Anthropic created a test marketplace for agent-on-agent commerce

In a recent experiment, Anthropic created a classified marketplace where AI agents represented both buyers and sellers, striking real deals for real goods and real money.

The company described the test, called Project Deal, as “a pilot experiment with a self-selected participant pool” of 69 Anthropic employees. Each participant was given a budget of $100, paid out via gift cards, to purchase items from their coworkers.

How Project Deal Worked

Despite its limited scope, Anthropic said it was “struck by how well Project Deal worked.” Over the course of the experiment, participants completed 186 deals totaling more than $4,000 in value.

Anthropic ran four separate marketplaces using different AI models. One marketplace was considered “real,” meaning all participants were represented by the company’s most advanced model and the deals were honored after the experiment concluded. The other three marketplaces were used for research and comparison.

Model Performance and Outcome Gaps

According to Anthropic, users represented by more advanced models achieved “objectively better outcomes.” However, participants did not appear to notice the disparity in performance.

This dynamic raises the possibility of what Anthropic described as “agent quality” gaps, where “people on the losing end might not realize they’re worse off.”

The company also found that the initial instructions given to the agents did not appear to affect the likelihood of a sale or the negotiated prices.

Anthropic shared additional details about the experiment on its website as part of its broader exploration of agent-driven commerce.