We analyzed ~40 million public GitHub pull requests from 2022–2025 to understand how AI agents show up in real-world code review.
In 2025, AI agents are involved in roughly 14% of pull requests, either as reviewers, commenters, or authors.
A few patterns stood out. CodeRabbit accounts for the largest share of AI review volume, touching the most PRs overall, while GitHub Copilot shows the broadest adoption across organizations. Most AI participation still happens during review rather than full PR authorship, suggesting a shift in how review attention is allocated rather than end-to-end automation.
We focused on observable PR activity rather than vendor-reported metrics, excluded non-LLM automation, and were conservative about bot classification. There are limitations, especially around private repositories and self-hosted tools, which we call out explicitly.
zak-mandhro•1mo ago
We analyzed ~40 million public GitHub pull requests from 2022–2025 to understand how AI agents show up in real-world code review.
In 2025, AI agents are involved in roughly 14% of pull requests, either as reviewers, commenters, or authors.
A few patterns stood out. CodeRabbit accounts for the largest share of AI review volume, touching the most PRs overall, while GitHub Copilot shows the broadest adoption across organizations. Most AI participation still happens during review rather than full PR authorship, suggesting a shift in how review attention is allocated rather than end-to-end automation.
We focused on observable PR activity rather than vendor-reported metrics, excluded non-LLM automation, and were conservative about bot classification. There are limitations, especially around private repositories and self-hosted tools, which we call out explicitly.