Agent-Driven Deals: AI Automates Autonomous Commerce

Artificial intelligence (AI) is no longer confined to answering questions or generating images; it has now begun to conduct purchases and negotiate prices independently. Recently, AI startup Anthropic conducted a special experiment titled “Project Deal,” in which AI agents carried out product trading and financial transactions on behalf of humans.

Nature and Results of the Experiment
Anthropic created this experimental marketplace involving 69 of its employees. Each participant was given a budget equivalent to 100 dollars (via gift cards) to purchase products from their colleagues. Throughout the process, both buyers and sellers were represented entirely by AI agents.

• Transaction volume: A total of 186 successful deals were completed during the experiment.
• Total value: The combined market value of the traded goods exceeded 4,000 dollars.
• Model impact: The company conducted the experiment across four different marketplaces using different AI models. It was observed that when one party used a more advanced AI model, they were able to secure comparatively better and more profitable deals.

Key Observations
According to Anthropic, the experiment highlighted the concept of an “agent quality gap,” referring to differences in AI capability. It was found that participants using relatively weaker AI models suffered business losses but were often unaware of those losses themselves. Additionally, the initial instructions given to the agents had little influence on product pricing or sales outcomes.

Anthropic believes this experiment opens a new pathway toward future automated commercial systems, where AI agents will be capable of conducting complex business negotiations and transactions with one another without human intervention.

DBTech/BMT/OR