The predicted artificial-intelligence jobs apocalypse has not materialized in the United States. Current evidence indicates that it is unlikely to materialize if businesses continue using AI to increase productivity, expand production, create new products and services, and employ people capable of implementing and managing these systems.

A recent article in The Economist, “The Jobs Apocalypse Is Postponed. An AI Jobs Boom Is Here,” provides an important corrective to the continuing assumption that every task performed by artificial intelligence represents a human job eliminated. The article’s central argument is well supported: AI is displacing some tasks and reducing hiring in certain occupations, but it is also generating substantial employment through data-center construction, electricity production and transmission, semiconductor and equipment manufacturing, software development, systems integration, professional services, and the implementation of AI throughout industry.

The Economist may overstate the precision of its estimate that AI has already created approximately one million American jobs. Nevertheless, the underlying evidence supports a broader and more important conclusion: AI is not merely an automation technology. Properly implemented, it is a productive-capacity technology that allows organizations to do more, serve more customers, enter new markets, develop new products, and employ people in new and expanded roles.

What The Economist Found

The Economist estimates that the AI expansion has created approximately one million jobs in the United States since the widespread introduction of generative AI. Its apparent calculation combines two principal categories.

The first consists of approximately 320,000 jobs above the pre-ChatGPT employment trend in industries associated with AI infrastructure. These include data-center construction, electrical installation, power generation and transmission, semiconductor manufacturing, server and communications equipment, cooling systems, generators, transformers, and related industrial supply chains.

The second consists of approximately 730,000 jobs above trend in professional and technical occupations considered closely connected to AI. These include mathematical and computer occupations, business-operations specialists, project managers, software developers, technical consultants, and people responsible for integrating AI into business operations.

The Economist compares those approximately one million estimated jobs with roughly 200,000 announced layoffs attributed to AI since mid-2023. On that basis, it argues that employment associated with AI has substantially exceeded the number of jobs visibly eliminated because of it.

That comparison is informative, but it is not a definitive measure of net job creation. The one-million figure is an estimate based largely on employment exceeding earlier trends. The 200,000 figure is based principally on layoffs that employers explicitly attributed to AI. The first is an expansive estimate of associated employment; the second is a relatively narrow count of announced job eliminations. They are not directly comparable measurements.

The defensible conclusion is therefore not that AI has conclusively created precisely one million net jobs. It is that substantial employment creation associated with AI is real, broadly distributed, and large enough to challenge the claim that AI is presently producing a net employment catastrophe.

The National Employment Evidence

The August 2026 employment report from the United States Bureau of Labor Statistics provides the clearest national perspective. American employers added 162,000 payroll jobs during August, while unemployment remained at 4.1 percent. Approximately seven million people were unemployed, but there was no evidence of the rapid, economy-wide employment collapse predicted by the most extreme AI forecasts.

The longer-term labor-market picture is less robust than the August result alone might suggest. The economy added an average of only approximately 31,000 jobs per month during the preceding 12 months. Employment growth had slowed, hiring was cautious, and the information industry continued to lose jobs. Nevertheless, the slowdown was not equivalent to an AI-driven employment collapse.

Research from Yale University’s Budget Lab reinforces this finding. Its analysis found no clear relationship between measured AI use and economy-wide employment or unemployment. It also found that changes in the occupational composition of American employment did not yet show a distinct break corresponding to the introduction of ChatGPT.

The Federal Reserve Board reached a similar result using job-posting information from Lightcast and AI-adoption information from the Census Bureau’s Business Trends and Outlook Survey. Researchers Jessica Liu and Douglas Webber found no evidence that industries or companies with higher AI adoption had reduced their total job postings. Their analysis suggested that companies may be changing the kinds of workers they seek, but the general post-pandemic decline in job postings could not be attributed to AI.

These independent findings strongly support The Economist’s central thesis: as of September 2026, artificial intelligence has not produced widespread net displacement across the American economy.

AI Is Changing Labor Demand Rather Than Simply Eliminating It

The most accurate model of AI’s employment effect is not a single line running from automation to unemployment. AI produces several effects simultaneously.

It can substitute for human labor by performing an existing task. It can augment human labor by helping a worker perform that task faster or better. It can increase demand by lowering the cost of a product or service. It can make previously impractical products economically viable. It can create entirely new business capabilities. It can also generate complementary employment in computing infrastructure, implementation, security, governance, maintenance, training, and management.

These effects can occur within the same occupation or company.

For example, AI may automate the preparation of a routine financial report while increasing the value of the employee who interprets the report, verifies its accuracy, explains its implications, and makes a business decision. It may reduce the number of people required to write basic software code while increasing demand for people who can design systems, integrate models, manage tools, establish deterministic controls, test outputs, protect data, monitor performance, and accept responsibility for operational results.

The important unit of analysis is therefore the task, not merely the job title. Most occupations are bundles of technical, administrative, interpersonal, judgmental, and accountability-related tasks. AI may automate some components while increasing the value or volume of the remaining work.

Productivity Can Produce Growth and Employment

The strongest positive employment mechanism is productivity-led expansion.

When a business uses AI only to reduce its payroll while maintaining the same output, employment is likely to decline. When it uses AI to increase capacity, improve quality, respond faster, reduce the cost of experimentation, develop new products, or reach additional customers, employment may grow.

PwC’s 2026 AI Jobs Barometer found that companies achieving the greatest AI-related productivity improvements were not using AI exclusively for cost reduction. The most productive AI-exposed companies were also experiencing comparatively strong headcount and wage growth. Productivity growth was reported to be 40 percent higher among the most AI-exposed companies than among the least exposed.

The skills required in highly AI-exposed jobs were changing more than twice as quickly as those in less-exposed jobs. New tasks appearing in AI-exposed occupations were also substantially more likely to require judgment, creativity, empathy, leadership, and other human capabilities that become more valuable when routine information processing is automated.

This is the mechanism that technical managers should understand. AI can reduce the labor required for each unit of output while simultaneously causing total employment to increase if production grows faster than the labor required per unit declines.

This pattern has appeared repeatedly in technological history. Improved tools reduce the labor required to perform a particular task, but falling costs and rising capabilities expand the range and volume of economically useful work. The final employment effect depends on the balance between substitution and expansion.

The Demand for AI Skills Is Real

The 2026 Stanford AI Index and Lightcast employment data show that AI is moving from experimentation into deployment.

AI skills appeared in approximately 2.5 percent of American job postings in 2025, an increase of 55 percent in one year and approximately 297 percent over the preceding decade. Agentic-AI skills appeared in approximately 90,000 job postings, increasing more than 280 percent in a single year.

The composition of demand is also changing. Employers are moving beyond general requests for chatbot familiarity. They increasingly seek people with skills related to deployment, workflow management, cloud platforms, scalability, integration, and operating AI systems in production.

This is especially important because a generative model is not a complete operational system. A model must be combined with a harness, orchestration, tools, context, memory, identity, authority, policies, deterministic controls, security, evaluation, observability, and human accountability. The more broadly AI is implemented, the more organizations require people who understand these surrounding systems.

Consequently, some of the most valuable positions will not carry the title “AI engineer.” They will include domain experts, process engineers, systems architects, data specialists, cybersecurity professionals, performance engineers, compliance personnel, product managers, technical salespeople, trainers, and executives who can redesign an organization around the productive use of AI.

The Physical Economy Behind Artificial Intelligence

Artificial intelligence appears to users as software, but its productive base is physical.

Training and operating advanced models require semiconductors, servers, memory, communications networks, buildings, electricity, cooling, water, generators, transformers, switchgear, transmission lines, and extensive construction. Each layer creates demand for labor and specialized suppliers.

Reuters documented how the data-center expansion is moving through American manufacturing supply chains. Generac planned a $250 million expansion of several factories and expected to add approximately 1,000 workers in response to a data-center generator backlog of approximately $1.6 billion. Manufacturers of cooling systems, transformers, construction machinery, cables, pipes, cement, bearings, and prefabricated building components were also reporting increased demand.

This demonstrates why counting only the permanent employees inside an operating data center understates the broader employment effect. A data center may employ a relatively small permanent operating staff, but its construction and continued operation create demand throughout the power, communications, manufacturing, maintenance, and professional-services ecosystems.

At the same time, claims about local employment must remain realistic. Research summarized by the Brookings Institution examined approximately 1,500 data centers and compared counties receiving completed facilities with counties where announced projects were canceled. It estimated that a typical data-center development produced approximately 100 to 200 continuing local jobs, depending on the facility type.

Hyperscale facilities produced stronger telecommunications and network effects than colocation facilities, but wages across the local economy did not increase significantly. The study also found that simplistic comparisons overstated employment effects because developers tended to locate facilities in counties that were already growing.

This distinction matters. Data centers create genuine employment, particularly during construction and throughout their supply chains, but they should not be represented as labor-intensive facilities comparable with major manufacturing plants. Their largest employment contribution may occur outside the facility itself.

The Entry-Level Employment Problem Is Real

The absence of an economy-wide jobs apocalypse does not mean that every group of workers is benefiting.

The strongest evidence of displacement appears among younger workers entering occupations containing codified, repeatable, digitally represented tasks. These workers historically performed research, drafting, data preparation, elementary coding, customer support, document review, and administrative work while accumulating the tacit knowledge required for more senior responsibilities.

Stanford researchers Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen found that employment among workers aged 22 to 25 in highly AI-exposed occupations was approximately 19 percent below where it would have been if it had kept pace with employment among comparable young workers in less-exposed occupations.

Experienced workers did not exhibit a comparable employment gap. The adjustment was also occurring primarily through reduced hiring rather than through layoffs.

This explains why national unemployment statistics can remain stable while recent graduates encounter serious difficulty. A position that is never created does not appear in a corporate layoff announcement. A company that hires three entry-level analysts instead of five has eliminated no existing jobs, but two employment opportunities have disappeared.

The Stanford researchers explicitly caution that their results are descriptive rather than definitive causal estimates. Interest rates, pandemic-era distortions, education, industry conditions, and other variables may explain some of the divergence. Nevertheless, the age pattern, its persistence, and its concentration in occupations where observed AI use substitutes for human tasks make AI a plausible contributor.

The Dallas Federal Reserve reached a related conclusion. Its analysis of millions of online postings found that Texas employers reduced openings for occupations containing tasks that generative AI could automate. The effect was particularly significant for new graduates and people trying to enter or change occupations.

Technical managers therefore face a strategic workforce problem. If AI eliminates the entry-level work through which employees traditionally acquired experience, organizations may eventually exhaust the supply of experienced people needed to verify AI outputs, exercise judgment, manage exceptions, and assume responsibility.

Businesses must deliberately redesign apprenticeships, mentoring, rotational assignments, simulations, and supervised AI-assisted work so that junior employees can still develop tacit knowledge. Eliminating every elementary task may produce a short-term efficiency gain while creating a long-term shortage of experienced professionals.

Employment and Wage Effects Are Dividing Into Two Tracks

Dallas Federal Reserve economist Scott Davis describes the emerging division in terms of codified and tacit knowledge.

Codified knowledge is documented, structured, teachable, and generally available in digital form. Current AI systems can often retrieve, combine, and apply this knowledge effectively. Tacit knowledge develops through experience, repeated exposure, relationships, institutional understanding, and judgment under imperfect conditions.

AI is more likely to substitute for workers whose value rests primarily on codified knowledge. It is more likely to complement workers whose value rests on tacit knowledge and responsibility.

Since the introduction of ChatGPT, employment in computer-systems design and related services declined, as did employment across the most AI-exposed industries. Yet wages in those sectors generally did not collapse. Computer-systems-design wages increased faster than the national average.

Falling employment combined with stable or rising wages is not consistent with a simple story in which AI makes all exposed workers less valuable. It suggests that some positions are being eliminated or consolidated while the remaining experienced and highly capable workers become more productive and valuable.

This creates a two-track labor market. Workers who can combine AI with domain expertise, judgment, customer understanding, systems knowledge, and accountability may command greater responsibility and compensation. Workers performing routine digital production face greater substitution risk.

Why the One-Million Estimate Requires Caution

The Economist’s one-million-job estimate should be treated as an informed approximation rather than an established fact.

Employment above a historical trend does not prove that AI caused the difference. Employment since late 2022 has also been affected by pandemic recovery, interest rates, immigration, semiconductor incentives, infrastructure spending, energy investment, defense demand, supply-chain relocation, tariffs, and the correction of excessive technology hiring during the pandemic.

Broad occupational categories also contain substantial non-AI employment. A project manager supervising a conventional construction project may appear in the same occupational classification as one directing an AI deployment. Business-operations and mathematical occupations similarly encompass activities that cannot all be attributed to AI.

The layoff comparison presents the opposite problem. Challenger, Gray & Christmas reported that AI had been cited in 116,175 announced job cuts during the first eight months of 2026. But employer announcements do not capture reduced hiring, contractor displacement, unfilled vacancies, or employment reductions attributed publicly to other causes. Conversely, a company may describe a broad restructuring as AI-related even when weak demand or ordinary cost reduction contributed to the decision.

The available evidence therefore cannot support a precise five-to-one comparison between jobs created and jobs eliminated. It can support the conclusion that AI-related employment creation is large, while displacement is concentrated and has not overwhelmed the broader economy.

What Technical Management Should Do

The practical question for management is no longer whether AI will eliminate all employment. It is whether an organization will use AI primarily to contract its existing operations or to expand its productive capabilities.

Organizations that treat AI merely as a labor-reduction tool may obtain short-term savings but risk weakening their capacity to innovate, serve customers, develop employees, and capture new markets.

Organizations using AI as a growth platform should:

* identify products, services, and customer groups that were previously too costly or technically difficult to address;
* redesign processes around combinations of human expertise, probabilistic model capabilities, and deterministic controls;
* recruit people who can implement, integrate, secure, evaluate, and operate AI systems;
* preserve human judgment, authority, and accountability at consequential decision points;
* develop new pathways through which early-career employees can acquire experience;
* measure productivity, quality, revenue growth, employment, and risk together rather than measuring payroll reduction alone; and
* invest in the physical infrastructure, electrical capacity, networks, and workforce required to sustain expanding AI use.

The employment outcome is not determined entirely by the technology. It is shaped by how management deploys the technology and what organizations choose to do with the resulting productivity.

The More Probable Direction

The evidence available through September 2026 indicates a good chance that artificial intelligence will never produce the generalized employment apocalypse that has been repeatedly predicted—provided that the United States continues expanding the computing, electrical, communications, and workforce capacity required to meet demand and businesses continue converting AI capabilities into productive commercial use.

That does not mean that no occupations will decline or that every worker will benefit. Writers performing routine content production, data-entry personnel, customer-service representatives, administrative workers, junior analysts, and some software workers already face measurable pressure. Some existing occupations will shrink, and many others will be substantially redesigned.

The decisive counterforce is expansion. If AI enables businesses to produce more, solve previously uneconomic problems, create new products, enter new markets, and provide better services at lower cost, total demand can grow faster than labor requirements decline within any single task.

The data-center and electrical-infrastructure expansion is the first physical phase of that process. Large capital expenditures are creating construction, manufacturing, engineering, and supply-chain employment before most industries have progressed much beyond initial AI experimentation.

The United States has only begun what is likely to become a journey lasting several decades. AI is the most powerful general-purpose technology for reasoning and computing yet developed, but most organizations have not comprehensively integrated it into their operations. Many have experimented with chatbots and isolated automation. Far fewer have redesigned complete products, services, workflows, organizational structures, and markets around AI-enabled productive capacity.

If sufficient data centers, power systems, networks, and trained workforces can be built—and if management directs the resulting capability toward expansion rather than cost reduction alone—the more plausible future is not one without human work. It is one in which work changes rapidly, productivity rises, new capabilities compound, and people who learn to develop, direct, constrain, verify, and apply AI become increasingly important.

References

1. The Economist. “The Jobs Apocalypse Is Postponed. An AI Jobs Boom Is Here.” The Economist, Finance & Economics, September 4, 2026. No individual byline was displayed; published under The Economist’s institutional authorship. Article link⁠. Access: Paid subscription or limited metered access.
2. United States Bureau of Labor Statistics. The Employment Situation—August 2026. United States Department of Labor, published September 4, 2026. Institutional statistical release based on the Current Population Survey and Current Employment Statistics survey. Report link⁠. Access: Free.
3. The Budget Lab at Yale University. Tracking the Impact of AI on the Labor Market. Yale University, published July 16, 2026; updated August 19, 2026. Institutional research analysis; no individual authors are identified on the public page. Report link⁠. Access: Free.
4. Jessica Liu and Douglas Webber. “AI Adoption and Firms’ Job-Posting Behavior.” FEDS Notes, Board of Governors of the Federal Reserve System, March 27, 2026. Analysis using Lightcast job postings and Census Bureau Business Trends and Outlook Survey data. Research note link⁠. Access: Free.
5. Samuel Dodini and Tucker Smith. “Job Postings Show Early Signs of AI Automation Impact.” Dallas Fed Economics, Federal Reserve Bank of Dallas, September 1, 2026. Analysis of Texas employment demand using Lightcast postings and occupation-level AI automation exposure. Article link⁠. Access: Free.
6. Scott Davis. “AI Is Simultaneously Aiding and Replacing Workers, Wage Data Suggest.” Dallas Fed Economics, Federal Reserve Bank of Dallas, February 24, 2026. Analysis of employment, wages, AI exposure, and the distinction between codified and tacit knowledge. Article link⁠. Access: Free.
7. Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen. Canaries in the Coal Mine? Six Facts About the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, revised August 2026. Research using ADP payroll records to examine employment by age, occupation, and AI exposure. Full paper⁠. Access: Free.
8. Stanford Institute for Human-Centered Artificial Intelligence, with labor-market data supplied by Lightcast. The 2026 AI Index Report. Stanford University, 2026. Institutional annual report examining global AI development, investment, adoption, employment, and skills. AI Index information and labor-market findings⁠. Access: Free.
9. PricewaterhouseCoopers. PwC 2026 Global AI Jobs Barometer: Two Futures for Jobs in an AI Era. PwC, published June 15, 2026. Institutional analysis of productivity, wages, headcount, job postings, and changing skill requirements in AI-exposed industries. Report link⁠. Access: Free.
10. Dany Bahar and Greg Wright. “New Evidence on Data Center Employment Effects.” Brookings Institution, originally published May 4, 2026; updated August 10, 2026. Summary of the authors’ research, Data Centers and Local Labor Markets, using approximately 1,500 facilities and comparison counties containing canceled projects. Research summary⁠. Access: Free.
11. Challenger, Gray & Christmas. August 2026 Challenger Report: August Job Cuts Up 58%, Consumer Products, Food Lead. Challenger, Gray & Christmas, published September 2026. Institutional report tracking announced United States job cuts, employer explanations, and hiring plans. Report summary and downloadable report⁠. Access: Free.
12. Timothy Aeppel. “The Unexpected Winners of America’s Data-Center Boom.” Reuters, August 19, 2026. Reporting on manufacturing, supply-chain, capital-investment, and employment effects associated with American data-center expansion. Article link⁠. Access: Generally free with possible registration, regional, or usage limitations.