“The proliferation of generative artificial intelligence (AI) has sparked a global debate about its potential impact on the labor market,” said a recent article, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” by Stanford Digital Economy Lab (SDEL) Director Erik Brynjolfsson, Postdoctoral Fellow Bharat Chandar, and Research Scientist Ruyu Chen in its Introduction. “This discourse, across academia, public policy, business, and popular media, spans utopian predictions of enhanced productivity, dystopian fears of widespread job displacement, and skeptical views that AI will have minimal effects on employment or productivity.”
Over the past two centuries, there’ve been periodic fears about the impact of technology-based automation on jobs. In the 1810s, for example, the so-called Luddites smashed the new machines that were threatening their textile jobs. But each time those fears arose in the past, technology advances ended up creating more jobs over the ensuing decades than they destroyed.
Automation anxieties have understandably accelerated in recent years, as AI-based innovations are now being applied to activities requiring cognitive capabilities that not long ago were viewed as the exclusive domain of humans. The concerns surrounding AI’s long term impact on jobs may well be in a class by themselves.
There’s a broad consensus that AI will have a major impact on jobs and the very nature of work, but it’s much less clear what that impact will be. Will AI play out like past technology innovations, — highly disruptive in the near term, but ultimately leading to the creation of new jobs, whole new industries, and a rising standard of living? Or will this time be different, as AI-based innovations end up replacing a large portion of the workforce, — leading to mass unemployment, economic dislocations and social unrest?
“Historically, technologies have affected different tasks, occupations, and industries in different ways, replacing work in some, augmenting others, and transforming still others,” the authors added. “These heterogeneous effects suggest that there may be canaries in the coal mine which are harbingers of more widespread effects of AI.”
Their working paper, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence,” aimed to shed light on whether AI is indeed supplanting human labor based on quantitative evidence by analyzing a large-scale, high-frequency administrative dataset from ADP, the largest payroll software provider in the United States. The data set consists of monthly payroll records from millions of individual workers across tens of thousands of firms.
Their analysis uncovered six facts about the employment impact of AI:
- Employment for young workers has declined in AI-exposed occupations.
- Though overall employment continues to grow, employment growth for young workers in particular has been stagnant.
- Entry-level employment has declined in applications of AI that automate work, with muted changes for augmentation.
- Employment declines for young, AI-exposed workers remain after conditioning on firm-time shock.
- Labor market adjustments are visible in employment more than compensation.
- Findings are largely consistent under alternative sample constructions.
Given that a lot of the recent media attention has been focused on the decline of job opportunities for the youngest workers, let me focus my discussion on the impact of AI on these early career workers. A related Substack article by co-author Bharat Chandar nicely summarized the paper’s findings.
According to Chandar, the paper’s key takeaways are:
- 22–25-year-olds are definitely experiencing declining employment in AI-exposed jobs like software development and customer service.
- While jobs automated by AI are seeing declining entry-level employment, jobs augmented by AI are not.
- Overall, the job market for experienced workers is holding up, but for entry-level workers it has been stagnant.
So, what’s actually happening? “Some note rising unemployment for college graduates, declines in job postings since 2022, or recent tech layoffs. Others point out that some of these trends preceded AI and that lots of other changes were happening in the economy.”
“Why don’t we have a definitive answer on how the job market is changing in the sort of work that AI is most capable of doing? The problem is that the workhorse data sets researchers use to track the labor market weren’t built to analyze such specific occupations or age groups in real time. There is no publicly available data that can specifically tell us about the condition of software developers between the ages of 22 and 25 in May 2025 with a reasonable amount of confidence, for example.”
The Stanford Digital Economic Lab has a relationship with ADP, so the research team decided to use ADP data to study the impact of AI on the labor market. “It turns out these anecdotes in the media are borne out in the ADP data.” While older workers continue to see employment growth employment for younger workers has been nearly flat since late 2022.
Entry-level employment has declined in applications of AI that most automate work, but not in ones that most augment it. The jobs most exposed to AI automation, such as software developers and customer service clerks, show a similar pattern: employment for the youngest workers aged 22-25 declined by nearly 20% between their peak around the time of ChatGPT’s launch in late 2022 and July 2025. Marketing and sales managers also showed a decline for this age group, albeit a considerably smaller one. Employment for 26 to 30 year olds also declined slightly, while there is no noticeable change in employment trends for older groups.
What about jobs that aren’t so exposed to AI? “A good example is nursing, psychiatric, and home health aides, whose work requires them to be in person and often perform physical tasks for patients. We actually see the opposite pattern. … For 22- to 25-year-olds, employment is rising in the least AI-exposed jobs like for health aides but notably declining in the most exposed jobs like software development or customer service. In contrast, for older age groups we see no meaningful divergence in employment patterns by AI exposure.”
While entry-level employment has declined in the applications of AI that most automate work, it has not done so in jobs that most augment work by helping employees learn new skills and become more productive in occupations like management or repairs that are generally based on human-AI collaboration. These jobs have actually seen robust employment growth.
Finally, Why might AI adversely affect exposed entry-level workers more than other age groups?, asked Brynjolfsson, Chandar, and Chen in “Canaries in the Coal Mine?”
“One possibility is that, by nature of the model training process, AI replaces codified knowledge, the book-learning that forms the core of formal education. AI may be less capable of replacing tacit knowledge, the idiosyncratic tips and tricks that accumulate with experience. As young workers supply relatively more codified knowledge than tacit knowledge, they may face greater task replacement from AI in exposed occupations, leading to greater employment reallocation. In contrast older workers with accumulated tacit knowledge may face less task replacement. These benefits of tacit knowledge may accrue less to non-college workers in occupations with low returns to experience. Furthermore, more experienced workers may be more skilled in other ways, making them less vulnerable to substitution by AI tools. An important direction for research is to further model and test these predictions.”