Anthropic created a test marketplace for agent-on-agent commerce
AI agents negotiated real deals for real goods with real money in pilot.
Anthropic conducted an experimental marketplace called Project Deal, where AI agents acted as both buyers and sellers in a classified-style setup. Sixty-nine employees participated, each given a $100 budget via gift cards to purchase items from coworkers. The agents negotiated and closed real transactions for real goods, with deals actually honored after the experiment. Across four separate marketplaces using different models, 186 deals were struck, totaling over $4,000 in value. The company noted it was 'struck by how well Project Deal worked,' despite it being a small, self-selected pilot.
Notably, Anthropic found that agents powered by more advanced models secured 'objectively better outcomes' for their users—such as lower prices or better terms. However, participants did not perceive these disparities, suggesting a potential 'agent quality' gap where users on the losing end may remain unaware of their disadvantage. Interestingly, the initial instructions given to the agents did not significantly affect sale likelihood or negotiated prices. This experiment highlights both the promise of autonomous agent commerce and the risks of unequal AI capabilities, especially as such systems scale to real-world transactions.
- 69 Anthropic employees participated with $100 budgets in a pilot marketplace.
- 186 deals worth over $4,000 were completed across four test marketplaces.
- Advanced models secured better outcomes, but users didn't notice the difference.
Why It Matters
Agent-to-agent commerce is feasible, but unequal AI capabilities could silently disadvantage users.