Most conversations about artificial intelligence and employment begin with a sweeping question: Will AI take jobs?
That question may be too broad to be useful. A better one is: Which jobs are changing first?
The emerging answer has a distinct shape. Exposure is not distributed evenly across the career ladder. It is concentrated near the bottom, where recent graduates and entry-level workers perform the research, drafting, summarizing, analysis, and production tasks that traditionally help them launch their careers. The immediate risk is not simply that AI may eliminate work. It is that AI may reduce access to the first job, which is often where graduates begin developing the judgment required for the next one.
The Automation Gap
Task-level estimates show a striking difference between junior roles and the positions above them. A Bloomberg analysis cited by the World Economic Forum estimates that AI could automate 53% of the tasks performed by market research analysts, compared with just 9% of those performed by marketing managers. For sales representatives and sales managers, the figures are 67% and 21%, respectively. For graphic designers and art directors, they are 50% and 24%.1
Across three different fields, the pattern is consistent. The more junior role faces substantially greater task exposure than the role responsible for directing, interpreting, or approving the work.
That gap reflects how jobs are constructed. Entry-level employees are commonly asked to assemble information, produce initial drafts, conduct preliminary analysis, and execute clearly defined tasks. More senior employees are more likely to supply context, manage relationships, recognize exceptions, make decisions, and accept responsibility for the outcome.
In other words, many of the tasks that make a position suitable for a beginner are also the tasks that are easiest to automate.
That does not mean an entire occupation disappears when some of its tasks can be performed by AI. It does suggest that employers may need fewer people to handle the foundational work that once justified hiring a large entry-level cohort. The consequences become more significant when those tasks are not merely outputs to be produced. They are also opportunities through which novices learn how experienced professionals think.
The Gap Is Widening
Task-exposure estimates describe what could happen. Payroll records are beginning to show what is happening. Researchers Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen analyzed ADP administrative payroll data to examine employment patterns in occupations with differing levels of AI exposure. Their August 2026 update found that employment among workers ages 22 to 25 in highly AI-exposed occupations was approximately 19% below where it would have been if it had kept pace with employment among similarly aged workers in less-exposed occupations.
The revision history is especially important. The researchers first documented a 13% gap in August 2025. That figure was revised to 16% in November 2025 and stood at approximately 19% in the August 2026 update. The gap has widened with each update rather than appearing as a single, temporary disruption.2
The underlying employment levels reinforce the pattern. Between November 2022 and June 2026, employment among 22- to 25-year-olds in the two most AI-exposed occupational quintiles fell by about 11%. For the same age group in the three least-exposed quintiles, employment grew by approximately 10%.3
This does not establish that AI is the only force affecting early-career employment. It does, however, provide a more specific signal than a survey of employer expectations or worker sentiment. It shows a growing divergence in actual payroll records, concentrated among younger workers in occupations most exposed to AI.
The Missing Signal
The mechanism behind that divergence may be even more consequential than the headline numbers. The decline appears to operate primarily through reduced hiring of young workers rather than increased separations. In plain terms, the effect is not showing up mainly as people losing jobs they already hold. It is showing up as fewer young people being hired in the first place. Experienced workers in the same occupations do not show a comparable gap.
This makes the change difficult to see.
Layoffs produce announcements, news coverage, and visible disruption. A position that is never posted creates no comparable signal. There is no departure to count and no displaced employee to interview. The door simply opens less often.
That distinction matters to anyone advising students or early-career workers. Watching layoff reports will not necessarily reveal where AI is affecting opportunity. By the time the shift becomes obvious in conventional labor-market headlines, a meaningful number of entry points may already have quietly narrowed.
Where Judgment Gets Built
In his recent blog article, When Intelligence Is No Longer Scarce, Jamal Khan, Chief Growth and Innovation Officer at Connection, framed the first job as more than an economic starting point. It is where a young worker begins learning the craft and building judgment.
The traditional apprenticeship exchange is simple. A novice performs necessary routine work while gaining proximity to experienced colleagues. Along the way, that person observes how experts frame problems, account for context, recognize exceptions, and take responsibility for decisions.
If AI absorbs much of the routine work, that exchange can break down. The higher-level role may remain, but the pathway to becoming qualified for it becomes less clear. As Jamal Khan writes, “The next rung on the ladder remains, but it may sit too high to reach from the ground.”
Schools therefore stand at one end of a transition that employers are already shaping at the other. A graduate needs more than a credential signaling readiness to work. That graduate also needs access to an environment where professional judgment can continue to develop.
The Outcome Is Not Predetermined
The same payroll research offers an important distinction. Entry-level employment declines were concentrated in occupations where AI use substitutes for human tasks. In occupations where AI complements workers, employment was flat or rising. The technology is the same. The employment outcomes are not.
That means the result depends partly on how organizations deploy AI. Employers can use it to remove junior work, or they can use it to help junior employees perform that work more effectively. This is a deployment decision, not an unavoidable property of the tool.
What Education Leaders Should Watch
Everyone is watching for layoffs. But the early effect does not appear primarily as layoffs. It appears as hiring that quietly does not happen.
For district leaders, the absence of bad news is therefore not evidence that nothing is changing. The widening gap in the payroll data suggests that waiting for a dramatic headline risks overlooking the place where the shift is already visible.
The useful question is not whether AI will take jobs in the abstract. It’s whether employers are using AI to replace entry-level job functions or to help junior employees become better at them—in the roles graduates actually enter. That is a question educators can bring into conversations with the workforce advisory boards and local employers they already interact with in the community.
None of this tells educators what to teach. It does tell them what to stop waiting for. AI adoption is already reshaping the workforce. It’s here. Now it’s time to start planning for what those changes mean to those choosing careers today. What does it mean for your curriculum? What does it mean for employers? Partner with the organizations and recruiters who fill the fieldhouse on career fair day. What does that first rung on the ladder look like to them—and how can your institution help students reach it?
1 https://www.weforum.org/stories/jobs-and-the-future-of-work/ai-jobs-international-workers-day/
2 https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf
3 https://digitaleconomy.stanford.edu/app/uploads/2026/08/Canaries_August2026.pdf
