Microsoft Copilot Splits Into Three Tabs With Two Billing Models

Microsoft shipped a substantial architectural overhaul of Copilot on September 25, 2026, consolidating what had been separate consumer and commercial products into a single three-tab application organised around Home, Code, and Autopilot. Each tab carries a distinct execution model and a distinct billing mechanism — making the billing architecture, not the AI capabilities alone, the substantive story for organisations evaluating whether to build on this platform.

Three tabs, three execution models

The Home tab combines Chat and Cowork. Chat is synchronous and conversational — fast lookups, drafts, in-context summaries. Cowork is explicitly asynchronous: the user defines a task and the system runs it end-to-end, returning a completed artifact such as an RFP response, launch kit, or financial close package. Microsoft reports that a model-routing layer called Auto sits at the centre of the user subscription license (USL) tier, weighing accuracy, speed, and cost to dispatch each request to the model it judges most appropriate — and that IT administrators can constrain which model families Auto is permitted to select from.

The Code tab is powered by the same underlying technology as GitHub Copilot, according to Microsoft. It targets knowledge workers rather than software engineers and generates apps, dashboards, trackers, and automations from natural-language descriptions. The execution environment is sandboxed and tenant-hosted via a new piece of infrastructure Microsoft calls Copilot Managed Runtime, now in preview. Managed Runtime also underpins apps produced by Cowork and Copilot Studio, and Microsoft is opening it to third-party and pro-code developers.

Autopilot — previously named Scout — is the cloud-persistent agent tab. Microsoft reports it operates with its own tenant identity, memory, cloud compute instance, and workspace, grounded in Microsoft IQ, the company's unified enterprise intelligence layer. It surfaces inside Teams, Outlook, chats, and documents via @mention and carries permission, audit, and governance controls. The key operational distinction from session-scoped agents is persistence: Autopilot continues executing recurring tasks and channel-watching workflows without requiring an active user session.

Billing architecture splits on workload type

Any workflow that crosses from conversational Chat into Cowork, Code, or Autopilot crosses a billing boundary. Organisations that have assumed flat per-seat pricing governs all AI usage will need to instrument and govern that boundary before deploying agents at scale. Microsoft's FinOps for AI tooling — which lets admins set model-family availability per user group and route credit requests into approval workflows — is the operational layer intended to manage that exposure. Cost management in Agent 365 is expanding to cover Code and Copilot Managed Runtime in the current rollout, with Copilot Studio agent support planned for October.

Context and plugin expansion

Microsoft is expanding Microsoft IQ with two grounding additions. Fabric IQ connects the 20-plus million semantic models in Power BI to Chat and Cowork, with Code integration coming through the Frontier program. Dynamics 365 and Power Platform data and workflows are rolling out in public preview over the following month. A unified plugin registry — consolidating Microsoft, partner, and custom-built plugins into a single catalog — is rolling out now and targeting general availability across surfaces within weeks. IT can approve plugins centrally; developers publish once and reach all supported Copilot surfaces. For teams that have been building agent workflows without a shared control contract, a centrally governed registry with publish-once reach reduces the per-surface integration overhead that has historically fragmented enterprise agent deployments.

AI Mastery analysis

The architectural bet Microsoft is making is that the platform boundary — tenant identity, audit trail, managed runtime, unified billing — is a stronger enterprise moat than raw model capability. The Code tab's sandboxed, tenant-hosted execution and Autopilot's persistent cloud compute instance address a real gap: most organisations cannot safely run long-lived agentic processes on infrastructure they do not control. But production agent failures tend to originate in architecture, not model intelligence, and Microsoft is shipping Managed Runtime in preview while Autopilot enters only private preview — meaning the runtime governing the most sensitive workloads is the least mature component in the stack.

The USL-versus-UBB billing split introduces a cost surface that is structurally difficult to predict. Auto's model routing is opaque to end users by design, and while admins can constrain model families, cost attribution only becomes visible after execution. The FinOps tooling acknowledges this, but its October timeline for Copilot Studio agent coverage leaves a window during which the most autonomous workflows lack the spend-visibility controls the less autonomous ones already have. Teams evaluating adoption should treat infrastructure governance as a prerequisite for safe agent deployment, not a follow-on concern.

Microsoft CEO Satya Nadella has described Copilot as "a new OS for work that spans every model, every form factor, and every task," and the company is explicitly comparing it to Office's influence on PC-era productivity. That framing is a deliberate choice to compete on workflow integration depth and enterprise control surface rather than model leaderboard position — consistent with the view that infrastructure rewrites, not model weights, will drive the next phase of enterprise AI gains. Whether it recovers ground lost to ChatGPT and Claude in the commercial segment depends less on any individual capability than on whether the governance and billing architecture holds up under real enterprise deployment conditions.

Sources

• Primary: Introducing the new Copilot with Home, Code and Autopilot - The Official Microsoft Blog
• Reported by: The Verge AI: Microsoft thinks its new Copilot ‘super app’ will be as influential as Office