Beyond the Principal-Agent Paradox: A Theory of Governance Failure and Mechanism Design in Agentic AI Systems

Principal-agent theory has provided the dominant framework for understanding governance failure in organizational settings for five decades. This paper argues that agentic AI systems produce a structural inversion of the classical principal-agent problem that renders conventional governance mechanisms categorically inadequate, not merely insufficient. We term this the Agentic Principal-Agent Inversion: a condition in which the human overseer retains nominal authority but loses informational parity, processing parity, and interpretive capacity simultaneously, creating a triple asymmetry that conventional principal-agent governance tools cannot resolve. Drawing on principal-agent theory, bounded rationality, and organizational control theory, we derive eight propositions addressing the technical conditions of governance failure, the institutional dynamics that amplify it, and the mechanism design principles that constitute a viable alternative. We illustrate the propositions through the context of financial services agentic AI deployment, where the conditions are most advanced and the accountability stakes are highest. The paper contributes a theoretical extension of principal-agent theory to agentic AI contexts, a typology of governance failure mechanisms, and a set of testable propositions constituting a research agenda for empirical IS research on agentic AI governance.