In a recent experiment, Anthropic created a classified marketplace where AI agents represented each buyers and sellers, striking real deals for real goods and real money.
The corporate admitted this test — which it called Project Deal — was only “a pilot experiment with a self-selected participant pool” of 69 Anthropic employees who got a budget of $100 (paid out via gift cards) to purchase stuff from their coworkers.
Nonetheless, Anthropic said it was “struck by how well Project Deal worked,” with 186 deals made, totaling greater than $4,000 in value.
The corporate said it actually ran 4 separate marketplaces with different models — one which was “real” (where everyone was represented by the corporate’s most-advanced model, and with deals actually honored after the experiment) and one other three for study.
Apparently, when users are represented by more advanced models, they get “objectively higher outcomes,” Anthropic said. But users didn’t seem to note the disparity, raising the opportunity of “‘agent quality’ gaps” where “people on the losing end may not realize they’re worse off.”
Also, the initial instructions given to the agents didn’t appear to affect sale likelihood or the negotiated prices.