Stanford study finds AI is gutting entry-level jobs

For the past two years, the tech world has debated whether AI is genuinely eliminating jobs or if white-collar layoffs were just routine post-pandemic belt-tightening.

Now, hard payroll data is finally in, and it confirms the damage is already real. But it is not targeting everyone equally: recent college grads are taking the hit first.

Stanford researchers analyzed ADP payroll data and found that for workers aged 22 to 25 in AI-exposed roles — such as software development and customer support — employment has dropped 16% since late 2022.

Erik Brynjolfsson, the Stanford economist leading the project, noted that the impact transitioned from theoretical projections to concrete numbers at historic speed. He called it the fastest, broadest shift he has observed outside of the sudden pivot to remote work during COVID-19.

Yet older and mid-career workers in those identical fields have not seen a measurable drop in opportunities.

Brynjolfsson explains that today's large language models mimic entry-level hires: they possess vast book knowledge but lack real-world reps. Senior employees rely heavily on tacit knowledge and unwritten tricks of the trade that models have not mastered yet.

Crucially, looking at aggregate job market data shows almost no net change from AI. The drop only becomes visible when filtering by specific roles and experience bands, and the team ruled out tech cycles and remote-work rollbacks as primary drivers.

Corporate strategy also splits outcomes. Companies using AI to augment staff are actually expanding headcount, while those deploying tools purely to cut human costs are slashing entry roles.

Brynjolfsson's team is now building a near-real-time AI economic dashboard to monitor hiring and wage patterns across industries.

Why it matters

This paper validates what junior engineers and fresh graduates have been venting about on social media: the bottom rung of the corporate ladder is disappearing.

For entry-level workers, textbook competence and basic syntax are no longer enough to secure a living. If an agent can churn out boilerplate code and draft Tier-1 support replies, companies will stop paying novices to learn on the job.

For the tech industry as a whole, this creates a terrifying pipeline crisis. Senior talent is still in demand, but if no one hires the junior cohort today, where will the next generation of senior engineers and leaders come from in seven years?

In our view, companies celebrating immediate margin gains from cutting entry-level budgets are borrowing against their own future. Younger workers appear to be the proverbial canary in the coal mine, and as reasoning models inch closer toward understanding contextual nuance, mid-level staff could be next on the chopping block.