The Bargaining-Verification Trap in AI-Mediated Work: An Integrated Theory of Productivity, Worker Voice, and Organisational Resilience

Artificial intelligence (AI) is changing not only the technological content of work but also the allocation of decision rights, the distribution of productivity gains, and organisations’ capacity to verify their own outputs. Existing literatures establish that AI can raise task productivity, algorithmic management can intensify control, and automation can alter labour shares. Few models integrate these mechanisms in one tractable analytical framework. This article introduces the bargaining-verification trap, a conditional regime in which greater AI delegation initially increases joint economic surplus but simultaneously reduces workers’ effective bargaining share, raises the congestion cost of human verification, and may erode the skills needed for intervention. We develop a two-sided model linking delegation intensity, queueing-based verification capacity, an opacity-sensitive bargaining rule, and a dynamic stock of human agency capital. We derive conditions under which worker remuneration falls while joint surplus rises; the firm’s preferred delegation exceeds the surplus-maximising level; and verification becomes less stable because review arrivals increase while independent human capacity deteriorates. A fully hypothetical numerical example separates the worker-income optimum, the joint-surplus optimum, and the firm-profit optimum. The theory is reconciled with divergent 2025-2026 labour-market evidence and translated into falsifiable propositions, measurement protocols, and governance principles. Its central implication is conditional rather than technologically deterministic: AI-mediated work can be both productive and inclusive when bargaining institutions, verification investment, and skill reproduction develop alongside automation, but increasing throughput cannot, by itself, establish that they have done so.