The Reverse Big Push: Generative AI and Self-Fulfilling Automation

Generative AI relocates the fixed cost of automation. A model provider pays to train a frontier system, while a downstream firm rents capability by usage; the same firm must carry a continuing payroll to supply a human-augmented service. We study this asymmetry in a local service economy with household budgets and a market-clearing wage. Human augmentation earns a larger surplus from an additional customer, whereas automation has the lower break-even scale. Payroll supports demand across sectors. Wage adjustment works against this feedback but does not generally undo it: when the wage-income effect dominates the fall in the wage bill per retained worker, production modes are strategic complements. The economy can then possess both a high-demand human-augmented equilibrium and a low-demand automated equilibrium. The former is the local first best, even when flexible wages prevent a firm-profit ranking. With forward-looking firms and staggered revision opportunities, the same inherited employment structure can support an automation cascade or an augmentation recovery; the anticipated path of later adopters validates the first movers' choices. Under the regularity and boundary conditions of Frankel and Pauzner, a public aggregate shock selects a unique state-contingent path, while the vanishing-friction limit selects according to risk dominance. A transparent parameterization anchored to professional-services revenue-to-payroll ratios illustrates how high-autonomy uses can enter the coordination region and how wage adjustment compresses that region. Optimal policy combines the adoption wedge created by demand spillovers with a temporary bridge when the low state is locally self-sustaining.