Microsoft has rebuilt Copilot around three new work modes: Home, Code and Autopilot. Announced on September 25, 2026, the redesign moves Copilot beyond a chat assistant and toward a broader work platform for asking questions, delegating multi-step work, building internal tools and running persistent agents.
The change matters because Microsoft is also drawing a clearer line between two kinds of AI spending: fixed-cost user subscriptions for everyday Copilot use, and usage-based billing for longer-running agentic work. For organizations already evaluating Microsoft 365 Copilot, this is not just a UI refresh — it changes how Copilot is positioned, deployed and budgeted.
Microsoft Copilot 2026: what changed?
The new Copilot app is organized around three capabilities:
| Capability | What it does | Availability |
|---|---|---|
| Home | Brings Chat and Cowork together, with Word, Excel and PowerPoint integrated through Office in Copilot. | Rolling out through Microsoft’s Frontier program over the coming weeks. |
| Code | Lets users describe apps, trackers, dashboards, automations and workflows in natural language and have Copilot build them. | Rolling out to Frontier at the end of September, with broader availability planned in the following weeks. Preview for Microsoft 365 Premium and Pro subscribers is planned later in 2026. |
| Autopilot | A persistent cloud-hosted agent that can continue recurring or delegated work even when the user is away. | Expanding to private preview at the end of September. |
Microsoft’s official announcement also adds a preview of Copilot Managed Runtime, expands grounding through Microsoft IQ, Fabric IQ, Dynamics 365 and Power Platform, and introduces additional FinOps controls for managing AI spend.
Home: Copilot becomes the starting point for work
Home is designed to consolidate the parts of Copilot that previously felt more separate. Chat remains the fast conversational layer for questions, drafting, lookups and analysis, while Cowork is intended for larger tasks that can be delegated end to end.
Microsoft gives examples such as producing an RFP response, building a launch kit, preparing a customer briefing or assembling a financial-close package. The longer-term direction is even more important: Microsoft says users eventually will not need to decide whether a task belongs in Chat, Cowork or Code. Copilot is expected to route the work to the appropriate capability automatically.
Office in Copilot is another major part of Home. Word, Excel and PowerPoint documents can be created or edited directly within the Copilot experience while remaining real Office files. Microsoft says edits remain synchronized between Copilot and the corresponding Office apps.
For users, that reduces one of the common friction points in AI workflows: generating content in a chat window and then rebuilding, reformatting or transferring it into the actual document that the team uses.
Code: natural-language app and workflow building
Code is Microsoft’s attempt to extend software creation beyond traditional developers. A user can describe an app, tracker, dashboard, automation or workflow in natural language, and Copilot can select an approach and build it.
Microsoft says Code uses the same underlying technology as GitHub Copilot. It runs in a sandboxed environment and can be hosted inside an organization’s Microsoft 365 tenant. Examples range from persistent desktop widgets and interactive dashboards to internal cloud-hosted apps that can be shared with coworkers.
This puts Code in an increasingly competitive category that includes AI app builders, coding agents and workflow-generation tools. The key Microsoft advantage is not simply code generation: it is the combination of generation, organizational context, tenant governance, Office data and deployment inside an existing Microsoft environment.
For organizations that already standardize on Microsoft 365, Code could reduce the number of small internal tools that need to be handed to a development team. For developers, GitHub Copilot remains the dedicated development environment; Code is positioned more broadly for knowledge workers.
Autopilot: Microsoft’s persistent cloud agent
Autopilot is the most consequential part of the announcement. Microsoft says the product was previously called Scout. Rather than waiting for one prompt at a time, Autopilot can be assigned a name, role and objective and then continue working over time.
It can watch channels, follow up on conversations, run recurring work and resume a project days later. Because it is cloud-hosted, it can continue operating when the user’s computer is off or when the user is doing something else.
Microsoft says each Autopilot instance lives inside the organization’s tenant with its own identity, memory, computer and workspace. It can appear in Teams, Outlook, chats, channels and documents, while enterprise permissions, auditing and governance remain in place.
This is materially different from a normal chatbot session. A persistent agent creates new operational questions: who owns the agent, what permissions it receives, what data it can access, what actions it can take, how long it retains context and how its ongoing consumption is controlled.
Those concerns are why persistent agents should be evaluated alongside broader AI agent security practices, not treated as just another productivity feature.
What happened to the old Copilot model?
The previous Copilot experience was primarily understood as an AI assistant embedded across Microsoft 365 and accessible through chat. That model is not disappearing, but Microsoft is broadening the product around four distinct types of work:
- Ask: quick questions, summaries, drafts and analysis.
- Delegate: hand a larger task to Cowork.
- Build: create an app, automation or workflow with Code.
- Run persistently: give ongoing work to Autopilot.
This makes Copilot look less like a single assistant and more like an operating layer for AI-enabled work across Microsoft 365.
Microsoft Copilot pricing: subscriptions versus usage-based billing
The most important pricing change is conceptual. Microsoft now separates everyday AI usage from long-running agentic usage.
| Billing model | Designed for | Examples |
|---|---|---|
| User subscription license (USL) | Everyday AI work at a predictable per-user price. | Copilot Chat, Word, Excel, PowerPoint, Outlook, Teams, model selection and everyday assistance. |
| Usage-based billing (UBB) | Advanced or long-running agentic workloads where consumption varies. | Cowork, Code, Autopilot and certain frontier models and agentic workloads. |
Microsoft says its Auto model-routing system is central to the subscription experience, choosing a model based on quality, speed and cost. Usage-based workloads provide more direct control over advanced model selection and allow organizations to match AI spending to the value of the task.
As of September 26, 2026, Microsoft’s U.S. business pricing page lists Microsoft 365 Copilot Business at a promotional starting price of $18 per user/month when paid yearly, down from the listed $21 starting price. The monthly subscription is listed at $25.20 per user/month. Pricing can vary by region, purchase channel, taxes and promotional eligibility, so organizations should confirm the current Microsoft pricing page before purchasing.
There is also an important CSP billing change coming on November 2, 2026. Microsoft says new Microsoft 365 Copilot Business purchases through the Cloud Solution Provider program will include usage-based billing enabled by default. Eligible metered experiences include Copilot Cowork, Work IQ APIs and GitHub Copilot Harness.
That does not mean every Copilot action suddenly becomes pay-as-you-go. The fixed user subscription remains the base for everyday Copilot use; usage-based billing applies to eligible metered experiences.
FinOps becomes part of Copilot administration
Once agents can run for extended periods, AI cost management becomes much more important. Microsoft is expanding FinOps controls across Agent 365, Insights and Copilot.
According to Microsoft, administrators will be able to manage spending policies, route credit requests into approval workflows, control which model families different user groups can access and review usage and outcomes from Cowork tasks. End users will also be able to see credit usage, balances and usage history inside Copilot.
For IT teams, this means deploying agents will increasingly resemble deploying other metered cloud infrastructure: permissions and security matter, but so do budgets, limits, observability and accountability.
Copilot Managed Runtime
Microsoft is also previewing Copilot Managed Runtime, a hosting layer for code and applications created through Copilot.
The runtime is designed to let generated applications execute inside the company’s Microsoft 365 environment under IT governance. Microsoft says it will support apps built through Cowork, Code and Copilot Studio, and will also be available to third-party and professional developers.
This could become strategically important because AI app builders often solve generation before they solve deployment. Microsoft is trying to package both: build the tool with Copilot, then run it in a managed enterprise environment.
Microsoft IQ and business grounding
The new Copilot also expands access to business context. Microsoft describes Microsoft IQ as the unified intelligence layer behind Copilot and its agents.
Fabric IQ can bring governed enterprise data and semantic models into Chat and Cowork, with Code integration planned through Frontier. Microsoft is also rolling out additional Dynamics 365 and Power Platform grounding and a unified plugin registry for Microsoft, partner and custom capabilities.
The practical difference is that Copilot is being designed to act not only on what a user types into a chat box, but on the organization’s approved data, applications and workflows.
Who gets access first?
Availability is staggered rather than universal on day one:
- Home: rolling out through Frontier over the coming weeks.
- Code: Frontier rollout begins at the end of September, followed by broader availability; Microsoft also plans a preview for Microsoft 365 Premium and Pro subscribers later in 2026.
- Autopilot: expands to private preview at the end of September.
- Copilot Managed Runtime: currently in preview.
- Today in Home: planned for private preview in October.
- @Copilot in Teams: planned for private preview by the end of September.
Because Microsoft is using previews and Frontier rollouts, two organizations with similar licenses may not see identical features at the same time.
Practical implications for businesses
1. Copilot is becoming a platform, not just a seat-based assistant
Organizations evaluating Copilot should now ask what they want employees and agents to do, not only how many users need chat access. Building apps, delegating work and running persistent agents introduce different governance and cost requirements.
2. AI budgets need two layers
A per-user subscription is no longer the whole cost model. Teams adopting Cowork, Code or Autopilot should plan for variable consumption and define usage thresholds before persistent agents become widely deployed.
3. Internal app building gets easier — governance gets harder
Code can reduce the barrier to creating internal software, but easier creation can also produce app sprawl. IT teams will need ownership rules, data-access policies, lifecycle controls and review standards.
4. Persistent agents deserve stronger security controls than chat
An agent with its own identity, memory, computer and workspace has a larger operational surface than a normal conversational assistant. Permissions should be intentionally scoped and actions should remain auditable.
5. Microsoft 365 customers gain a stronger integration advantage
Companies already using Teams, Outlook, SharePoint, Office and Power Platform may get more value from Copilot than organizations using Microsoft only as a document suite, because the new experience depends heavily on organizational context and integrated workflows.
How this compares with the broader AI-agent trend
Microsoft’s direction reflects a broader shift across the AI industry: assistants are moving from responding to prompts toward executing work. Coding agents, browser agents, automation platforms and persistent cloud agents are increasingly overlapping.
The competitive question is becoming less about which chatbot writes the best paragraph and more about which platform can safely combine reasoning, tools, business data, execution, memory, permissions and cost controls.
If you are comparing that broader category, see AI-XBlog’s guide to AI computer-use and browser agents and our practical overview of AI tools for small business.
Bottom line
The September 2026 Copilot overhaul is one of Microsoft’s clearest attempts yet to turn Microsoft 365 into an agentic work platform.
Home centralizes conversational and delegated work. Code gives non-developers a path to build apps and automations. Autopilot introduces persistent agents that can continue operating in the cloud. At the same time, Microsoft is separating fixed subscription usage from metered agentic workloads and adding FinOps controls to make that model manageable at scale.
For current Microsoft 365 Copilot customers, the most important next steps are to watch Frontier availability, understand which workloads are metered, define agent permissions early and decide where persistent automation genuinely creates enough value to justify variable usage costs.
Sources
- Microsoft: Introducing the new Copilot with Home, Code and Autopilot
- Microsoft Learn: September 2026 Partner Center announcements
- Microsoft: Copilot plans and pricing for business
Last updated: September 26, 2026. Microsoft is rolling these capabilities out in stages, and pricing or availability may change. AI-XBlog will update this page as Microsoft publishes broader release details.
AI-XBlog Weekly Brief
Keep up with AI that actually works
Join the AI-XBlog Weekly Brief for major AI updates, practical workflows, useful tools, and editor’s picks. No daily noise.
Reader discussion
Join the discussion
Have you tried this tool or workflow? Share your experience, corrections, or questions. Useful reader feedback may help us improve this article.
All comments are reviewed before publication. Your email address will not be published. Promotional links and low-value spam are removed.
