Imagine a company with no mass layoffs, no closed departments, and uneventful quarterly reports. This year it stops hiring junior analysts. Next year it cuts internships.
For existing employees, that may look like efficiency. For newcomers, the first rung of the career ladder is narrowing.
Debates on AI and employment often ask how many jobs will disappear. Another question matters just as much:If machines take the tasks beginners used to practice on, where will the next generation of experts develop?
There are signals worth watching, but no answer we should rush to declare.
What is happening to entry-level hiring?
An August 2026 Stanford Digital Economy Lab update uses ADP payroll data through June. It finds no broad economy-wide employment displacement. Within its sample, however, employment among 22–25-year-olds in highly AI-exposed occupations was 19% below a comparison benchmark tracking less-exposed peers, largely through reduced hiring.Research update and methods
This does not mean AI made 19% of young people unemployed, nor is it a causal estimate attributable to AI alone. The study notes that education controls weaken some differences, some trends predate generative AI, and sample differences exceed those in national surveys. These limits are essential to interpreting the headline number.
These findings suggest we should watch who gets hired, as well as who leaves. Ages 22–25 define an age group, not a job level: not everyone in it is a newcomer, while older career changers may also need entry-level roles. This is a relevant signal about career entry, not a substitute for directly measuring job seniority, experience requirements, and hiring practices.
Layoffs produce announcements; hiring freezes often do not. A position left vacant through attrition may never appear in technology-displacement headlines. Firms can change their age and experience mix without large headcount changes. Today's savings may surface as missing talent pipelines years later.
Junior tasks also teach judgment
Junior work has traditionally served two purposes: getting tasks done and developing people who can handle harder tasks.
Gathering research, inspecting components, writing first drafts, and recording customer needs can be repetitive or inefficient. They also expose beginners to constraints: why does a client reject logically sound advice? Why does a lab-approved part fail on site? Is an anomaly in a report noise, or a reason to stop?
Judgment often develops through doing, making mistakes, receiving feedback, and gradually bearing consequences. Retaining final signatories while removing that learning process can make delivery faster and the supply of experience more fragile.
Not all repetition is educational. Keeping junior staff busy with routine sorting can waste their abilities. AI can accelerate learning, comparisons, and detection of omissions. The learning path needs redesign: which exercises can go, which feedback must remain, and which responsibilities can beginners take sooner?
ILO research on generative AI takes a task-based view and distinguishes exposure from actual replacement. A job contains many tasks; automating some does not eliminate demand for the occupation.ILO research overview
When machines enter the workplace, where does experience come from?
We need care when extending this argument to Physical AI. Office-software employment studies do not directly establish equivalent changes in repair, manufacturing, or care, where safety, equipment costs, and environments differ. The shared organizational question is how people learn to handle nonstandard situations once machines take more standard tasks.
A future factory might need fewer people repeatedly moving goods and more people diagnosing exceptions, planning maintenance, and changing processes. Firms hiring only experienced people must explain who trains them initially.
Put talent development in the AI deployment plan
Managers should review AI deployment and talent development together.
Map the learning chain: where beginners start, who gives feedback, and what qualifies them for independent responsibility. Mark the steps AI will take over. Remove wasted waiting; replace lost opportunities to meet customers, encounter failures, and exercise judgment with meaningful practice.
That practice might include mentored real projects, reviews of past failures, comparisons of human and AI approaches, and progressively harder tasks within safety limits. AI can tutor, but an AI-generated score alone should not establish competence. Assess reasoning, uncertainty, and knowing when to seek help if tools fail.
Give mentoring an explicit budget. If senior staff must maintain unchanged output, an AI apprenticeship can become an unattended online course. Training belongs in project costs, rather than depending on colleagues' unpaid goodwill.
Look beyond output per person to junior hiring, internship conversion, time to independent responsibility, and quality after promotion. A deployment's return is incomplete if current efficiency comes at the expense of long-term training capacity.
Preserve entry points without inventing pointless work
The strongest objection is that hiring may fall because of interest rates, industry cycles, or earlier overhiring. Retaining unnecessary posts could turn training into a subsidy.
I agree that firms should not invent pointless work to preserve old staffing. The better goal is an entry point to real responsibility. Industry bodies, vocational schools, and employers could jointly design paid projects and rotations, reducing the problem of one firm training while others poach. Assess employment and capability outcomes, beyond participant counts.
For young people, learning AI matters. So does demonstrable judgment: defining problems, verifying results, rejecting a tool's advice when warranted, and taking responsibility for real delivery.
Schools and employers should also update assessment. Tools make it easier to finish work quickly. Interviews can focus more on the choices made, how results were checked, and what the candidate learned. That acknowledges tools while letting people without prestigious credentials prove themselves through concrete work.
Tomorrow's experts may use entirely different tools. They still need practice, feedback, and opportunities to be trusted.
Has AI let your team give newcomers harder work sooner, or reduced the need for newcomers? If the latter, where will your experts come from in five years?
Sources current to October 20, 2026. Company scenarios illustrate mechanisms and are hypothetical.