In August, researchers at Stanford's Digital Economy Lab reported that young workers in occupations highly exposed to AI were falling behind their peers. The change appeared to come mainly through reduced hiring, while experienced workers showed no comparable gap.[1] A company could therefore look much the same to the people already working there, even as it became harder for someone new to get through the door.
That is a difficult kind of change to recognize. A layoff has a date and a group of people who know what happened to them. A job that is never offered leaves an applicant wondering whether the problem is the economy, their qualifications or something they should have done differently. If AI is contributing to that change, its effects may become part of ordinary life before we can confidently separate them from everything else affecting work.
The Stanford findings give us reason to examine how a career begins, and what happens when the work that once gave a beginner experience can be completed by a machine.
Learning through the work
Consider the kind of assignment a junior employee might receive: read several reports, compare their conclusions and explain the differences. The finished summary is what matters. Preparing it can also teach the employees how to follow an argument, question an assumption and recognize when two apparently similar numbers describe different things. A colleague's corrections help connect the result to the reasoning behind it.
If an AI system prepares the summary, the employee may be able to move on to more demanding work. However, someone still has to decide whether the summary is sound. Asking a beginner to check it assumes that the beginner already knows enough to recognize what went wrong, including errors that arrive in confident, orderly prose.
This is where the promise of automation becomes more complicated. An employer may want the judgment of an experienced person while having less reason to pay for the work through which that person would have become experienced. The immediate decision can be reasonable: a smaller team can complete the assignment. The consequence may take years to appear, when the organization needs people who understand more than the answer in front of them.
There is no reason to preserve every tedious assignment. Repetition can waste time as easily as it can build skill, and a busy supervisor is not automatically a good teacher. The question is whether removing the assignment also removes an opportunity to learn, and whether anything takes its place.
The evidence points in both directions
Stanford's researchers used ADP payroll data through June 2026. Employment among workers ages 22 to 25 in highly AI-exposed occupations stood about 19 percent below where it would have been if it had kept pace with employment among similarly aged workers in less-exposed occupations.[1] That comparison measures a relative shortfall. It does not mean AI eliminated 19 percent of young people's jobs.
The authors also reported no widespread, economy-wide displacement associated with AI.[1] Education differences, earlier employment trends and differences between their sample and the broader economy leave substantial uncertainty about how much of the gap AI explains.[1]
There is evidence that AI can help beginners, too. Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied a conversational assistant used by customer support agents. In their working paper, the largest productivity gains went to novice and lower-skilled workers, with suggestive evidence that the system helped newer employees learn from practices associated with stronger workers.[2]
That possibility deserves as much attention as the hiring gap. An employee could receive guidance while dealing with a problem, ask for an explanation without waiting for a colleague and reach a level of competence sooner. For an employer, faster learning could make hiring someone inexperienced more attractive. For the employee, it could open work that would otherwise have been beyond reach.
However, better performance with assistance leaves a further question about what the person has learned. Completing more work and developing independent judgment may happen together, although one does not establish the other. The customer support study offers encouraging evidence in a particular setting, rather than an answer for every profession.[2]
The same technology could therefore improve the experience of people who get hired while reducing the number who receive that opportunity. We will need to watch both to understand what is happening.
Who gets to become experienced?
A company can choose to use AI as part of an apprenticeship. A new employee could examine the evidence behind an answer, explain a disagreement and take on responsibility with someone experienced close enough to correct a mistake. That would require time and attention even if the underlying work became faster. Access to a model would be only part of the arrangement.
If fewer employers make that investment, the burden could fall more heavily on the people trying to enter a profession. They would be expected to arrive with judgment that they had fewer opportunities to develop. People with money, connections or time to work without dependable income might find another route. Those who need a first job to pay their bills would have less room to experiment.
This is also where the argument reaches beyond learning. Paid work can give someone a way to save, gain independence and eventually own a home or part of a business. If that route narrows while machines produce more of the economy's output, greater abundance could coexist with fewer ways to acquire a share of it. The question of who owns the systems and receives their income would become more pressing.
That is a larger argument than this evidence can resolve. It begins, however, with a practical concern about who gets an opportunity to participate before they have proved what they can do.
Which day are we on?
We are at a point where AI can help inexperienced people perform better, while employment data show some young workers falling behind in occupations exposed to the technology.[1][2] The evidence remains insufficient to assign the hiring gap entirely to AI. It is sufficient to question whether our ways of bringing people into working life are keeping pace with what the tools can do.
A technological singularity may still be distant or may never arrive in the form people imagine. Changes in the beginning of a career can matter well before then. Companies can reorganize around the capabilities available to them today, and the effects of those decisions can accumulate while the larger debate remains unresolved.
The next thing to watch is whether employers continue to give beginners paid opportunities to learn as more routine work is automated. If AI helps those people become capable sooner, it could widen access to experience. If the openings disappear, we will have made it harder to become the experienced person whose judgment we still expect to rely on.
Sources
[1] Stanford Digital Economy Lab, August 2026 employment update