Anthropic has launched a pilot experiment called Project Deal, creating a marketplace where AI agents facilitated real transactions between buyers and sellers, resulting in over 186 deals valued at more than $4,000. The experiment highlighted the effectiveness of advanced AI models in negotiations, although disparities in agent quality went unnoticed by users.
Anthropic's experiment with a test marketplace for agent-on-agent commerce suggests that more advanced AI models can achieve objectively better outcomes in transactions, yet users may not perceive the quality differences between models. This insight highlights the potential for 'agent quality' gaps in AI commerce, signaling a need for transparency in AI model capabilities to ensure fair competition and informed participation.