This full-day, hands-on workshop is for people who have already built agents with Copilot Studio and are comfortable with its core features. If you want to move beyond the basics into more reliable and maintainable, production-ready approaches, this workshop is for you. It is not intended for first-time users of Copilot Studio.

The workshop is built around in-depth technical labs that focus on how agents should be designed, evaluated and extended. You’ll work with advanced prompting and instruction approaches, including agent- and topic-level guidance, and grounding strategies to reduce hallucinations and unpredictable behaviour.

Participants will explore agent design techniques such as componentised agents and agent-to-agent collaboration. MCP servers are introduced as a key integration approach for agent-based solutions. You’ll learn how to use out-of-the-box MCP servers, including Microsoft-provided MCP servers such as Dataverse, and where they fit architecturally when connecting agents with data and tools.

Testing and evaluation are covered as a core theme, with practical techniques for validating agent behaviour, tracking changes over time and identifying regressions. Prompts are treated as versioned and testable assets rather than static text.

By the end of the workshop, attendees will have a clear, practical framework for designing, testing and evolving Copilot Studio agents they can trust in real-world scenarios.


Key takeaways:
- Design Copilot Studio agents using approaches that improve reliability and maintainability
- Use advanced prompting and grounding to reduce hallucinations and unpredictable behaviour
- Test and evaluate agents with repeatable techniques that support change over time
- Understand and apply MCP servers as an integration approach