Agent Derby

Agent Derby settles the "which coding agent is best" argument the dumb way: it makes AI coding agents race each other. You type one task, pick who races, and every agent gets the exact same prompt at the exact same moment in its own sandbox — a private copy of the project, so they can't see each other's work or write to your real code. While they run, you see every lane at once: what each one is reading, editing, or running, with the clock, tokens, and cost ticking. When one finishes, its result just starts up in its lane — web apps open in a frame, terminal programs get a real terminal. Then you get the numbers lined up next to each other, and you can keep any result as a new branch or folder.

The first race says a lot about why this exists: asked to "build a playable snake game in the browser," Sonnet 5.5 finished in 17 seconds, Haiku 4.5 in 31, and Opus 5.5 took two and a half minutes at ten times the cost — but Opus shipped a start screen, sound effects, and a pause button, six times the code nobody asked for. Fastest and best were two different models, and the only way to know was to play all three.

And for full disclosure — the author's own words from the dev.to launch post (2026-10-05): "I built this with Claude Code. Which means an AI coding agent helped write the tool that judges AI coding agents. Make of that what you will." It ships as an npx CLI (no API keys — it uses the agent CLIs you're already logged in to) with a built-in demo agent that runs a scripted race for free, plus a terminal version and a desktop app. MIT licensed.

Two honest notes: the author says only Claude Code has been properly tested — Codex CLI and Gemini CLI are wired up but untested with real logins; and there's no sandbox on Windows yet, so Windows is basically untested.