The loudest voices on the internet right now are telling you that artificial intelligence is coming for your job. Not eventually. Now. Every week there is a new think piece, a new research paper, a new podcast episode in which someone who builds AI tools for a living warns, in carefully calibrated tones, that AI tools are going to hollow out the workforce and leave millions of people economically stranded.
There is also a counter-argument. It is less exciting, harder to make go viral, and has the unfortunate quality of being supported by most of the historical evidence. The argument goes like this: every major technology in the past two centuries - the steam engine, the tractor, the electric motor, the spreadsheet - was supposed to make large categories of human work obsolete. Every time, it did eliminate specific jobs. Every time, the total number of jobs grew. Not because anyone planned it that way. Because the same force that destroyed old work also made the remaining human effort so much more productive that entirely new categories of work became possible.
This is not a new argument. It is, in fact, a 160-year-old argument about coal.
The background
In 1865, a British economist named William Stanley Jevons was studying the coal industry and noticed something counterintuitive. Engineers were making steam engines dramatically more efficient - the same machines were burning less coal to do the same work. By simple logic, that should have reduced total coal consumption. Instead, coal consumption was exploding upward.
The reason: cheaper, more efficient engines made it economical to use steam power for applications that had previously been too expensive. New factories opened. New shipping routes became viable. New industries were born. The efficiency gain did not shrink the market for coal. It detonated demand for it.
Economists now call this the Jevons Paradox - the counterintuitive principle that making a resource cheaper or more efficient tends to increase total consumption of that resource, not decrease it. When a thing gets cheaper, people use more of it. When people use more of it, the economy built around it expands.
The question running through every serious economic discussion of AI right now is whether the same paradox applies to human labor. If AI tools make each hour of human work more productive, does that reduce the total demand for human workers? Or does it make human effort more economical to deploy - and therefore, more consumed?
The intuition says the first. Most of the history says the second.
What is actually happening
On May 6, 2026, Andreessen Horowitz general partner David George published an essay declaring that the AI job apocalypse is "a complete fantasy" - what he called "unhelpful marketing, bad economics and worse history," rooted in the lump-of-labor fallacy: the mistaken assumption that there is a fixed amount of work to be done, and that if AI does more of it, humans must do less.
George cited four large datasets to back the claim. An Atlanta Fed survey covered roughly 6,000 corporate executives across the United States, the United Kingdom, Germany, and Australia. Over 90% of business managers reported no AI-related impact on employment. A separate US Census Bureau working paper found that only about 5% of firms using AI reported effects on workforce size, and the proportions reporting increases and decreases were nearly equal. The Yale Budget Lab concluded in April 2026 that while anxiety about AI's impact on the labor market was widespread, the data suggested it remained "largely speculative."
The aggregate picture - across millions of jobs, dozens of industries, multiple countries - shows, so far, stability. Not the dramatic destruction that dominates the headlines.
The Jevons logic is showing up in specific sectors already. The cost of AI inference has dropped roughly 92% since early 2023. Rather than shrinking the market for intelligence, that collapse has detonated an explosion of demand. Software engineer job postings, which had been declining alongside overall job postings through early 2025, have since reversed - software engineering job postings are now up 11% year-over-year as companies race to build the systems, integrations, and products that cheaper AI makes newly viable.
The data center construction boom tells the same story in steel and concrete. A company called Sterling Infrastructure, which used to build highways and pivoted to building data centers, went from 3,200 employees in the first quarter of 2025 to 4,400 employees in the first quarter of 2026 - a 37% increase in headcount, driven almost entirely by AI infrastructure demand. Meanwhile, call center employment in the Philippines nearly doubled over the past decade to 2 million workers, through the entire period of AI's rise.
This is the pattern historical disruptions have consistently produced. In the video that inspired this piece, Brazilian economist Fernando Ulrich walked through the agriculture example: in 1850, roughly 70% of the US workforce worked in farming. Mechanization made each agricultural worker vastly more productive. Farm employment collapsed as a share of the workforce. Food prices fell dramatically. And all of that released labor was eventually absorbed into entirely new industries - manufacturing, services, healthcare, technology - that barely existed or did not exist at all in 1850. The economy did not shrink to fill the gap. It grew to fill the gap, and beyond.
The money trail
The economic logic behind the optimists' case runs through a concept called the lump-of-labor fallacy - the incorrect belief that the economy has only so many jobs to offer, and that automation therefore takes jobs away from humans permanently. Most professional economists reject this framing because it treats the economy as a fixed pie. Economies grow. Pies get larger.
When AI reduces the cost of performing a task - legal research, code review, financial analysis - two things happen simultaneously. The direct labor cost of that task falls, which means some workers doing that task may lose their jobs. But the overall cost of the service or product falls too, which means it becomes economically viable in new contexts, at new scales, for new customers who could not afford it before. The lawyers who lose work to AI document review may find that same AI has made legal services cheap enough that millions of new clients can now afford them. The net effect on legal employment is unclear. The net effect on legal services delivered - more people getting legal help - may well be positive.
This is how spreadsheets worked. The video cites data showing that from the early 1980s onward, bookkeeping clerks - workers who manually entered and organized numbers - saw employment fall as spreadsheet software automated their work. But the number of accountants, auditors, and financial analysts grew substantially over the same period, eventually exceeding 1 million jobs in the United States. The spreadsheet did not destroy accounting. It destroyed the boring, repetitive part of accounting and made the analytical, judgment-intensive part economically viable at much larger scale.
The same pattern appears in the Goldman Sachs analysis cited in the video: the jobs most exposed to AI substitution are telephone operators, insurance clerks, and payroll processors. The jobs most likely to benefit are industrial engineers, production managers, lawyers, surgeons, and company executives - roles where AI augments human judgment rather than replacing it. Lower-value, repetitive cognitive tasks get automated. Higher-value, judgment-intensive cognitive tasks get amplified.
There is one important wrinkle in this story that the optimists tend to understate. The Dallas Fed published a study in January 2026 finding that workers aged 22-25 in the most AI-exposed occupations have experienced a 13% decline in employment since 2022. The overall labor market is stable. But bachelor's degree holders now represent a full quarter of all unemployed Americans, a historic high, while high school graduates are finding jobs faster than college-educated workers, which is unprecedented.
The Jevons Paradox can be true at the macro level - more AI, more total work - and still be brutal at the micro level, for the specific cohort of young, educated workers who were counting on entry-level professional jobs to build their careers. More demand for senior engineers does not help a 23-year-old who cannot get the entry-level role that would eventually make them senior.
What people are doing about it
The most direct response happening across the workforce right now is adoption. Workers who have integrated AI tools into their daily work are reporting significant productivity gains - not by doing the same work faster, but by taking on categories of work that were previously beyond their reach or too expensive to pursue.
A technology infrastructure worker quoted in the comments of Ulrich's video described using AI to handle data analysis, code improvement, and process refinement tasks that previously required either more time or more specialized skills. The key phrase: "AI doesn't take my job. It helps a good professional be faster and more productive." That pattern - AI as amplifier for skilled workers - is consistent with what the data shows at the macro level.
Companies are also restructuring roles. Rather than hiring fewer people to do less, many firms are hiring more senior engineers and fewer junior ones, using AI to handle the work that entry-level hires would previously have done while deploying experienced workers on higher-complexity problems. For the companies, this is a productivity gain. For young workers trying to enter the field, it is a blocked door.
Some workers are responding by starting businesses. The video cites a correlation between AI tool adoption since the launch of ChatGPT in November 2022 and a surge in new business applications in the United States - the argument being that AI has lowered the cost of starting and operating a business enough to make entrepreneurship viable for people who would otherwise have taken a salaried job.
Governments are mostly watching. Labor ministries in Europe and the United States have published papers tracking AI's impact on employment, but formal policy responses - retraining programs, transition support, regulation of AI-driven layoffs - remain fragmented and largely inadequate to the pace of change.
The bottom line
The historical evidence and the current data both point in the same direction: AI is not destroying the labor market in aggregate. The Jevons Paradox - cheap resources get used more, not less - is a real force, and it is already showing up in software engineering job postings, data center construction employment, and call center headcount in countries that were supposed to be automated out of existence. But the distribution of that disruption is deeply unequal. Entry-level workers, recent graduates, and anyone in a role defined by repetitive cognitive tasks are bearing the costs of a transition whose benefits are flowing primarily to experienced workers and capital. The macro story is optimistic. The micro story, for a significant number of people, is not.
Timeline
- 1865 - William Stanley Jevons publishes The Coal Question, documenting what would become known as the Jevons Paradox: that efficiency gains in coal-powered engines increased, rather than decreased, total coal consumption in England.
- 1850-2020s - US agricultural employment falls from roughly 70% of the workforce to under 5%, driven by mechanization - without reducing total employment; labor migrates to manufacturing, services, and new industries.
- 1980s - Spreadsheet software begins displacing bookkeeping clerks while simultaneously expanding the market for accountants and financial analysts.
- 1993 - Microsoft introduces VBA (Visual Basic for Applications) into Excel, automating further layers of financial work; US accounting sector employment subsequently rises from under 700,000 to over 1 million jobs by the 2020s.
- November 2022 - OpenAI launches ChatGPT, triggering a surge in new business applications in the United States and a sharp increase in AI infrastructure investment.
- 2023-2026 - The cost of AI inference falls roughly 92%, consistent with the Jevons Paradox dynamic: cheaper intelligence increases total demand for AI-powered services.
- January 2026 - The Dallas Fed publishes research showing workers aged 22-25 in high-AI-exposure occupations have seen a 13% employment decline since 2022 - the clearest data point yet that the disruption is concentrated among young, entry-level workers.
- April 16, 2026 - Yale Budget Lab publishes tracking data on AI's labor market impact, concluding that disruption remains "largely speculative" at the economy-wide level.
- May 6, 2026 - Andreessen Horowitz general partner David George publishes an essay arguing that the AI job apocalypse is a "complete fantasy" rooted in the lump-of-labor fallacy, citing four major surveys showing over 90% of firms reporting no AI-related employment impact.
- May 14, 2026 - Brazilian economist Fernando Ulrich publishes a video applying the Jevons Paradox to AI and labor, drawing on the a16z essay and historical data to argue that AI will increase demand for human work, not destroy it.
- May 17, 2026 - Fortune reports that call center employment in the Philippines has nearly doubled over the past decade to 2 million workers, consistent with the Jevons employment effect during the AI era.
Summary
Who: Workers across all skill levels, with the sharpest impact on entry-level professionals aged 22-25; economists including David George (Andreessen Horowitz) and Torsten Slok (Apollo); researcher Fernando Ulrich.
What: A 19th-century economic paradox - the Jevons Paradox - is being applied to explain why AI tools, by making human labor more productive and therefore cheaper, are likely to increase total demand for human work rather than eliminate it. Current data mostly supports this view in aggregate, with a critical exception for young workers.
When: The debate is intensifying in May 2026, anchored by new research from Yale, the Dallas Fed, and Andreessen Horowitz, published alongside a surge in AI infrastructure investment and a recovery in software engineering job postings.
Where: The United States, where the macro labor data is most complete; the Philippines, where offshore call center employment has continued rising through the AI era; and globally, as AI tools become accessible in any economy.
Why: AI inference costs have fallen 92% since 2023, triggering a Jevons-style expansion in demand for AI-powered services - and, by extension, for the human labor needed to build, manage, and apply them. The question is not whether the total job count will grow, but whether the transition costs are being borne fairly, and whether the people losing entry-level work today will have a path into the higher-skill roles being created.