Anthropic Turned Its Own Office Into an AI Negotiation Marketplace — And Published What Went Wrong
Project Deal put Claude agents in charge of buying and selling goods between Anthropic employees, revealing the specific failure modes that emerge when AI agents negotiate with real stakes.
Anthropic published results from an internal experiment called Project Deal, in which Claude agents were tasked with buying, selling, and negotiating on behalf of employees in a real marketplace inside the company's San Francisco office. As @AnthropicAI described, the setup was straightforward: real goods, real employees, real transactions — but with AI agents as intermediaries.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.