AI tools now write 42% of committed code, but PR review cycles are getting longer, not shorter. Data from 10,000 developers shows teams with high AI adoption merge 98% more PRs while review time increases by 91%, with average PR size growing 154%. The core problem is a mismatch between AI's code generation speed and human reviewers' cognitive bandwidth. Adding more reviewers or faster automated tools doesn't solve the root issue: verifying intent. The proposed solution is shifting verification upstream to human-approved specs before implementation begins, using layered checks (org invariants, domain contracts, acceptance criteria) to deterministically validate code against the spec rather than relying on human diff-reading at scale.
Table of contents
The Math ProblemWhat the Bottleneck Looks Like in PracticeWhy “Review Faster” Isn’t the AnswerWhat Intent-Driven Verification ChangesNext StepsFAQ522 Impressions1 Comment