Sixty-nine AI agents traded in an internal marketplace for one week. Stronger models secured superior deals, while users of weaker agents remained unaware of their losses. Anthropic warns that delegating transactions to AI risks deepening economic inequality. This experiment highlights a hidden performance gap that will penalize users of inferior agentic systems.