As part of a recent experiment, Anthropic created a private marketplace where artificial intelligence representatives are both buyers and sellers. This study, called Project Deal, aimed to test a trading environment using real products and real money.

The company stated that this initiative was a limited pilot with the participation of 69 employees. Participants were assigned a budget of $100 via gift cards so they could shop with their co-workers.

Results of Project Deal Experiment

Anthropic announced that the experiment was very successful, with 186 transactions totaling over $4,000. In this process, the company established four separate marketplaces where different models were used and examined the functioning of the system.

At one stage of the experiment, all participants were represented by the company’s most advanced models, and the agreements were made confirmed at the end of the exercise. The other three marketplaces were used for research and observation purposes only.

According to the research data, users achieve objectively better results when they are represented by more advanced models. However, participants often do not notice this performance difference, creating an agent quality gap.

The company points out that people who are disadvantaged by this gap may not realize that they are actually receiving a worse service. It was also observed that initial instructions given to agents had no significant impact on the likelihood of sales or negotiated prices.

The Future of AI-Powered Commerce

Anthropic states that this pilot study provides important data on what role artificial intelligence representatives can play in future commercial activities. The ability of agents to negotiate autonomously and conclude deals could herald a new era in the functioning of digital commerce.

The results of the experiment prove that the capabilities of artificial intelligence models directly affect trading processes. In particular, the advantages provided by more advanced models may become one of the key elements determining the user experience in future commercial platforms.

The company emphasizes that such experiments are critical to understanding the behavior of AI agents in real-world scenarios. The resulting data will guide the development of more complex and comprehensive trading systems in the future.

What do you think about the idea of ​​​​AI representatives shopping or negotiating on your behalf?