We brought AI into our incident response workflows to move faster—quicker debugging, smarter fixes, less operational stress.

At first, it worked.

AI suggested fixes, generated runbooks, and helped us respond faster than before. It felt like we had upgraded our entire DevOps workflow overnight.

Until it didn’t.

We started seeing fixes that looked correct but failed in production, recommendations that didn’t fully make sense, and engineers quietly losing trust in the system. Incidents weren’t just technical problems anymore; they became judgment calls between human intuition and AI suggestions.

In this 5-minute Ignite talk, I’ll share what actually happened when we let AI into incident workflows, the wins, the failures, and the unexpected lesson: AI doesn’t reduce complexity, it shifts it.