When you think about AI replacing jobs, the story feels settled. Stanford economists found something more specific. The useful question is which workers and which jobs, and the answer turns on one distinction that most hiring strategies ignore.
What did the Stanford research actually find?
That the same AI exposure produces opposite employment effects depending on how the work uses it. Stanford's payroll-data analysis, covering millions of US workers, found that employment of workers aged 22 to 25 in the occupations most exposed to AI now sits roughly 19% below where it would be had it kept pace with their less-exposed peers, measured through June 2026, while experienced workers in the same occupations show no comparable gap (Brynjolfsson, Chandar, & Chen, 2025, revised 2026). The authors call these early descriptive indicators rather than causal estimates. And the shortfall is not uniform. It clusters.
Employment fell in jobs where AI primarily automates work. In roles where AI augments human effort, young workers saw employment hold steady or grow. That is the difference between AI replacing workers and AI changing what those workers do.
What separates automation from augmentation?
Two different relationships between the tool and the person.
Automation means AI substitutes for labor. The system completes tasks humans previously performed. A generative model that writes code replaces the coder's work. A model that sorts customer service tickets replaces the triage.
Augmentation means AI complements human effort. It handles parts of the work while the person stays central. An accountant using AI to pull data from documents still reconciles and validates it. A recruiter using AI to sort resumes still runs the interviews and makes the call.
The employment data splits along that line. Occupations where AI involvement centered on complete task delegation, with minimal human interaction, saw entry-level employment decline. Customer service, tax preparation, and data processing all showed the pattern. Young workers with strong foundational knowledge and limited experience faced the most disruption, because the AI was doing the work those junior employees would have done.
Occupations where the AI work was collaborative, with feedback loops and iterative problem-solving, looked different. Employment stayed stable or grew. Young workers were being folded into a workflow that combined human judgment with machine capability.
What does this mean for the roles you are hiring?
It gives you a way to sort your open roles before you write the job description. Three moves follow from the data.
First, identify which tasks in each open role are likely to be automated rather than augmented over the next 12 to 24 months. A software engineer using AI to handle routine scaffolding is still needed. A junior data analyst whose entire job is data extraction faces real disruption.
Second, recognize that augmentation creates new skill requirements. Roles that work well in an augmentative setup need judgment, synthesis, and the ability to catch and correct AI output. Junior employees need different training for it. Senior employees often need new capabilities too.
Third, automation compresses entry-level opportunity rather than erasing it. Companies still need junior talent. They need it for the work the AI can't do.
What should early-career candidates focus on?
Judgment over recall. The Stanford data points to a plain reality: raw technical knowledge is becoming less differentiating, because AI improves at codified knowledge every quarter.
Candidates who do well will be the ones who develop judgment, verification instincts, and the ability to synthesize complex information, which is the same profile that makes adaptable candidates win in the age of AI.
For hiring teams, that changes the questions you ask junior candidates. The point isn't to test whether they can perform a routine technical task any more. It is to find out whether they can think critically about AI output, catch errors, and decide well with incomplete information.
How do you build a role for augmentation?
Four practices, and the strongest hiring teams are already running them.
Redefine roles around oversight. Instead of hiring someone to execute a process, hire someone to oversee and refine AI-assisted execution. That takes different skills and different questions.
Identify what stays distinctly human. What does the role actually need from a person rather than a model? Lead with that in the job description.
Build a structured assessment that tests judgment and synthesis rather than memorization or procedure. Can the candidate think critically about information? Decide with incomplete data? Question and verify? Structured interviews are the #1 predictor of job performance, more than twice as predictive as unstructured ones (Sackett et al., 2022), and Google reached the same conclusion from its own hiring practice (Google re:Work).
Build the skill hierarchy deliberately. Pattern recognition, judgment under uncertainty, and stakeholder communication are becoming more valuable. Rote data processing and routine task execution are becoming less valuable. Your interview kit should say so.
The honest reality
The Stanford research doesn't say AI and employment are fine. Young workers in exposed occupations faced real headwinds. What the data suggests is that the severity depends on whether companies are building for augmentation or betting on automation.
Companies investing in augmentative AI are hiring more junior talent, not less. They are asking for different skills and applying them differently.
So the choices hiring teams make now decide which kind of organization they become: one that extends what people can do, or one that automates it away. When you break a role into specific skills and mark which ones AI will augment and which it will automate, you know what to assess. A structured interview built on that breakdown gives you comparable evidence and a decision you can defend.
References
- Brynjolfsson, E., Chandar, B., & Chen, R. "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence." Stanford Digital Economy Lab, 2025, revised 2026. link
- Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. "Revisiting meta-analytic estimates of validity in personnel selection." Journal of Applied Psychology, 107(11), 2040-2068, 2022. https://doi.org/10.1037/apl0000994
- Google re:Work. "Use structured interviewing." link
Frequently asked questions
What is the difference between AI automation and AI augmentation?
Automation means AI substitutes for labor, completing tasks humans previously performed, like generating code or triaging customer service tickets. Augmentation means AI complements human effort: it handles parts of the work while humans remain central for judgment, validation, and final decisions. Employment data shows these two scenarios play out very differently for workers.
Is AI reducing jobs for young workers?
In some roles. Stanford's payroll-data analysis found that employment of workers aged 22 to 25 in the occupations most exposed to AI now sits roughly 19% below where it would be had it kept pace with their less-exposed peers, measured through June 2026, while experienced workers show no comparable gap (Brynjolfsson, Chandar, & Chen, 2025, revised 2026). The shortfall clusters in roles where AI primarily automates work, such as customer service, tax preparation, and data processing. In roles where AI augments human effort, employment for young workers stayed stable or grew. The authors describe the findings as early descriptive indicators, not causal estimates.
What skills matter most in AI-augmented roles?
Judgment, synthesis, and the ability to catch and correct AI output. Raw technical knowledge is becoming less differentiating because AI gets better at codified knowledge every quarter. Interviews for these roles should test whether candidates can think critically about AI output and make sound decisions with incomplete data.
How should hiring teams prepare roles for AI augmentation?
Redefine the role around oversight rather than execution: hire someone to oversee and refine AI-assisted work instead of someone to run a process by hand. Identify what the role distinctly needs from a person, lead with that in the job description, and build a structured assessment that tests judgment and synthesis rather than memorization or procedural knowledge.
Was this useful?