Three ways to work, one destination
Microsoft announced a new structure for its Copilot app on September 25: Home, Code and Autopilot. Home brings Chat and Cowork together and incorporates Word, Excel and PowerPoint work into the experience. Code is intended to let people build solutions using technology related to GitHub Copilot. Autopilot is described as a persistent agent that can continue working when the user is away. These are product descriptions from Microsoft, not a claim that all three modes are available to every customer today.
The rollout matters. Microsoft says Home and Code will begin reaching its Frontier program in the coming weeks. Autopilot is expanding to private preview at the end of September. That makes this a direction of travel for workplace software, not an immediate universal upgrade. A team deciding what to buy should check its tenant, licensing and data policies before planning a workflow around any announced feature.
Why the boundaries matter
The interesting change is the division of labor. A conversational assistant answers and drafts; a delegated worker completes a defined project; a coding environment builds a tool. If Copilot can route a request among those modes, the user may spend less time translating one task between apps. Microsoft says that automatic routing is coming, but the experience and safeguards need to be evaluated in the actual environment where work occurs.
Persistent agents raise a different question from chatbots: what can they do without a person watching? A good deployment needs clear permissions, visible progress, an audit trail and confirmation before an irreversible action. The useful metric is not how impressive an agent looks in a demo. It is whether a colleague can understand what it changed, recover from an error and stop it in time.
What to test first
Start with a low-risk workflow, such as turning a meeting brief into an editable draft. Record the input documents, expected output, time spent correcting it and whether the final file preserves formatting. Then test a handoff between chat and a delegated task. Do not put sensitive material into a preview environment without approval. Microsoft's announcement also discusses spend-management features; budget owners should measure both subscription cost and agent activity. NEURA has not independently tested the new modes. Their value will be determined by reliability, control and actual availability, not the number of names in the menu.




