Nous Research is showing how Hermes Agent can run as an autonomous coworker in Microsoft Foundry, with its own Microsoft 365 account rather than borrowing an employee’s login. The setup could let the agent send email, work with shared files and run scheduled tasks, while giving administrators a named account to manage and audit.
Nous/Hermes Watch analysis
What happened
Nous’s cookbook walks organisations through deploying Hermes in their own Microsoft 365 tenant. An agent instance can be created from Teams, with its own account, mailbox and OneDrive; scheduled routines can run without anyone signed in. The agent’s access depends on what people share with it, and its actions are recorded under its own name.
The setup has two deployment routes: a guided installer or a more manual process. In either case, an administrator must approve the agent blueprint and choose who can create instances. Nous says the feature is in Microsoft’s Frontier preview programme, and describes the repository as a proof of concept. Its guide recommends using a test or pilot tenant, not production.
Why it matters
Agents acting through a worker’s account can blur who did what and risk giving the software access to everything that worker can reach. A separate identity offers a more legible alternative: grant access by sharing specific resources, and keep the agent’s activity distinct in the audit trail.
That is a useful change in the plumbing, not proof that an autonomous agent is ready to be treated like a dependable colleague. Organisations will still need to decide what it may do, who supervises it and how mistakes are handled. There is also a practical gate: the guide lists Frontier access, an eligible Microsoft 365 licence and an available preview agent seat among the prerequisites.
Our read
The strongest idea here is not that an agent gets a charming new job title. It is that it gets its own account, permissions and audit identity. That makes the agent easier to contain and its actions easier to attribute than software quietly acting as an employee.
For now, Nous’s own instructions put this in pilot territory. IT teams curious about the model can learn how the deployment works, but should take the guide’s advice and keep experiments out of production.
What to watch
- Whether Frontier preview access expands and the capability becomes generally available.
- What permissions administrators must approve, and whether they can keep them narrowly scoped.
- How reliably scheduled tasks and account-level controls work in real deployments.
Discussion spark: Would you trust an AI agent with its own work account and narrowly shared files more than one acting through an employee’s login, or does autonomy remain the bigger concern?
Sources and evidence
- Source update (20 November 2026, 14:00 UTC)
Nous/Hermes Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Nous Research.