Artificial intelligence is often described as a jobs story.

Machines will replace workers.

New AI companies will hire engineers.

Some professions will disappear while others emerge.

But one of the biggest workforce challenges facing the United States may be more complicated than either mass unemployment or an AI hiring boom.

It is the possibility that millions of Americans will need to move from one type of work into another, often requiring completely different skills.

An older McKinsey Global Institute estimate put the number of U.S. workers who could need to switch occupations at about 17 million by 2030 as automation, artificial intelligence, e-commerce, remote work and other structural forces reshape the labor market. McKinsey described those workers as being in occupations that could see declining demand, not as 17 million people who would necessarily become unemployed.

That distinction is crucial.

The 17 million figure should not be interpreted as a current AI job-shortage number or a prediction that 17 million people will lose their jobs directly to artificial intelligence. It is better understood as a measure of the potential scale of occupational disruption and the retraining challenge associated with it.

And the issue is becoming more urgent because AI is arriving at a time when the U.S. labor market is already changing.

For decades, technological change has tended to eliminate some tasks while creating new ones.

The challenge with AI is the speed at which the transition can happen.

A worker who has spent 15 years processing documents, handling routine customer inquiries, preparing reports or performing administrative work may not be able to move automatically into an AI-related position simply because the economy needs more technology workers.

The new job may require data skills.

Programming.

AI-system supervision.

Cybersecurity.

Project management.

Advanced technical training.

Or a combination of technical and interpersonal skills.

That creates what economists increasingly describe as a skills mismatch.

McKinsey has previously emphasized that labor shortages are not simply about the number of workers available. Employers can simultaneously have large numbers of vacancies and large numbers of people looking for work because the available workers do not have the skills required for the available positions.

That problem could become more pronounced as AI adoption accelerates.

Today's AI systems can generate text, write software, analyze information, summarize documents, produce images and automate portions of customer service and administrative work.

That does not necessarily eliminate an entire occupation.

Instead, it can change how much human labor is required to perform the occupation.

A company may need fewer junior analysts but more senior reviewers.

Fewer manual documentation specialists but more people who manage automated workflows.

Fewer routine programmers but more engineers who can supervise AI-generated code and validate complex systems.

That process can create what looks like contradictory labor-market data.

An occupation may still exist, but employers may hire fewer people for it.

At the same time, other occupations can expand.

The result can be a healthy aggregate job market alongside severe disruption for particular groups of workers.

That phenomenon is already visible.

Axios reported earlier this year that employment in core white-collar sectors — financial activities, information, and professional and business services — had fallen about 2% from its 2023 peak despite continued strength in the broader economy.

The same analysis noted that those sectors account for roughly 34 million jobs, or about 22% of U.S. employment.

The overall labor market can therefore look relatively healthy while a large segment of professional workers experiences weaker hiring.

One of the most closely watched parts of the AI labor story is what happens to younger workers.

A Census Bureau working paper published in April 2026 found evidence of an immediate and persistent decline in early-career hiring among 22- to 24-year-olds in industries and states with higher AI exposure following the introduction of ChatGPT. The study found that regression-adjusted employment for those early-career workers fell by 12% over the subsequent 10 quarters in the most AI-exposed group, although hiring later recovered to some extent.

That finding matters because entry-level positions are traditionally where workers acquire experience.

A company might automate some junior-level work while continuing to need experienced managers and specialists.

But that creates a pipeline problem.

Where do tomorrow's experienced workers come from if fewer people are given the chance to become beginners today?

AI can therefore disrupt careers without immediately creating mass unemployment.

The first symptom can be fewer opportunities to enter a profession.

The other side of the story is the dramatic increase in demand for AI-related skills.

The Bipartisan Policy Center's analysis of Lightcast job-posting data found that employer demand for AI skills remained strong and accelerating in 2026. The center reported that postings containing AI skills had more than doubled year over year by May and continued rising through the summer.

That means America is not simply losing technology jobs.

It is changing what employers want from workers.

AI skills are appearing in occupations outside traditional technology companies as well.

Finance departments need employees who can work with automated analysis.

Marketing teams need people who can manage AI-generated content.

Healthcare organizations need staff who understand AI-assisted workflows.

Manufacturers need technicians who can operate increasingly automated equipment.

Law firms need professionals who understand how AI changes document review and research.

The result is a labor market where "AI worker" is becoming an increasingly broad category.

The central challenge is not simply creating new jobs.

It is building a bridge between old jobs and new ones.

That requires training.

But traditional education can move slowly.

A four-year degree is not always the right answer for a worker who needs a new skill within six months or one year.

McKinsey has argued that companies and governments need more modular, targeted training and skills-based approaches to hiring, rather than relying exclusively on traditional credentials.

That could mean employer-sponsored training.

Apprenticeships.

Short professional certificates.

Community-college programs.

Industry-specific credentials.

Or internal mobility programs that allow an employee to move from a declining role into a growing one without leaving the company.

The irony is that AI's biggest employment opportunity may not exist inside the most glamorous software companies.

The physical infrastructure supporting AI is creating demand for electricians, technicians, construction workers, engineers and energy specialists.

The Federal Reserve's Beige Book has already captured reports of worker shortages affecting AI-related infrastructure projects. Axios reported that Philadelphia-area businesses cited a shortage of electricians delaying data-center projects, while Dallas contacts said oilfield-services companies were losing skilled workers to power-generation and AI infrastructure projects.

That creates a strange labor-market reversal.

AI may reduce demand for some forms of office work while simultaneously increasing demand for people who build and maintain physical infrastructure.

Someone who never writes a line of AI code could still have a career strongly connected to the AI boom.

The workforce cannot simply be told to "learn AI."

That advice is too vague.

A warehouse worker cannot necessarily become a machine-learning engineer after taking a short online course.

An administrative assistant cannot automatically become a data scientist.

But those workers may be able to move into adjacent occupations that use different levels of automation.

An office worker might move into project coordination, compliance, healthcare administration or AI-assisted operations.

A technician could move into data-center maintenance.

A salesperson might use AI tools to become more productive rather than disappear altogether.

The important question is not whether every worker becomes an AI specialist.

It is whether workers can move into occupations where their human judgment becomes more valuable alongside AI.

This may be the most important lesson from previous technological transitions.

The United States added millions of jobs after earlier waves of automation.

That does not mean every worker affected by the transition recovered equally.

Axios' comparison with manufacturing illustrates the problem. U.S. manufacturing employment fell dramatically during the early 2000s even though the overall labor market later recovered, leaving many workers who lost factory jobs facing long-term occupational and geographic adjustment.

AI could produce a similar distinction.

The headline unemployment rate could remain relatively low.

Total employment could rise.

GDP could grow.

And yet specific communities, professions and generations could experience severe disruption.

That is why the 17 million figure remains relevant even though it is not a literal forecast of AI unemployment.

It illustrates the possible scale of the transition.

The United States has invested enormous amounts of money in AI models, chips, data centers and electricity infrastructure.

But machines do not build themselves.

Data centers need workers.

Semiconductor plants need technicians.

AI companies need researchers.

Factories need electricians.

Companies deploying AI need managers capable of reorganizing workflows.

And workers need opportunities to acquire the skills demanded by all those industries.

That means workforce policy and AI policy are becoming inseparable.

The country can build the world's most powerful AI infrastructure and still struggle to capture its economic benefits if millions of workers cannot move into the jobs being created around it.

The most important labor question may therefore not be:

How many jobs will AI destroy?

It may be:

How many workers can successfully move from shrinking occupations into growing ones?

That is a much harder metric to track.

It requires measuring wages.

Training.

Career mobility.

Geography.

Education.

Age.

Industry.

And the quality of the new jobs.

The 17 million figure provides a sense of the possible scale, but the real outcome will depend on whether companies, schools and workers can build the pathways connecting the old economy to the new one.

AI could create enormous productivity gains.

It could also produce an economy with more high-skill jobs and fewer routine roles.

Whether that transition becomes an opportunity or a prolonged labor-market disruption will depend less on the intelligence of the machines than on the speed with which humans learn how to work alongside them.

America has already built the AI revolution's computers.

Now it has to build the workforce capable of living through the transition.