A team of four AI agents coordinated through in-game chat to plan, code, and debug their way to victory in Minecraft without human intervention or pre-written game logic.
Researchers placed four Gemini 4 Argon agents in a randomly generated Minecraft survival world with one objective: defeat the Ender Dragon. Each agent controlled a player via Mineflayer, an open-source library, and communicated with teammates through in-game chat and shared memory. No human assigned roles or wrote game logic.
The agents completed the task in 27 hours and 14 minutes, writing 2,783 programs and dying 605 times. According to the researchers, this is the first end-to-end autonomous completion of Minecraft by AI agents using only standard Mineflayer without pre-scouted seeds or human-written game-specific tools.
How coordination worked
The agents divided labor, shared what they learned, and recovered each other’s items when one died. Researchers watched the run on a live dashboard and could pause or stop it, but they had no cause to intervene. The agents maintained a coherent plan across thousands of dependent steps over more than a day.
Why this matters for software engineering
The task mirrors challenges in real-world software engineering and enterprise work: long-horizon coding, multi-step dependencies, and coordination across multiple agents. The agents had to plan, write code, debug failures, and adapt without human guidance.
Constraints
The run used Easy difficulty and a sandboxed environment with standard tools, not custom integrations or domain-specific optimization.
Go deeper
Get the latest AI updates in the Tech AI Magazine app
iPhone · Android
