Anthropic, a leading artificial intelligence research company, has recently conducted an internal experiment named “Project Deal” to explore the potential of AI agents in managing real-world transactions. This initiative involved 69 employees, each allocated a $100 budget, who engaged in a digital marketplace where AI agents represented them as buyers and sellers. Over the course of the experiment, 186 deals were completed, totaling more than $4,000 in transactions.
The experiment was structured into four distinct marketplace environments, each utilizing different versions of Anthropic’s AI models. Notably, the most advanced model, Opus 4.5, facilitated more favorable outcomes for participants, even though they were unaware of which AI version was representing them. This suggests that the sophistication of AI agents can significantly influence transaction results. ([techsifted.com](https://techsifted.com/posts/anthropic-agent-commerce-april-2026/?utm_source=openai))
Participants expressed a willingness to pay for such AI-driven negotiation services in real-world scenarios, indicating a potential shift in how contracts and transactions could be managed in the near future. ([completeaitraining.com](https://completeaitraining.com/news/anthropic-tests-ai-agents-that-negotiate-deals-on-behalf-of/?utm_source=openai))
While the results of Project Deal are promising, they also highlight the need for further research into the ethical and legal implications of AI agents handling transactions. Ensuring fairness, transparency, and accountability in AI-mediated deals will be crucial as this technology continues to evolve.