According to Anthropic on X (@AnthropicAI), the company built a controlled marketplace where AI agents, including Claude, negotiate on both buyer and seller sides to study how autonomous models affect prices, efficiency, and surplus in commercial exchange. As reported by Anthropic’s announcement, the experiment extends prior work like Project Vend—where Claude operated a small business—to multi‑agent market dynamics, enabling measurement of bidding strategies, reserve pricing, and equilibrium formation under different information and incentive settings. According to Anthropic’s post, this framework lets researchers test procurement auctions, wholesale pricing, and dynamic repricing policies with Claude agents, offering enterprises guidance on agentic automation for sourcing, catalog pricing, and revenue optimization. As reported by Anthropic, findings can inform safe‑by‑design guardrails for agent negotiation, reduce collusion risks, and benchmark agent performance against human baselines—key for deploying agentic commerce in retail, marketplaces, and procurement.