New PM wants AI-generated root cause analysis. Am I overreacting to the quality?
Just started building out an agentic workflow for incident response and our new PM is fully bought in on AI-generated RCA reports. Says it'll cut toil and catch patterns we miss manually. Sounds great in theory.
Then I see the POC output and it's flagging random correlations, like "high number of firefox texture limit events may indicate frontend rendering issues" showing up in a backend latency incident.
I pushed back saying we need proper data correlation, not just anomaly pattern-matching, but he wants everyone committing AI outputs to the runbooks directly. I'm the platform lead and this feels like it's going to create more review work, not less.
Has anyone dealt with AI RCA tools that actually reduce MTTR without creating a mountain of garbage to sift through first? Or is manual still king for complex incidents? Starting to lose faith in this whole direction.