The idea that artificial intelligence would immediately eliminate jobs has been repeated for months in corporate presentations, consultancy reports, and alarmist headlines. But the aggregated data from the U.S. labor market still do not confirm this scenario. The most suggestive insight comes from Apollo Global Management: if AI makes certain professional tasks cheaper, something like Jevons’ paradox may be occurring, where efficiency gains do not reduce resource consumption but increase it.
The graph shared by Apollo, based on weekly private employment data from ADP, shows an acceleration in job creation in the United States during spring 2026. It does not alone prove that AI is creating jobs, but it challenges the simpler thesis of mass substitution. If technology were destroying jobs broadly and rapidly, the data should start reflecting this more clearly.
What the Jevons paradox really says
William Stanley Jevons observed in the 19th century that efficiency improvements in steam engines did not reduce total coal consumption in the UK. Quite the opposite occurred. By making coal use cheaper and more profitable, new industrial uses, factories, and activities that were previously unviable emerged. Efficiency reduced the unit cost but multiplied the total demand.
Applying this to AI, the reasoning is similar. If generating code, analyzing documents, producing reports, reviewing contracts, preparing campaigns, or automating processes costs less, it does not necessarily mean less human labor will be needed. It may happen that many more things are done: more software, more services, more audits, more content, more analysis, more startups, and more customized products.
This is the point that Torsten Sløk, Chief Economist at Apollo, advocates. His thesis is that AI is lowering the cost of certain professional tasks and, in doing so, expanding the market. The consequence would not be an automatic job decline but an increased demand for profiles capable of using AI to produce more.
The comparison with PCs in offices helps to understand this. In the late 1980s and early 1990s, it was also said that computers would destroy a huge portion of administrative work. The reality was more complex: tasks disappeared, functions changed, new professional categories emerged, and the ability to produce information, manage data, and coordinate companies increased. Employment was not frozen; it was reorganized.
Data still do not support a labor collapse due to AI
Apollo has been especially firm in stating that there is “zero evidence” of employment losses attributable to AI in the aggregated data. Meanwhile, ADP recorded 109,000 additional private sector jobs in April 2026, and its weekly NER Pulse indicator showed a positive moving average in May. These figures should be interpreted with caution, as the labor market depends on many factors: interest rates, consumption, business investment, public spending, productivity, demographics, and economic cycles.
Nevertheless, the message is important. AI is permeating companies, but the aggregated labor impact many anticipated does not clearly emerge in general statistics. Some companies have cited AI in restructuring or adjustment plans, but that alone does not prove a massive net job destruction caused by this technology.
It’s also important to differentiate automation of tasks from job destruction. AI can replace specific parts of a job without eliminating it entirely. A lawyer might review documents faster, a programmer might generate draft code, an analyst might prepare reports more quickly, and a support team could respond better with the help of assistants. Often, the question is not whether the position disappears but how the content of the work changes.
Available academic evidence points in this nuanced direction: generative AI can lead to significant productivity improvements in certain tasks, but its impact on employment depends on the sector, the level of adoption, work organization, and whether technology is used to replace people or to expand output.
The risk doesn’t disappear: it changes form
Saying there is no evidence of a broad labor collapse doesn’t mean denying the risks. Some profiles may come under pressure, especially in repetitive office tasks, basic writing, first-level support, simple translation, document review, or routine programming. It might also be more challenging for junior positions if companies try to cover part of the work that previously served as an entry point to the market with AI.
Jevons’ paradox also does not guarantee that everyone will benefit. It could increase the overall demand for work while displacing those who cannot adapt. Efficiency often redistributes value. Faster benefits tend to accrue to companies with capital, data, processes, and technological integration capacity. Workers and small businesses that do not adopt AI may face worse conditions compared to those that leverage it effectively.
The key point for managers and professionals is not to wait for the market to decide. It’s understanding that AI can raise the minimum expected productivity level. If a person working with AI can accomplish in one morning what previously took two days, the company may not always reduce staff; it might also ask for more projects, more analysis, more clients served, or more improvement cycles.
That’s Jevons applied to intellectual work: when the cost of producing a unit of knowledge decreases, the demand for knowledge can grow. Fewer programs are not created because programming is more efficient. The goal is to produce more software. Fewer analyses are not performed because a model accelerates them; instead, more scenarios are examined. Less content is not generated because tools assist; the competition is to publish more, better, and faster.
Productivity will truly be the new frontier
Discussions about AI and employment are often framed as a binary battle between job substitution and creation. The reality will be more complex. Some sectors will see AI reduce employment, others will see it increase, and many will experience changes in tasks, wages, entry profiles, and internal organization.
For companies, the key will be to measure actual productivity, not just adoption. Using AI does not automatically mean improvement. What matters is whether it reduces times, increases quality, lowers errors, accelerates sales, improves support, or enables launching products previously unviable. Jevons’ paradox works when efficiency creates new demand; it does not when only a costly tool is added to poorly designed processes.
For workers, the message is straightforward. AI does not eliminate the need for judgment, context, communication, responsibility, and business knowledge. But it does reward those who know how to use it to enhance their capabilities. Just as the PC didn’t replace all office workers but penalized those who failed to learn to work with it, AI may become a basic tool for professional productivity.
Current data do not justify the narrative of an imminent labor apocalypse. Nor do they justify complacency. AI is changing the market from within, but for now, it appears more as a technology expanding the capacity to do work than a machine that eliminates all work instantly.
The question is no longer whether AI will destroy jobs in the abstract. The real question is who will know how to turn that efficiency into more activity, more products, more services, and more value.
Frequently Asked Questions
What is Jevons’ paradox?
It is an economic idea suggesting that efficiency improvements can increase the total consumption of a resource because they make it cheaper and open new uses.
How does it apply to artificial intelligence?
If AI reduces the cost of professional tasks, it can increase the overall demand for analysis, software, content, automation, and services rather than automatically reducing employment.
Is there evidence that AI is causing massive job destruction?
Available aggregated data do not yet show a broad labor collapse attributable to AI, though there are specific impacts on tasks, profiles, and companies.
What should professionals do in response to this change?
Learn to use AI as a productivity tool, strengthen judgment, specialization, communication, and business knowledge. The advantage will come from combining human experience with automation.