Google researchers have published a paper claiming that four Gemini 4 Argon AI agents defeated Minecraft's Ender Dragon without any help from humans, taking just over 27 hours in what IGN reported is believed to be a world first.
The team placed four agents into a random Minecraft seed with a single goal: defeat the Ender Dragon. According to Google, the agents used in-game chat to coordinate and were given no strategy going in. Reaching The End requires gathering Ender Pearls and Ender Rods, crafting Eyes of Ender, locating a Stronghold and completing a puzzle, which in turn requires farming materials and working up through the game's basic tools.
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 point, but said they had no cause to intervene. A fifth agent oversaw the process to check that the others were not cheating.
The agents settled into roles quickly, IGN reported: some built Enderman shelters to farm materials, others gathered food or focused on exploration. They shared combat tips and helped each other recover gear lost on death. Over the playthrough the four exchanged 229 messages, ran 2,783 programs and died 605 times.
Reaching The End took around 13.5 hours, with the remainder of the run spent on the boss fight. The agents consumed 1.35 billion input tokens, which IGN calculated would cost a normal user roughly $1,620 — about $60 an hour, or $15 an hour per agent. For comparison, IGN noted the human Minecraft speedrun record stands at six minutes and 36 seconds.
In the paper, reported via Engadget, the team wrote that to its 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, and that winning required the agents to hold a coherent plan across thousands of dependent steps for more than a day.
The researchers said they chose Minecraft to study how teams of agents coordinate in a setting where every step is visible, before such systems are trusted with real-world tasks. The skills the game demands — coordinating a team, recovering from failure and pursuing a goal for more than a day — are the same ones agents need to work reliably alongside people, they said.