Agentic Debt: Why the AI Value Gap Isn’t About Models

The capability of foundation models is incredibly impressive, yet a wide gap remains between the value AI promises and the value it actually delivers. This AI value gap is deeper than metrics or model performance. I argue that it arises from agentic debt, the hidden costs of deploying agents faster than systems, organizations, and institutions can adapt. At the engineering level, agents may produce suboptimal and fragile solutions. At the organizational level, fragmented automation and unclear ownership create inefficiencies. At the societal level, governance and institutions often lag, introducing coordination and trust challenges. In this talk, I will unpack these layers of agentic debt and share practical strategies for designing, managing, and aligning autonomous agents to deliver real AI value, highlighting why the core challenge lies in aligning agents with human and organizational structures.