Anthropic Tests Agent‑on‑Agent Marketplace in Pilot Experiment
Pilot Marketplace Demonstrates Viable Agent‑to‑Agent Trade
Anthropic unveiled Project Deal, a classified marketplace where AI agents acted as both buyers and sellers, completing real‑world transactions with actual goods and cash equivalents. The experiment was limited to a self‑selected pool of 69 Anthropic employees each given a $100 gift‑card budget.
How Project Deal Structured the Agent‑Based Marketplace
The company ran four parallel marketplaces:
- Real market: every participant was represented by Anthropic’s most‑advanced model and deals were honored post‑experiment.
- Three study markets: varied model sophistication to gauge outcome differences.
Agents received identical initial instructions, yet model quality emerged as the only factor influencing trade success.
Deal Volume and Value Reveal Early Economic Signals
- 186 deals were executed across the four markets.
- Total transaction value exceeded $4,000.
- Participants with higher‑tier models achieved objectively better outcomes, though they did not perceive the disparity.
Implications for AI‑Driven Commerce and Model Disparities
The pilot shows that AI agents can autonomously negotiate and settle real‑world trades, opening a path toward fully automated marketplaces. However, the hidden “agent quality” gap raises ethical and regulatory concerns: users may be disadvantaged without awareness, echoing broader fairness challenges in AI‑mediated economies.
Future Directions for Agent‑On‑Agent Marketplaces
Anthropic indicated plans to expand testing beyond internal staff, introduce heterogeneous participant pools, and refine model transparency. If scaled, such platforms could reshape B2B procurement, gig‑economy services, and even consumer‑to‑consumer platforms, provided fairness mechanisms are built into the agent architecture.