AI & agents
Anthropic shares Project Swap, an experiment in agents trading for people
- Published
- Source
- Anthropic research blog
Summary
Anthropic has published Project Swap, an experiment in which 201 employees each sent a Claude-powered agent onto a digital trading floor to barter books on their behalf. From a roughly five-minute chat, an agent's ranking of the books matched its owner's actual preferences on 61% of pairs, and market shortfalls traced mostly to missing information about participants rather than weak negotiation. Re-running the same market dozens of times, the team found that which model an agent ran on shaped trading outcomes more than the instructions it was given.
Why it matters for our work
It offers early experimental data on how far agents can go when they search and negotiate for people. Everyday deals that never happen because they are too tedious to arrange, such as shift swaps or secondhand trades, are exactly the kind of work agents could take over to speed up collaboration.
Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.