I have built a harness in which Codex (which I consider a very detail-oriented AI) serves as the code reviewer for Claude's code. In fact, it proved pretty good at catching error scenarios Claude never considered. For reference, I use Claude Sonnet 5.1 Medium and GPT-6-Fast Luna - Medium) In complex situations, my harness switches to Fable 5.1 and Astra. I tested the reverse situation: Codex coding and Claude reviewing; the review cycles were less frequent. I am not jumping to any conclusion, because my tests (building a mundane weather app) might be biased.So my question: who should review what? Comments URL: https://news.ycombinator.com/item?id=50243565 Points: 1 # Comments: 0