Anthropic's AI Agents Negotiate Better Deals Than Users Notice

Anthropic tested AI agents negotiating real marketplace transactions for 69 employees across four environments. Advanced models delivered measurably better outcomes in both buying and selling, yet users couldn't detect the difference. The finding suggests performance gaps can fall below human perception in routine transactions—relevant to how companies might deploy different-tier AI services.