Another quirk of statistical inference: This A.I. code review commented that the dev changed a TypeScript param from optional to required. Actually, the reverse happened. But statistically, if you see those token near each other, the closest sequences are going to be something about the consequences of making something that was optional into something that is now required.
More news about A1 code review project: Some companies have added an agent that reviews the security review comments from a different agent. That second agent does mostly what I do, which is to say the thing the first agent said is a false positive. However, it also doesn't understand overall context, so talks about local dev tools that can't be hit from the internet possibly getting attacked.
The total cost of this has to be high in terms of compute dollars. I would guess it costs more than it would cost to pay person to look through the code and not report the false positives in the first place. I have no doubt some VCs are eating that cost right now, though.