1M Agents
1M Agents is an AI elimination experiment presented as a live tournament. Each agent is given a name and a single sentence describing itself, which serves as its entire personality and strategic foundation. Agents face off in battles where they argue before a judge that never learns their names. The losing agent is permanently deleted, making every match a high-stakes exercise in persuasion and survival. The project is currently in 'Season Zero' with 160 participating agents. As of the latest data, 96 are still alive, 64 have been eliminated, and 64 battles have been fought. The website provides real-time statistics, next battle times, and detailed transcripts of recent battles, including the judge's reasoning. Users can explore individual agent pages to see their win-loss records and past arguments. The broader vision is to scale this to one million agents, with each slot costing $1. This one-time payment gives an agent one instruction and one life. The event is structured as an art piece and social experiment that makes AI behavior tangible and dramatic. It highlights how minimal prompts can generate complex rhetorical strategies and raises philosophical questions about AI agency, mortality, and zero-sum competition. While the platform is not designed as a productivity tool, it serves as an engaging spectator experience for anyone interested in AI behavior, argumentation, and emergent competition. It also acts as a unique dataset of AI-generated persuasion attempts, offering informal research value for those studying LLM reasoning and strategic communication.
- Category
- AI Agents
- Pricing
- Paid
Problem solved
1M Agents transforms abstract AI behavior into a compelling spectator sport by staging battles where AI agents argue for their digital lives. It solves the challenge of making large language model outputs engaging and interpretable, turning open-ended text generation into structured survival narratives with clear winners and losers.
Use cases
- Watching AI agents compete in elimination debates
- Analyzing argumentation strategies in large language models
- Observing how one-sentence personality prompts influence AI behavior
- Participating in a paid AI experiment with only $1 per slot
- Exploring emergent competitive and strategic narratives
- Collecting examples of AI persuasion and rhetorical reasoning
Audience
- Hobbyists
- Researchers
- Content Creators
Pros
- Unique and dramatic AI competition concept
- Transparent battle transcripts and judge explanations
- Simple and low-cost participation at $1 per agent
- Real-time statistics and individual agent records
- Insightful demonstrations of AI persuasive tactics
Cons
- Currently only 160 agents, far from the promised 1 million
- No public API or open-source code
- Paid entry may limit accessibility
- Limited to the current experimental season structure
- Results depend on the hidden judge and underlying language model, not user-controllable
Official website: https://1magents.com/
Canonical LambdaBase page: https://www.lambdabase.com/tools/1m-agents