
Claude Duet works in the background, connects to apps like Gmail and Slack, and requires approval for sensitive actions.
A concise outline of The Decoder’s report on four Gemini 4 Argon AI agents completing Minecraft autonomously, including coordination, coding activity, deaths, runtime, and cost.

Four AI agents running on Google’s Gemini 4 Argon beat Minecraft without human input, according to The Decoder. The run ended with the agents slaying the Ender Dragon, the game’s central completion milestone. The report frames this as the first fully autonomous Minecraft completion of its kind.

The Google Deepmind team placed the agents into a randomly generated world without preset roles, strategies, or game-specific tools. The agents split up tasks on their own, warned one another about threats, recovered items from dead teammates, and logged mistakes in shared memory. The key takeaway is not just that the agents finished the game, but that coordination developed during the run.
The session lasted 27 hours and 14 minutes. Across the run, the four agents wrote 2,783 JavaScript programs, exchanged 229 chat messages, and died 605 times. The Decoder reports the session cost $1,620 at the standard Gemini 4 Argon price that takes effect in 2027.
Minecraft is a useful stress test because progress requires planning, adaptation, recovery from setbacks, and multi-step execution. The report places this achievement in a broader trend of agentic AI models beating video games that previously required purpose-built AI systems trained from scratch. For readers tracking AI capabilities, the practical signal is improving autonomy under open-ended conditions.

Claude Duet works in the background, connects to apps like Gmail and Slack, and requires approval for sensitive actions.

Taste-Bench finds top AI models often fail to pick the better path in long-running tasks.

AI agents are taking on more work as junior technical hiring falls sharply at Anthropic.

Six AI agents failed to win a refund while exposing risks around privacy, honesty, and message overload.