In 1865, British economist William Stanley Jevons observed that improvements in the efficiency of steam engines were increasing Britain’s coal consumption instead of decreasing it.
As engines became more efficient and coal production became cheaper, more industries adopted coal and expanded their operations.
More than 160 years later, Jevons’ observation, now known as Jevons’ paradox, has become one of the most frequently cited frameworks in the debate over artificial intelligence and employment.
Applying this argument to knowledge work, if AI lowers the cost of a cognitive task, the demand for that task may increase enough to increase the total amount of work performed. Thorsten Slok, chief economist at Apollo Global Management, made this claim directly in an April client note, calling it the “Jevons jobs effect.”
“When the steam engine made coal more efficient, Britain burned more coal rather than less,” Throck wrote. “The same pattern is happening with cheaper legal services, consulting services, and financial services.”
Aaron Chatterjee, chief economist at OpenAI, made a similar point last month.
“When the price of something goes down and demand is elastic, people might buy more of it,” Chatterjee said at the European Central Bank’s ECB Forum on Central Banking on June 30.
Corporate credit card company Ramp and workforce analytics company Revelio Labs have found empirical evidence for this hypothesis. They published a study, “A New Look at AI’s Impact on Jobs,” which looked at 21,559 U.S. companies from January 2021 to February 2026 and correlated observed AI spending with employee records.
The study found that companies with active adoption, defined as those that spend an average of $33.67 per month per employee on AI in the first three months, experienced a 10.2% increase in employee headcount over the two years following adoption. The number of entry-level employees increased by 12%. Low-adoption companies spent an average of $2.78 per employee per month, but did not see any statistically significant changes in employment. The results were seen across engineering, sales, management, and customer service, rather than focusing on a single function.
Also read: Companies spending heavily on AI are adding employees, not cutting them.
Companies that spend the most on AI also hire the most
Ramp Rebellio’s findings contain important warnings, the Human Resources Director (HRD) reported on July 3. Almost all of the headcount growth was concentrated in companies in the technology sector, and the study looked only at white-collar workers. The authors focused on correlation rather than causation. Companies that are actively adopting AI are already large, rapidly growing, and technologically sophisticated before they do so.
“Even when comparing companies that are growing at a similar pace, growth accelerates after implementing AI compared to companies that have not yet adopted AI,” said Ara Karazian, chief economist at Lamp and co-author of the study, in the report.
PwC’s 2026 Global AI Job Barometer analyzed more than 1 billion job ads across 27 countries and found similar patterns on a larger scale. Companies most exposed to AI saw a 52% increase in employee headcount compared to the 2018 baseline, compared to a 36% increase in companies least exposed to AI, and wage growth of 24% compared to 17%. The top 20% of companies that adopted AI achieved an average 163% increase in labor productivity over the same period.
PwC described two outcomes. “Specialized” roles, where AI handles routine tasks and human judgment is central, are growing twice as fast as roles where AI primarily simplifies tasks.
data is consistentbut the debate is not settled
Jevons’ employment effect is a theory of long-term dynamics, and short-term data are mixed. This year, companies have announced around 102,000 AI-related layoffs, with technology departments accounting for a third of those cuts.
A Stanford University study found that workers ages 22 to 25 in the jobs most exposed to AI have seen a 16% decline in employment since the spread of generative AI, a finding cited by the Dallas Fed in a January research report.
Dallas Fed economists Tyler Atkinson and Shane Yamko conducted their own analysis using Census Bureau survey data and found a small but consistent effect. The employment share of young workers in the occupations most exposed to AI fell from 16.4% to 15.5% from the end of 2022 to September 2025, but this decline only explains about 0.1 percentage point of the broader rise in the unemployment rate.
Both studies tracked patterns in delays in hiring new entrants, rather than in layoffs.
The Ramp Rebellio study said a 24-month period may be too short to capture long-term reallocation effects, and the sample may be biased toward larger and more technologically sophisticated companies.
Forecasts from established institutions point in different directions depending on the time period and methodology. Data from Ramp, Revelio, and PwC shows that among the companies spending the most on AI, employment is not declining. Areas where AI augments, rather than replaces, human judgment are seeing the fastest growth in areas with the most investment.
For all of our coverage of PYMNTS AI, subscribe to our daily subscription AI Newsletter.
