A 2D ARPG
GaLo
A game I started building to see if Claude Code was any good.
Fighting monsters, saving the world.
GaLo came into being because I've always wanted to build a game; it seemed like a 2D ARPG that makes affectionate fun of the genre ("why do these snakes have gold coins?!" and other such mysteries) might be a fun and relatively easy* thing to build, and I needed a project to see if the latest AI models (Opus 4.7 was current at the time) were as good as people claimed.
The answer was "mostly, but they can still be really dumb". A small example: the first implementation of the game had a single enemy type to get it running as quickly as possible. The standard implementation of enemies in games is an interface that all enemy classes implement; the agent built a single hard-coded enemy type as part of the first run, so when we added other enemy types I couldn't hit any of them (but they could hit me); the AI assumed that there would only ever be a single type of enemy, which is an incredibly dumb assumption in an ARPG.
I still plan to finish the game; I have the basic story planned out, I bought some assets from Humble Bundle, and I plan to compose and create the music myself.
One thing that building it did teach me was the importance of a disciplined process when working with AI agents. Quite apart from stupid assumptions, at one point the agent spent several hours blaming the Godot engine for a seemingly random hang, which I eventually tracked down to an infinite loop caused by invalid parameters to a method if a specific monster was attacking the player.
So, the experience taught me to be less sceptical of agentic workflows; it also taught me a lot about their limitations.
*I was wrong about this. Even the simplest ARPGs are complex; underlying systems like inventories, stashes, NPCs, balancing legendary powers across multiple classes are complex to build on their own merits, even with a game engine doing the heavy lifting.