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

Anthropic launched 'Project Deal', an experimental marketplace where 69 employees utilized AI agents to conduct real-world commerce using $100 budgets. The experiment involved 186 successful transactions totaling over $4,000. Anthropic discovered that users represented by more advanced AI models achieved objectively better financial outcomes, yet participants often failed to notice these disparities. This highlights potential 'agent quality' gaps where individuals might be disadvantaged by their AI's performance without realizing it. Interestingly, initial prompt instructions appeared to have little impact on the final negotiated prices or the likelihood of sale success.

(Source：TechCrunch)