Everyone's focused on AI writing code. But the shift to AI reviewing code moved faster - 1 in 7 active PRs now have an AI reviewer, up from 1 in 100 eighteen months ago.
A few things that surprised us:
- Copilot overtook CodeRabbit in November despite CodeRabbit being purpose-built for review - native GitHub integration matters more than we expected
- Gemini went from near-zero to #3 in 10 months (43X growth). Google isn't marketing this aggressively yet
- AI agents authored 99K PRs but participated in 5.9M -automation is still heavily weighted toward review, not generation
- Some early entrants like Korbit have already shut down
Data is GitHub Archive + our anonymized corpus. Happy to answer questions on classification or edge cases.
amnaanwar•28m ago
I met the Korbit founder and team at AI Dev World last year - smart, thoughtful group. The founder was deeply enthusiastic about their approach to PR analysis. They had a very academic, research-driven team behind the product, which showed in how rigorously they were approaching the problem.
It was exciting work, and a good reminder of how much serious experimentation has been happening in AI code review over the past couple of years, even among teams that didn’t ultimately make it.
zak-mandhro•1h ago
Everyone's focused on AI writing code. But the shift to AI reviewing code moved faster - 1 in 7 active PRs now have an AI reviewer, up from 1 in 100 eighteen months ago.
A few things that surprised us:
- Copilot overtook CodeRabbit in November despite CodeRabbit being purpose-built for review - native GitHub integration matters more than we expected
- Gemini went from near-zero to #3 in 10 months (43X growth). Google isn't marketing this aggressively yet
- AI agents authored 99K PRs but participated in 5.9M -automation is still heavily weighted toward review, not generation
- Some early entrants like Korbit have already shut down
Data is GitHub Archive + our anonymized corpus. Happy to answer questions on classification or edge cases.
amnaanwar•28m ago