hey have u guys seen what LinkedIn just built for code review
they made a multi agent AI system that reviews pull requests at scale. like multiple agents working together, each one focused on different things. one checks bugs, one checks security, one checks performance
already running on 10,000 repos and doing 79,000 reviews a week. and the crazy part is 63.9% of the AI comments actually got accepted by developers. thats not just noise, thats real signal
i have been using CodeRabbit on some of my repos and its decent but single agent. the multi agent approach sounds like a big upgrade because each agent has a narrower focus and less room to hallucinate
my question is, at what point does AI review replace human review entirely? or do we always need that second pair of eyes?
anyone here tried multi agent code review on their repos yet?
Top comments (0)