This full-day workshop is aimed at technical makers, developers and solution architects who already use Copilot Studio and want to build agents that work reliably in real-world scenarios. We’ll start with agent design, focusing on prompt guidance, context management and grounding techniques that help reduce hallucinations and improve consistency from day one. Attendees will leave with clear design patterns they can apply immediately in their own projects.

From there, we’ll move into the challenges that appear once agents move beyond demos and into production. Based on real-world projects, we’ll cover authentication and authorisation, cost control and usage limits, and how to design effective human-in-the-loop flows. You’ll also learn how to use Dataverse with an MCP server and we will show how to bring services like Microsoft Foundry and AI Search into Copilot Studio in a controlled and practical way. The day follows the natural lifecycle of an agent, iterating through design, to testing, to operation and evaluation.

The workshop will be delivered by Microsoft MVPs Andrew Bibby and Cathrine Bruvold and is structured into clear modules, each supported by hands-on labs and guided exercises. Labs will cover refining prompts at agent and topic level, securing agent tools, monitoring behaviour in Copilot Studio and the Agent 365 workspace, and refining agent functionality over time. By the end of the day, you’ll have practical experience and a clear framework for designing, running and improving production-ready agents.

Key takeaways from this workshop:

• Design agents with clear boundaries and guidance to improve agent accuracy and ensure consistent results
• Apply governance, security and human-in-the-loop patterns to run agents safely in production
• Use a lifecycle approach to test, monitor and continuously improve agent behaviour over time