not a generally available replacement for an existing automation today. A pilot should start with a narrow task and measure whether the agent reports uncertainty, handles a changed instruction and avoids silently repeating an action whose outcome is unknown.
Context and plugins connect the pieces
Microsoft IQ supplies business context across the platform. The company says Fabric IQ context from Power BI semantic models is generally available in Chat and Cowork, with Code integration to follow through Frontier. Dynamics 365 and Power Platform grounding is scheduled for public preview over the next month. Those distinctions determine which demonstrations are possible now.
A new plugin registry will gather Microsoft, partner and custom plugins into one catalogue. IT can approve and manage them centrally, while developers can publish once for supported Copilot surfaces. This can reduce duplicate integration work, though each plugin still needs a permission and data-flow review. A central listing is not proof that an extension is suitable for every user.
The company also promises that Copilot will eventually route a request to Chat, Cowork or Code without asking the user to select a mode. This is described as coming soon. Organisations should test the routing when it arrives, especially where a conversational request could become a longer task or a generated application with different cost and approval implications.
Rollout and practical evaluation
Home and Code are starting in the Frontier programme over coming weeks, while the Autopilot private preview expands at month-end. Managed Runtime is already in preview. A responsible adoption plan should track each capability separately and document which users, regions and licences can actually access it.
Pricing has its own separate policy announcement: everyday Copilot remains tied to a per-user subscription, whereas advanced agentic capabilities use consumption billing. This article concerns the product redesign; administrators should read the dedicated pricing post before enabling a costly workload. A successful demonstration is not a cost estimate.
The new Copilot could shorten the distance from an idea to an editable document, a small app or an ongoing agent. The key question is whether the platform maintains clear ownership and review as that distance shrinks. Test collaboration, permissions, auditing and task completion in the real environment before scaling beyond a controlled group.