Gen Z will be hurt—and helped—the most by generative AI in these types of jobs

For years, discussion of how AI might replace tasks, the office, and even employees themselves has dominated the cultural zeitgeist. But until now, there hasn’t been much overarching data on how AI is actually changing the workforce.

On August 26, researchers at Stanford published what they’re calling the “largest scale, most real-time effort” to calculate that impact—and it includes some bad news for Gen Zers.

The study was conducted by Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, all researchers at Stanford’s Institute for Human-Centered AI. They used an administrative dataset from ADP, the largest payroll software provider in the U.S., to track monthly payroll records across tens of thousands of firms from late 2022 to July 2025.

That data was supplemented by a calculation of “occupational AI exposure” in a diverse array of fields—essentially using data from prior studies (see here and here) to categorize how much various professions are being impacted by AI technology.

Based on this analysis, the researchers reached six key conclusions on how AI is changing the workforce. Here are the most important findings for employees to understand.

Bad news for Gen Zers in AI-exposed fields

The study shows that, for some early career professionals, AI may indeed be making it harder to find a job—despite the fact that, overall, employment “continues to grow robustly.”

Among workers aged 22 to 25, the researchers uncovered a 13% decline in employment in occupations most exposed to AI, like software developers and customer service representatives, even after controlling for firm-level impacts. For example, by July 2025, employment for software developers aged 22 to 25 declined by nearly 20% compared to its peak in late 2022.

In an interview with Fast Company, Brynjolfsson clarified that his team can’t necessarily draw a hard causal line between companies’ AI exposure and this decline in employment. However, he says, the pattern did remain consistent even when testing against a range of variables, which suggests a likely connection.

“We didn’t do an experiment where we gave some companies LLMs and not other companies, so we’re just observing what happened,” Brynjolfsson says. “But we could rule out some of the main alternative hypotheses. Even if you take out the entire tech industry, for instance, you still see this phenomenon: It’s something going on beyond just tech. Or if you take away all remote work and just look at jobs that are not remote, you see the same thing.”

Good news for older workers

While jobs for young workers in AI-exposed fields have been declining, older workers have been largely shielded from those negative impacts. Employment for mid-level and senior employees in AI-exposed fields has actually increased by 6 to 9 percent from 2022 to 2025, consistent with the growth across the job market at large.

“I think one of the things that the senior people may have had an advantage in is they had a lot of on-the-job learning tips and tricks about how to use AI that aren’t necessarily taught in school,” Brynjolfsson says. “That was something that gave them a differentiation from what the LLMs were able to do. But for all of them—mid-level, senior, junior—I think you have to keep an eye on the rapidly evolving capabilities. This is not the end of AI advances.”

Good news for everyone in an occupation that doesn’t rely on AI (yet)

In jobs that are less exposed to the impacts of AI tech—like nursing aides, maintenance workers, and taxi drivers—employment remained stable or continued to grow. In fact, young workers in these kinds of professions actually saw higher employment growth than their older counterparts.

Not all AI use is created equal

According to the researchers, not all uses of AI are associated with declines in employment. Employment declined for young workers in occupations that largely rely on AI to automate tasks (i.e. complete them in their entirety), whereas employment actually grew in occupations where AI was primarily augmentative (i.e. helping the user complete a task or learn a new skill).

“The reality is that there are a lot more benefits in augmenting and allowing people to do new things,” Brynjolfsson says. “[Automating and augmenting] can be productive, but I think complementing workers, rather than substituting for workers, has been underrated. It was striking to see the data show the same thing.”