Archive — US — techcrunch
Anthropic created a test marketplace for agent-on-agent commerce
Anthropic tested an AI-driven marketplace, Project Deal, where AI agents acted as buyers and sellers in a pilot with 69 employees, generating over $4,000 from 186 deals. The experiment showed advanced AI models led to better outcomes for users, though they were unaware of disparities. The study ran four marketplace versions, indicating that initial agent instructions barely influenced outcomes.
Key points
- Anthropic launched Project Deal, testing agent-on-agent commerce with AI agents trading.
- 69 employees participated, each given $100 to spend, resulting in 186 completed deals.
- Total transactions exceeded $4,000 in value, indicating real engagement.
- Four marketplace models were tested, including one with advanced AI models which produced better user outcomes.
- Participants were often unaware of the quality gaps between different agent performances.
Why it matters
- This experiment aligns with 2025 IT trends emphasizing AI in commerce and business automation.
- Historically, AI-driven transactions have grown, indicating a shift towards automation in market functions.
- Concerns about AI transparency and performance disparities mirror broader discussions in AI ethics.
Anthropic's Project Deal marks a significant exploration into AI-driven commerce, revealing both the potential and the ethical challenges of using AI agents, with implications for future automated marketplaces.