How we review 4x more AI code without reading every line

To handle a fourfold increase in AI-generated pull requests without sacrificing code quality, engineering team Aviator stopped reviewing diffs and instead shifted focus to verifying intent, acceptance criteria, invariants, decisions, and evidence. By mining past review history, codifying repeated comments into automated invariants checked by tools like ast-grep, and having agents generate proof for each scenario, the team reduced median time to merge and dropped reverts by 40%. Open article →