<![CDATA[Agentic artificial intelligence is shifting automation from prediction and recommendation toward autonomous execution. This transition creates an economic externality that is not captured by conventional firm-level investment models. A firm receives much of the productivity benefit from delegating decisions to an artificial intelligence agent, while part of the expected loss from correlated errors, shared model dependencies, common data suppliers, and synchronized actions can be transmitted to customers, counterparties, markets, and public institutions. This paper develops a welfare-economic model of the autonomy externality. Firms jointly choose an autonomy level and an assurance effort that reduces residual operational risk. Systemic loss is represented by a quadratic network term whose strength depends on cross-firm failure correlation. The model yields closed-form private and socially optimal choices, establishes that correlated failure raises privately excessive autonomy and depresses assurance intensity, and derives an assurance-adjusted residual-risk levy that decentralizes the social optimum. A stylized numerical calibration shows that the welfare wedge is small when failures are nearly idiosyncratic but expands sharply as common dependence rises. At a cross-firm correlation of 0.80, the socially optimal autonomy level falls from 0.973 to 0.813, assurance rises from 0.404 to 0.688, and residual risk declines by approximately 59.5 percent. Monte Carlo robustness analysis across 193,446 admissible parameter draws confirms the direction of all principal results. Policy comparisons indicate that autonomy caps and assurance floors are inferior to instruments that price residual risk and reduce common dependencies. The paper contributes a tractable economic foundation for governing agentic artificial intelligence through risk-sensitive levies, assurance credits, provider diversification, interoperability, incident reporting, and sector-specific systemic-risk supervision.]]>
Copyrights © 2026
