Anthropic has recently embarked on a groundbreaking experiment by establishing a classified marketplace where AI agents acted as both buyers and sellers, successfully completing transactions for actual goods and currency. This initiative, known as Project Deal, involved a select group of 69 employees from Anthropic, each provided with a budget of $100 in gift cards to purchase items from their colleagues.
The company expressed surprise at the success of Project Deal, which resulted in 186 transactions amounting to over $4,000 in total sales. Anthropic conducted four distinct marketplace models during this pilot, including one "real" marketplace where transactions were honored post-experiment, alongside three additional models designed for research purposes.
Findings indicated that when users were represented by the company's most sophisticated AI models, they achieved "objectively better outcomes." However, participants did not seem to recognize the differences in agent quality, suggesting the existence of potential "agent quality gaps." This raises intriguing questions about awareness and performance in AI-assisted transactions.
Interestingly, the initial guidelines provided to the agents did not significantly influence the likelihood of sales or the prices negotiated, highlighting the complexity of agent interaction in a marketplace setting.
As AI continues to evolve, experiments like Project Deal pave the way for future innovations in commerce, potentially transforming how transactions are conducted across various sectors.