Who captures the productivity gains — labor or capital — when AI makes execution abundant across the economy?

The shift

Skilled execution — turning intent into a finished artifact, whether that's code, a brief, an analysis, a design, a support resolution, or a marketing campaign — goes from scarce (years of training, a specific person's hours) to abundant (near-zero marginal cost through a model). The scarcity of execution skill was what gave a broad swath of workers bargaining power; it's that scarcity, not scarcity as such, that flips.

The axioms

• Skilled execution labor is scarce, so workers who supply it capture a wage premium over the cost of producing them.
• That scarcity gives labor bargaining power: a worker can credibly threaten to leave, and replacing them is slow and expensive.
• Human capital — skill built over years — is an asset the worker owns and rents to the firm, so the returns to it accrue to the person.
• Production needs both capital and labor, and the surplus splits between them roughly by relative scarcity and replaceability.
• Capital owners take the residual after paying labor, because they own the means of production and bear the risk.
• A broad, general-purpose productivity technology eventually raises wages across the economy — the pattern set by mechanization, electrification, and computing.
• Owning the means of production means owning physical plant, brand, and capital; the model of production sits inside the firm.

Invalid axioms

1. Skilled execution labor is scarce, so the worker who supplies it captures a premium. Execution was the scarce, rentable good for most of the professional middle — the reason a competent drafter, analyst, coder, or paralegal earned well above the cost of training them. When the marginal unit of competent execution trends toward the price of an API call, the premium that rested on that scarcity has nothing under it. The habit-trap: firms, schools, and comp bands still price and promote for execution throughput — output per person — the exact thing that's decoupling from scarcity.
2. Scarcity of skilled labor gives the worker bargaining power. The leverage was never the skill in the abstract; it was that the firm couldn't cheaply replace you. Abundant execution weakens the "I'll walk" threat for any role whose value was mostly execution, because the fallback is no longer "hire someone slower" but "route it to a model." The habit-trap: labor-market and retention thinking still assumes replacement is slow and costly by default.
3. A general-purpose productivity technology eventually raises wages broadly. This held for technologies that augmented scarce human labor and needed more of it to realize the gains — the worker stayed in the loop as a complement. It is not a law; it was a consequence of the complementarity. A technology that substitutes for the execution itself, rather than amplifying a human doing it, breaks the mechanism that historically fed the gains back into wages. Treating the broad-wage-lift as automatic is the habit-trap — it was contingent on labor being complementary, not incidental.

Unchanged axioms

1. Someone accountable must own the outcome, and accountability stays human. A model produces the artifact; it can't be answerable for the misdiagnosis, the mispriced contract, the shipped defect, the regulatory breach. The person or firm that carries that liability still captures rent for carrying it — and vendors structurally decline it. This is why abundant execution doesn't flow value straight to whoever holds the capability: capability without accountability doesn't command the same return as accountability-bearing labor.
2. Judgment on novel, high-stakes, ill-specified problems stays scarce and human. Models are strong inside a well-framed problem and weak at knowing when the frame is wrong — what's worth doing, when to say no, when the confident answer is confidently wrong. As execution commoditizes, this is where the residual labor premium concentrates. It is real, but it's a narrower base than the execution premium it's replacing, so it doesn't automatically re-employ everyone the first bucket displaced.
3. Trust, relationships, and the standing to make commitments stay with humans. Whoever owns the customer relationship, the demand, and the reputation still captures value the model can't — and this is often the firm (as brand and contract), sometimes the individual (as the trusted name clients follow). This is a large part of why the surplus tends to accrue to ownership rather than to the execution layer once execution is cheap.
4. Physical action in the world stays scarce. Anything that isn't producing tokens — the procedure, the install, the inspection, the in-person negotiation — doesn't get cheaper because drafting did. Labor whose value is physical or transactional keeps its footing longest; this is a real limit on how economy-wide the flip is, and one that could narrow if robotics closes the gap, which is moving far slower than the software side.
5. Capital owners take the residual, and now more of the residual is theirs to take. This axiom doesn't break — it strengthens. When execution stops requiring a scarce, well-paid complement, more of the surplus per unit of output lands with whoever owns the means of production and the demand. The load-bearing question is only which capital captures it — the deploying firm, or the party that sits between the firm and the labor it replaced (next section).

New axioms

1. A third party now sits between labor and capital, and can capture the surplus both assumed was theirs. The model vendor supplies the execution that used to come from the firm's workers. If frontier capability consolidates, the vendor can price to capture much of the productivity gain — extracting from the firm's residual, not just from displaced wages. Labor assumed the surplus was contestable between it and the employer; the employer assumed the residual was theirs. Neither priced in a supplier of cognition with pricing power over the input that replaced the workforce. Whether this materializes hinges on whether frontier execution commoditizes (many near-equal models, price to marginal cost) or consolidates (one or two durably ahead) — the fastest-moving variable in the whole picture, and today's multi-lab parity leans toward commoditization.
2. When execution is abundant, what returns the surplus to labor at all — and through what mechanism? Historically the split tracked bargaining power and mobility, not a neutral default. If the execution premium erodes and the judgment/accountability premium is a narrower base, there's no automatic channel returning the gains to the displaced. Whether that channel is a genuine shortage of verifiers and judgment-holders, policy, ownership broadening, or nothing — is unresolved, and "nothing" is a live outcome, not a pessimistic flourish.
3. The judgment premium is a narrower gate than the execution premium it replaces. The value moves to framing problems and verifying output — but far fewer roles are pure judgment than were pure execution, and judgment has historically been built through years of doing the execution that's now automated. If the ladder's bottom rungs (execution apprenticeship) disappear, the pipeline that produced judgment-holders may not refill, which tightens the scarce good over time rather than democratizing it.
4. Ownership, not wages, becomes the channel to the gains — and access to ownership is unevenly distributed. As returns shift from the execution layer to capital and demand-ownership, the people who capture AI's productivity gains are increasingly those who own — equity, the model deployment, the customer relationship — rather than those who supply labor to it. This reframes the labor-vs-capital question as a who-owns question, and the distribution of ownership is far more concentrated than the distribution of skilled labor was.

Where it breaks

Firms are compressing skilled-execution headcount on the belief that the execution premium was theirs to reclaim (INVALID #1, INVALID #2), while a model vendor with pricing power can quietly capture that same reclaimed surplus as the price of the input that replaced the workforce (NEW #1). The firm optimizes to strip out labor's share and can end up handing it to a supplier that sits upstream of both — the residual it thought it was capturing routes past it. Whether this happens is entirely gated by vendor consolidation, so it's the call most likely to move fast.

Separately, the economy is removing the execution work that judgment was historically built on (INVALID #1) at the same time the scarce, still-human good is the judgment that used to come from doing that work (STILL HOLDS #2, NEW #3). Cutting the apprenticeship rungs to save on execution starves the pipeline for the exact capability that's supposed to be labor's remaining claim on the surplus — the system optimizes against the scarce good it's counting on to re-employ people.

ai-axioms.vercel.app/axioms/society/economy-labor-vs-capital-ai-axioms