Jevons Paradox is about to hit the workforce — and most founders aren't ready for what it means.
In 1865, British economist William Stanley Jevons noticed something counterintuitive. James Watt had built a steam engine that was dramatically more fuel-efficient than earlier designs. The obvious expectation was that Britain would burn less coal. The opposite happened. Coal consumption exploded. Why? Because cheaper energy per unit didn't reduce consumption — it made energy economically viable for entirely new applications. Factories, mills, and railroads that couldn't justify the old cost suddenly could. Efficiency didn't conserve the resource. It detonated demand for it.
That's Jevons Paradox. And it's coming for knowledge work and the pattern has repeated itself for 160 years
When ATMs arrived, the obvious prediction was that bank tellers were done. ATMs did reduce tellers per branch. But because they cut operating costs, banks opened dramatically more branches — urban locations increased 43% over the following two decades. Net result: teller employment rose. The job didn't disappear. It evolved from cash handling to relationship banking, to the work that machines couldn't do.
Spreadsheets tell the same story. When VisiCalc could do in seconds what took a skilled accountant a full day, roughly 400,000 bookkeeping positions disappeared. But 600,000 new accountant positions were created in their place. The BLS counted about 339,000 accountants in 1980. By 2022 it counted 1.4 million. Spreadsheets didn't eliminate demand for financial expertise — they made financial analysis so cheap that every business wanted dramatically more of it.
The mechanism is consistent: when you make something 10x cheaper, you don't get the same amount of activity at one-tenth the cost. You get 50x the activity.
The AI version is already playing out
The clearest live data comes from software development. GitHub Copilot studies show developers finish tasks 55% faster. The reasonable expectation is fewer developers needed. What's actually happening: GitHub reported 43 million pull requests merged monthly in 2025, up 23% year over year. Projects that previously weren't worth starting cross the viability threshold when AI cuts the cost of building. Developers aren't building one thing faster and going home — they're building six.
The World Economic Forum's Future of Jobs Report 2025 projects 170 million new roles globally by 2030, against 92 million displaced — a net gain of 78 million positions. AI and machine learning specialist roles grew 143% year over year in 2025. Not because companies are adding overhead. Because lower costs of execution are unlocking entire categories of work that weren't viable before.
What this means for operators
The Jevons insight isn't that AI won't change anything — it's that the nature of the change is routinely misunderstood. AI absorbs execution. What it can't absorb is judgment, institutional knowledge, accountability, and the ability to own outcomes. As Genpact CEO BK Kalra put it recently, frontier AI handles roughly 80% of a business process effectively. The remaining 20% — edge cases, regulatory nuance, decisions that don't follow patterns — is where actual business value lives.
Klarna learned this the hard way. Their AI chatbot initially handled the equivalent of 700 customer service agents. They cut headcount. Customer satisfaction declined. They began hiring humans back. IBM told a similar story, tripling entry-level hiring after finding the limits of AI deployment. When AI commoditizes basic execution, the premium on genuine human judgment rises.
The founders who get this win bigger
The trap is treating AI as a margin play — a way to do the same things with fewer people. The actual opportunity is using AI-driven efficiency to do things that previously weren't possible. More products shipped. More customer segments served. More experiments run.
Every historical precedent points the same direction: efficiency gains create more total demand, not less. The question for founders isn't how many roles can be eliminated. It's what was previously impossible that you can now build, serve, or solve — and how fast you can staff up to capture the demand that follows.
Jevons didn't predict coal would run out. He predicted it would be consumed faster than anyone imagined. The same dynamic is coming for human knowledge work. The founders who understand that will build something much larger than the ones treating AI purely as a cost-cutting tool.
