A Utility-Driven Mathematical Framework for Agent-Centric AI Adoption