Microsoft Rebuilds Copilot With Home, Code and Autopilot: ‘OS for Work’ Explained

Microsoft Copilot Autopilot is the most ambitious part of Microsoft’s latest Copilot overhaul. Announced alongside Copilot Home and Copilot Code, Autopilot is designed as a persistent, proactive cloud agent that can keep working across projects even when the user is not actively prompting it.

The broader redesign points to a bigger strategy. Microsoft is bringing Chat, Cowork, Office, app building and autonomous agents into one Copilot experience instead of treating AI as a feature layered on top of separate productivity apps. In an interview reported by The Verge, Microsoft CEO Satya Nadella described that direction as an “OS for work” — not a formal product name, but a useful description of the company’s ambition.

Microsoft’s September 25 announcement divides the new Copilot into three major modes: Home for everyday work and delegated tasks, Code for creating apps and automations, and Autopilot for recurring, ongoing work that should continue without constant user attention. Home and Code are rolling out through Microsoft’s Frontier program, while Autopilot is moving into private preview.

Quick take: Microsoft is trying to turn Copilot from an assistant inside Office into the central layer where people ask questions, delegate work, create apps, collaborate on documents and manage autonomous agents. The company is also introducing Copilot Managed Runtime for governed internal apps and expanded FinOps controls for usage-based AI spending. Microsoft announced the new Copilot on September 25, 2026.

What changed in Microsoft Copilot?

The new Copilot reorganizes Microsoft’s AI work experience around three capabilities rather than a long list of disconnected features.

Microsoft’s design reflects three different kinds of interaction. Chat handles immediate questions. Cowork handles tasks the user wants to delegate end-to-end. Code creates software solutions. Autopilot takes responsibility for work that persists over time.

That distinction is important because it moves Copilot beyond “generate a document” or “summarize this email.” Microsoft is building a system where the AI can create the document, build the internal app that tracks the process, and assign an always-running agent to keep the process moving.

Copilot Home brings Chat, Cowork and Office together

Copilot Home is the new starting point. Microsoft says it combines Chat and Cowork so users can move between quick conversational work and fully delegated tasks without leaving the Copilot experience.

Chat remains the fast, reactive layer: lookups, drafts, summaries and short questions. Cowork is for work that can run end-to-end, such as preparing an RFP response, customer briefing, launch kit or financial close package.

The bigger change is Office in Copilot. Word, Excel and PowerPoint are being brought directly into the Copilot experience. A user can ask Copilot to draft a brief, build a spreadsheet model or create a presentation, then continue editing the real file rather than switching into a separate generated artifact.

Microsoft says the files remain live and synchronized with the underlying Office apps. That means edits made in Copilot or in Word, Excel or PowerPoint can remain aligned rather than creating multiple disconnected copies.

This is strategically important. Microsoft is not removing Office; it is changing where users begin. The starting point increasingly becomes Copilot, while Office apps become specialized editing environments inside or alongside that broader AI workflow.

Microsoft Copilot Code turns natural language into internal software

Copilot Code extends the new Copilot beyond documents and chat. Users can describe an app, tracker, dashboard, automation or workflow in natural language, and Copilot chooses an approach and builds it.

Microsoft says Code uses the same underlying technology as GitHub Copilot, but the audience is broader than professional developers. The company wants solution-building to become a normal knowledge-worker skill — similar to creating a spreadsheet or presentation.

Examples Microsoft gives include:

• Persistent desktop widgets for quick-reference information.
• Interactive dashboards for exploring business data.
• Internal apps shared with a team.
• Automations that connect data and workflows.
• Purpose-built tools grounded in organizational context.

Code is designed to run inside a sandboxed environment and can be hosted securely within an organization’s Microsoft 365 tenant. Microsoft says plugins and work context can be incorporated automatically, while developers can continue using GitHub Copilot for deeper engineering work.

This overlaps with the broader trend covered in Vynula’s AI workflow automation guide: software creation is moving closer to the people who understand the business process, while IT retains responsibility for security and governance.

What is Microsoft Copilot Managed Runtime?

Making it easy for employees to generate apps creates a governance problem: organizations do not want hundreds of untracked internal applications running with unknown permissions. Copilot Managed Runtime is Microsoft’s answer.

Managed Runtime is a preview hosting environment for line-of-business apps built through Cowork, Copilot Studio or Microsoft’s developer SDK. Microsoft says apps created there automatically inherit organizational controls rather than requiring each maker to build authentication, compliance and lifecycle management from scratch.

According to Microsoft Learn, Managed Runtime includes:

• Microsoft Entra authentication and authorization by default.
• Connections to more than 1,500 data sources.
• Microsoft 365 tenant governance policies.
• Data loss prevention and connector policies.
• Centralized app inventory in the Microsoft 365 admin center.
• Usage, health and lifecycle monitoring.
• Git-based development workflows for professional developers.

Every app can be visible to administrators regardless of whether it was created through Cowork, Copilot Studio or the SDK. This gives Microsoft a way to encourage “vibe coding for work” without telling enterprises to abandon traditional control structures.

Microsoft’s documentation describes the feature as preview software, so availability, limits and implementation details may still change before general release.

Microsoft Copilot Autopilot is the persistent agent layer

Microsoft Copilot Autopilot is the most autonomous of the new capabilities. It was previously called Scout, according to Microsoft’s own announcement, and is designed as a digital teammate rather than a one-off task runner.

The user gives Autopilot a name, a role and a goal. From there, Microsoft says it can watch channels, follow up on threads, run recurring work and resume a project days later without waiting for another prompt.

The key technical idea is persistence. Autopilot is cloud-hosted and has its own:

• Identity
• Memory
• Computer
• Workspace

Microsoft says Autopilot lives inside the organization’s tenant and is built on Microsoft IQ so it can use work context while remaining subject to enterprise permissions, audit and governance.

The company’s example is a supplier-review process. Autopilot could build a schedule, prepare meetings, follow up with stakeholders, collect updates and continue managing the process over time.

That moves the product closer to the persistent-agent model Vynula recently covered in OpenAI Dots. Both companies are moving beyond agents that wait for a task toward agents that maintain a standing responsibility.

Autopilot can work inside Teams, Outlook and documents

Microsoft says Autopilot appears in the tools employees already use, including Teams, Outlook, chats, channels and documents. Users can @mention the agent like a colleague while the organization controls what it is allowed to access.

This matters because enterprise agents are unlikely to succeed if they require employees to constantly switch into a separate AI application. Microsoft’s advantage is that many organizations already conduct work inside Teams, Outlook, SharePoint and Office.

The tradeoff is that an always-running agent can potentially touch a much larger surface area than a normal chat session. Identity, permission design and audit logs therefore become core product features rather than optional enterprise extras.

That connects directly with Vynula’s broader AI agent governance coverage: the more persistent and capable an agent becomes, the more important it is to define who owns it, what it may access and how its actions are reviewed.

Copilot Home vs Cowork vs Code vs Autopilot

Microsoft says a future version of Home will reduce the need to manually choose between these modes. The user will describe the goal, and Copilot will route work to Chat, Cowork or Code depending on what the task requires.

That routing concept is another sign that Microsoft wants Copilot to behave less like a collection of AI features and more like a control layer over work.

What is Today in Copilot Home?

Microsoft also previewed Today, a proactive command center inside Home. Today is designed to summarize what changed across email, calendar, Teams threads, meetings and tasks, then prepare the next actions.

Microsoft’s examples include a drafted message waiting for review, a proposed schedule change and a forgotten follow-up prepared with context.

Today is expected to enter private preview in October and later appear in Outlook and Teams. This is distinct from Autopilot: Today is primarily a proactive overview and action surface, while Autopilot is a persistent agent with an assigned role.

@Copilot in Teams gets shared team context

Another upcoming feature is a more collaborative form of @Copilot in Teams. Microsoft says that when Copilot is invoked in a channel, group chat or meeting, it can use the shared context and permissions available to that group.

For example, a team could ask why a product decision was made and have Copilot retrieve supporting context from previous discussions rather than requiring one individual to restate everything.

This design highlights an enterprise problem that consumer assistants rarely face: whose context should an agent use? A personal assistant has the user’s permissions; a team assistant needs a shared model of what the group is allowed to know and do.

Microsoft Copilot Autopilot vs OpenAI Dots

Microsoft Autopilot and OpenAI Dots share a broad goal: both are persistent agents intended to keep working after the initial conversation ends. But their product context differs.

This table describes documented product characteristics, not a ranking. Real-world usefulness will depend on reliability, integrations, governance requirements and the software environment an organization already uses.

The strategic similarity is more important than the feature-by-feature difference: both Microsoft and OpenAI now believe the next major AI interface is a persistent agent with continuing responsibility.

Microsoft is adding FinOps for AI

Persistent agents and long-running tasks create a new cost problem. A normal productivity subscription is predictable; an agent that runs for hours, invokes premium models and creates applications can generate variable usage.

Microsoft is therefore expanding FinOps for AI, its framework for monitoring and controlling AI spending. The company says long-running capabilities including Cowork, Code and Autopilot use usage-based billing.

Microsoft’s new FinOps controls include expanded cost management across Agent 365, Code and Copilot Managed Runtime. Administrators can create spending policies, route credit requests through approval workflows and restrict which model families groups may use.

Users can also see credit usage, balances and history inside Copilot.

This is an important part of the architecture because “autonomous” software can create costs without a human manually starting every operation. Enterprise adoption therefore requires not only security guardrails but financial guardrails.

Why Microsoft is moving Office into Copilot

The traditional Microsoft 365 workflow starts with an application: open Word to write, Excel to analyze, PowerPoint to present and Outlook to communicate.

The new Copilot model starts with an objective. The user describes what needs to happen, and the system decides which capability, application, document or agent should be involved.

That is why bringing Office into Copilot is more strategically significant than simply adding another sidebar. It changes the navigation hierarchy. Word, Excel and PowerPoint remain important, but the user increasingly enters through an AI layer that spans all of them.

The Verge reports that Satya Nadella described Copilot as an “OS for work.” That phrase captures the ambition, but it should not be confused with a literal replacement for Windows or a formal product name. Microsoft is describing an orchestration layer over work rather than a conventional computer operating system.

Why enterprise governance could decide whether Autopilot succeeds

Persistent agents promise convenience, but they also magnify mistakes. An incorrect answer in a chat is one event; an incorrect decision inside an agent that continues acting over several days can compound.

Microsoft’s product architecture puts unusual emphasis on governance:

• Tenant-scoped identities and permissions.
• Auditability.
• Centralized app inventory.
• Entra authentication.
• Data-loss-prevention policies.
• Usage-based spending controls.
• Admin visibility into managed apps and agents.

These mechanisms do not prove that an autonomous workflow will be safe or reliable. They provide the organizational controls needed to deploy one responsibly.

This is also why Managed Runtime matters. If employees can generate applications and agents with natural language, IT needs visibility by default rather than discovering those systems after they become critical to a business process.

What the new Copilot does not prove yet

Microsoft’s announcement describes an ambitious end state, but several questions remain open.

• Reliability: How consistently can Autopilot maintain context and complete recurring work over days or weeks?
• Permission design: How easy is it for organizations to give agents enough access to be useful without over-permissioning them?
• Agent sprawl: What happens when teams create dozens or hundreds of autonomous workers?
• Cost predictability: Can FinOps controls keep long-running usage aligned with business value?
• Human oversight: Which recurring actions should always require review?
• App quality: How robust are Code-generated internal apps compared with traditionally engineered software?
• Preview limitations: How much will the products change before general availability?

These questions matter more than whether a demo looks impressive. The new Copilot is attempting to become infrastructure for real work, where failure has organizational consequences.

Vynula analysis: Microsoft wants Copilot to become the control plane for work

The most important part of this launch is not Autopilot alone. It is the way Microsoft is connecting documents, applications, agents, organizational context and spending controls into a single system.

Home gives users a starting point. Cowork executes delegated tasks. Code turns natural language into software. Managed Runtime hosts those applications under governance. Autopilot owns recurring work. FinOps controls the cost of all of it.

Taken together, that looks less like an AI assistant and more like a control plane for knowledge work.

Microsoft has a structural advantage here: many businesses already store identity, communication, files, meetings and collaboration inside Microsoft 365. An AI layer does not need to convince those organizations to move all their work into a completely new ecosystem if it can operate across the systems they already use.

The challenge is trust. Earlier generations of workplace copilots could be ignored if their suggestions were mediocre. A persistent agent cannot be ignored in the same way because it may be acting while the user is elsewhere.

That changes the standard Microsoft must meet. Autopilot needs to be more than clever; it needs to be predictable, observable and governable.

In Vynula’s view, the “OS for work” idea is useful precisely because it reframes the competition. The next enterprise AI platform may not be judged mainly by which chatbot produces the most elegant answer. It may be judged by which system can coordinate documents, software, agents, permissions and budgets across an organization without creating more chaos than it removes.

What to watch next

• Autopilot private preview: what early enterprise users report about persistent-agent reliability.
• Frontier rollout: how quickly Home and Code reach more Microsoft 365 tenants.
• Managed Runtime adoption: whether business users actually build useful internal apps with natural language.
• Today in Home: whether proactive briefings reduce or increase notification overload.
• Usage costs: how expensive Cowork, Code and Autopilot become under sustained workloads.
• Agent 365 governance: whether centralized agent management scales cleanly as companies create more autonomous workers.
• Competition with OpenAI: how Autopilot evolves relative to OpenAI Dots and other persistent-agent products.
• Office behavior: whether users begin more work in Copilot rather than opening Word, Excel or PowerPoint directly.

Frequently asked questions

What is Microsoft Copilot Autopilot?

Microsoft Copilot Autopilot is a cloud-hosted persistent AI agent designed to keep working on recurring responsibilities without waiting for a new prompt each time. Microsoft says it has its own identity, memory, computer and workspace inside the organization’s tenant.

Is Copilot Autopilot available now?

Microsoft says Autopilot is expanding into private preview. It is not yet a broadly available production feature for all Microsoft 365 customers.

What is Copilot Home?

Copilot Home is Microsoft’s new starting point for AI work. It combines Chat, Cowork and Office in Copilot so users can ask questions, delegate work and create or edit Word, Excel and PowerPoint files from one interface.

What is Copilot Code?

Copilot Code lets users describe an app, dashboard, tracker, automation or workflow in natural language and have Copilot build it. It uses technology related to GitHub Copilot and can deploy apps through Microsoft’s governed runtime environment.

What is Copilot Managed Runtime?

Copilot Managed Runtime is a preview Microsoft-hosted environment for internal business apps. It provides Entra authentication, enterprise governance, data connections, admin visibility and lifecycle controls.

Is Microsoft Copilot really an operating system?

No. “OS for work” is a strategic description reported from Satya Nadella, not the name of a literal operating system. The idea is that Copilot becomes an orchestration layer over Microsoft 365 work, documents, apps and agents.

How is Microsoft Copilot Autopilot different from OpenAI Dots?

Both are persistent AI agents, but Microsoft Autopilot is designed deeply around Microsoft 365 tenant identity, Teams, Outlook and enterprise governance. OpenAI Dots are centered on the ChatGPT ecosystem and connected apps. Both products are still evolving, so practical differences will become clearer as deployments expand.

Primary and independent sources

• Microsoft — Introducing the new Copilot with Home, Code and Autopilot
• Microsoft Learn — What is Copilot Managed Runtime (preview)
• Microsoft Learn — Copilot Managed Runtime governance overview
• Microsoft 365 — AI at Work Roadmap
• The Verge — Inside Microsoft’s big Copilot rethink

Reporting note: Product capabilities, availability and governance details are attributed to Microsoft unless otherwise stated. The phrase “OS for work” is attributed to Satya Nadella via The Verge and is not presented as a formal product name. Vynula Analysis represents editorial interpretation. Images are illustrative Unsplash photography, not official Microsoft product artwork.