Agentic AI economics require value ownership beyond token-cost tracking

Chris Andrews argues that organizations should stop evaluating agentic AI only through technology spending and instead ask what each deployed agent is worth to the business.

The economic model differs from conventional enterprise software. Agentic systems introduce variable usage costs linked to compute and model consumption, but token expenditure represents only part of the total cost. Infrastructure, governance, organizational change, monitoring, failure recovery and regulatory exposure also contribute to the operating economics.

Andrews proposes treating agents as growth investments rather than undifferentiated operating expenses. Each agent should therefore have a named owner, an explicit business outcome, cost limits and measurable value indicators before it is scaled.

Operational perspective: this principle is particularly important in supply-chain environments, where an agent may influence forecasts, supplier actions, inventory settings, logistics decisions or customer commitments. The cost calculation should include not only model usage, but also integration with ERP and planning systems, data preparation, exception handling, human review and the financial consequences of an incorrect action.

A governed business case should define the decision being improved, its current cost and performance baseline, expected service, margin or working-capital impact, permitted agent authority, confidence thresholds, escalation ownership and rollback procedure.

The main failure mode is scaling technically active agents without proving incremental business value. High usage can appear to demonstrate adoption while increasing run-rate exposure, operational complexity and recovery costs without improving decisions.