The Experiment

How fast can you go from zero to a playable multiplayer game using only AI code generation? That was the question behind Bloburb — an Agar.io-style browser game built entirely through AI prompting. Specifically, I wanted to put my local OpenClaw multi-agent setup to the test and see what could be produced with a solid spec and minimal intervention from me.

Motivation

I recently hopped on the OpenClaw hype train, repurposing a spare gaming PC with a 3090 with the intention of running models both locally and through APIs. After some failed setup attempts (OpenClaw is great at breaking itself if you try to rush things), I finally got an environment going that didn't require constant babysitting.

To put it to the test, I needed a somewhat complex project idea to see how far it could take things on its own. Since my kids have become recently obsessed with the YouTube Playables game "Hole.io" (very similar to the popular Agar.io game), that was the first idea that came to mind. I wanted to put together something very quickly that I could actually play with them, and that they could bring their own ideas to.

The Result

A fully functional multiplayer blob game with bot players and a leaderboard, using real-time WebSocket connections. The majority of the game was built with a few detailed prompts to get OpenClaw going over while I slept. Then dialing in some of the specific mechanics and other polish happened through iterative prompting via Claude and Codex.

The visuals are composed of flat shapes, but my plan is to take these same mechanics and bring them to a more fleshed out 3D game using Three.js when I get some more time (and API tokens, RIP).

Check it out at blob-game.fly.dev.