When the price of one production input falls, producers substitute toward it and away from more expensive inputs—but simultaneously, inputs that complement the cheaper input become more valuable. This dual dynamic, fundamental to economics since the marginal revolution, explains why AI produces both compression and expansion in knowledge-work labor markets. AI substitutes for execution (writing code, drafting documents, building models), putting downward pressure on wages for workers who primarily execute. It complements judgment (deciding what to build, evaluating quality, identifying problems), putting upward pressure on compensation for workers who primarily direct. Both effects operate simultaneously on different populations, and sometimes on different components of a single worker's job, producing the experiential vertigo of feeling both threatened and empowered by the same tool.
The substitution effect is the more visible and the more politically inflammatory. When factories substituted machines for manual labor in the Industrial Revolution, the displaced workers could see their replacement in physical form—the power loom that one operator could run, doing the work of twenty handweavers. The emotional and political response was immediate, organized, and violent in some cases. The complementarity effect was real but harder to see: cheap textiles expanded the market, increased demand for designers, pattern-makers, logistics coordinators, and retail workers. The total labor demanded by the textile industry eventually exceeded pre-mechanization levels, but the complementarity gains accrued slowly, over decades, to populations distinct from the displaced weavers. The gap between visible substitution losses and invisible complementarity gains is where the Luddite response was born.
Cowen's application to AI identifies the specific task-level movements. AI substitutes most directly for 'routine cognitive work'—tasks that follow explicit rules, that can be specified completely, that involve pattern-matching rather than genuine novelty. These were the tasks David Autor identified in the 2000s as most vulnerable to computerization, and AI has now automated them at a scale that validates Autor's framework. The complementarity shows up in non-routine cognitive work: strategic decision-making, cross-domain integration, the evaluation of outputs under genuine uncertainty. These tasks require not just intelligence but stakes—consequences, responsibility, the willingness to be wrong—and AI possesses competence without consequences. The premium migrates to the humans who can bear the weight of real decisions.
The organizational evidence comes from early adopters. The Berkeley study documented that AI-equipped workers took on wider scope—crossing into domains that were previously other people's jobs—because the execution barrier had fallen. They did not work less; they worked more, on a greater variety of tasks, with higher output per hour. This is the complementarity effect in action: when execution becomes cheap, the demand for the human judgment that directs execution expands to fill the freed capacity. Whether this expansion is sustainable without burning out the humans providing the judgment is an empirical question the economic theory cannot answer in advance, but the direction of the effect—more judgment demanded, higher returns to judgment—is consistent with every historical complementarity case.
The policy prescription is surgical: accelerate the substitution in domains where it produces genuine efficiency gains, but build institutional structures that help displaced workers transition to complementary roles rather than exiting the market. Retraining programs that teach execution skills (coding bootcamps, paralegal certification) are investing in the substituted input and will produce diminishing returns. Programs that develop judgment—case-based business education, cross-disciplinary liberal arts, apprenticeships pairing juniors with experienced decision-makers—are investing in the complementary input and will produce expanding returns. The challenge is that judgment is harder to teach than execution, develops more slowly, and resists the standardized credentialing that institutional education is built to provide. The mismatch between what institutions can teach and what the market will pay for is the core human capital challenge of the AI transition.
The substitution-complementarity framework is fundamental to neoclassical production theory, formalized by John Hicks in the 1930s with the elasticity of substitution concept. When one input becomes cheaper, firms substitute toward it (substitution effect) while demand for complementary inputs rises (complementarity effect). The classic case is capital and labor: when capital becomes cheaper, firms substitute machines for workers in specific tasks while simultaneously expanding operations that require human-machine complementarity, often increasing total employment. Cowen's application to AI draws on this century-old framework but sharpens it with the recognition that the AI substitution is happening at cognitive task-level within individual jobs, not merely at the job level within firms—making the dual effect experientially immediate rather than statistically abstract.
Substitution and complementarity are not alternatives but simultaneous. The same tool that replaces your execution work increases the value of your judgment work—producing both threat and opportunity in the same cognitive domain.
The gap between visible loss and invisible gain breeds resistance. Displaced execution skills generate immediate, concentrated pain; complementary judgment gains accrue slowly, diffusely, to different people—creating political asymmetry.
Historical lag between effects determines transition cost. The Industrial Revolution's complementarity gains took generations to distribute; whether AI's gains arrive faster depends on institutional adaptation speed.
Position yourself in the complementary domain. The only reliable individual strategy is ruthless honesty about which components of your work are substitutable and aggressive investment in the components that are complementary.