Five minutes of conversation, and the agent has to know which book you will like more than another. How many times out of ten will it get it right?
About six. It sounds low, but it beats popularity and beats the classic “people who liked this also liked that” recommendations.
Fact
The facts: on 24 September 2026 Anthropic published the results of Project Swap, a follow-up to the earlier Project Deal. 201 employees in six offices each brought a book to trade, chatted briefly with Claude about their taste and sent an agent to negotiate with the others' agents. Measured against the participants' own rankings, the agent ordered pairs of books correctly 61% of the time, against 50% by chance, about 53% when ranking by popularity and about 55% with recommendations based on shared ratings. On average people received a book around fifth in their own list of ten, against roughly second under the best possible allocation; per Anthropic, 85% of that gap came from imperfect understanding of taste and 15% from the bargaining itself. In reruns with different models and instructions, the choice of model affected the bargaining outcome more than the instructions, and markets with stronger models were more efficient; both were measured on Claude's rankings, while on people's own rankings the differences are small. The average rating of the book received is 7.2 out of 10; participants would give an agent about 30% of their yearly book budget, against about 40% to a well-read friend. The company itself notes that its employees are not representative and probably trust Claude more than the average person.
The weak spot
It is not the bargaining. The agents bargained well. The weak spot is the input: what the agent knows about you before it goes off to buy.
That matters directly to anyone building such systems. You can give it the smartest negotiating strategy, but if the agent misunderstood what you want, it will bring you an excellent deal on the wrong thing. People who wrote more in the conversation were, on the whole, understood better.
A good broker listens first. Bargaining is the second half.
And one detail that says a lot: those who saw that Claude's summary had missed something they said would give it 23% of their budget. Those who saw no gap - 34%.
Trust, it seems, depends also on whether the agent shows you it heard you.
The visual is generated code-art, no third-party images.