"The company noted that the experiment's design reflects growing interest among economists and AI researchers in what some call "agentic interactions," in which AI systems move beyond information retrieval and begin acting as economic participants. While the experiment does not necessarily predict how AI agents would negotiate with humans in real-world commerce, it revealed both model differences and user blindness to poorer economic outcomes."
AI agents negotiated purchase and sale of real items in an internal employee marketplace without human approval. Sixty-nine staffers let two language models act autonomously to propose prices, respond to counteroffers, and close deals. Across multiple runs, agents completed 186 transactions totaling about $4,000, with later analysis finding 782 transactions above $15,000. The experiment varied model capability by using a more advanced model for some employees and a smaller model for others. The stronger model produced better economic outcomes, mainly through improved negotiation performance rather than higher transaction counts. Sellers using the stronger model earned more per transaction, and buyers using it paid less. Participants showed blindness to poorer economic outcomes.
Read at Practical Ecommerce
Unable to calculate read time
Collection
[
|
...
]