When people imagine artificial intelligence replacing workers, they usually imagine someone getting fired.
A company installs a new system. The machine does the work. The employee gets a pink slip.
That certainly may happen.
But it may not be where the disruption begins.
A company does not have to announce an “AI layoff” to reduce its dependence on human labor.
A senior employee leaves and is not replaced. A department that once hired six college graduates hires three. An experienced programmer using AI performs work that once required two junior programmers. A marketing manager produces drafts, research, graphics and analysis that previously gave several younger employees their first opportunity to learn the business.
Nobody necessarily gets fired.
There may be no headline. There may be no press release saying: “Artificial intelligence eliminated these jobs.”
The jobs simply stop appearing.
And if that happens across enough companies and occupations, artificial intelligence could begin altering the labor market long before anything resembling mass unemployment becomes visible.
I call this problem The Missing First Rung.
Every Career Ladder Has a Bottom
We talk constantly about experienced workers: senior engineers, senior accountants, portfolio managers, attorneys, executives, software architects, master technicians and experienced buyers.
But experienced workers do not simply appear.
They become experienced.
The first rung is often inefficient.
A young employee asks questions. Makes mistakes. Needs supervision. Takes longer. Handles routine assignments. Performs research that a senior employee could probably complete faster. Writes a first draft that someone else must revise.
That looks inefficient if we examine only today's productivity.
But there is another way to look at it.
Today's inexperienced employee is potentially tomorrow's expert.
Remove enough of the work performed by inexperienced people and we create a strange career ladder:
Where exactly do the experienced people come from?
AI May Be Particularly Good at the Work That Used to Train Beginners
A great deal of entry-level professional work consists of tasks that are relatively structured, repetitive, research-heavy or draft-oriented.
Summarize these documents. Prepare the first version of this presentation. Research these competitors. Write some basic code. Categorize these transactions. Prepare an initial financial analysis. Draft this correspondence. Search these cases. Create several advertising concepts. Compile the data.
Those tasks have economic value.
But they also have educational value.
They expose inexperienced workers to the information, judgment, mistakes and repetition through which expertise develops.
AI is becoming increasingly capable of performing precisely these kinds of tasks.
The International Labour Organization estimates that roughly one in four workers worldwide is employed in an occupation with some exposure to generative AI. It currently expects transformation to be more common than complete job elimination, and clerical work remains especially exposed.1
That distinction is important.
We May Already Be Seeing the First Signs
This is where the discussion moves beyond speculation.
Researchers at Stanford's Digital Economy Lab have been examining payroll records covering millions of American workers through June 2026.
Their updated analysis contains two findings that deserve attention.
First, they find no evidence of widespread economy-wide job displacement from AI so far.
That should be stated clearly.
The AI employment apocalypse has not happened.
But their second finding is harder to dismiss.
Among workers ages 22 to 25 in highly AI-exposed occupations, employment is about 19% below where it would be if it had kept pace with similarly aged workers in less-exposed occupations. Experienced workers in the same highly exposed occupations do not show the same pattern.2
And perhaps most important for the Missing First Rung hypothesis, the researchers report that the adjustment appears to occur mainly through reduced hiring of young workers rather than increased separations of existing workers.2
The researchers themselves caution that these are descriptive patterns, not proof that AI caused the entire divergence.
That does not prove AI caused every missing job. It does not prove the pattern will continue. It certainly does not prove mass unemployment is inevitable.
But it looks remarkably similar to what we should expect if the first stage of AI labor displacement occurs through the bottom of the career ladder.
The First Signal May Not Be Layoffs
Suppose a company employs 20 senior professionals, 30 mid-career professionals and 20 junior employees.
AI substantially increases everyone's productivity.
Management does not necessarily call twenty junior employees into a conference room and fire them.
Instead, over several years, four junior employees leave and only two are replaced. Then another four leave and one is replaced. Next year's graduate recruiting class is reduced. An experienced employee who retires is replaced internally.
The organization slowly changes from:
to:
There was never a dramatic AI layoff.
Yet twelve career-entry opportunities disappeared.
Multiply that by thousands of employers and the aggregate effect becomes very different from the story visible inside any one company.
An employer can truthfully say, “We haven't laid anybody off because of AI,” while simultaneously employing fewer people because of AI than it otherwise would have.
Those statements are not contradictory.
Hiring Matters as Much as Firing
Public discussion tends to focus on job destruction.
Economically, however, employment changes through both ends of the pipe.
People leave jobs. People enter jobs.
If hiring slows sufficiently, employment can decline even if very few existing employees are directly terminated.
This is especially important for young workers because they are disproportionately dependent on the entry side of the labor market.
A 45-year-old accountant already has fifteen or twenty years of experience to sell. A 22-year-old accounting graduate does not.
If technology makes an experienced accountant substantially more productive, the company may conclude it can maintain output while hiring fewer junior accountants.
The senior worker benefits. The employer benefits. Productivity rises.
And the young worker discovers that the first rung has become harder to reach.
Nothing about that outcome requires anyone to behave maliciously.
It may simply be economically rational.
The Company Has an Incentive to Remove the Rung
Consider the decision from the employer's perspective.
You have a senior employee earning $150,000 and a junior employee earning $60,000.
Historically, the junior employee performed research, drafted documents, assembled information and handled repetitive work. The senior employee reviewed it.
Now AI allows the senior employee to perform much more of that work directly.
From the company's immediate perspective, perhaps you shouldn't.
That is the same competitive mechanism we have discussed elsewhere in the Paradox of Automation.
One company reduces costs. Its competitor must respond.
Nobody needs to conspire. Nobody needs to hate young workers. Nobody even needs to believe AI should replace people.
Each employer can make a perfectly rational decision.
But society has a different problem to solve:
The Company Saves Money Today. Who Produces the Expert Tomorrow?
This is where the Missing First Rung becomes more than an employment problem.
It can become an expertise pipeline problem.
Imagine an accounting firm in 2036 complaining: “We cannot find enough accountants with ten years of experience.”
The obvious question should be:
The same applies to software, engineering, law, finance, procurement, journalism, management, design and countless other skilled occupations.
A young attorney becomes a better attorney partly by performing junior legal work. A programmer becomes better by writing and debugging actual software. An engineer develops judgment by working under experienced engineers. A buyer learns through negotiations, supplier failures, quality problems, production delays and thousands of decisions that cannot be fully reproduced in a classroom.
AI can provide extraordinary assistance in all of these fields.
But if organizations use AI to eliminate much of the work through which beginners become professionals, they may eventually discover that they have optimized away their own training system.
We Have Seen Versions of This Problem Before
Technology is not the only reason organizations underinvest in training.
There is a longstanding incentive problem.
Company A spends years developing a worker. Company B hires that worker away.
Company A paid much of the training cost. Company B receives much of the benefit.
That can encourage employers to look for experienced employees rather than create them.
Anyone who has searched modern job postings recognizes the result.
Then everyone asks: how am I supposed to get experience if every employer requires experience?
AI could make that old problem substantially worse.
Employers may simultaneously become less willing to hire inexperienced workers and more demanding that candidates already possess experience.
That is not an efficient career pipeline.
It is a bottleneck.
This Also Affects People Who Are Not Young
“Entry-level” can be misleading.
A 48-year-old person changing industries may have decades of professional experience and still be entry-level in the new field.
That matters greatly if AI displacement forces people to change occupations.
We routinely answer technological displacement with: Retrain.
But retraining is only one part of occupational transition.
The worker must subsequently persuade an employer to provide the first opportunity in the new field.
A person can acquire a certificate, complete coursework, learn new software and study a new industry.
None of that guarantees an employer will hire someone without the precise experience the employer has decided it wants.
This creates a potentially nasty interaction:
That is not merely an education problem.
It is a labor-market design problem.
There Is Important Evidence Against the Strongest Version of This Argument
We should resist the temptation to turn early signals into certainty.
Research from Denmark provides an important counterweight.
Economists Anders Humlum and Emilie Vestergaard studied widespread adoption of AI chatbots using surveys linked to administrative labor-market records. Their March 2026 revision finds substantial changes in tasks and work organization, but no measurable average effect on earnings or recorded hours large enough to exceed roughly 2% during the first two years after ChatGPT's launch.3
The International Labour Organization likewise emphasizes that exposure measures are signals of possible change, not forecasts of job losses, and that transformation remains more likely than outright redundancy for most exposed occupations.14
This matters.
The Missing First Rung is not a declaration that entry-level work is disappearing everywhere.
And we should want the hypothesis to be wrong.
AI Could Also Make Young Workers More Productive
There is another possibility.
Perhaps companies will discover that AI makes junior workers so productive that hiring them becomes more attractive, not less.
A new graduate equipped with powerful AI tools may perform work that once required several years of experience.
That could reduce the cost of training. It could democratize expertise. It could allow smaller businesses to employ young people who become productive almost immediately.
The World Economic Forum notes that more than one in three young workers globally are in occupations with medium-to-high exposure to AI-driven task change, while also emphasizing the opportunity to redesign entry-level roles rather than simply eliminate them.5
So once again, technology does not dictate the future.
How organizations choose to deploy it matters.
An AI system capable of replacing junior work may also be capable of accelerating junior learning.
Those are radically different uses of the same technology.
Perhaps We Should Stop Treating Training as Waste
For decades, businesses have tried to eliminate inefficiency.
That is generally sensible.
But not every apparent inefficiency is waste.
Children are extraordinarily inefficient learners.
Apprentices are inefficient craftspeople.
Residents are inefficient physicians compared with experienced doctors.
Junior employees are inefficient professionals.
If we evaluate every activity solely by whether a machine can perform today's task more cheaply, we may optimize a system while inadvertently destroying the process through which tomorrow's human capability develops.
That would be a particularly ironic use of artificial intelligence.
The Missing First Rung Is a Systemic Problem
One employer does not have to worry about maintaining the entire economy's supply of experienced workers.
Its responsibility is its own organization.
It can hire experienced people from somewhere else.
That works as long as somewhere else continues producing experienced people.
But if enough companies adopt the same strategy, we encounter the same structural problem repeatedly appearing throughout the Paradox of Automation:
One company can stop training beginners and hire experienced workers from competitors.
Every company cannot.
One employer can eliminate its first rung.
The economy cannot eliminate the first rung indefinitely and still expect people to arrive magically on the third.
What we know
Young workers in highly AI-exposed U.S. occupations are showing a widening relative employment gap in Stanford's payroll data. The researchers find that reduced hiring appears to be an important mechanism, while experienced workers in the same exposed occupations show no comparable gap. AI exposure is also substantial globally, especially in clerical and highly digitized cognitive work.
What we don't know yet
We do not yet know how much of the young-worker gap is caused specifically by AI, whether it will persist, whether new occupations will absorb affected workers, or whether AI will ultimately increase demand for junior employees by making them more productive. Early Danish evidence also shows significant task reorganization without large average earnings or hours effects.
And This Is Before Mass Unemployment
That may be the most important point.
We should not wait for unemployment to reach 15%, 20% or 30% before asking whether AI is changing the labor market.
By that time, the earliest structural changes may have been underway for years.
Watch entry-level hiring.
Watch graduate recruitment.
Watch job vacancies.
Watch how many people companies need to produce a given amount of output.
Watch the age distribution within AI-exposed occupations.
Watch whether junior responsibilities disappear while senior employment remains stable.
Watch whether people changing careers can actually enter new industries.
And watch the career pipeline.
Because the first major labor-market effect of artificial intelligence may not look like millions of workers walking out of buildings carrying boxes.
It may look much quieter.
A graduate sends another application.
And another.
And another.
Nobody calls.
The company is doing fine.
Productivity is rising.
Profits may be rising.
No layoff occurred.
But somewhere between education and experience, a rung of the ladder has disappeared.
Endnotes & sources
- International Labour Organization, Generative AI and Jobs: A 2025 Update, May 20, 2025. The ILO estimates one in four workers worldwide are in occupations with some degree of generative-AI exposure and says transformation is more likely than redundancy for most jobs.
- Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, Stanford Digital Economy Lab, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, revised August 12, 2026. Uses ADP payroll data through June 2026. The authors explicitly describe the evidence as descriptive rather than causal.
- Anders Humlum and Emilie Vestergaard, National Bureau of Economic Research, Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI, revised March 2026.
- International Labour Organization, Workers' Exposure to AI: What Indicators Tell Us — and What They Don't, April 17, 2026.
- World Economic Forum, Artificial Intelligence and the Future of Entry-Level Work, June 22, 2026.