Multi-AI Code Review at the Merge Gate: 28 Codex Runs, 14 PRs, What the Data Says
Single-provider AI review has known bias. Claude rarely flags Claude's bad code. Here is what happened when I ran Codex (OpenAI) and Gemini (Google Vertex) as peer reviewers across 28 runs and 14 PRs, and what the codex-finds-bugs data actually shows.