Session
When Copilot Studio Agents Go Wrong: Designing for Observability, Incident Response, and Recovery
Your Copilot Studio agent is live.
The flow succeeds. Dataverse is updated. The API returns 200 OK.
One problem remains: the agent did the wrong thing.
When that happens, the worst time to discover that you don't have the data needed to understand why is after the incident.
Operating AI agents in production requires a different mindset. Observability, traceability, and recoverability shouldn’t be added once things go wrong. They need to be part of how we design, build, test, and deploy our agents from the start.
In this session, we’ll follow a Copilot Studio agent from design to production and explore how to make it ready for the day when something doesn’t behave as expected. We’ll look at what information we need across the complete execution chain: the user’s conversation, agent instructions, knowledge and context, orchestration decisions, tool calls, agent flows, APIs, identities, and the underlying business systems.
Then we’ll put that design to the test.
A fictional production incident has occurred. The agent ran successfully, the tool completed, and the backend accepted the transaction, but the business outcome was wrong. We’ll use the observability we designed into the solution to reconstruct what happened, isolate the failure, contain its impact, and identify the root cause.
Finally, we’ll close the loop by turning the production failure into an evaluation and regression test so that today’s incident becomes part of tomorrow’s quality gate.
Because production readiness for AI agents isn’t just about making sure they work. It’s about making sure that when they don’t, you have the evidence to understand why.
Holger Imbery
Microsoft MVP Business Applications - Copilot Studio & MCT & PowerPlatform | Azure | Dynamics - principal architect @NTTData Inc.
Berlin, Germany
Links
Please note that Sessionize is not responsible for the accuracy or validity of the data provided by speakers. If you suspect this profile to be fake or spam, please let us know.
Jump to top