McKinsey: 11 Million U.S. Workers May Be Forced Into New Occupations by 2035 Due to AI

Roughly 11 million U.S. workers — about 6.5% of the labor force — may need to change occupations entirely by 2035 as AI reshapes what work gets done and who gets hired to do it. That is the base-case estimate from a McKinsey report released September 29, which describes the coming decade as the largest sustained workforce transformation in U.S. history. The finding is significant not because the number is catastrophic, but because it is precise — and it arrives when aggregate labor data still shows no economy-wide shock.

Bloomberg reported the McKinsey findings on September 29, three and a half years after ChatGPT's launch in late 2022. The timing matters: the absence of a visible shock so far is not evidence that one is not coming. It is evidence of a lag.

The Numbers in Detail

McKinsey's model separates gross displacement from net employment:

• 36 million roles face reduced labor demand due to AI automation — tasks that currently require humans will be handled by AI systems, shrinking the headcount those functions need
• 40 million roles face increased labor demand — new jobs created by AI adoption, including AI oversight, deployment, and entirely new categories that don't yet fully exist
• Net result: roughly 4 million more jobs than lost, but 11 million workers whose specific occupations shrink enough that they will need to change fields entirely, not just update skills within their current role

The distinction between "update skills" and "change occupations" is critical. Retraining a graphic designer to use AI generation tools is a skills update. Asking a data entry clerk whose entire role has been automated to retrain as an AI systems auditor is an occupational change — a different career path, different credentials, different job search.

Who Is Most Exposed

McKinsey identifies the most exposed workers as those in:

• Entry-level office and administrative roles — scheduling, data entry, document processing, basic research tasks
• Customer service and support — call center agents, live chat operators, help desk roles
• Content production support — copy editing, transcription, basic writing and summarization roles
• Recently hired workers — a particularly sharp finding: workers who entered their current field in the last three to five years have had less time to build the specialized judgment and institutional knowledge that makes experienced workers harder to automate

The pattern is consistent across McKinsey's historical AI impact research: automation hits the bottom of occupational hierarchies first, then climbs. Junior roles are most exposed because they involve the most standardized, codifiable tasks. Senior roles involve more judgment, relationship management, and non-routine decision-making — harder targets for current AI.

The Lag That Explains the Disconnect

If 11 million workers need to change occupations by 2035, why don't current unemployment statistics show it? McKinsey's report addresses this directly.

First, AI adoption at the workflow level is still incomplete. Many companies have deployed AI tools but have not yet restructured headcount around them. They are capturing productivity gains from existing workers rather than reducing headcount. That changes when contracts expire, attrition creates vacancies that are not refilled, and the next hiring cycle reflects the new AI-augmented staffing model.

Second, the workers most at risk are concentrated in roles with relatively low turnover. Layoffs show up in the data immediately; slow attrition shows up in labor statistics over years.

Third, new job creation in AI-adjacent roles is happening in parallel. The net number is growing, which buffers the aggregate statistics even as specific occupational categories contract.

What the 4 Million Net Gain Actually Means

McKinsey projects 40 million demand-growth roles against 36 million demand-shrinkage roles — a net of 4 million jobs. That sounds like good news. It should be read carefully.

The 40 million growth roles are not the same jobs in the same places as the 36 million shrinking roles. A warehouse manager whose facility automates receiving and sorting does not automatically qualify for an AI operations role at a software company. Geographic concentration, credential requirements, and salary levels of growth roles versus shrinkage roles can differ significantly.

This is the core structural risk: the economy produces net jobs, but the workers displaced from shrinking roles are not necessarily positioned — by location, education, or economic means — to transition to growing ones. Workforce displacement is not just a count problem; it is a matching problem.

What to Watch

Watch Q1 2027 hiring data in sectors with the highest entry-level AI automation rates — financial services, customer support, and business process outsourcing. If the McKinsey projection is tracking correctly, those sectors should show measurable reductions in entry-level headcount relative to output growth over the next 12 months. Also watch whether Congress incorporates any of these projections into the Farm Bill, CHIPS Act extensions, or any workforce reauthorization legislation — 11 million is a number that tends to find its way into policy arguments.