AI Is Displacing Employment Opportunities for Young Workers

Economists at Stanford University have uncovered compelling evidence that artificial intelligence is beginning to displace certain jobs. However, the situation is not entirely straightforward: while younger workers in some sectors face replacement by AI, older workers are finding new opportunities emerging.

Erik Brynjolfsson, a professor at Stanford University, along with Ruyu Chen, a research scientist, and Bharat Chandar, a postgraduate student, analyzed payroll data from ADP, the largest payroll service provider in the US, spanning from late 2022, when ChatGPT was introduced, to mid-2025.

The team uncovered several significant trends in the data, notably that the rise of generative AI was linked to a reduction in job opportunities for younger workers in industries already identified as being at high risk due to AI automation (like customer service and software development). Employment for workers aged 22 to 25 fell by 16 percent in these sectors.

This new research paints a complex picture of AI’s influence on the labor market. Although advancements in artificial intelligence have often been accompanied by grim predictions about job losses, there has been little data to substantiate these claims. For instance, the relative unemployment rate for young graduates began declining around 2009, well ahead of the current surge in AI. Furthermore, sectors that were thought to be at risk, like translation, have actually experienced job growth in recent years.

“It’s always challenging to understand [the situation] when you’re focusing solely on a particular company or just hearing stories,” Brynjolfsson states. “Thus, we aimed to approach it in a much more systematic manner.”

By analyzing payroll information, the Stanford researchers found that AI’s impact is more closely related to a worker’s experience and skills rather than the specific nature of their job. More seasoned employees in fields where generative AI is becoming prevalent were largely shielded from job loss, with their opportunities either remaining stable or slightly increasing. This aligns with what some software developers have previously indicated about AI’s influence on their sector—specifically, that monotonous, repetitive tasks, such as coding to connect to an API, have become more amenable to automation. The study also suggests that while AI is indeed displacing jobs, it has not, at least so far, led to wage reductions.

The research team took into account various potentially confounding elements, including the Covid pandemic, the increase in remote work, and recent layoffs in the tech sector. They discovered that AI still exerts an influence even when these factors are considered.

Brynjolfsson emphasizes that the study provides insights into maximizing the advantages of AI throughout the economy. He has long advocated for reforms in the tax system to avoid incentivizing companies that replace human labor with automation. He also suggests that AI firms create systems that foster collaboration between humans and machines.

In a paper published in June, Brynjolfsson and fellow Stanford researcher Andrew Haupt argued that AI companies should establish new “centaur” benchmarks for measuring human-AI collaboration, encouraging greater emphasis on augmentation rather than simple automation. “I believe there are still numerous tasks where humans and machines can outperform [AI alone],” Brynjolfsson asserts.

Some experts predict that increased cooperation between humans and AI could characterize future labor markets. Matt Beane, an associate professor at UC Santa Barbara who studies AI-driven automation, envisions that the AI revolution will spur demand for augmentable work, as managing AI output becomes increasingly vital. “We’ll automate as much as possible,” Beane states. “But that doesn’t imply that there won’t be a vast amount of augmentable work remaining for humans.”

However, AI is advancing rapidly, and Brynjolfsson warns that its effects on younger workers may extend to those with more experience. “What we need is to create an early-warning dashboard to monitor this in real time,” he advises. “This is a highly consequential technology.”

This is an edition of Will Knight’s AI Lab newsletter. Read previous newsletters here.