Microsoft's Copilot gets Autopilot, an AI agent that works without repeated prompts
At Ignite 2026, Microsoft began giving some business customers an agent that follows up on work by itself, plus tools to track and limit what agents do.
The short version
- Autopilot checks ongoing work and acts on it, using its own memory and tool access.
- It works inside Teams and Outlook on email follow-ups, repeat tasks and project schedules.
- Each Autopilot gets its own digital identity and a named human manager.
- Agent 365 now tracks agent usage, spending and success rates, and can cap spending.
- Autopilot and Copilot Workspaces are available only to some enterprise customers so far.
Why it matters
Most people still use AI by typing a request and waiting for an answer. Autopilot shifts that toward an assistant that keeps working in the background, which raises practical questions about access, cost and who is responsible. If your workplace uses Microsoft tools, this is the kind of feature your I.T. team will soon need a policy for.
An assistant that keeps working
Microsoft used its Ignite 2026 event in San Francisco on November 17 to start rolling out 'Autopilot' inside Copilot, its AI assistant. Microsoft first revealed the idea in September. As aitimes.com reported, it is now reaching some business customers.
An agent is an AI that takes actions rather than only answering questions. What sets Autopilot apart is that it does not wait for a fresh command every time. It looks at the work already in progress and does what needs doing.
It has its own memory and its own permission to use certain tools. It lives in everyday work programs such as Teams and Outlook, where it handles email follow-ups, repeat chores and project scheduling.
What it looks like in practice
Microsoft's example is vendor evaluation. Give Autopilot that job and it builds a schedule, prepares the relevant meetings, checks in with the people involved on progress, and manages the follow-up tasks.
In Teams you can call on it much as you would a colleague and hand it a task. Microsoft also announced Copilot Workspaces, a shared space where several employees and several agents work on one project. People can draft documents there and see changes and remaining tasks. This is also in the first rollout to some business customers.
A third piece, Copilot Code, lets people build work apps and dashboards by describing them in plain language. Microsoft plans a full release within the coming weeks.
Guardrails for a self-directed tool
A tool that acts on its own needs limits, and Microsoft built several in. Every Autopilot is given a separate digital identity and a named human manager. Companies can restrict which data and tools it can reach, and they can switch it off.
- Its work history is recorded.
- Important actions can be set to require a person's approval.
The newsletter put it this way: always-on takes trust, and tracking something is not the same as it being right. A log shows what an agent did. It does not show whether the agent did the right thing.
Managing agents like a workforce
Microsoft also expanded Agent 365, its platform for managing company agents. Since its launch in May, the number of agents registered on it has passed the tens-of-millions mark, according to Microsoft.
A new analytics feature covers agent usage, spending, task success rates, hours of work supported and user satisfaction. Companies can use it to judge whether their AI spending pays off and to forecast future costs. They can also set spending limits by user, department or agent. This analytics tool is still a preview for some business customers.
The management reaches beyond Microsoft's own products. Agents from OpenAI, Anthropic, Google and Salesforce can be given separate identities, access rights and security policies. Support differs by platform and is being added step by step.
Data, security and the bigger strategy
Microsoft IQ is a platform that connects company documents, work procedures and databases so agents can use them. It supports connections to more than three hundred outside services. In security, Microsoft Perception is now generally available. It has several specialist agents cooperate to detect, investigate and respond to cyberattacks, with people approving the important steps.
The aim is to turn Copilot from a chatbot into a platform that does and oversees company work. Microsoft offers its own models alongside those from OpenAI and Anthropic. It can pick a suitable model for each task to balance quality and cost.
Judson Althoff, a Microsoft sales leader, told The Information that Copilot can do what Claude or ChatGPT does at higher quality and lower cost. That is a company claim, not a tested result. Microsoft's real advantage is its installed base: Word, Excel, PowerPoint and Teams already hold a great deal of company data. The competition with OpenAI and Anthropic is increasingly about who controls the workplace, the data and the agent management layer, not only whose model is smartest.
What to watch
- Which tasks Autopilot handles reliably once more companies use it, and how often people must step in.
- Pricing, which was not stated, and when Autopilot reaches more customers.
- How well Agent 365 manages agents from other companies, since support differs by platform.
Try it
If you use Copilot at work, ask your I.T. team what Autopilot would be allowed to read and do, and who the named manager would be, before anyone switches it on.
Sources and links
- Reportingaitimes.com: the article we sourced
First reported by aitimes.com. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.