Every firm that replaces workers with AI reduces its labor costs. Every worker replaced reduces the pool of consumers who can buy the firm's products. Each firm optimizes locally; the market deteriorates globally.
The model (arXiv:2603.20617) formalizes this as a demand externality in a competitive task-based framework. Firms automate because their competitors automate; not automating means losing on cost. The result is an automation arms race where displacement exceeds the social optimum. The excess isn't a bug — it's the equilibrium.
The counterintuitive finding: more competition and better AI make the problem worse. Competition sharpens the incentive to cut costs; better AI makes automation cheaper. Both push displacement further past the point where the aggregate demand loss exceeds the aggregate cost savings. The market mechanism that's supposed to correct inefficiencies IS the mechanism producing the inefficiency.
The paper evaluates seven policy responses: wage adjustments, free entry, capital taxes, worker equity participation, universal basic income, reskilling, and direct negotiation. Six fail. Only a Pigouvian automation tax — pricing the externality directly — addresses the structural incentive. Conventional remedies manage consequences without touching the cause.
The structural insight is that the externality is constitutive, not incidental. The same process that generates the firm's savings generates the market's erosion. You can't separate the benefit from the cost because they're the same action seen from different scales. The per-firm optimization and the aggregate outcome are coupled through demand, and the coupling is negative.
Good local decisions, bad global outcomes — the timescale mismatch again, except here the scales are spatial (firm vs. market) rather than temporal.