> More broadly, though, if you've gotten to the point where you're relying on AI code review to catch bugs, you've lost the plot.

> The point of a PR is to share knowledge and to catch structural gaps.

Well, it was to share knowledge and to catch structural gaps.

Now you have an idea, for better or for worse, that software needs to be developed AI-first. That's great for the creation of new code but as we all know, it's almost guaranteed that you'll get some bad output from the AI that you used to generate the code, and since it can generate code very fast, you have a lot of it to go through, especially if you're working on a monorepo that wasn't architected particularly well when it was written years ago.

PRs seem like an almost natural place to do this. The alternative is the industry finding a more appropriate place to do this sort of thing in the SDLC, which is gonna take time, seeing as how agentic loop software development is so new.