Stanford data shows how AI is reshaping the talent pipeline and why leaders must act now.
Stanford has released one of the first large-scale, real-time studies on how AI is shaping jobs. The paper,
For years, the conversation has swung between visions of endless productivity and warnings of mass layoffs.
This new data reveals a structural shift in how companies will develop talent.
1. Entry-Level Jobs Are the First to Go
Since late 2022, employment for software developers and customer service workers aged 22 to 25 has dropped nearly 20 percent.
The problem is that these roles rely on codified knowledge like documentation, standard procedures, basic troubleshooting. This is exactly what AI systems now replicate at scale. The knowledge young graduates bring is displaced before they even finish customizing their Slack avatars.
Companies still need the work done. They just don't need as many people to do it.
2. The Economy Looks Fine, Until You Zoom In
Employment growth looks healthy in aggregate. But underneath the surface, younger workers in AI-exposed jobs are flatlining.
The headline numbers mask the fact that an entire generation of entry-level talent is thinning out.
This is what makes the Stanford data important. It disaggregates by age and occupation. When you look at the full workforce, things look stable. When you filter for workers under 26 in software development or customer service, the picture changes completely.
3. Automation Cuts While Augmentation Boosts
The impact of AI depends on how it is used.
Jobs that rely on automation are shrinking. But jobs that use AI to augment work are actually expanding.
The distinction matters for workforce planning. Leaders need to ask: are we designing roles where AI helps people do more, or where AI replaces what people used to do?
4. The Decline Is Not Driven by the Economy
One of the most important findings in the Stanford paper is what it rules out.
The drop in entry-level roles persists even after adjusting for firm-level shocks such as industry cycles or interest rates. The decline is specific to AI-exposed occupations and younger age groups. It shows up consistently across firms that adopted generative AI tools after late 2022.
In other words, algorithms are quietly removing the lowest rungs of the ladder.
This distinction matters for how leaders respond. If this were economic, you'd wait it out. But structural changes require redesigning how talent moves through the organization.
5. Pay Data Masks the Reality
Compensation levels appear steady. In some cases, they're even rising. But headcount is declining. Payroll data hides the fact that fewer workers are getting the chance to start climbing the ladder.
This is a version of survivorship bias. The people who still have junior roles are doing fine. But there are far fewer of them than there would have been three years ago.
The question isn't whether current employees are paid well. It's whether the next generation will have a way into the organization at all.
6. Education Pipelines No Longer Match the Work
The decline cannot be explained by pandemic schooling or remote work. The issue is more fundamental. Universities cannot prepare graduates to be resilient against tools like Claude or ChatGPT.
Traditional education trains people to perform tasks. But when AI can perform those same tasks faster and cheaper, the value of that training collapses.
What graduates need now is judgment, systems thinking, and the ability to work alongside AI tools. But most curricula still focus on foundational skills that AI can replicate.
Companies will need to create new apprenticeship-style models that combine AI literacy with hands-on experience.
The Risk of a Broken Ladder
Debates about AI and employment often focus on unemployment rates. The more significant risk is a broken talent ladder.
Tacit knowledge, which only comes from years of experience, cannot be automated. Without entry-level roles, no one has the opportunity to build it.
At Microsoft, I watched teams struggle after leaning too heavily on contractors for junior roles. Years later, when systems needed maintenance or adaptation, no one inside the company understood the details.
It was not intentional neglect, but the gap forced costly rebuilds. The same risk is playing out now at a broader scale.
The Stanford study is an early warning. AI has already altered how companies hire and train. The real question is whether organizations will adapt their systems to withstand it.