McKinsey research shows AI creating more jobs than automation displaces - Fortune

TEXT START: While debates about AI’s impact on the workforce dominate headlines, the biggest challenge ahead will be helping workers find the most direct pathway into growing jobs.

The Dissection

This is a labor-reallocation memo disguised as evidence of economic survival. It shifts the central question from “Will human labor remain economically necessary?” to “Can displaced workers be retrained into whatever occupations are still expanding?” Those are not the same question.

The article’s own numbers expose the rupture: 80 percent of administrative-support hours are exposed to automation; 11 million Americans may need to change occupations; 770,000 workers per year must cross occupational lines; 85 percent of growing jobs require credentials; and more than 70 percent of workers need new skills. That is not friction around a healthy labor market. It is a mass sorting operation.

The text treats job counts, skill overlap, and credential pathways as proof of durable human employment. Under the Discontinuity Thesis, that is the wrong unit of analysis. A project manager role can grow temporarily while its planning, scheduling, documentation, forecasting, and coordination functions are simultaneously absorbed by AI. A “growing job” is not necessarily a durable human niche. It may be a lag artifact before the same automation reaches that occupation.

The Core Fallacy

The core error is the job-slot fallacy: assuming that more occupational slots means the mass employment-to-wage-to-consumption circuit remains intact.

The article never establishes that the projected jobs preserve comparable human work-hours, wages, bargaining power, autonomy, or long-term necessity. It counts categories while ignoring the economic substance inside them. AI can increase measured output and even create new job titles while reducing the amount of human labor required per unit of output. That is precisely how productive participation collapses without every job disappearing at once.

It also mistakes task adjacency for economic security. If an administrative assistant shares 58 percent of a project manager’s tasks, that may indicate transferability—or it may indicate that both roles contain automatable coordination work. The overlap is not a moat. It can be a shared target.

The proposed solution—remove credentials, subsidize training, and make employers hire on demonstrated skills—may reduce transition friction. It cannot solve P1, P2, or P3. It does not create enough human demand, prevent AI from entering the destination occupations, or transfer ownership of the productive systems replacing workers.

Hidden Assumptions

• “More jobs by 2035” means more necessary human labor, rather than more titles, thinner roles, or AI-supervised residue.
• Growing occupations will remain human-intensive after AI capabilities diffuse into them.
• Skill overlap predicts successful transition better than wages, geography, caregiving obligations, time without income, health, or available vacancies.
• Credentials are the binding constraint. The deeper constraint is whether employers need humans at scale.
• Employers will voluntarily absorb the cost of a social transition whose profitable alternative is labor substitution.
• Employer demand for AI fluency, adaptability, and resilience signals durable complementarity rather than a temporary requirement to operate inside an increasingly automated system.
• Aggregate job growth translates into stable employment for the particular workers displaced, on a tolerable timeline and at tolerable pay.
• A decade of institutional and physical lag represents recovery rather than delayed obsolescence.

The article also assumes that the occupations presented as destinations are terminal destinations. Under DT logic, they are merely the next exposed layer. Physical care, fabrication, licensing, and public-trust requirements can delay automation. They do not reverse the trajectory.

Social Function

This is a partial truth serving as transition management and ideological anesthetic.

It correctly identifies credential barriers, unpaid training time, employer hiring conventions, and the brutality of forcing workers to finance their own displacement. But it relocates a structural ownership problem into a worker-mobility problem. If the transition fails, responsibility can be assigned to inadequate training systems, rigid licensing, or insufficient worker adaptability rather than to the replacement of human labor by machine-controlled capital.

Its reassuring claim is not that disruption is absent. It is that disruption can be administered. That preserves institutional legitimacy: employers can automate, governments can subsidize retraining, colleges can redesign curricula, and workers can be told to cross another occupational gap. The machinery keeps moving while the displaced are instructed to become compatible with its next configuration.

The Verdict

The article does not refute the Discontinuity Thesis. It documents its early phase.

There may be more nominal jobs in 2035, but that does not prove that mass human participation survives. The text describes millions pushed through expensive credential funnels into occupations that are either low-paid, physically constrained, legally protected, or temporarily insulated from AI. It mistakes a longer runway for a restored system.

This is not evidence that AI creates durable employment. It is evidence that the system can generate new job labels while stripping labor of necessity, bargaining power, and economic centrality. The transition pathways are real. They are also the corridors through which the old labor order is being herded toward obsolescence.