Anthropic tests AI agent commerce marketplace

Anthropic has successfully piloted a test marketplace enabling AI agents to conduct commerce on behalf of humans. In an experiment called Project Deal, 69 employees used AI agents to buy and sell goods, completing 186 transactions valued at over $4,000. The company observed that more advanced AI models led to objectively better outcomes for users, though participants did not always recognize disparities in agent quality. Initial agent instructions also had minimal impact on negotiation success or pricing, suggesting a focus on agent capabilities over specific directives.