The labor-pool fallacy
Labor replacement is one of the largest apparent AI TAMs. If a profession costs $500 billion per year and AI can automate half of it, the superficial market estimate is $250 billion. But the $250 billion is the historical cost of a production method, not necessarily the equilibrium price of the replacement method.
The first machine solution may price against labor because labor is the alternative. Once multiple machine solutions are sufficient, pricing shifts toward machine substitutes.
Benchmark₀ = human cost
Benchmark₁ = competing AI cost
Benchmark₂ = cheapest sufficient mechanism
The final mechanism may not even be an AI model. It may be a deterministic workflow, a smaller specialized model, a sensor, an API, a rule, a redesigned process or a combination. The labor pool is therefore best interpreted as the amount of cost available for destruction, not as guaranteed recurring AI revenue.
This is the first place where the paper's inversion becomes explicit: the larger the cost pool AI can eliminate, the larger the social value created—but potentially the weaker the argument that the old cost pool survives as a monetizable market.