Google DeepMind announced that four autonomous artificial intelligence agents successfully completed a challenging Minecraft game from scratch in just 27 hours. Unlike the company's earlier gaming achievements with AlphaGo and AlphaStar, which required systems specifically trained for individual games, this milestone was accomplished using general-purpose coding assistants - the same type of software tools that programmers use daily.
How the agents worked together
The AI team operated inside a sandboxed survival world with human oversight limited strictly to safety monitoring. The four agents wrote code in real time, divided labor through text chat, and debugged thousands of errors without any pre-programmed game logic or human intervention. Over the course of 27 hours, they created 2,783 custom programs and ultimately defeated the Ender Dragon, the game's final boss.
DeepMind stated: "Four Gemini 4 Argon agents beat Minecraft from scratch in 27 hours. Operating in a sandbox using only code and text, they wrote 2,783 programs, divided up jobs over chat, shared notes, and debugged errors together - all with no human help."
Completing Minecraft in a randomly generated survival world is considered one of the most difficult challenges in computer science because it demands maintaining a coherent long-term plan across thousands of interdependent tasks. Players must harvest and process raw materials, explore dangerous alternate dimensions to obtain rare resources, use spatial mathematics to locate hidden underground fortresses, and defeat a flying boss that regenerates health continuously. Any mistake can result in death, which strips the player of inventory and forces hours of recovery. Previous community attempts at this challenge relied on shortcuts such as playing on peaceful mode, using pre-scouted maps, starting with active portals, or hard-coded strategies.
This achievement demonstrates that general-purpose AI coding agents can coordinate autonomously to solve complex multi-step problems that stretch over many hours - a capability that has potential applications far beyond gaming.