AI code review (Claude, maybe Codex)

[Jonas Lochmann]

Am Mon, Apr 13, 2026 at 02:37:19PM +0000, schrieb Valent at MeshPoint:
> Any LLM is as good as person using it, yes they hallucinate, but if you know
> how to use  LLMs properly and have domain knowledge in which you are using
> it they it can't produce "bad code". Also if you give it your perferences
> and coding style it can follow it quite nicely.

I do not see yet the process to use this - how will it improve the
contributor experience? If I want an AI review, I know where I can get
it. It could be helpful if the reviewers and the submitter see the same
AI review. Still, there needs to be a reviewer that does something.
Or the AI system gets the permission to merge. However, for now, I do
not consider this a realistic option.