Abstract for: When Task Structure Is Ignored: Centralization Spirals and Delegation Treadmills in Generative AI Adoption

Amid intense generative-AI hype, boards face pressure to show rapid progress, and markets increasingly reward firms that can delegate work to autonomous systems such as AI agents. This paper examines how such scaling pressure can outpace task appropriation—understanding, specification, instrumentation, and workflow redesign—creating endogenous dynamics where unbounded delegation leads to risk buildup, governance clampdowns, and disappointing impact. We develop a system dynamics model that allocates bounded organizational resources between task appropriation and delegation enablement. The model separates attempted delegation, realized usage, and effective execution, and endogenizes trust, local flexibility, risk-debt accumulation, delayed governance responses, and central enablement backlogs. We use the model as a reference-mode generator and map regime boundaries via 2D parameter sweeps and Latin Hypercube sampling. Simulations reproduce three dominant regimes: appropriation-first safe scaling, a centralization spiral driven by governance backlash and backlog formation, and a high-activity/low-yield delegation treadmill. Regime maps show that increasing the appropriation share expands the safe-scaling basin, especially under high hype or low governance tolerance. Across robustness sampling, impact is non-monotonic in delegation: it often peaks at intermediate allocation levels. The model suggests that "more AI usage" does not systematically yields to "more value". When delegation scales faster than task appropriation, organizations can accumulate latent risk that triggers restrictive governance and centralization, reducing learning and impact. Prioritizing appropriation—task structure analysis, specification, and evaluation—widens the conditions for safe scaling, while still leaving an interior optimum where excessive appropriation can slow beneficial scaling. Only marginally for spelling checks and cosmetic mistakes