How to roll out AI agents in a Microsoft school, without losing control of them
Microsoft Copilot Studio lets a school build its own AI agents, trained on its own documents, policies and knowledge, and bring them straight into Teams. The tool itself isn't the hard part. Getting the rollout right is.
Start with a shared model, not a solo build
The agents that work well aren't built by one person in isolation. The strongest approach is a small working group involving the appropriate SLT, sitting down together to agree the prompt, the desired outputs, and which knowledge files the agent should be grounded in, whether that's a behaviour policy, a curriculum document, or a set of safeguarding procedures. SLT involvement here isn't a formality. It's what gives the agent oversight and keeps it accountable to what the school actually wants it to say.
Test small before you go wide
Once that group is happy with how the agent behaves, it can be shared with a small trial group of named staff first, not the whole school. That's the point to properly stress test it: does it answer the way you'd expect, does it stay within the knowledge you've given it, are there gaps. Only once that's been through proper oversight and testing should it move to a broader rollout, and even then, it can be targeted to a specific department or team rather than switched on for everyone at once.
The licensing point that catches people out
This does require more than standard Microsoft Copilot licensing. To build and publish your own agents, a school needs Copilot Studio added on top of that, not just Copilot itself. It is worth checking with your IT lead or Microsoft partner before you plan a rollout around it, so it doesn't stall at the budget conversation.
If you've got questions about any of this, licensing, the shared build process, or how to structure the trial, that's very much what Taught AI is there for. Sign up for a free 1:1 call below.
Ben
Comments