Marginal Utility and the Diamond-Water Paradox

In the AI Story

The diamond-water paradox troubled classical economists for centuries. Adam Smith noted the contradiction but could not resolve it within his labor theory of value, which held that a good's worth derives from the labor required to produce it. Water requires minimal labor yet sustains life; diamonds require extensive mining yet serve vanity. The marginalists' insight was that both classical questions—'what is water worth?' and 'what are diamonds worth?'—were malformed. The correct question is always marginal: what is the next unit of water worth to this person in this situation? A man dying of thirst in a desert values the next cup of water infinitely. A man standing beside a river values it at zero. Context and quantity determine marginal value, not intrinsic properties or labor inputs.

Cowen extends marginalist thinking into the AI moment by asking what happens when a general-purpose technology makes one category of input radically abundant while leaving its complement scarce. Execution—writing code, drafting briefs, building models—becomes the abundant input. The market, which prices marginal units, reprices execution toward water-level values. Judgment—the decision of what code to write, which legal strategy to pursue, which model captures relevant dynamics—remains scarce, perhaps becomes scarcer as execution abundance expands the option space requiring evaluation. Judgment becomes the diamond, and the marginal hour spent exercising judgment commands premiums that the marginal hour spent executing can no longer justify.

The unbundling mechanism amplifies the repricing. Before AI, execution and judgment were bundled into single professional roles: the software engineer both decided what to build and built it, the lawyer both formulated strategy and drafted documents. The market paid for the bundle, and the component values were obscured. AI splits the bundle by handling execution independently, exposing each component's marginal value. The discovery—uncomfortable for many practitioners—is that execution was consuming eighty percent of professional time while contributing perhaps twenty percent of the marginal value. The judgment that took twenty percent of the time was generating eighty percent of the value, but the bundle pricing hid the asymmetry. When the bundle splits, so does compensation, and the split is not equal.

For individual workers, the marginal analysis is surgical: identify which components of your work are execution (substitutable) versus judgment (complementary), then ruthlessly shift investment toward the latter. The senior engineer whose expertise is writing performant algorithms is holding a depreciating asset; the engineer whose expertise is knowing which performance trade-offs matter for which users is holding an appreciating one. The marginal revolution's lesson for the AI age is the same as its lesson for the 1870s: ignore the total, interrogate the margin, and position yourself where the next unit is scarce.

Origin

The marginal utility concept emerged independently in the early 1870s from three economists working in isolation: Jevons in England, Menger in Austria, Walras in Switzerland. Each formalized the principle that subjective value depends on the quantity possessed and the intensity of the next want to be satisfied, not on aggregate necessity or labor content. The near-simultaneous discovery is itself evidence of intellectual readiness—the problem had been accumulating pressure for decades, and the solution arrived when economic thinking had developed the conceptual tools (diminishing returns, substitution, equilibrium) to receive it. Cowen's application to AI is original, appearing in his 2024-2025 essays and lectures as he recognized that the substitution-complementarity dynamics governing factor inputs map precisely onto the execution-judgment split AI is producing.

Key Ideas

Value lives at the margin, not the aggregate. Essentials are cheap when abundant, luxuries expensive when scarce—and AI has made execution the essential-but-cheap while elevating judgment to the scarce-and-expensive.

Bundling hides component values. Professional roles bundled execution and judgment into single jobs; AI splits the bundle, exposing that execution consumed most time while contributing least marginal value.

The market prices the next unit, not your sunk cost. The years invested in execution skills do not obligate the market to maintain their premium—only the marginal utility of one more hour of your labor determines its price.

Scarcity determines premium, always. Identify where you operate on the abundance-scarcity spectrum, then migrate toward scarcity before the market forces the migration.