Anthropic created a test market for agent-to-dealer trading

In a recent experiment, Anthropic created a sorted marketplace where AI agents represent both buyers and sellers, concluding real deals for real goods and real money.

The company admitted that this test — which it called Project Deal — was merely a “pilot experiment with a self-selected participant group” of 69 Anthropic employees who were given a budget of $100 (paid for via gift cards) to buy things from their coworkers.

However, Anthropic said it was “amazed at how successful the Deal Project was,” with 186 deals worth a total of more than $4,000 concluded.

The company said it actually ran four separate markets with different models — one that was “real” (where everyone was represented by the company’s most advanced model, with trades actually honored after the trial) and three more for study.

Apparently, when users are represented by more advanced models, they get “objectively better results,” Anthropic said. But users do not appear to notice this disparity, raising the possibility of “gaps in agent quality” where “people on the losing end may not realize they are worse off.”

Initial instructions given to agents also appear to have no effect on the likelihood of a sale or negotiated prices.