When something breaks, our first instinct is to focus on the thing that failed. But in complex systems, the failure is usually the result of many things going right, changing, and interacting in ways we didn't anticipate. Learning From Incidents has taught us to look beyond the failure itself; to understand the decisions, adaptations, and conditions that made it possible. But what happens when the system keeps changing while we're trying to understand it?

AI is making this problem harder. Models are updated, agents get new tools and permissions, prompts change, and system behavior has shifted. By the time we've investigated an incident and acted on what we learned, the system may already be different.

This talk looks at what we can carry forward from years of Learning From Incidents practice, where those approaches start to strain, and what it means to build a learning process that can keep up with systems that don't sit still.