I wanted to see how close a small DIY review team (a bug-hunter, a keeper, a sweeper) could get to full-fledged review products. On a public benchmark of 50 real PRs it beat Cursor Bugbot and CodeRabbit, and the trick was not the personas
Summary
A small DIY review team using personas (bug-hunter, keeper, sweeper) achieved higher performance on a 50-PR benchmark than Cursor Bugbot and CodeRabbit, demonstrating the effectiveness of role-based review strategies.
Similar Articles
@RayFernando1337: The bugs that cause churn almost never show up in a diff, and you only really catch them when you stop reviewing code a…
A developer shares a workflow using Cursor's Opus 4.8 Max Thinking model with subagent harness, and introduces a GitHub repository with installable skill files for AI coding agents, including a 'running-bug-review-board' skill that performs live QA testing.
Built a 6x cheaper CodeRabbit alternative using open source models
Built a cheaper alternative to CodeRabbit using open source models, achieving similar or better PR review accuracy at 6x lower cost, with features like auto-fix and prompt-based bug fixing.
Built an AI PR review workflow that generates actual fix PRs instead of just comments
The author built an AI-powered PR review workflow that generates actual fix PRs instead of just comments, claiming to be 6x cheaper than CodeRabbit and more accurate on large PRs.
Faster Code Review with Cursor's Bugbot (3 minute read)
Cursor's Bugbot code review tool is now over 3x faster, 22% cheaper, and finds 10% more bugs, with most runs finishing under three minutes. The update also adds new features like running reviews before pushing and only reviewing new changes.
I let Codex review the PR Claude wrote
An individual shares an experience where Codex reviewed a pull request written by Claude, finding a real bug that another tool also flagged, raising questions about the efficacy of having multiple AI agents check each other's work.