Session Abstract

Autonomous agents are powerful — sometimes "a little too" powerful. When you give an agent the ability to make decisions, call connectors, and take multiple steps toward a goal, you quickly discover a new world of weird edge cases, misunderstandings, and surprisingly creative AI problem‑solving. In this fun but practical session, I will shares reallife lessons from building semi‑ and fully autonomous agents in Copilot Studio. Through anonymized examples and field-tested patterns, you’ll learn where autonomous agents can go off‑track, from over‑planning, misunderstood goals, unexpected connector calls, odd reasoning loops, or hallucinated steps, and how to prevent, mitigate, or monitor these behaviors. We’ll also explore the guardrails you need in place before letting an agent loose inside Teams or the wider Microsoft 365 environment. This is not just theory: it’s the session where you hear what actually happens in production and what you can do to be prepared.

Kai Stenberg

I have been a Microsoft MVP since 2022 in M365 and Teams. One of my key mantra is as all MVP's "Sharing is Caring". This I predict both at work and in the community. For some years I have also helped organize MVP-Dagen and Collabday Oslo together with several MVP's. For 9 seasons I have run the Podcast TeamnsCastAway, a Norwegian podcast that started to just speak about Teams and Teams addoons. During the years, this is now also speaks about M365 Copilot and Copilot Studio. Without working as Copilot SME in Capgemini I have long experience as a consultant for more than 25 years. To be on stage share my knowledge and experience is one of the things I really like.