Hot take: The AI code review tool made me a worse reviewer

Got feedback from a senior dev last Tuesday after I flagged a false positive that the AI caught. He said I was leaning on the tool instead of reading the pull request context, which was 100% right. I started using an AI assistant to scan for style issues and security gaps, but it made me skim past the logic flow and edge cases, the stuff that matters. I stopped using auto-summaries for two weeks and forced myself to trace the data paths manually, and I caught a race condition the AI missed. Anyone else found that AI review tools push you toward lazy habits if you don't set hard rules for yourself?

2 comments

2 Comments

craig.viola2d agoMost Upvoted

Damn, that checklist rule actually makes total sense, totally changing my view on this.

jessica9212d ago

My team ran the same experiment in March, 42 devs total, and the data showed the opposite. People who used the AI tool with strict checklists, like requiring a logic trace before flagging style stuff, caught 18% more edge cases than the manual-only group. The problem isn't the tool, it's that nobody told you how to use it right. You stopped using auto summaries, but you could've kept them and just added your own pass on the data flow. The senior dev calling you out was fair, but that's a workflow issue, not a tool issue. I keep mine on, but I set a rule that I can't write a comment until I've mapped out the input, state, and output for every changed function.