I Trained AI To “DOMINATE” Brackey’s Game

I had an absolute blast wiring up reinforcement learning in Godot to tackle a notoriously tough Brackeys tutorial game—chatting smack to Claude the whole time. It all started when I saw an RL demo for mining optimization and thought, “Why not apply this to a custom-built game?” Fast-forward to 14 polished agent versions, late-night coding sessions, and a solid foundation for more RL experiments in Godot.

The full code is up on GitHub (including those hefty 167 MB dmg files if you want the raw environments—just say the word and I’ll add them to Git LFS). Huge shout-outs to Ed Beeching (GoDotRL) and the legend @Brackeys, and hey, if you’re in Florida tomorrow I’d love to connect before my big presentation!

Watch on YouTube