Why the optimists are right — until they're not
March 26, 2026
In 1865, a twenty-nine-year-old English economist accidentally wrote the most dangerous argument against everything this book will propose.
His name was William Stanley Jevons. Born in Liverpool to an iron merchant family that went bankrupt in 1847, he had crossed the world at twenty to work as an assayer at the Sydney Mint — because his family needed money more than he needed a degree. When he returned to England and published his first serious work, it made, by his own admission, "no noise."
But the British public had been waiting for someone to say what Jevons said next.
Coal exhaustion was the anxiety of the age. Britain ran on coal the way a modern economy runs on data — it powered everything, and the question of when it would run out was existential. Jevons, still in his twenties, still unknown, sat down to write a short book called The Coal Question.
It turned out to be something else entirely.
James Watt had improved the steam engine so dramatically that it consumed roughly a quarter of the coal the old Newcomen engines required — from forty to forty-five pounds per horsepower-hour down to less than ten. A fourfold gain in efficiency. The numbers should have meant less coal burned.
The numbers said the opposite.
Between Watt's engine entering wide use and the time Jevons was writing, Britain's total coal consumption had increased sixteenfold. Population had merely quadrupled. The country was burning coal at a rate that dwarfed anything the Newcomen era could have imagined — not despite the efficiency gains, but because of them. By 1861, the United Kingdom was producing eighty-six million tons per year and still accelerating.
Jevons wrote the line that would outlive him by more than a century: "It is wholly a confusion of ideas to suppose that the economical use of fuel is equivalent to a diminished consumption. The very contrary is the truth."
That sentence is the entire paradox. Make a resource more efficient to use, and you will use more of it, not less. Efficiency does not conserve. It unleashes.
The Coal Question was published in April 1865. William Gladstone, Chancellor of the Exchequer, devoted half his budget speech to its argument. Parliament appointed a Royal Commission to investigate whether Britain was burning through its prosperity.
And Jevons was wrong.
Not about the mechanism. About the conclusion. The Coal Question argued that Britain was hurtling toward coal exhaustion. It didn't happen. UK coal production kept climbing for another half century, peaking around 1913 — and the decline, when it came, was driven by economics and competing energy sources, not by running out of coal. The mines didn't empty. The market moved on.
Jevons proved that efficiency creates demand. He was so convinced by his own proof that he overshot the implication. The mechanism was airtight. The timeline was not. Keep that distinction. It will matter later.
A hundred and sixty years later, the paradox found a new fuel.
On January 27, 2025, a Chinese Artificial Intelligence (AI) lab called DeepSeek released an open-source language model that matched the performance of America's best systems — at a fraction of the cost. Silicon Valley panicked for about forty-eight hours. Then it did what Silicon Valley always does: it found the upside.
Satya Nadella, the Chief Executive Officer (CEO) of Microsoft — a company that had just invested thirteen billion dollars in OpenAI — posted three words before the dust settled: "Jevons paradox strikes again." His full argument: "As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can't get enough of." This was not spin. It was Jevons, translated into the language of cloud computing. Cheaper AI means more AI. More demand means more jobs building, deploying, and maintaining the systems that run on it. Every tech CEO in the Valley made the same argument within forty-eight hours: efficiency creates demand.
And the current data supports them.
Software engineer job postings were up eleven percent year over year in early 2025, according to a Northeastern University analysis — this despite a year of headlines about AI replacing programmers. Demand is rising, not falling.
Zoom out further. The World Economic Forum (WEF) published its Future of Jobs 2025 report projecting a net gain of seventy-eight million jobs globally by 2030 — one hundred and seventy million new roles created against ninety-two million displaced. The WEF exists to make the future of work sound manageable. That is its institutional function, and you should weight its conclusions accordingly. But the direction of the numbers is consistent with the historical pattern.
David Autor, the MIT economist who wrote the definitive study of automation and employment in 2015, documented two hundred and fifty years of the same result. The title of his paper asked the question directly: "Why Are There Still So Many Jobs?" His answer was that automation creates complementary tasks faster than it eliminates existing ones. The economy does not have a fixed amount of work to distribute. It generates new work in response to new capabilities.
This pattern is not a theory. It is the most thoroughly documented regularity in economic history.
Now look at where AI stands today. Anthropic — which builds Claude, one of the most capable AI systems in existence — published data showing that business and finance professionals face a theoretical automation exposure of 94.3 percent. The current actual automation rate is twenty-eight percent. Software engineers: 94.3 percent theoretical, 35.8 percent actual. The gap between what AI can do and what it is doing is enormous. That gap is where the growth lives — and historically, more productive businesses hire more people, not fewer.
By most industry estimates, fewer than ten percent of companies have deployed generative AI in production. The boom is happening on the strength of the early adopters — and it has already driven AI-related job postings up over two thousand percent in five years, according to Indeed. Prompt engineers, AI safety researchers, machine learning operations specialists, data curators: these roles did not exist a decade ago. When the other ninety percent of companies catch up — and they will, because the cost of not catching up is competitive death — the demand curve steepens further.
The optimists are not wrong. They are early.
Their argument deserves better than dismissal. It deserves a shelf life.
The shelf life of that argument — and when it expires — that's for the book.