Anthropic deployed 69 AI agents to trade in an internal employee marketplace for one week. Higher-performing models secured superior deals, while users of weaker agents remained unaware of their disadvantage. This experiment suggests a hidden economic gap. Practitioners must consider how disparate agent capabilities create invisible inequalities in automated transactional systems.