A team of Gemini 4 Argon agents beats Minecraft
Maxence Faldor, Satinder Singh and Roberta Raileanu (Google DeepMind) report the first end-to-end autonomous completion of Minecraft: four Gemini 4 Argon agents, coordinating through chat and shared memory, wrote their own control programs and defeated the Ender Dragon in 27 hours and 14 minutes.
Ask this paper
Setup. Each agent controls one player through the stock Mineflayer library. No roles were assigned, no game logic was written by humans, and no seed was pre-scouted. The agents planned, wrote and debugged every program.
Scale of the run. Starting on a random Easy-difficulty world, the team wrote 2,783 programs and died 605 times before beating the dragon. Researchers watched a live dashboard and never intervened.
Coordination behavior. The team divided work, shared what each agent learned, and recovered each other's items after deaths, all through in-game chat and a shared memory store.
Long-horizon claim. Winning required keeping one coherent plan across thousands of dependent steps for more than a day. The authors argue this is the same capability long software-engineering and enterprise tasks need.
Limits. The paper is a single-run report on Easy difficulty and documents the limits of the result alongside the strategies.
Abstract
We placed four Gemini 4 Argon agents in a sandboxed Minecraft survival world and gave them one goal: defeat the Ender Dragon. Each agent controlled a player via an open-source library called Mineflayer and coordinated with its teammates through in-game chat and shared memory. No one assigned roles or wrote game logic. The agents planned, wrote, and debugged every program themselves. Researchers could watch the run on a live dashboard and pause or stop it at any time, but they had no cause to intervene. Starting from a randomly generated world on Easy difficulty, the team wrote 2,783 programs and defeated the dragon in 27 hours and 14 minutes, dying 605 times along the way. To our knowledge, this is the first end-to-end autonomous completion of Minecraft by AI agents, achieved using only the standard Mineflayer library without pre-scouted seeds or other game-specific tools written by humans. Winning required the agents to keep a coherent plan across thousands of dependent steps for more than a day. Complex real-world software engineering and enterprise tasks demand the same long-horizon coding and multi-agent coordination. We describe the setup, how the run unfolded, the team's strategies (dividing the work, sharing what they learned, recovering each other's items), and the limits of the result.