Dealt with a code review today where the reviewer admitted to using AI to review the PR because it was quite large. This is technically within policy, because the reviewer in question did not include any of the AI's output; it just gave them a place to start. They discovered a regression, which they then personally verified before submitting it.
The regression was *sort of* real. But it had the same … flavor that certain AI slop security reports have. It was *almost* valid. But not quite.
I feel like this reviewer was earnestly just following what is now best practice in the industry, "saving time" with a tool they probably have to use all the time anyway. But I just … don't believe that it saves time. I think they would have understood the issue better if they'd just looked at the code in sequence, rather than getting a defect report from the bot and then laboriously attempting to reverse-engineer what was actually going on