AI + ML
Machine learning models are automating some entry-level roles
AI-pocalypse Researchers at the Stanford Digital Economy Lab found that workers ages 22 to 25 in jobs most exposed to AI, such as software developers, experienced a 13 percent relative decline in employment compared to other occupations.
This is due to stable or increasing employment of experienced workers in the same jobs and workers of all ages in jobs with less contact with AI.
The findings, detailed in a research paper entitled “Canaries in the coal mine? Six facts about artificial intelligence’s recent employment impact,” echo previous reports of a tough job market for recent graduates in computer-related fields.
For example, according to data released by the Federal Reserve Bank of New York in February 2025, the unemployment rate for new graduates was 5.8 percent, while the average unemployment rate for workers was 4.0 percent. Additionally, the U.S. unemployment rate for people who studied computer science was 6.1 percent and jumped to 7.5 percent for computer engineering.
According to SignaFire’s May report, graduate hiring has fallen to 50% of pre-pandemic levels. The report attributes this shift in part to AI and described the situation as a hiring reset, with companies seeking more experienced workers and hiring fewer entry-level workers.
However, this seems to have remained the status quo for some time. Last September, James O’Brien, a computer science professor at the University of California, Berkeley, published a post on LinkedIn confirming a Wall Street Journal report about technology jobs drying up or moving overseas, as some commenters pointed out.
“Previously, Berkeley CS graduates received multiple job offers that were attractive in terms of job type, location, salary, and employer, even if they were not the top students,” he wrote. “However, I have had high-achieving students, such as those with a 4.0 in-major GPA, reach out to me concerned about zero job offers. I believe this trend is likely irreversible and part of a broader trend impacting nearly every employment sector.”
A new Stanford Digital Economy Lab paper by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen is based on an analysis of monthly administrative data provided by US payroll company ADP from January 2021 to July 2025, covering millions of workers and tens of thousands of businesses.
The study focuses on six findings and relies on data from not-totally-independent AI companies like Anthropic. [PDF] OpenAI defines occupational exposure to AI. Microsoft, another company actively participating in this race, has also published a hit list of AI roles. The jobs most at risk are said to include customer service representatives, accountants and auditors, software developers, secretaries and administrative assistants.
The researchers say the diminishing prospects for jobs exposed to AI are contributing to slower overall employment growth for the 22-25 age group, even as employment for other workers is growing at a higher rate. The authors also argue that not all uses of AI involve job losses. The impact will depend on whether the use of AI automates roles or simply enhances and supports jobs.
They also note that AI currently appears to be having a bigger impact on employment than wages. So just because the job market for computer science students has softened recently doesn’t mean salaries are going down overall.
Finally, they say their findings cannot be explained by other factors such as offshoring or outsourcing of entry-level computers or support roles.
“We find that our results are not driven solely by computing occupations or occupations susceptible to remote work or outsourcing,” the authors said.
The authors acknowledge that other researchers have reached different conclusions. For example, an April study (A Humlum) of Danish workers found that generative AI has no real impact on jobs or wages, and a February study (M Hampole) found that the impact of AI is limited, balanced by lower demand for some roles and higher productivity for others. But the Stanford University authors say these papers are flawed.
“These previous papers used data that were not granular or immediate enough to reliably study changes in employment with exposure to AI and age,” the authors argue. “In contrast, this paper uses large-scale, near real-time data to take a step toward resolving the ongoing debate about the employment implications of AI for young workers.” ®
