Image Gami | iStock | Getty Images
Job growth in August was stronger than expected, but wage growth was below the latest inflation rate, adding urgency to a question that economists are only beginning to grapple with: whether AI could squeeze wages before it puts workers out of work.
The latest non-farm payrolls report is not the only source of government statistics showing slowing wage growth. Inflation-adjusted wages and salaries fell 0.4% year over year through June, according to the Bureau of Labor Statistics’ Employment Cost Index. There are also worrying long-term trends in the country’s data. According to the BLS Productivity Report, labor’s share of nonfarm output/income was 52.8% in the second quarter of 2026, the lowest level since the first quarter of 1947. Some researchers attribute this to decades of automation, which AI could accelerate.
There are reasons to avoid hasty decisions. First, the rate of wage growth during the coronavirus era is unusual and reflects an extremely tight labor market, with current wage growth rates approaching recent historical norms. High-wage sectors such as technology and professional services are also losing jobs, while lower-wage sectors such as hospitality and health care are driving job growth, pushing down average wages.
But the current state of the labor market is causing more people to focus on trends in employment income growth rather than Bill Gates’ recent dire warning of widespread job losses. While what we have learned so far does not allow us to answer this question authoritatively, it does move this topic to a more central place in the debate about the employment implications of AI.
AI research moves to a new stage
A recent study by Torsten Slok, chief economist at Apollo Global Management, and co-author Sania Edrich provides some evidence consistent with AI contributing to slower wage growth. Their study found that workers in occupations classified as having high exposure to AI had real wage growth 6.7 percentage points slower than workers in occupations with low exposure to AI from 2023 onwards. At the same time, the study found no statistically significant effects on employment. The authors say the results suggest that companies may capture some of the productivity gains from AI through wage compression, rather than through layoffs.
While the findings are surprising, labor market experts warn that the data is also very limited.
“There’s definitely the potential for AI to impact the demand for certain types of jobs,” said Ben Zipperer, senior economist at the left-leaning Economic Policy Institute, adding that this demand dynamic could put downward pressure on wages. But he was quick to add that the Apollo study had too small a sample size to be convincing.
Zipperer said this research approach could exaggerate the negative effects of AI. Consider the example of coding, a specialty with high exposure to AI. Even if AI reduces the demand for software developers and makes software development cheaper, the money saved won’t disappear. This money goes elsewhere, such as being spent on hiring elsewhere, thereby increasing the demand for other workers. “That makes highly exposed jobs look worse by comparison, even though some of that measured loss is just an increase in income for other workers,” Zipperer said.
He added that recent job losses in the technology industry are another factor. The tech hiring slowdown is related to early overhiring due to the pandemic. So lower wage growth and weaker employment may partly reflect a similar post-pandemic normalization, not just AI. “Labor demand for computer programmers and related jobs has been relatively slow due to non-AI-related rehiring effects from the pandemic,” Zipperer said.
Apollo said the study takes into account differences in occupations and annual trends in the labor market. But he also cautioned that the study was “early evidence” based on a limited subset of BLS occupations. Only 321 of the approximately 800 occupations in the BLS were available, and only 11 met the study’s high exposure threshold. The authors also noted that, in addition to delving into the wage implications, the study is fundamentally important as it demonstrates that “AI research has entered a new phase in which labor market impacts can be measured from observations of implementation, rather than predicted from theoretical exposure.”
Daron Acemoglu, an economics professor at the Massachusetts Institute of Technology (MIT), said the limited number of job-specific AI models continues to skew data on the impact on jobs. “AI models are still in their infancy and have not yet been widely adopted by many professions and tasks, so some of the displacement effects may be overstated at this point,” he said.
Acemoglu said there was still no convincing evidence that it would significantly impact wages in specific regions or demographic groups, but there was “growing evidence that there is some impact on entry-level jobs,” adding that given the structure of the economy, it is reasonable to argue that wages will ultimately be the most visible impact of AI. “Ultimately, we expect the impact on wages to be larger than the impact on employment, given the relative flexibility of the U.S. labor market and fairly weak social safety net,” he said.
Acemoglu’s research on the impact of robots on wages and employment supports his views on how the labor market will evolve as AI plays a larger role.
“If AI continues to be developed as an automation technology (particularly under the AGI banner), there will be even more impacts. Currently, there are not many easy-to-use applications relevant to a large number of tasks/industries. As these are developed, the impact on the labor market will be multiplied,” Acemoglu said.
Why the “AI exposure” narrative may be the wrong one for work
Economists remained concerned that the terminology “exposure to AI” itself was limiting.
“The problem is that knowing that an occupation is exposed to AI doesn’t tell us anything about what will happen to that occupation, whether there will be more or less employment, whether wages will go up or down,” said David Orter, a labor economist and chair of the economics department at the Massachusetts Institute of Technology. He says the more relevant questions he has researched focus on human expertise and whether AI will take on the expert or non-expert aspects of a given job.
In a paper he co-authored with Neil Thompson, an AI researcher at the Massachusetts Institute of Technology (MIT), he looked at two jobs that have been similar over the past few decades: accounting clerks and inventory clerks. Both “seem destined to become obsolete in the computer age,” he says.
“Both performed many jobs that economists would classify as ‘routine’; that is, jobs that followed explicit rules and procedures, exactly the kind of work that could be codified in software and executed by computers. … Fast forward to the present, and the trajectories of these occupations diverge.”
According to their research, wages for accounting clerks have increased by 39% compared to 40 years ago, even though employment has declined by 32%. Meanwhile, inventory clerk wages fell by 13%, but the job market grew by 175%.
“These occupations faced the same technological capabilities but experienced opposite outcomes,” the co-authors wrote. One profession (accounting clerk) became more specialized and paid higher wages, while the other (inventory management) could employ more workers but became less profitable.
According to Auter, this study runs counter to the standard automation and AI exposure narrative. “As jobs become more automated, the conventional wisdom is that workers in more automated occupations will be relegated to lower-wage, less specialized jobs. As these two occupations illustrate, the reality is more nuanced,” the researchers wrote.
While Apollo is not alone in trying to address concerns from an AI exposure and pay perspective, other companies have come to different conclusions. A Dallas Fed analysis in February found no direct correlation between overall wages and exposure to AI, but there is one very big caveat. Younger workers with less experience – or, as the bank puts it, less of a “premium” for experience – may actually already be under wage pressure.
“For occupations with very low experience premiums, exposure to AI has an even more negative effect on wage growth, as AI replaces both new and experienced workers. A low experience premium suggests that there is less tacit knowledge required for the occupation, so experienced workers are more easily replaced by AI,” the researchers wrote.
While this may be good news for experienced workers, it’s a clear warning for those entering the workforce for the first time.
“The current model for white-collar career advancement involves taking an entry-level job right out of school and slowly learning tacit knowledge while completing codifiable tasks to become an experienced worker. Companies will find that AI makes this method of developing employees less cost-effective, at least in the short term,” the Dallas Fed concluded. “Of course, removing new employees from the job ladder is not sustainable in the long term. In the long term, AI adoption will require rethinking how entry-level employees gain experience on the job.”
Human-versus-robot reconfiguration
Ultimately, that hints at what Acemoglu hopes to see change in this debate, specifically that the adversarial picture, AI versus human workers, be abandoned. Otherwise, he worries, the opportunity to think about what he calls “pro-worker AI” will be lost.
“The most important thing is that AI doesn’t have to be a pure automation technology,” he said, adding that AI can also create new tasks and new expertise for workers.
But, he acknowledged, “that’s not the direction we’re going in,” at least for now.
The transition to pro-worker AI will require “the right investment from the technology sector and the right policy framework,” both of which will help produce “better outcomes than the constant competition to replace workers.”
Jennifer Huddleston, senior technology policy fellow at the right-wing Cato Institute, said she sees signs of a more aggressive approach from both government and workplaces. He pointed to the Department of Labor’s efforts to encourage AI education in the workforce, including easy ways for workers to gain basic knowledge as they encounter this new technology, as key to reframing the conversation. “Such an approach is likely to be helpful by helping workers transition, rather than protecting or targeting specific industries or jobs,” Huddleston said.
Additionally, concerns about AI in the workplace often overlook some of the most widely used applications of AI, such as email summarization and cybersecurity, which “will change office culture to some extent, but not necessarily lead to major changes in employment,” he said.
“One element that is often underestimated is how AI is potentially ushering in new categories of jobs and entrepreneurship opportunities. This could mean that AI is creating jobs through such new opportunities, even if those jobs are not directly related to AI itself,” Huddleston added.
—CNBC’s Jeff Cox contributed to this report
