Firms should not automatically hand over control to AI: higher demand uncertainty makes delegation to agentic systems optimal, but hybrid human-in-the-loop governance often outperforms full automation or pure human control. Buying AI and granting it authority are separate choices, so companies may invest in AI while deliberately restricting its autonomy to manage risk.

Main Finding

Delegating pricing and inventory decisions to agentic AI is not always optimal; instead, optimal delegation depends on demand uncertainty, risk asymmetries, and organizational structure. An analytical model shows (1) there is a demand-variance threshold above which AI delegation is optimal even if AI is imperfect; (2) partial (human-in-the-loop) delegation can strictly dominate both full automation and full human control; and (3) when AI investment is endogenous, firms may adopt AI but deliberately limit its authority, producing a three-region equilibrium (no-AI, AI-with-restricted-authority, AI-with-high-autonomy). Extensions reveal several counterintuitive comparative statics (e.g., faster learning can delay autonomy).

Key Points

• Trade-off introduced by agentic AI:
  • Pro: greater responsiveness via state-contingent (real-time) pricing and allocation.
  • Con: reduced buffering and accelerated execution increase operational exposure and downside risk.
• Demand-variance threshold:
  • Above a critical level of demand variance, delegation to AI becomes optimal, even when AI is imperfect.
• Superiority of partial delegation:
  • Hybrid governance (human-in-the-loop) can strictly dominate both extremes by combining AI responsiveness with human risk management.
• Endogenous AI investment uncouples adoption from autonomy:
  • Firms may invest in AI capabilities but intentionally restrict authority—creating three behavioral regions across parameter space.
• Robustness / counterintuitive effects from model extensions:
  • Faster learning by AI can postpone granting early autonomy.
  • Better pricing coordination via AI can exacerbate inventory imbalances.
  • Stochastic lead times, asymmetric service-level penalties, endogenous oversight, and firm scale materially reshape delegation incentives.
  • Larger firms may favor autonomy even at moderate uncertainty levels.

Data & Methods

• Modeling approach:
  • Analytical, theoretical model at the operations–marketing interface.
  • Firm jointly chooses pricing and inventory under demand uncertainty.
  • Decision modes modeled: full human control, full AI autonomy, and human-in-the-loop governance.
• Key ingredients / assumptions:
  • Agentic AI enables state-contingent (real-time) decisions but tends to reduce traditional buffers and speed execution.
  • AI is imperfect (no assumption of perfect predictions).
  • Demand uncertainty (variance) is a central parameter driving trade-offs.
  • Extensions incorporate endogenous AI investment, learning dynamics, asymmetric service-level costs, stochastic lead time, endogenous oversight costs, and organizational scale.
• Methods / analysis:
  • Closed-form characterization of optimal policies and threshold conditions where delegation is preferred.
  • Comparative statics to study how parameters (variance, learning speed, scale, asymmetry) affect delegation equilibria.
  • Equilibrium analysis when AI investment and autonomy are endogenous, yielding a three-region partition of strategy space.
  • (Paper reports robustness checks / extensions to show qualitative persistence of results.)

Implications for AI Economics

• Delegation is an economic — not purely technological — decision:
  • Firms should evaluate responsiveness gains versus increased operational exposure when setting AI authority.
• Governance design matters:
  • Hybrid (human-in-the-loop) governance can be welfare-improving for the firm and reduce downside risk; it is a rational equilibrium outcome under many realistic settings.
• Adoption strategy:
  • Investment in AI and granting of autonomy are distinct managerial choices; firms may benefit from staged adoption (build capabilities first, limit authority initially).
• Policy and regulation:
  • Regulators and standard-setters should recognize heterogeneity in optimal delegation and tailor guidance to risk asymmetries and firm size/scale.
• Organizational strategy and competition:
  • Larger firms may adopt higher autonomy sooner, potentially altering competitive dynamics and labor allocation across firms.
• Future empirical directions:
  • Testable predictions include (a) a positive relation between demand variance and AI autonomy, (b) prevalence of hybrid governance in intermediate uncertainty regimes, and (c) nonmonotonic effects of learning speed on autonomy adoption timing.