Moreover, this is a breach of trust between the reviewer and the author. Our assumption when we review is that we’re talking to a real person who wants to do their best work.
I think this is a really excellent point and gets at a general feeling I’ve had towards LLM-generated PR comments for a while now. I can understand the desire to use LLMs to generate changes (my company pushes AI super hard and it genuinely is a helpful tool in many cases), but to delegate discussion of those changes to an AI model feels extremely antisocial to me. I don’t really like the idea of vibe-coding in general, and this practice really distills down everything I don’t like about it with an added layer of deception.
- Making code easier to write exacerbates our existing issues with review bandwidth.
Man, so much this. LLM written code is, most of the time, both excessively verbose, and overly complicated.
Also, I’ve found myself asking more about “why” than “how” in my reviews. Why is this loop here, why is this one liner method used only once, why do these five tests do the same thing. I’ve reached a point where I feel like I’m gaslighting myself into thinking that I’m constantly missing something, because nothing makes sense from an engineering perspective anymore.
direct link to the policy
