Will AI create more jobs than it destroys?

"Every prior automation wave created more work than it destroyed."

From the spinning jenny to the spreadsheet, every productivity revolution has destroyed specific jobs while creating more total work, higher real wages, and entire industries no one foresaw. AI is following the same script: it lowers the cost of cognitive work, which expands demand for it. Programmers using AI assistants ship more code, which generates more product, which requires more designers, marketers, support staff, and yet more programmers. The pattern of new categories — prompt engineers, AI safety researchers, model evaluators, AI product managers — is already visible just three years in. The pessimists made the same arguments in 1811, 1933, and 1995, and they were wrong every time.

"This time the substrate of automation is cognition itself."

Every prior wave automated muscle or routine logic, freeing humans to do what only humans could: judge, reason, communicate, create. AI automates exactly those capabilities. The set of work humans can credibly claim as 'safe' is shrinking each model generation, and the re-skilling pipeline that smoothed past transitions assumed the destination jobs existed and were learnable in a few years. When the frontier itself moves faster than mid-career workers can retrain, the historical analogy breaks. The fact that we cannot today name the jobs of 2035 is not reassuring — it is exactly what 'this time is different' would look like from the inside.

"Labor market effects depend on diffusion speed and complement vs. substitute dynamics."

The empirical literature is clear that whether a technology raises or lowers wages depends on whether it complements or substitutes for the worker, and on how fast capital can be redeployed relative to labor. Early data from AI-augmented call centers, software development, and legal research shows productivity gains of 15–55% concentrated in lower-skilled workers, which is a complement-pattern and wage-compressing in a good way. But in domains where AI is a substitute — translation, basic copywriting, mid-tier coding — wages are already falling. The aggregate effect is the sum of these dynamics across the economy, and right now it is genuinely indeterminate.

"Transitions historically harmed a generation before lifting the next."

The aggregate-progress story is true but obscures the cost in human lifetimes. English handloom weavers between 1800 and 1850 saw real wages fall by half and life expectancy drop; their grandchildren were better off, but they themselves never recovered. American manufacturing communities hollowed out by globalization and automation between 1980 and 2015 have not, four decades later, reached previous levels of employment, family formation, or trust in institutions. Aggregate gains tell us nothing about whether the people who bear the costs are the same people who eventually receive the benefits — and historically, they are not.

"Outcomes are policy-determined, not technology-determined."

The same wave of automation produced very different outcomes in different countries. Germany's apprenticeship system, Denmark's flexicurity model, and Singapore's continuous learning credits all softened transitions that devastated less-prepared economies. The instruments that matter — portable benefits, antitrust on AI labor markets, public investment in retraining with real wage replacement, sector-by-sector labor councils, and tax treatment of capital vs. labor — are well known. Whether AI looks like the postwar productivity boom or the Rust Belt depends on whether governments use these instruments in the next five to ten years, not on the models themselves.