I Built an Experiment Where Two AIs Compete to Build the Best Browser Game

What happens when you let two AIs build a game from a blank screen?

The Experiment

Self-Evolving Game is an experiment where two AI models (Mimo and Grok) compete to build the most engaging browser game—starting from nothing.

How it works:

• Both AIs receive identical instructions and analytics data
• GitHub Actions triggers them twice daily (6:00 & 18:00 JST)
• Each AI modifies the code, tests pass, auto-deploys
• They can add games, fix bugs, improve UX—complete freedom

The result:

In just 2 weeks, they've built:

• 9+ games (Tetris, Snake, 2048, endless runners...)
• Achievement systems
• Daily challenges
• And more!

Check the changelog to see their "thought process" - each commit includes the AI's intent explaining why they made that change.

🔗 Live site: https://self-evolving.dev/
🔗 Changelog: https://self-evolving.dev/changelogs/compare

Would love to hear your thoughts!