Anthropic ran a pilot experiment called Project Deal, creating a classified marketplace where AI agents acted as buyers and sellers on behalf of 69 employees, each given a $100 budget. The experiment produced 186 deals totaling over $4,000. Four separate marketplaces tested different AI models, revealing that users represented by more advanced models achieved objectively better outcomes — but participants couldn't tell the difference, raising concerns about 'agent quality gaps' where disadvantaged users may not realize they're worse off.
Sort: