The Coordination Compression Trap: A Dynamic Model of AI-Enabled Productivity, Verification Debt, and Organisational Resilience in Global Knowledge Firms

Abstract

Aims: This study develops a dynamic explanation for the gap between task-level artificial intelligence productivity and durable enterprise value. It introduces verification debt, the accumulated stock of unverified assumptions, weak review, concealed dependencies, deferred validation, and capability loss. Study Design: Analytical model development with longitudinal Monte Carlo simulation, policy-space search, and structural-heterogeneity robustness analysis. Place and Duration of Study: Singapore; model development, simulation, and revision were conducted in July 2026. Methodology: The model links automation, governance, verification, task fit, uncertainty, expertise, resilience, and shocks. Five stylised regimes were simulated for 5,000 firms over 60 periods. A three-dimensional search evaluated 3,375 fixed policies before boundary refinement. The analysis also incorporated moderate firm-specific heterogeneity and paired comparisons using common random numbers. Results: Verified autonomy produced the highest mean cumulative value among the five named regimes at 454.93 units, compared with 419.97 for debt-responsive governance and 3.77 for acceleration-first adoption. Acceleration-first peaked in period 23, generated fragility in 82.94% of base trajectories, and ended with negative values in 41.40%. The full search did not confirm verified autonomy as optimal over the expanded grid. The highest evaluated fixed policy was , with a mean value of 608.92 and no fragile trajectories under the base model. Under synthetic cross-firm heterogeneity, the five-regime ranking persisted and paired comparisons remained strongly favourable to verified autonomy over the other named regimes. Conclusion: The coordination compression trap is conditional on inadequate assurance, not automation alone. Extensive automation can remain valuable when verification is strong, while weakly verified acceleration creates debt, deskilling, and shock sensitivity. Numerical optima remain model-dependent and require empirical calibration.