Four Google Gemini 4 Argon AI agents have successfully defeated Minecraft's Ender Dragon without any human assistance, a feat reported by Google researchers. The AI team took just over 27 hours to achieve this world first.
To reach The End and defeat the Ender Dragon, players must first gather essential items like Ender Pearls and Ender Rods to craft Eyes of Ender. These are then used to locate a Stronghold, where a puzzle must be solved to open the portal to the final boss. This process typically involves extensive farming and progression through Minecraft's various tools.
The experiment involved placing four AI agents into a randomly generated Minecraft seed with the sole objective of defeating the Ender Dragon. According to Google, the agents coordinated their actions through in-game chat and shared memory, without any pre-assigned roles or strategies.
"Each agent controlled a player via an open-source library called Mineflayer and coordinated with its teammates through in-game chat and shared memory," the researchers stated. "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."
The agents quickly established distinct roles within the team. Some focused on building Enderman shelters for material farming, others gathered food, and the rest dedicated themselves to exploration. They shared combat strategies and aided teammates who lost gear upon death.
During the 27-hour playthrough, the four agents exchanged 229 messages and executed 2,783 programs, incurring 605 deaths. The journey to The End took approximately 13.5 hours, with the remaining time dedicated to the final boss battle. In comparison, the current Minecraft speedrun record stands at six minutes and 36 seconds.
The AI agents consumed 1.35 billion input tokens, which researchers estimate would cost a typical user around $1,620, or $60 per hour. The researchers noted this achievement as 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." The task required the agents to maintain a coherent plan across thousands of dependent steps over more than a day.
The primary goal of this research, according to the scientists, was to study how teams of agents coordinate in a visible environment. This understanding is intended to help in developing AI systems that can be reliably trusted for real-world applications. The skills demonstrated, such as team coordination, failure recovery, and long-term goal pursuit, are considered essential for AI agents working alongside humans.