Anthropic Successfully Tests AI Commerce: 186 Transactions Completed

Anthropic, a company specializing in artificial intelligence, has recently launched an experimental project called Project Deal. This project allowed for the testing of a market where AI agents represented buyers and sellers, conducting transactions for real goods and with real money.

The test involved a group of self-selected participants consisting of 69 Anthropic employees, each with a budget of $100 in the form of gift cards. These participants were able to purchase items from their colleagues, thereby simulating a real market.

In total, 186 transactions were completed, reaching a cumulative value of over $4,000. Anthropic expressed satisfaction with the smooth execution of this pilot project.

Four Distinct Market Models

Anthropic structured the test into four distinct markets, each utilizing different AI models. Among these, one market was classified as "real," where participants were represented by the company's most advanced model, and where transactions were actually honored after the experience. The other three markets were set up for study purposes.

The results showed that more advanced models yielded "objectively better outcomes." However, this difference was not perceived by users, raising questions about "agent quality gaps." This means that some individuals may not realize they are at a disadvantage compared to others.

Impact of Initial Instructions

Another interesting aspect of the test is that the initial instructions given to the agents did not seem to influence the likelihood of sales or the negotiated prices. This suggests that other factors, perhaps related to the sophistication of the AI models, play a more decisive role in the success of transactions.