The new labor warning from McKinsey Global Institute’s “Workforce in motion” report is easy to misread as another AI replacement story. It is more useful than that. McKinsey’s base case says roughly 11 million U.S. workers — about 7% of current employees — may need to move from declining occupations into different ones by 2035, while more than 70% may need some degree of role reinvention as tasks change inside existing jobs.
Why that matters is the part executives, workers, and policymakers cannot dismiss with a headline about “net new jobs.” McKinsey’s model says the U.S. could still have more jobs available in 2035 than it has today. The problem is mobility: whether a person leaving a shrinking role can realistically reach the growing one without a long unpaid training period, a sharp pay cut, a new license, or a move to a different labor market.
That is the real reader question here: if AI and automation can coincide with overall job growth, why could millions still struggle? Because job creation and job access are not the same thing.
Why more jobs can still feel like a labor-market shock
McKinsey models labor demand from a 2025 baseline through 2035 across 22 broad occupational groups and about 1,800 detailed occupations. In its base case, automation could affect the equivalent of about 54% of current U.S. work hours by 2035. After accounting for offsetting forces — including reduced overtime, productivity-driven demand, new AI-related tasks, demographic change, and broader economic growth — the firm estimates a net labor-demand reduction equal to about 21% of current work hours.
That is still a large reshuffling. McKinsey translates the reduction into labor demand equivalent to about 36 million jobs, while growth in the AI value chain and the broader economy could create demand for more than 40 million. Those are modeled demand equivalents, not layoffs, and that distinction matters. A task can be automated or augmented without eliminating the occupation if the business expands output, adds oversight work, or keeps people in the loop for judgment and accountability.
But even a non-apocalyptic forecast can produce painful career disruption when the losses and gains land in different places. McKinsey expects automation adoption to move fastest in office and administrative support, computer and mathematics, and sales. It estimates that about 80% of current work hours in office and administrative support could face automation adoption by 2035, versus roughly 26% for healthcare professionals.
The wage split is just as important as the industry split. More than 70% of declining employment is projected to be concentrated in the bottom two wage quintiles, especially in office and administrative support, retail and sales, and transportation and logistics. Roughly 60% of growing employment is projected to be in the top two wage quintiles, with healthcare, construction, and management standing out.
That is why aggregate comfort can mask individual strain. As Axios noted in its summary, the pace implied by the report is about 770,000 occupational moves a year, several times the long-run historical average of roughly 215,000. It also helps explain why worker anxiety is already showing up in sentiment data. Glassdoor’s latest employee-confidence release showed only 42.9% of employees giving their employer a positive six-month business outlook, a series low, while mentions of AI in reviews jumped 164% year over year. That does not prove AI-driven job loss. It does show that workers are processing the change as a job-security issue now.
The pathway problem is bigger than a “learn AI” slogan
McKinsey’s most useful contribution is not the 11 million headline. It is the idea that transitions rise or fall on pathway quality.
The report evaluates whether moves are usable by four tests: destination demand, skill adjacency, wage preservation, and the time needed to obtain legally required credentials. Under that framework, only about one in seven workers has a direct pathway into a growing occupation with limited retraining and no wage loss. Nearly half may face what McKinsey calls an unpaved pathway, where skill gaps, credential requirements, lower pay, or long timelines make the move technically possible but operationally unrealistic.
Credentials are the hardest bottleneck in the model. McKinsey says about 85% of growing jobs require a credential or certification. That does not mean every posting reflects a legal requirement; advertised credentials can be employer preferences. But for workers deciding what to do next, that difference is crucial. A preferred certificate is a screening hurdle. A license is a hard gate.
This is also why generic advice to “upskill” is too vague to be useful. The report finds demand for AI fluency has risen roughly 11 times since 2022, while demand for adaptability has increased fivefold and demand for resilience, curiosity, and willingness to learn has tripled. Those are real labor-market signals. They are not, by themselves, a pathway. A worker with adjacent skills still needs an employer willing to recognize them, a training option that fits around income and caregiving, and a destination job that exists in the same region or at least the same pay neighborhood.
What would make the pathways real
For employers, the report shifts the question from “Will AI replace jobs?” to “Which tasks will change, and what happens to the people attached to them?” That is a management choice as much as a technology choice. McKinsey notes that adoption depends on workflow maturity, data quality, implementation cost, regulation, business criticality, reputational risk, and the cost of errors. Smaller organizations may keep broader human roles because one employee handles too many mixed tasks to separate cleanly. Larger organizations often have more room to redesign jobs and reallocate headcount.
The practical test is whether AI deployment comes with internal mobility that workers can actually use. That means mapping declining roles to growing ones before cuts happen, paying for the needed training where credentials are real barriers, and rewarding managers for releasing people into new roles rather than hoarding talent. If productivity gains simply remove routine work without financing the bridge to what comes next, the “pathway” stays theoretical.
For workers, the most valuable response is less dramatic than “learn to code” and more specific than “be adaptable.” It is a skills-and-credential audit. Identify which parts of your current role are routine enough to be automated or heavily augmented. Map adjacent occupations with clear demand. Check whether the destination requires a legal license, a short certification, or just an employer preference. Then compare not only wages, but schedule, location, and the length of time you would need to train without losing income.
For policymakers and educators, the unresolved question is speed. If millions of occupational moves are possible on paper, can training supply, credential rules, and financing move quickly enough to support them in practice? The report does not answer who will pay for credentials, who will support workers through income gaps, or which industries will redesign roles before reducing headcount. Those questions now matter more than the familiar debate over whether AI creates jobs in aggregate.
A labor market can add jobs overall and still strand people. Between now and 2035, the decisive variable is likely to be pathway quality: whether a worker can move from a changing role into a growing one without an unaffordable detour.
By
By
By
By
By
