When Claude, Codex, and Gemini Disagree on the Same Code

Four real cases from our AI code review logs where three frontier models read the same diff and arrived at different verdicts — and what the disagreements reveal about the limits of single-model review.