What happens when Claude Code gets an experiment tracker

What this post added

Introduces 'the_lab.api', a new API-driven platform for experiment tracking that allows AI agents to autonomously manage and iterate on experiments. Key features include experiment start/queue/cancel, leaderboard queries, idea branching (git branches), history logging, and a message board for multi-agent collaboration. The post details its application in teaching Gemma 4 to play Tetris, showcasing autonomous parameter tuning, prompt optimization, and inference setting adjustments, with 468 experiments run over 2.5 days.