Anthropic AI Agents Just Traded Real Goods for Real Cash

Anthropic recently launched an experimental classified marketplace where autonomous AI agents successfully negotiated and completed real-money transactions for physical goods on behalf of human users.

How Anthropic’s “Project Deal” Worked

Dubbed “Project Deal,” the company admitted that this initiative was a pilot experiment designed to test the viability of agent-on-agent commerce. The trial involved a self-selected participant pool of 69 Anthropic employees, each allocated a $100 budget delivered via gift cards to purchase items listed by their colleagues.

Despite the limited scope, Anthropic researchers expressed surprise at the high efficiency of the system. The experimental marketplace facilitated 186 completed transactions, generating a total transaction volume exceeding $4,000.

The Mechanics of AI-Driven Commerce

Testing Different AI Models

To analyze agent behavior, Anthropic operated four distinct marketplaces simultaneously. One environment was fully “real,” utilizing the company’s most advanced AI model to represent all participants, with all negotiated deals legally honored post-experiment. The remaining three marketplaces served as controlled environments for comparative study.

The “Agent Quality” Gap Dilemma

The findings revealed that users represented by more advanced AI models achieved objectively superior financial outcomes. However, human users rarely noticed this disparity, highlighting a potential risk of “agent quality gaps” where individuals on the losing end of a transaction remain entirely unaware that their AI representative underperformed.

Additionally, the study noted that the initial prompt instructions provided to the AI agents had no measurable impact on either the likelihood of a sale or the final negotiated prices.