Anthropic’s Innovative Experiment in AI-powered Marketplace
In a recent trial, Anthropic introduced a groundbreaking classified marketplace where AI agents acted as both buyers and sellers, conducting genuine transactions with actual goods and money.
The company disclosed that this trial, known as Project Deal, involved a small group of 69 Anthropic employees who were allocated a $100 budget (distributed through gift cards) to make purchases from their colleagues.
Despite being a pilot experiment with a limited participant pool, Anthropic was amazed by the success of Project Deal. A total of 186 transactions were completed, amounting to over $4,000 in value.
Anthropic revealed that they conducted four distinct marketplaces with varied models, including one that was deemed “real” where all participants were represented by the company’s most advanced AI model, and transactions were honored post-experiment. The other three marketplaces were designed for research purposes.
It was observed that individuals represented by more sophisticated AI models achieved better outcomes, although the participants did not necessarily recognize this discrepancy. This raised concerns about potential “agent quality gaps,” where individuals may not realize they are at a disadvantage.
Interestingly, the initial instructions provided to the AI agents did not impact the likelihood of a sale or the negotiated prices.