Microsoft has reorganized its Copilot application around three modes of work: Home, Code and Autopilot. The company announced the redesign on 25 September 2026 as a single place where people can ask questions, delegate longer assignments, create small software tools and manage agents that continue working in the background.
The launch is significant because it moves Microsoft’s workplace AI pitch beyond a chat box attached to familiar applications. Home is designed as the starting point for everyday work. Code lets non-developers describe an app, dashboard, tracker or automation in natural language. Autopilot is Microsoft’s name for a persistent agent that can monitor work and act over time within assigned boundaries.
There is an important qualification: the announcement is a rollout, not a declaration that every feature is immediately available to every Microsoft 365 customer. Microsoft says Home and Code will reach its Frontier early-access program in stages, while Autopilot is expanding to private preview at the end of September. Broader access comes later.
Independent reporting from CNBC and GeekWire confirmed the main product changes and placed them in a wider competitive context. Both outlets described Microsoft’s effort to combine chat, coding and agentic work in one application as it competes with increasingly integrated offerings from Anthropic and OpenAI.
| Key takeaway: Microsoft is turning Copilot into a work surface for conversations, files, lightweight software and long-running agents. The product direction is clear, but availability, control and real-world value still need to be tested organization by organization. |
|---|
What Microsoft announced on 25 September
The new Copilot is built around three named areas rather than a single undifferentiated assistant.
Home brings Chat and Cowork together. Microsoft says Word, Excel and PowerPoint can also operate within the Copilot experience, allowing a user to create or update an editable file while continuing the surrounding conversation. The goal is to reduce context switching between a prompt, a document and the apps used to finish it.
Code is for building purpose-specific tools. A user can describe a tracker, dashboard, internal app or automation, and Copilot can select an approach and construct it. Microsoft says the feature uses underlying technology from GitHub Copilot, while the resulting code can run in a sandboxed environment and be hosted in the organization’s Microsoft 365 tenant.
Autopilot is the long-running agent layer. Previously introduced under the name Scout, it can be given a role, goal and operating boundaries. Microsoft’s examples include monitoring channels, following up on work and resuming a project after time has passed.
The company also announced Copilot Managed Runtime, hosting infrastructure intended to let generated code run inside a managed Microsoft 365 environment. Microsoft describes the runtime as governed by IT and usable by apps created through Cowork, Code, Copilot Studio and third-party development tools.
Home combines chat, delegated work and Office files
Home separates quick interaction from delegated outcomes without forcing users to begin in a different application. Chat is positioned for questions, research, summaries and first drafts. Cowork is positioned for a more complete assignment, such as assembling a briefing, preparing a launch kit or developing a response package.
Office in Copilot is central to that design. Instead of producing a flat answer that must be copied into a file, Microsoft says Copilot can work on live Word documents, Excel workbooks and PowerPoint presentations. Colleagues can continue editing through the regular Office apps, while changes remain synchronized.
That integration could reduce handoff friction, provided people can see what changed, confirm the source data and reverse unwanted edits.
Code brings app building to knowledge workers
Code extends Copilot from writing files to creating small pieces of software. Microsoft’s examples include internal dashboards, desktop widgets, trackers, cloud-hosted apps and automations.
Business teams should treat a Copilot-built app like any operational asset: assign an owner, document inputs, test error states, control access and keep a rollback path.
Microsoft says Code operates in a sandbox and can use Copilot Managed Runtime. Those are meaningful design claims, but they do not eliminate the need for organization-specific security review. A sandbox limits some risks; it does not decide which data a workflow should access or which action should require approval.
Autopilot handles persistent work
Autopilot is designed to continue working after the initial prompt. Microsoft says each agent can operate inside the organization’s tenant with its own identity, memory, computer and workspace. It can appear in Teams, Outlook, chats, channels and documents, and administrators can apply permissions, audit and governance controls.
Persistence changes the risk profile. A normal chat response is visible at the moment it is produced. A long-running agent may encounter new messages, changing data and exceptions hours or days later. Its objective can stay the same while the environment changes around it.
That makes boundaries more important than eloquent prompts. Teams should define what the agent may read, what it may write, which people it may contact, how much it may spend and when it must stop for human review.
What is available now and what is still in preview
Microsoft’s announcement contains several different rollout stages:
| Capability | Announced state |
|---|---|
| Home | Starting to roll out through the Frontier program in the coming weeks |
| Office in Copilot | Planned for consumers and Frontier customers in the coming weeks |
| Code | Frontier rollout at the end of September, broader availability later; Premium and Pro preview planned later in 2026 |
| Copilot Managed Runtime | In preview |
| Autopilot | Expanding to private preview at the end of September |
These labels matter. A preview can change, have limited support or exclude some regions and tenants. Procurement and implementation plans should use the availability shown in the customer’s own admin center and contract, not a headline.
Why the pricing model matters
Microsoft is dividing Copilot work into two commercial patterns. A user subscription license covers familiar, frequent tasks such as chat and assistance in Microsoft 365 apps. More intensive agentic work—including Cowork, Code, Autopilot and access to certain frontier models—uses usage-based billing.
That creates a new management problem. Per-seat software is relatively easy to budget. An agent that runs for hours, invokes several models and repeats work can produce variable consumption. Microsoft is adding FinOps for AI controls, but organizations still need task-level measurement.
A useful pilot should record cost per completed outcome, not merely prompts or tokens. If an agent prepares a supplier review, for example, the comparison should include the quality of the result, human review time, rework and total consumption.
The PILOT evaluation framework
Before assigning real work, evaluate the new Copilot through five gates:
| Gate | Question | Evidence to collect |
|---|---|---|
| Permissions | Does it have only the access required for the task? | Role, data sources, write actions and approval points |
| Information | Can reviewers trace outputs to current, authorized sources? | Citations, file versions and data freshness |
| Limits | Does the agent stop safely when inputs, costs or conditions exceed a boundary? | Budget caps, timeouts, escalation tests and denied-action logs |
| Ownership | Is a named person accountable for the workflow and its failures? | Owner, backup owner, review cadence and rollback procedure |
| Total value | Is the outcome better after cost, review and rework are counted? | Cost per accepted result, review minutes and defect rate |
Do not advance because a demonstration looks polished. Advance when the evidence for all five gates is strong enough for the task’s risk.
What the launch means for Microsoft and enterprise AI
Microsoft is betting that the AI interface for work will combine conversation, files, software creation and persistent agents. Its advantage is proximity to systems many organizations already use. The competitive interpretation is an inference, not a launch fact: Microsoft is defending its position as the front door to workplace software while specialist AI companies build integrated work environments. CNBC and GeekWire both framed the announcement in that context.
What remains unproven
The announcement cannot prove that Code will create maintainable tools, that Autopilot will handle exceptions correctly or that usage-based economics will work for a specific team. Microsoft’s examples are product demonstrations. Independently confirmed facts are limited to the three-part structure, rollout stages and mixed subscription-and-consumption model.
Practical next steps for organizations
- Pick one reversible, low-consequence workflow with clear inputs and an objective output.
- Establish a manual baseline for time, cost, quality and error rate.
- Give the pilot the minimum permissions it needs.
- Require human approval before external messages, record changes or financial commitments.
- Test missing files, contradictory instructions, stale data and service interruptions.
- Set a hard consumption cap and an automatic stop condition.
- Compare accepted results—not raw generations—with the manual baseline.
- Expand only after the owner can explain the workflow, evidence and rollback path.
The redesigned Copilot could make workplace AI more coherent. The strongest early adopters will be the teams that pair that convenience with disciplined permissions, measurable outcomes and clear human accountability.
FAQ
Is the new Microsoft Copilot available to everyone?
No. Microsoft announced staged access. Home and Code are beginning through the Frontier program, Managed Runtime is in preview and Autopilot is expanding to private preview. Customers should verify availability in their own tenant.
Is Copilot Code the same as GitHub Copilot?
No. Microsoft says Code uses the same underlying technology as GitHub Copilot, but it is positioned inside the broader Copilot app for knowledge workers to create small apps, dashboards and workflows. Developers continue to use GitHub Copilot for software-development work.
Does Autopilot work without supervision?
Microsoft describes it as persistent and able to continue working, but organizations still need objectives, permissions, audit, limits and human approval points. Persistent does not mean risk-free or infallible.
How will the new capabilities be billed?
Microsoft says everyday assistance remains covered through a user subscription, while Cowork, Code, Autopilot and certain advanced models use usage-based billing. Contract terms and availability can vary.
