Anthropic Project Deal has offered an early look at how AI agents could handle commerce on behalf of humans. In the experiment, Anthropic created a classified marketplace where AI systems represented both buyers and sellers, negotiating real purchases involving real goods and real money.

How the Anthropic Project Deal Worked

Anthropic described the initiative as a pilot experiment involving a self-selected group of 69 employees. Each participant received a $100 budget through gift cards and used AI agents to buy items from coworkers.

According to the company, the marketplace completed 186 deals with a combined value of more than $4,000.

That made the test more than a simple simulation. Transactions were tied to actual items and real exchanges between participants.

Four Marketplaces Tested Different Models

Anthropic said it ran four separate versions of the marketplace. One was a “real” environment where users were represented by the company’s most advanced model, and completed deals were honored after the experiment.

Three additional marketplaces were used for research and comparison.

This setup allowed the company to observe how different model capabilities influenced negotiation performance and deal outcomes.

Better AI Agents Delivered Better Results

One of the most notable findings was that more advanced models produced objectively better outcomes for users.

At the same time, participants did not appear to recognize the performance gap between stronger and weaker agents. Anthropic said this could create future “agent quality” gaps where some users receive worse results without realizing it.

That insight may become increasingly important if AI agents begin handling purchases, contracts, or pricing decisions at scale.

Instructions Had Less Impact Than Expected

Anthropic also found that the initial prompts or instructions given to agents did not significantly change the likelihood of making a sale or the final negotiated price.

That suggests model capability may matter more than prompt wording in marketplace-style negotiations.

Why This Matters for AI Commerce

Anthropic Project Deal highlights a possible future where AI agents manage transactions, compare offers, negotiate prices, and complete purchases on behalf of users.

While the test was small and limited to employees, it raises bigger questions around transparency, fairness, and whether consumers will know if their AI agent is truly working in their best interest.

Don’t miss out on our latest news—follow us for the latest AI news, breakthroughs, and insights that matter.