Imageegami | Istock | Getty Images
Job growth beat expectations in August, but wage growth has lagged the latest inflation readings, adding urgency to a question economists are only beginning to wrestle with: Could AI pressure workers’ pay before it costs them their jobs?
The latest nonfarm payrolls report isn’t the only source of government data showing deceleration in wage growth. The Bureau of Labor Statistics’ Employment Cost Index showed that inflation-adjusted wages and salaries decreased 0.4 percent year over year through June. And there is also a concerning longer-term trend in the national data: labor’s share of nonfarm business output/income was 52.8% in the second quarter of 2026, the lowest in the series beginning in the first quarter of 1947, according to the BLS productivity report. Some researchers are attributing that to decades of automation, which AI may accelerate.
There are reasons to avoid a rush to judgment. For one, the wage growth of the Covid era was atypical and reflected an extremely tight labor market, with the current level of wage gains closer to the recent historical norm. Higher pay sectors, such as tech and professional services, are also losing jobs while lower-pay sectors such as hospitality and health care have been leading job gains, which pushes down the average pay.
But the current labor market situation is leading more people to focus on the job earnings growth trend line rather than dire warnings like the recent one from Bill Gates about widespread job losses. What’s been learned to date can’t answer this question with authority, but it is moving the topic into a more central place in the AI jobs impact debate.
AI research enters a new phase
A recent study from Apollo Global Management’s chief economist Torsten Slok and his co-author Sania Edlich offers some evidence consistent with AI contributing to slower wage growth. Their research found that workers in occupations classified as highly exposed to AI experienced real-wage growth that was 6.7 percentage points slower after 2023 than workers in less-exposed occupations. At the same time, the study found no statistically significant effect on employment. The authors say the results suggest companies may be capturing some of the productivity gains from AI through wage compression rather than workforce reduction.
The study’s findings are striking, but labor market experts caution that the data is also very limited.
“It’s absolutely the case that AI could be affecting the demand for certain types of jobs,” said Ben Zipperer, senior economist at the left-leaning Economic Policy Institute, and he added that this demand dynamic could be applying downward pressure on wages. But he was quick to add that the Apollo study had too small of a sample size to be convincing.
Zipperer says this research approach can overstate negative effects of AI. Consider the example of coding, a high AI-exposure professional field. If AI causes it to become cheaper to build software by reducing demand for software developers, the saved money doesn’t disappear. It will go elsewhere, including being spent on hiring elsewhere and thereby boosting demand for other workers. “That makes the highly exposed jobs look worse by comparison, even though some of that measured loss is just income increases for other workers,” Zipperer said.
He added that the recent job losses in tech are another factor. The slowdown in tech-related hiring related to earlier over-hiring coming out of the pandemic. That means lower wage growth and weaker employment could partly reflect the same post-pandemic normalization rather than AI alone. “There was a relative slowdown in labor demand for computer programmers and related jobs in the wake of pandemic rehiring that had nothing to do with AI,” Zipperer said.
Apollo said its study did take into account differences in occupations and annual trends in the labor market. But it also caveated its study as “early evidence” that was based on a limited subset of BLS job categories — only 321 of roughly 800 BLS occupations could be used, and only 11 met the study’s high-exposure threshold. Its authors also noted that in addition to digging into the wage effects, its study was fundamentally important as a demonstration that “AI research has entered a new phase, one in which labor market impacts can be measured from observed adoption rather than predicted from theoretical exposure.”
Daron Acemoglu, professor of economics at MIT, says the limited number of occupation-specific AI models remains a factor skewing data on job impacts. “AI models are still developing and they are not widely adopted for many occupations or tasks yet. So some of the displacement effects, as of now, may be exaggerated,” he said.
While Acemoglu said there is no convincing evidence yet for meaningful effects on wages in any given area or for any demographic group, he added that there is certainly “mounting evidence that there is some impact on entry-level jobs” and given how the economy is structured, it is reasonable to argue that wages may ultimately be where the AI impact most visibly materializes. “Ultimately, given that the U.S. labor market is relatively flexible and has a fairly weak social safety net, I expect the impact on wages to be bigger than those on employment,” he said.
Acemoglu’s work on the impact of robots on wages and employment underpins his view of how the labor market will evolve as AI plays a greater role.
“If AI continues to be developed as an automation technology (most notably under the lodestar of AGI), there will be more impacts. Right now we don’t have many easy to use applications relevant for a large number of tasks/industries. Once these are developed, the labor market effects will be multiplied,” Acemoglu said.
Why ‘AI exposure’ narrative may be the wrong one for jobs
Economists remained concerned that the lingo of “AI exposure” is itself limiting.
“The problem is knowing that an occupation is exposed to AI tells you nothing about what will happen to it; not whether employment rises or falls; not whether wages rise or fall,” said David Autor, a labor economist and head of MIT’s economics department. He says the more relevant question, one he has studied, centers on human expertise, and whether AI is taking on an expert or non-expert aspect of a particular job.
He looked at two occupations that were similar in decades past — accounting clerks and inventory clerks — both “seemingly destined for obsolescence in the computer era,” according to a paper he co-wrote with MIT AI researcher Neil Thompson.
“Both performed many job tasks that economists classify as ‘routine’: tasks that follow explicit rules and procedures, exactly the kind that can be codified in software and executed by computers. … Fast forward to the present, and the trajectories of these occupations have diverged.”
Their research showed that accounting clerks experienced wage gains of 39% compared to 40 years ago, even though employment fell by 32%. Inventory clerks, meanwhile, experienced wage declines of 13% but an employment market that grew by 175%.
“These occupations faced the same technological force but experienced opposite outcomes,” the co-authors wrote. One occupation (accounting clerk) became more specialized and better paid, while the other (inventory) became open to more workers but less lucrative.
This research, Autor says, defies standard automation and AI exposure narratives. “The conventional wisdom is that as tasks are automated, workers in more automated occupations are pushed downward into lower-paid, less expertise-intensive jobs. The reality, as these two occupations illustrate, is more nuanced,” they wrote.
Apollo is not alone in attempting to address the concerns from the perspective of AI exposure and pay, but others have reached different conclusions. A Dallas Fed analysis from February found no direct correlation between overall wages and AI exposure, with one very big caveat: youngers workers with less experience — or as the bank refers to it, less of an experience “premium” — may in fact be subject to wage pressure already.
“For occupations with a very low experience premium, AI exposure has a more negative effect on wage growth since AI substitutes for both entry and experienced workers. The low experience premium suggests there is not much tacit knowledge required for the occupation, so experienced workers are easily substituted by AI,” its researchers wrote.
That may be good news for experienced workers, but is a clear warning for those first entering the workforce:
“The current model of white-collar career progression involves taking an entry level job right out of school and doing codifiable tasks while slowly learning the tacit knowledge to become an experienced worker. Firms are going to find that AI is making this method of employee development cost-ineffective, at least in the short run,” the Dallas Fed concluded. “Of course, leaving new employees off the job ladder is not sustainable in the long run. In the long run, AI adoption will require rethinking how entry-level employees gain experience on the job.”
Reframing human vs. robot
Ultimately, that alludes to what Acemoglu hopes will change in this debate — specifically, the adversarial framing, AI vs. human workers, being jettisoned. If it is not, he worries that we lose an opportunity to think in terms of what he calls “pro-worker AI.”
“The most important thing is that AI does not need to be a pure automation technology,” he said, adding that AI can also create new tasks and new expertise for workers.
But he conceded that at least for now, “that’s not the direction we are heading in.”
A move toward pro-worker AI would require “the right investments from the tech sector and the right policy framework,” both of which would help to produce “better outcomes than this ceaseless race to replace workers.”
Jennifer Huddleston, senior fellow in technology policy at the right-leaning Cato Institute, says she does see indications of a more positive approach from both government and from within the workplace. She pointed to efforts by the Department of Labor to encourage AI education in the workforce, including simple ways to help workers educate themselves on basics as they encounter this new technology, are key to a reframing of the debate. “Such an approach is likely to serve by helping workers with such a transition rather than protecting or targeting a certain industry or jobs,” Huddleston said.
In addition, fears about AI in the workplace often overlook some of the most widely used applications of AI, such as email summary and cybersecurity, “which change to some degree office culture but don’t necessarily result in massive changes to employment,” she said.
“One often underappreciated element is the way AI is leading to potentially new categories of jobs and opportunities for entrepreneurship. This can mean AI is creating jobs through such new opportunities even if the jobs are not directly related to AI itself,” Huddleston added.
—CNBC’s Jeff Cox contributed to this report
