Microsoft Copilot Autopilot and Today: From Meetings and Email to Autonomous Follow-Up

Key Takeaways

• Microsoft introduced Home, Code and Autopilot on September 25, 2026, reorganizing Copilot around asking, delegating, building and persistent agent work.
• Autopilot is now listed by Microsoft as available in Private Preview and is designed to keep working across channels, threads, recurring work and multi-day projects without a fresh prompt for every step.
• Today is scheduled to enter Private Preview in October as a Home experience that combines mail, calendar, Teams threads, meetings and tasks into a proactive personal work queue.

Microsoft’s September 25 Copilot update moves the center of gravity from summarizing work to carrying work forward. The new Copilot still answers questions and creates documents, but Microsoft is now separating three different jobs: Home helps users orient around current work, Today prepares the next actions that deserve attention, and Autopilot is designed to keep executing a longer-running objective after the user moves on.

That distinction matters for anyone following AI meeting notes. A transcript or recap answers what happened. A follow-up system decides what should happen next. A persistent agent goes one step further: it can continue pursuing that objective across later messages, meetings and recurring tasks.

What Microsoft Actually Announced on September 25, 2026

Microsoft announced three core additions to the Copilot app―Home, Code and Autopilot―while separately previewing Today in Home and @Copilot in Teams. The September 25 Copilot announcement is important because Today is easy to misread as a fourth peer to Home, Code and Autopilot. Microsoft presents Today as an upcoming Home experience, not as a fourth top-level pillar.

Home is the navigation layer. It brings Chat and Cowork together so a user can review recent activity, pick up unfinished work, ask a quick question or delegate a larger task. Office in Copilot also brings Word, Excel and PowerPoint work into the same experience.

Code is the building layer. Microsoft describes it as a place where knowledge workers can describe the dashboard, tracker, application or automation they need and have Copilot build it using the same underlying technology family as GitHub Copilot.

Autopilot is the continuity layer. Previously called Scout, Autopilot is designed as a cloud-hosted agent that can keep working when the user is no longer actively prompting it. That is the architectural change that separates this release from another Copilot interface refresh.

Autopilot vs Today: They Solve Different Parts of the Work

Today and Autopilot overlap around follow-up, but they operate at different levels. Today is designed to surface and prepare work for a person. Autopilot is designed to own a continuing objective inside the boundaries that person and the organization set.

Today finds and prepares the next action

Today is the closer match to an intelligent morning briefing that can also prepare work. Microsoft says Today will combine signals from email, calendars, Teams threads, meetings and tasks, then distinguish what was missed, what needs attention now and what can wait. The difference from a static dashboard is that Today can prepare editable drafts, propose schedule changes and surface forgotten follow-ups with their surrounding context.

The user remains visibly in the loop in Microsoft’s examples. A draft is ready to edit; a schedule change is proposed; a forgotten follow-up is prepared to send. Today therefore reduces the work required to notice and reconstruct an obligation, but the published examples do not describe it as independently owning that obligation until completion.

Autopilot keeps owning the work

Autopilot is designed for work that remains open after the current session ends. A user gives the agent a name, role, goal and boundaries. Microsoft says Autopilot can watch channels, follow up on threads, run recurring work and return to a project days later without waiting for another prompt.

Microsoft’s supplier-review example shows the intended scope more clearly than a generic “AI assistant” description. Autopilot can build the review schedule and workback plan, handle preparation, coordinate meetings, perform follow-up and contact stakeholders for updates. The important unit of work is the ongoing supplier-review process, not one email, one recap or one chat session.

That makes the cleanest mental model simple: Today says, “You missed this follow-up; here is the draft and context.” Autopilot says, “This review process is my assigned objective; I will keep the work moving within the permissions and boundaries you set.”

The category boundary becomes clearer when the unit of work changes from a message to an ongoing obligation.

Microsoft Copilot Today is a proactive work-prioritization layer that combines mail, calendar, Teams, meetings and tasks, while Copilot Autopilot is a persistent cloud-hosted agent designed to continue multi-step objectives across time. Today prepares the next action; Autopilot continues the assigned work.

For users, the practical difference is whether the AI is preparing the next decision or remaining responsible for the workflow after that decision.

From Meeting Notes to Follow-Up: Where Each Microsoft AI Layer Fits

Microsoft already had several meeting-specific AI layers before Autopilot. The new release matters because those layers can now be understood as upstream inputs to broader follow-up systems rather than the endpoint of the workflow.

Teams users who need the meeting layer itself separated can use the Microsoft Teams meeting notes guide, which distinguishes Collaborative Notes, transcription, Intelligent Recap, Copilot and Facilitator instead of treating them as one feature.

Facilitator sits inside the live meeting rather than after it. Microsoft documents real-time AI notes, key decisions, open questions, agenda support and in-meeting collaboration as core Facilitator functions. The Facilitator support guide also makes clear that Facilitator is a meeting agent, not the same thing as a persistent cross-day process owner.

Post-meeting automation already existed in narrower business workflows. Microsoft’s Sales agent, for example, can use a Teams meeting recap to draft a post-meeting email, create a CRM task and save AI-generated meeting notes to CRM. Autopilot broadens the concept by making persistence itself a first-class agent behavior rather than tying follow-up to one sales-specific meeting flow.

What Changes When Email, Meetings and Tasks Become One Context

The practical change is not that Copilot can read more apps; it is that work can remain connected as it moves between those apps. A meeting decision can create an email obligation, an email can expose a scheduling conflict, and a recurring project can require both before the original conversation is still top of mind.

Meeting to follow-up

A meeting recap becomes more useful when its open actions survive the meeting window. The sequence is no longer limited to record → transcribe → summarize. A decision can become an assigned action, the action can trigger a draft or scheduling change, and a persistent agent can keep checking whether the dependency has been resolved.

This does not make the recap irrelevant. The transcript remains the evidence layer for what was actually said, especially for names, figures, commitments or disputed decisions. The follow-up layer should consume that evidence rather than silently replace it.

Email thread to ongoing work

Email follow-up is a natural test case because the failure is often not drafting but remembering. Today is aimed at the moment when an unanswered thread has fallen out of working memory. It can surface the missed obligation with context and prepare a response. Autopilot is aimed at the longer process in which the thread is only one checkpoint inside an assigned objective.

That difference matters in procurement, recruiting, customer onboarding and project management. A single polished email is not the completed job if the next stakeholder still has to respond, a review still has to be scheduled, or the same check must happen again next week.

Recurring process to persistent agent

Recurring work exposes the limit of prompt-by-prompt assistants most clearly. Weekly status collection, supplier reviews, launch readiness and account follow-up are not one-shot tasks. They require memory of the objective, awareness of what changed, repeated contact with people and a way to resume after days of inactivity.

Autopilot is Microsoft’s attempt to make that continuity explicit. The agent has its own identity, memory, computer and workspace inside the tenant, so the organization can reason about what the agent is allowed to access and which actions belong to it instead of pretending that every action came from the human user.

@Copilot in Teams Adds a Third Kind of Context

@Copilot in Teams is neither the same as Today nor the same as Autopilot. Its defining feature is shared team context. Microsoft says the capability can use the context and permissions of a channel, group chat or meeting so a team can ask why a decision was made and have Copilot assemble the background across prior conversations.

Microsoft also describes a more operational use case: when one decision affects manufacturing, packaging and launch timing, @Copilot can offer to prepare a cross-functional review that organizes dependencies, owners, open questions and decisions. The value is shared coordination around the team’s existing context, not a private personal inbox.

The distinction is therefore useful: Today is personal prioritization, @Copilot is shared team context, and Autopilot is persistent delegated execution. Microsoft announced @Copilot in Teams for Private Preview by the end of September; availability can still vary by tenant and preview enrollment.

How Autonomous Is Autopilot, Really?

Autopilot is not described as an unbounded worker that can act anywhere without control. Microsoft says the agent lives inside the customer tenant with its own identity, memory, computer and workspace. It operates with permissions, audit and governance, while the user defines the objective and boundaries.

That design is more important than the word “autonomous.” Enterprise software needs an accountable identity for the agent, a permission model for the resources it can reach, and an audit trail for the actions it takes. A persistent agent without those controls would create a larger governance problem than the manual work it replaces.

The autonomy claim is therefore better understood as persistence inside a governed operating envelope.

Microsoft Copilot Autopilot is a persistent enterprise agent with its own tenant identity, memory, computer and workspace. The agent can continue recurring and multi-step work in the cloud, but execution remains constrained by organizational permissions, audit controls, governance policies and user-defined objectives.

For buyers, that means the real evaluation is not “Can the AI act alone?” but “Which actions can it take, under whose authority, with what review and at what cost?”

Independent coverage has focused on the same enterprise boundary. Reuters coverage of the launch highlighted Autopilot’s separate corporate identity and customizable permissions as part of Microsoft’s response to security, compliance and governance concerns around agentic AI.

What Is Actually Available as of September 30, 2026?

Announcement date and actual tenant availability are not the same thing. Microsoft’s current product page now lists Home and Code in Microsoft Frontier and Autopilot in Private Preview. Today remains scheduled for an October Private Preview, while @Copilot in Teams was announced for a Private Preview by the end of September.

The current Microsoft Copilot page is the better source for present-tense availability than launch-day wording because Microsoft can update product status as previews open. Even then, preview labels should not be read as general availability.

Why This Matters for AI Meeting Tools

The competitive question around AI meetings is shifting from transcription quality toward what happens after the transcript exists. Accurate capture still matters, but the more valuable workflow increasingly spans five layers: capture → transcript → interpretation → business action → continued execution.

That shift changes how meeting tools should be compared. A note taker can be excellent at transcription and still leave users copying decisions into task managers. A CRM assistant can automate structured sales follow-up but remain narrow to the sales system. A persistent agent can continue the work longer, but it introduces a larger permission, governance and cost surface.

Third-party meeting tools therefore do not become irrelevant simply because Microsoft can automate more of the Microsoft 365 stack. Cross-platform meetings, physical-room capture, specialized compliance requirements and non-Microsoft systems still create gaps. The more useful comparison is architectural: where the conversation is captured, where the evidence lives, which system turns it into action, and which agent―if any―owns the unfinished work.

The separate sales call notes workflow shows the same distinction at a smaller scale: one customer conversation can produce an evidence record, CRM changes, owned next steps and a client-facing follow-up. Microsoft’s September Copilot update extends that logic from a single post-call handoff toward longer-lived agent ownership.

Bottom Line

Microsoft’s September 25 Copilot update is significant because it changes the unit of AI work from a prompt toward an ongoing job. Home organizes the starting point. Today is designed to surface and prepare the work that needs attention. @Copilot brings shared context into the team. Autopilot is the component designed to keep a delegated objective moving after the user stops actively prompting it.

That does not mean every follow-up is now autonomous or every Microsoft 365 customer has these capabilities. Autopilot is still in Private Preview, Today is scheduled for October preview, and governance boundaries remain central to the design. The direction, however, is clear: meeting summaries and email drafts are becoming intermediate artifacts in a larger system that tracks obligations and continues work across time.