Ever since ChatGPT burst onto the scene in late 2022, we’ve been hearing increasingly dire predictions about AI’s impact on jobs. As AI systems became capable of writing software, analyzing documents, answering questions, and performing other cognitive tasks once thought to require humans, warnings of massive technological unemployment have become increasingly common.
These fears have a long history. Over the past two centuries, virtually every major wave of technological innovation has raised concerns about the destruction of jobs. In the 1810s, the so-called Luddites famously smashed textile machinery they believed threatened their livelihoods. Yet over the ensuing decades, technological advances repeatedly created new jobs and industries even as they eliminated or transformed existing ones.
Could AI be different?
There is broad agreement that AI will have a major impact on jobs and on the very nature of work. What’s much less clear is what that impact will ultimately be. Will AI follow the historical pattern — highly disruptive to particular occupations and workers, but ultimately creating new jobs, transforming industries, and raising living standards? Or will AI’s ability to perform increasingly sophisticated cognitive tasks enable it to replace such a large portion of human labor that this time will really be different?
We’re beginning to get some answers. And so far, the evidence points to something considerably more complicated than an AI jobs apocalypse.
Prepare for an AI Jobs Apocalypse?
“The launch of ChatGPT in 2022 ignited the artificial-intelligence boom — and elicited a chorus of warnings from AI bosses of an impending jobs apocalypse,” wrote The Economist in “Prepare for an AI jobs apocalypse,” the lead article in its May 16, 2026 issue. “Never mind that they have reason to talk up the disruptiveness of their products, or that rich-world employment is near all-time highs — the dark message has landed.”
Public anxiety about AI-driven unemployment has been rising. Seven in ten Americans believe AI will make it harder to find work, while nearly one-third fear for their own jobs. The concerns have been amplified by a slowdown in hiring for recent college graduates, especially in technology-related occupations such as software development.
But economists are considerably less pessimistic. “Labour markets constantly change,” noted The Economist. “Today’s offices would be unrecognisable to a worker from 50 years ago. Never in modern history has technological progress hurt the overall demand for human labour.”
“If technology displaces workers from some occupations,” economists argue, “it enriches others, who then spend their gains on goods and services that create new employment.”
“Mass unemployment induced by AI would be unprecedented.”
So, should we expect an AI jobs apocalypse?
Canaries in the Coal Mine
To shed light on this question, Stanford Digital Economy Lab Director Erik Brynjolfsson, Postdoctoral Fellow Bharat Chandar, and Research Scientist Ruyu Chen published “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence” in August, 2025.
The paper attracted considerable attention because it offered quantitative evidence of AI’s employment impact based on a large-scale administrative dataset from ADP, the largest payroll software provider in the United States.
Their research found that although overall employment continued to grow, employment among young workers ages 22–25 had declined in occupations highly exposed to AI. Particularly noteworthy, the declines were concentrated in applications of AI that automated work, while changes were much more muted when AI augmented human capabilities.
A year later, in August 2026, the Stanford authors released a revised version incorporating additional data.
“Historically, technologies have affected different occupations in heterogeneous ways, replacing work in some, complementing others, and transforming still others,” they wrote. “However, the specific capabilities of generative AI, particularly its proficiency with codified knowledge work, suggest there may be canaries in the coal mine which will be affected first.”
Their updated findings are both reassuring and concerning. They found no evidence of widespread, economy-wide job displacement. But beneath the aggregate numbers they uncovered a striking divergence: employment of young workers in AI-exposed occupations now stands 19% below where it would be had it kept pace with their less-exposed peers, with no comparable gap for experienced workers.
Moreover, this divergence has widened steadily since they first documented it in August 2025 and appears primarily through reduced hiring rather than increased separations.
This is an important distinction. The first visible employment effects of AI may not arrive in the form of millions of workers suddenly losing their jobs. They may appear much more quietly, as companies discover that experienced employees equipped with AI can perform work that previously required additional junior employees.
The researchers also found that employment declines are concentrated in occupations where AI substitutes for human tasks. Where AI complements human workers, employment has generally been flat or rising.
Codified versus Tacit Knowledge
Why might AI disproportionately affect entry-level workers?
“One possibility is that AI more effectively substitutes for knowledge that has been digitized and codified — the core of formal education — while acting as a complement to the tacit, experience-based knowledge held by senior employees,” wrote the Stanford researchers. “Consequently, AI may be automating the checkable process-intensive tasks that historically justified entry-level headcount, while increasing the leverage of experienced staff.”
The distinction between codified and tacit knowledge may prove particularly important in understanding AI’s impact on work.
Codified or explicit knowledge can be articulated, written down, digitized, and taught in schools and universities. It’s precisely the kind of knowledge on which large language models have been extensively trained and at which generative AI increasingly excels.
Tacit or implicit knowledge is different. It’s difficult to articulate and is generally acquired through experience, insight, and judgment. Experienced workers understand not only the formal rules of their occupations but also when those rules apply, how organizations actually function, and how to deal with ambiguous situations for which there may be no clearly defined answer.
This creates an intriguing paradox. For decades, a college education offered considerable protection from technological displacement because knowledge-intensive jobs were generally harder to automate than routine manual work. Generative AI may partially reverse that relationship because some of the knowledge acquired through formal education is precisely the explicit, codified information that AI systems are increasingly capable of reproducing.
Meanwhile, the tacit knowledge accumulated by experienced workers may become more valuable when combined with AI.
Automation versus Complementarity
The Stanford findings reinforce an important economic distinction about the direction of AI innovation.
Employment declines among young workers are concentrated in occupations where AI is primarily used to automate tasks. In occupations where AI is used to complement human capabilities, employment has generally been stable or increasing.
This suggests that asking whether “AI will destroy jobs” may be the wrong question. A more important question is whether companies will primarily use AI to automate tasks previously performed by people or to give workers new capabilities that enable them to perform more valuable work.
Both will undoubtedly happen. Companies have always adopted technologies that automate tasks and reduce costs. But throughout history, consequential technologies have also complemented human capabilities, enabling workers to perform new tasks and companies to create new products and services. The balance between automation and complementarity may therefore play a major role in determining AI’s ultimate impact on employment.
The Stanford findings raise another important longer-term question.
Experienced workers didn’t begin their careers with tacit knowledge. They acquired it through years of working, solving problems, making mistakes, and learning from more experienced colleagues.
Traditionally, entry-level positions have provided the first rung of that career ladder. But what happens if AI increasingly performs the routine cognitive tasks that have historically been assigned to junior employees?
This could also pose a challenge for universities. If AI can increasingly perform tasks based on the codified knowledge traditionally acquired in college, preparing students for work will have to involve more than transmitting that knowledge. Young workers will need opportunities to develop the judgment, context, interpersonal skills, and tacit knowledge that become increasingly valuable in human-AI collaboration.
The Jobs Apocalypse Is Postponed
Four months after warning readers to “Prepare for an AI jobs apocalypse,” The Economist reached a considerably more optimistic conclusion.
“The jobs apocalypse is postponed. An AI jobs boom is here,” declared an article in its September 5 issue, concluding that “the initial effects of the technology on employment look positive.”
“Perhaps machines will make many humans unemployable eventually — but there is no sign of it yet,” said the article.
The U.S. unemployment rate was 4.1%, lower than in almost 90% of months over the previous half-century. Young workers, often viewed as AI’s first victims, also appeared to be doing relatively well in the aggregate: the gap between unemployment among 20–24-year-olds and the overall rate was close to a multi-decade low.
Some occupations once believed particularly vulnerable to AI have actually been growing. Between 2023 and 2025, employment among paralegals rose by about 11% and among market-research analysts by 6%, compared with national employment growth of around 2%.
At first glance, these findings might appear to contradict the Stanford research. But both can be true. The overall labor market can remain healthy while particular groups experience significant disruption. Employment in an occupation can grow overall while entry-level hiring within that occupation weakens.
History also suggests that AI will create jobs we cannot yet imagine. More than six million Americans now work in computer-related occupations that didn’t exist in the pre-computer era, noted The Economist. Millions more work in industries such as e-commerce, the gig economy, and content creation that barely existed a few decades ago.
What Happened to the AI Jobs Apocalypse?
So, what happened to the AI jobs apocalypse?
For now, it’s been postponed. Employment remains high, unemployment remains relatively low, and occupations once predicted to be among AI’s earliest victims continue to employ millions of people. If history is any guide, AI will also create new jobs and occupations that we cannot yet imagine. But that doesn’t mean that AI isn’t already changing the labor market.
The Stanford evidence suggests that its effects may first appear not as mass layoffs but as slower hiring for particular groups — especially young workers in occupations where AI can automate the codified knowledge and routine cognitive tasks traditionally performed by entry-level employees. At the same time, workers whose experience and tacit knowledge complement AI may become considerably more productive and valuable.
The future of work will therefore depend not only on what AI can do, but on how companies choose to leverage AI in the workplace — whether primarily to replace human capabilities or to complement them and create new products and services.
And there’s an important challenge hidden inside today’s relatively reassuring employment numbers. Every experienced worker was once an inexperienced worker. If AI begins removing the lower rungs of the career ladder, we’ll have to find new ways for young people to acquire the tacit knowledge and experience that AI cannot easily replicate.
The aggregate numbers tell us there’s no AI jobs apocalypse — at least not yet. But the more granular evidence tells us something equally important: the transformation of work has already begun.
