McKinsey: Automation could absorb half of U.S. work hours by 2035. The bigger challenge is skills

TEXT START: Good morning.

The Dissection

This is a labor-market anesthesia memo disguised as transition analysis. It admits that AI could absorb more than half of U.S. work hours and displace the equivalent of 36 million jobs, then neutralizes the threat by forecasting 41 million new jobs. A control problem—who owns AI and captures the freed labor—is converted into a skills problem: can workers retrain quickly enough?

The revealing question is what companies will do with saved time. The article treats redeployment, internal mobility, and training as plausible defaults while barely examining the obvious power outcome: fewer workers producing more, with the gains captured by owners. The CFO framing makes displacement legible as ROI, productivity, talent, and credential management—not as a collapse in mass bargaining power.

The Core Fallacy

It treats gross job creation as a substitute for productive participation. The projected 36 million displaced-equivalent jobs and 41 million new jobs are not interchangeable units. New jobs may be geographically, credentially, or skill-inaccessible to displaced workers. They may also contain far less human labor than the forecast assumes.

The distinction between automating work hours and eliminating jobs is technically true but structurally evasive. A job can survive while its hours, pay, autonomy, and headcount collapse. A role retaining 20% of its former human content is not evidence of labor resilience; it is a human checkpoint attached to a machine-dominated system.

Under DT, this is the P1-to-P3 chain: cognitive automation dominates tasks; institutions cannot preserve human-only economic domains at scale; productive participation collapses. Growth may create niches. Niches do not restore mass employment.

Hidden Assumptions

• AI-driven growth will create 41 million sufficiently human-intensive jobs rather than expand output with limited labor demand.
• Displaced workers can retrain at 3.6 times the historical occupational-transition rate.
• Companies will reinvest productivity gains in workers instead of converting them into layoffs, wage compression, surveillance, or owner returns.
• Credentials can be relaxed without employers replacing formal filters with informal ones.
• Workers can move across occupations, regions, firms, and income levels without destructive delay.
• Aggregate demand will remain strong enough to absorb the additional output.
• AI-related growth will distribute income broadly rather than concentrate it among owners of models, data, compute, energy, and platforms.
• Judgment, coordination, and accountability will remain durable human moats rather than temporary lag defenses.
• A projected net gain of five million jobs cancels the social damage of making 36 million jobs redundant. It does not.

The article’s own figure—that 85% of growing employment requires a credential or certification—undercuts its easy-transition narrative. The bottleneck is not merely skill. It is access to the institutions that recognize and monetize skill.

Social Function

Primary classification: partial truth serving as transition management and ideological anesthetic.

This is not pure propaganda. It records the 11 million occupational exits, the 770,000 annual transitions, the credential bottleneck, and firms’ weak reinvestment in workforce development. But it domesticates those facts. Structural displacement becomes “skills”; ownership becomes “AI ROI”; distribution becomes “internal mobility”; mass redundancy becomes a CFO optimization question. The worker is presented as a training project. The owner of AI capital disappears.

The Verdict

The report is a useful early warning wrapped in an obsolete equilibrium model. It correctly identifies the blast radius, then mislabels the crater as a skills shortage. Its 41 million jobs is the escape hatch preserving the employment-growth myth; its own figures show that automation will move faster than historical adaptation, credentials will throttle access, and firms have weak incentives to distribute the time and gains they capture.

Under Discontinuity Thesis mechanics, this is not evidence that the system survives. It is evidence of the lag phase: human jobs persist in altered, lower-bargaining-power forms while AI capital owners decide how much labor remains necessary. The central question is not what businesses do with saved time. It is who owns the machines—and whether everyone else is still economically needed when the answer is settled.