2,783 JS Programs Wrote Way to Ender Dragon: 4 AIs Spend $1,620 in 27h
Google DeepMind agents autonomously defeated Minecraft’s final boss, writing thousands of programs and dying hundreds of times without human input.
Key points
- Four Gemini 4 Argon agents defeated the Ender Dragon in a fully autonomous run.
- The session lasted 27 hours and generated 2,783 JavaScript programs.
- The total cost was $1,620 based on projected 2027 pricing.
Four artificial intelligence agents running on Google’s Gemini 4 Argon model defeated the Ender Dragon in Minecraft without any human assistance, marking the first fully autonomous victory in the game. According to The Decoder, the Google DeepMind team placed the agents into a randomly generated world without assigning specific roles, strategies, or game-specific tools. The agents coordinated their own tasks, warned each other about threats, and recovered items from fallen teammates while logging mistakes in a shared memory system.
The risk
The agents operated without predefined roles or strategies, splitting up tasks on their own. According to The Decoder, they warned each other about threats and recovered items from dead teammates. The session involved significant computational activity, with the agents writing 2,783 JavaScript programs and exchanging 229 chat messages. The agents died 605 times during the 27-hour and 14-minute run. This demonstrates the ability of agentic AI models to coordinate complex actions without human input.
Who is exposed
This event involves AI agents running on Google’s Gemini 4 Argon model. The Google DeepMind team conducted the demonstration. The agents were dropped into a randomly generated world. They coordinated on their own, logging mistakes in a shared memory. This is an example of agentic AI models beating video games that previously required purpose-built AI systems trained from scratch. The exposure is limited to the specific demonstration conducted by the research team.
Cost and compliance
The entire session cost $1,620 at the standard Gemini 4 Argon price that takes effect in 2027, according to The Decoder. The agents wrote a total of 2,783 JavaScript programs. The session lasted 27 hours and 14 minutes. The agents exchanged 229 chat messages and died 605 times. The cost reflects the usage of the AI model during the autonomous run. No other financial or compliance details were provided in the source material.
What should you do?
Managers should monitor autonomous AI experiments for resource usage. Teams should define clear objectives for agent deployments. Ensure that AI systems have appropriate safeguards during testing.
The background in plain language
Agentic AI refers to models that can take independent actions to achieve goals. In this context, the agents functioned as a team, dividing tasks and communicating without human direction. This approach allows AI to handle complex, multi-step problems. Teams usually implement safeguards to manage agent behavior. Understanding how agents coordinate is essential for evaluating their capabilities.
Background: AI agents
AI agents are software systems that perceive their environment, make decisions, and execute actions to achieve specific goals. Unlike standard models, they maintain internal state and interact with external tools, creating unique security risks around tool execution, memory retention, and autonomous decision loops that require specialized controls.
Read the full guide: AI Agents Explained: Answers to Security and Architecture Questions
What to do and how to stay safe: Google
- Monitor AI model usage during autonomous operations to track resource consumption.
- Define clear objectives and termination criteria for agent deployments.
- Review agent communication logs to understand coordination patterns.
- Ensure appropriate safeguards are in place during AI testing.
Step-by-step guide: AI Governance Explained: Building Rules for Autonomous Systems
General security guidance from the Vector Tribune newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
How long did the agents take to beat Minecraft?
The agents took 27 hours and 14 minutes to defeat the Ender Dragon.
How much did the session cost?
The session cost $1,620 based on projected 2027 pricing for Gemini 4 Argon.
Did humans help the agents during the run?
No, the agents operated fully autonomously without any human input or predefined strategies.
