AI and entry-level jobs: The data tells two stories

What this is about

Young college graduates in the United States are having an unusually difficult time entering the workforce. According to the Federal Reserve Bank of New York, their unemployment rate was 5.7 percent in June 2026, compared with 4.1 percent for all workers. An NPR report published on August 18, 2026 shows why AI is quickly blamed: two graduates describe sending roughly 450 and more than 500 applications respectively without receiving an offer.

The obvious explanation is that language models now perform typical entry-level tasks. Yet the available research does not tell one consistent story. It points both to displacement in specific occupations and to additional growth at companies spending heavily on AI.

What the labor-market data actually shows

A Stanford analysis of payroll data found a 16 percent relative employment decline since late 2022 among workers aged 22 to 25 in highly AI-exposed occupations. Software development and marketing are among the examples. Older workers in the same fields and workers in less automatable occupations developed more steadily.

A different analysis by Ramp and Revelio Labs examined more than 21,000 US companies between 2021 and early 2026. At firms with the highest AI spending, entry-level headcount grew by 12 percent in the two years after adoption. This does not prove AI caused the new jobs. It does contradict the assumption that intensive AI use automatically means fewer junior positions.

Timing creates another problem. New York Fed research places part of the decline in junior hiring before ChatGPT was released. Remote work is another proposed cause because companies may find it harder to train inexperienced employees from a distance.

Why it matters

The distinction matters in practical terms for people starting their careers. If AI eliminates entire entry-level occupations, broad retraining would be necessary. If it mainly changes tasks within an occupation, evidence of judgment, collaboration and safe AI use becomes more valuable.

Employers also need to be careful. Removing junior positions may save training costs in the short term, but it weakens the pipeline for future specialists and managers. The evidence suggests examining which tasks can be automated, augmented or still need to be learned by people instead of making blanket decisions based on job titles.

In plain language

Imagine a bakery buying a machine that kneads dough. It may need fewer hands at the mixer, but it still needs people to adjust recipes, inspect quality, serve customers and eventually lead a shift. If the bakery grows at the same time, it may hire more trainees despite the machine. That is why AI and junior employment can fall or rise together depending on the task and the company.

A practical example

A software company employs 100 developers, including 20 junior staff. A coding assistant cuts the time spent on simple tests and documentation in half. Management could eliminate five junior positions. Alternatively, it could use the capacity to support more products and hire five additional graduates who inspect errors and analyze customer problems alongside experienced colleagues.

Both companies would appear in a dataset as intensive AI users. Only evidence about hiring, tasks, product growth and timing reveals whether AI replaced work or expanded capacity. That distinction is missing from many simplified headlines.

Scope and limits

• The findings mainly concern the United States and cannot automatically be transferred to Germany or the European Union.
• Observational data shows relationships, not a single cause. The business cycle, interest rates, remote work and industry-specific cuts operate at the same time.
• An occupation being “AI-exposed” does not mean a specific job was replaced by AI. Occupations contain very different tasks.
• The 12 percent growth among heavy AI users may also reflect fast-growing companies buying more software while hiring more people.

The current evidence therefore does not show that AI is harmless for graduates. It shows that an effect may already be visible but cannot yet be cleanly separated from other labor-market changes.

SEO & GEO keywords

AI labor market, entry-level jobs, college graduates, artificial intelligence, future of work, New York Fed, Stanford Digital Economy Lab, Ramp, Revelio Labs, unemployment, junior hiring