Testing AI agents and clarifying the rules they work within

How can an idea for an AI agent become a test that others can follow and understand? At Patrimonium, Theia combined a first day of Copilot Studio training with governance work on the Power Platform. The collaboration brought together testing by internal AI Champions and questions from the supporting IT team.

Patrimonium wanted to give employees a practical introduction to Copilot Studio. Its own use cases and suitable exercise materials were the starting point. Preparation therefore also covered sample documents from SharePoint, the access required and suitable training environments.

The technical possibilities could only be assessed in the context of the existing environment. Data policies define which connections may be used together in an automation. Their scope can therefore affect both new exercises and existing workflows.

Theia delivered the first Copilot Studio training day in Crissier on 23 September 2026. Preparation included agreed use cases and adapted exercises. Discussion of the examples was planned to cover the data required, implementation steps, effort and technical dependencies. Afterwards, participants received the training materials for their further work.

The governance follow-up addressed a specific issue: when reviewing the exercises, it became clear that a newly configured data policy also applied outside the training environments. Theia used the Power Platform inventory to review the affected flows and limited the policy's scope to the AI Champions' environments. The people responsible received guidance on which existing automations needed checking and, where necessary, reactivation.

The participants' own agent tests also raised further questions. One example was analysing a generated investment document as a PDF. These experiments showed that file processing, agent configuration and policies need to be considered alongside the idea itself. They remained part of the testing and subsequent technical clarification.

Feedback on the first training day was positive: participants wanted to apply what they had learnt themselves. The next step was therefore their own work on specific examples. A further training day was planned for October, while additional development assignments were discussed separately.