Microsoft Splits Copilot Into Three Layers: Home Owns the Entry, Code Builds, Autopilot Works While You're Away
The chatbot-era Copilot answered when you asked. In this redesign the keyword changed — Autopilot is defined as "a persistent, proactive and personal agent that keeps working even when you're not."
Three key facts
On September 25, Microsoft formally launched the new Copilot, split into three capability layers: Home (the new starting point, merging Chat and Cowork with Word, Excel and PowerPoint fully built in), Code (anyone can build their own solutions, powered by the same underlying technology as GitHub Copilot, with a safe runtime), and Autopilot (a persistent, proactive, personal agent that keeps working offline). Home and Code begin rolling out through the Frontier program within weeks; Autopilot expands to private preview at the end of the month.
Teams gets a parallel upgrade: invoke @Copilot in a channel, group chat or meeting and it receives that context's shared permissions — ask "why was this product decision made" and it pulls the background across conversations, no digging and no re-explaining. When a decision ripples into manufacturing, packaging and launch timing, Copilot proactively offers to prepare a cross-functional review with dependencies, owners, open questions and decisions. @Copilot lands in Teams private preview by month's end.
Alongside it, Microsoft shipped new FinOps for AI capabilities to help enterprises manage Copilot and agent spend. It also teased more Copilot news at Ignite, November 17–20 in San Francisco — this redesign is clearly paving the road to the year-end push.
Wang Dou's Take
Of the three layers, Home and Code are familiar faces; what really deserves the spotlight is Autopilot's line — "keeps working even when you're not." That's a product-definition boundary crossing: the old assistant waited to be summoned; the new agent schedules its own shifts. Enterprise software's billing unit is quietly sliding from "seats" toward "work-hours" — once one person commands several agents that never clock out, how long can seat-based pricing survive? FinOps for AI launching the same day is no coincidence: Microsoft knows better than anyone that once proactive agents run loose, token invoices become the new IT budget black hole. Sell the shovels first, then the gold-panning tools — the order has never changed.
@Copilot entering Teams is the move to chew on. What it receives is "the channel's shared context and permissions" — translated: your meeting transcripts, group-chat history and decision trails all become readable corpus for the agent. The efficiency narrative flows smoothly: no digging, no re-explaining. But once the permission model shifts from "person grants person" to "person grants agent," the granularity of audit has to shift with it — who read what, which decisions the agent made on whose behalf, and who signs when it goes wrong. Microsoft didn't talk about any of that this time; it only talked convenience.
Lining up the three giants makes it more interesting: Meta's Muse is grabbing the personal entry point, OpenAI's voice-plus-plugins is grabbing the interaction layer, and Microsoft's Autopilot is grabbing the enterprise workflow itself. The battle of entry points has run for a year; now the fight is over "who owns your eight working hours." Copilot's biggest advantage isn't technology — it's that it already ships on every corporate PC. Distribution is the moat. The real test: will IT departments dare to grant a proactive agent cross-channel read permissions? If they dare, Microsoft wins; if they don't, Autopilot is just a more expensive calendar reminder.
Source: Microsoft Official Blog · Futurum Group