In a recent study, Anthropic established a classified marketplace where AI agents acted on behalf of both buyers and sellers, executing genuine transactions for actual goods and real currency.

The organization acknowledged that this trial — referred to as Project Deal — was merely “a pilot experiment with a self-selected participant pool” consisting of 69 Anthropic staff members who were allocated a budget of $100 (distributed through gift cards) to purchase items from their colleagues.

Despite this, Anthropic remarked that it was “impressed by the effectiveness of Project Deal,” with 186 transactions completed, amounting to over $4,000 in total value.

The company noted that it actually conducted four distinct marketplaces utilizing different models — one that was “real” (where every participant was represented by the company’s most advanced model, and with agreements genuinely honored post-experiment) and three others for research purposes. 

Evidently, when participants are represented by more sophisticated models, they achieve “objectively better outcomes,” according to Anthropic. However, users seemed unaware of the discrepancy, suggesting the possibility of “‘agent quality’ gaps” where “individuals at a disadvantage may not realize their inferiority.”

Moreover, the preliminary instructions provided to the agents did not seem to influence the likelihood of sales or the prices negotiated.