Yeah. We have an AI reviewer. Just now I had a PR where I didn't normalize paths in some configuration and then compared them. I.e. let's say the configuration had

    file = /foo/bar/

    if file == other_file:
        ...

    if normalized(file) == normalized(other_file):
        ...

I've encountered a lot of attempted "fixes" of this kind. So, in my experience, it's good to have AI look at the PR because it's just another free pair of eyes, but it's not very good at fixing the problems when it does find them. Also, it tends to miss conceptual problems, concentrating on borderline irrelevant issues (eg. an error raised by the code in the use-case beyond the scope of the program like when a CI script doesn't address the case when Git is not installed, which is of no real merit).