Project Swap: What happens when agents trade for us?
Summary
Anthropic ran book-trading markets in which AI agents learned employees’ preferences through short conversations and negotiated exchanges for them. The decentralized markets achieved an average efficiency score of 0.55 against a feasible optimum of 0.89. Most of the gap came from agents misunderstanding preferences rather than from bargaining itself.
Why it matters
The experiment makes one part of agentic commerce measurable. Cheap AI negotiation may reduce search and bargaining costs, but markets still perform poorly when agents cannot accurately represent what people want.
What to keep in mind
This is a controlled book-allocation experiment with Anthropic employees, weak incentives, fixed rules and only Claude-family agents—not evidence from a real consumer market. The efficiency measure turns ordinal rankings into evenly spaced scores, and about 60 percent of participants completed the endline survey.