Run a swarm of agents in Minecraft

A team tested 50 AI agents in Minecraft using a custom harness, then watched them self-organize around survival and food.

Done with Custom agentยท Advanced

What they did
They built an open-source setup with Pi handling the agent loop and Mineflayer managing in-game actions. In Minecraft, they dropped 50 GPT-6.1 Sol agents into a survival world for 3 hours with a single objective: create a working civilization. The agents had a directory for roles, a shared sandbox, and private messages, which helped them form a leader, track rations, and split up fishing duty.
How it went
The agents exchanged about 5,900 messages and recorded 34 deaths, with 22 happening on the first night. After that, they became more careful and focused on food, and a few deliveries missed their target. The Mindustry run was not reviewed.
Tips
Feeding a large swarm became the main bottleneck, so check food availability before scaling up. Even when deaths only cost gear, early losses can make the group overly cautious.

Try it yourself with Claude Code

Set up a swarm of [number] AI agents in [Minecraft or Mindustry] using a working game harness. Give each agent a clear role such as [gatherer], [builder], or [defender], and have them coordinate toward [goal]. Use a shared channel for group planning, private messages for one-on-one coordination, and a simple log of deaths, resources, and deliveries. Before any irreversible action like spending in-game currency, deleting files, or submitting changes, pause and ask me to approve it. Run the swarm for [time period], then show me what each agent did, what problems came up, and the exact steps to run the setup again.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on labs.ycombinator.com โ†—

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.