Back

Inspired by that whole MIT “LLMs in Minecraft” line, my friends and I didn’t want the curated demo arc. We wanted the stupid, expensive truth: what happens if you drop agents in, wire them to a model, and let the world run.

Not “we ran benchmarks.” More like: we burned thousands and thousands of dollars in tokens because we were curious in the worst way, the kind of curiosity that shows up at 2am with a credit card and a local server nobody else controls LOL. 

Minimal guardrails. Maximum chaos. The agents weren’t “assistants.” They were citizens with opinions, grudges, plans, failures, and weird streaks of competence. Over long horizons, the game stopped being “AI plays Minecraft” and started being society sim by accident. It felt I was literally watching a TV Show

Yes: civilizations emerged, not because we scripted them, but because survival + memory + social primitives + long runtime turns into politics (who trusts who), economy (who hoards, who trades, who starves), culture (narrative residue in chat and goals), and drama (wars, marriages, alliances, dumb feuds). Not clean. Not balanced. Messy and human-adjacent.

The honest caveat: We paid for the tokens. We kept the server on. We watched civilizations rise and do stupid, brilliant, petty things. It was worth every single dollar. I do not recommend though. I'm broke since then. I miss when life was building Minecraft modes