Microsoft introduces autonomous AI agents as workforce members

Key Takeaways

Microsoft is preparing to introduce what it describes as a new class of AI agents that will function as independent users within enterprise workforces, according to company documentation and industry experts who have reviewed the upcoming offering.

AI agents with full organizational identities

According to Microsoft's product roadmap document 518220, these "agentic users" represent a significant shift in how AI integrates with business operations.

The document states that "Each embodied agent has its own identity, dedicated access to organizational systems and applications, and the ability to collaborate with humans and other agents.

These agents can attend meetings, edit documents, communicate via email and chat, and perform tasks autonomously."

Microsoft MVP João Ferreira, who has examined Microsoft's internal documentation, noted that these agents will receive their own email addresses, Teams accounts, entries in enterprise directories through Entra ID or Azure AD, and even placement on organizational charts.

Writing on his technical blog, Ferreira explained that "They can participate in meetings, send and receive emails and chats, access enterprise data, and learn from interactions to improve over time."

The agents will be available through the M365 Agent Store and discoverable within Microsoft Teams collaboration tools.

According to Microsoft's documentation, the targeted release rollout begins in mid-November 2025, positioning the announcement to potentially coincide with the company's annual Ignite conference scheduled for November 18-21 in San Francisco.

Licensing concerns and cost unpredictability

Microsoft licensing specialist Rich Gibbons has raised concerns about both the pricing structure and governance challenges these autonomous agents may present.

According to documentation he reviewed, administrators will assign a required "A365" license at the time of approval, with "No additional Microsoft 365 or Teams license is required."

However, Gibbons expressed concern about the consumption-based pricing model Microsoft has increasingly adopted.

Writing on his licensing analysis blog, Cloudy with a Chance of Licensing, he stated: "Microsoft recently launched the Copilot Credit Pre-Purchase Plan (P3), noting that its base tier offers customers the chance to buy 300,000 credits. He says the software behemoth is moving more and more to a consumption-based pricing model, which is inherently much harder to forecast for customer organisations. Here, where we're going to have AI agents doing things on their own back – how are you supposed to predict usage/consumption in those scenarios?!"

Management and security questions

Beyond pricing concerns, Gibbons also highlighted potential governance challenges with autonomous AI systems operating within corporate environments.

"As well as the licensing and cost concerns, I am also wondering how an organization manages these agents," he wrote. "If they can join meetings and send emails/messages to people, what happens if they go rogue? It could be sending sensitive data to the wrong people, providing incorrect information, or it could be sending strange or offensive messages…how is that to be prevented, monitored, and acted upon?"

Microsoft's documentation indicates that only users approved by tenant administrators will be able to create agents from templates, with all users in a tenant able to discover and view agent templates.

The company has stated that agents can be paused or deleted, with resources remaining available for 30 days post-deletion.

The agentic users will be provisioned as full-fledged user objects in organizational directories, distinguishing them from traditional chatbots or automation tools.

Read more: