The Economics of Agentic AI: Runtime Governance as a Distinct Determinant of Enterprise AI Cost

Per-token inference prices are falling, while agentic workflows increase the number of model interactions, tool calls, and downstream actions required to complete a task. Established optimization techniques lower unit cost and can reduce computation under technical and budget conditions. Runtime semantic authorization adds a distinct control question: whether an exposed Operational Interpretation carries institutional authority to proceed. This paper argues that runtime governance is a distinct determinant of enterprise AI cost, because it governs the volume of execution-dependent activity admitted beyond a governable checkpoint, not only the cost of each computational unit. The principal contribution is Preventable Computation, defined as additional reasoning, tool use, downstream execution, and failure-related organizational effort that would not arise if an institutionally available runtime control held an unauthorized or materially divergent trajectory at the earliest governable checkpoint. Governing meaning and governing cost are the same control action, because the authority boundary that protects authorized meaning also determines which execution-dependent cost categories arise. The Governance Cost Stack and the Economic Blast Radius organize where those costs accumulate and propagate. Externalized Institutional Knowledge and Governed Semantic Reuse provide a separate architectural mechanism for maintaining authorized meaning outside broad prompt context and supplying it selectively at runtime. The economic effect is net of the measurement, evidence, maintenance, latency, and exception-handling cost required to operate the control. Runtime governance can determine whether execution-dependent cost categories arise at all, with the magnitude and net effect in any deployment open to measurement.