How Badly Is AI Cutting Early-Career Employment?

As AI tools become more common in people’s everyday work, researchers are looking to uncover its effects on the job market—especially for early-career workers.

A paper from the Stanford Digital Economy Lab, part of the Stanford Institute for Human-Centered AI, has now found early evidence that employment has taken a hit for young workers in the occupations that use generative AI the most. Since the widespread adoption of AI tools began in late 2022, a split has appeared, and early-career software engineers are among the hardest hit.

The researchers used data from the largest payroll provider in the United States, Automatic Data Processing (ADP), to gain up-to-date employment and earning data for millions of workers across industries, locations, and age groups. While other data may take months to come out, the researchers published their findings in late August with data through July.

Although there has been a rise in demand for AI skills in the job market, generative AI tools are getting much better at doing some of the same tasks typically associated with early-career workers. What AI tools don’t have is the experiential knowledge gained through years in the workforce, which makes more-senior positions less vulnerable.

These charts show how employment over time compares among early-career, developing, and senior workers (all occupations). Each age group is divided into five groups, based on AI exposure, and normalized to 1 in October 2022—roughly when popular generative AI tools became available to the public.

The trend may be a harbinger for more widespread changes, and the researchers plan to continue tracking the data. “It could be that there are reversals in these employment declines. It could be that other age groups become more or less exposed [to generative AI] and have differing patterns in their employment trends. So we’re going to continue to track this and see what happens,” says Bharat Chandar, one of the paper’s authors and a postdoctoral fellow at the Stanford Digital Economy Lab. In the most “AI exposed” jobs, AI tools can assist with or perform more of the work people do on a daily basis.

So, what does this mean for engineers?

Software Developers Among Most AI-Exposed

With the rise of AI coding tools, software engineers have been the subject of a lot of discussion—both in the media and research. “There have been conflicting stories about whether that job is being impacted by AI, especially for entry-level workers,” says Chandar. He and his colleagues wanted to find data on what’s happening now.

Since late 2022, early-career software engineers (between 22 and 30 years old) have experienced a decline in employment. At the same time, midlevel and senior employment has remained stable or grown. This is happening across the most AI-exposed jobs, and software engineering is a prime example.

Since late 2022, employment for early-career software developers has dropped. Employment for other age groups, however, has seen modest growth.

Chandar cautions that, for specific occupations, the trend may not be driven by AI alone; other changes in the tech industry could also be causing the drop. Still, the fact that it holds across industries suggests that there’s a real effect from AI.

The Stanford team also looked at a broader category of “computer occupations” based on the U.S. Bureau of Labor classifications—which includes hardware engineers, Web developers, and more—and found similar results.

Growth in employment between October 2022 and July 2025 by age and AI-exposure group. Quintiles 1–3 represent the lowest AI-exposure groups, which experienced 6–13 percent growth. Quintiles 4 and 5 are the most AI-exposed jobs; employment for the youngest workers in these jobs fell 6 percent.

Augmentation vs. Automation

Part of the analysis uses data from the Anthropic Economic Index, which provides information about how Anthropic’s AI products are being used, including estimates of whether the types of queries used for certain occupations are more likely to automate work, potentially replacing employees, or augment an existing worker’s output.

With this data, the researchers were able to estimate whether an occupation’s use of AI generally complements employees’ work or replaces it. Jobs in which AI tools augment work did not see the same declines in employment, compared with roles involving tasks that could be automated.

This part of the analysis was based on Anthropic’s index alone. “Ideally, we would love to get more data on AI usage from the other AI companies as well, especially Open AI and Google,” Chandar says. (A recent paper from researchers at Microsoft did find that Copilot usage aligned closely with the estimates of AI exposure the Stanford team used.)

Going forward, the team also hopes to expand to data on employment outside of the United States.