🔍 Read the full analysis: AI Employment: 80% Is All Services on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
US services accounted for about 79.6% of employment in 2025, according to International Labour Organization estimates cited by ThorstenMeyerAI.com. The November 9 report connects the long-term shift to Baumol’s cost disease and cites a study in which an AI assistant helped customer-support agents resolve 15% more issues per hour on average. How productivity gains will affect service employment remains unsettled.
A November 9 report from ThorstenMeyerAI.com says services accounted for about 79.6% of US employment in 2025, while examining whether artificial intelligence can change productivity in work such as customer support, health care and education. The report connects the service sector’s long growth to economist William Baumol’s theory that wages and costs can rise in jobs where human labor is difficult to reduce, even as productivity improves elsewhere.
The employment figures show that the shift toward services predates the decline in factory jobs. The report’s calculation from Bureau of Labor Statistics payroll data puts service-providing industries at 62% of nonfarm payroll jobs in 1939, 67% in 1967, 72% in 1979 and 86% in 2025. These figures exclude farm workers and self-employed people. A separate International Labour Organization estimate, which counts all workers, puts services at 79.6% of US employment in 2025.
The report uses Baumol’s cost disease to explain why some services can become more expensive without becoming proportionately more productive. A string quartet still needs four musicians to perform a roughly 45-minute piece; reducing the number of performers or the time spent playing could change the product itself. By contrast, manufacturing methods can increase output per worker. If wages rise across the economy, labor-intensive services may face higher costs even when their output per worker changes little.
Prices cited in the report illustrate the contrast. Since 2000, the official, quality-adjusted US price of a television fell 98%, while hospital services rose 274%, college tuition 188% and car repair 143%. Those figures cover different goods and services and do not establish that one cause explains every increase. The report also notes that costs vary between countries: it says the United States spent 18.0% of its economy on health care, compared with 12.4% in Germany.
Infographic 02 · Baumol's cost disease
Why the workers moved to services
Baumol's mechanism in four steps, and what it looks like in US jobs.
- 1966Baumol and Bowen: the quartetA 45-minute Schubert quartet still takes about three hours of human work
- 1967Two kinds of workWhere labour is a tool, machines cut it again and again; where labour is the product, they can’t
- WagesPay rises everywhereTeachers, nurses and musicians must earn more too, or they leave
- ResultCosts rise, people moveIf we keep buying the same mix, more and more workers end up in the slow sector
Service industries, share of US nonfarm payroll jobs
US jobs, annual average 2025 (millions)
Payroll shares exclude farm workers and the self-employed (BLS establishment survey). Health spending: national health expenditure as a share of gross domestic product.
Sources: W. J. Baumol, “Macroeconomics of Unbalanced Growth”, American Economic Review 57(3), 1967, BLS Current Employment Statistics (own calculation of shares), CMS, National Health Expenditure, Table 1.
Powered by Thorsten Meyer AI
AI Meets Labor-Heavy Services
The economic question is not simply whether AI can complete a task. It is whether the technology can raise output without reducing the quality or personal attention that people expect from a service. In manufacturing, a machine may produce more units with fewer hours of labor. In teaching, medical care or live performance, more output per worker can be harder to define: a larger class or shorter appointment may not be treated as an improvement.
The report points to a study of 5,172 customer-support agents in which an AI assistant increased the number of issues resolved per hour by 15% on average. That result suggests AI can help in at least some service tasks. It does not show that the same gain applies across the service sector, that jobs will disappear, or that customers will receive better service. The study’s measured outcome was issues resolved per hour.
For workers and employers, the distinction matters. If AI helps employees handle routine work, organizations could serve more customers or let staff focus on harder cases. If productivity gains instead lead to fewer staff, the effects on employment and service access could differ. The source presents that as an open question, not a settled forecast.
Products Worth Considering
EcoFlow DELTA 3 Portable Power Station, 1024Wh, 1800W
Bose Lifestyle Ultra Soundbar and Ultra Subwoofer Bundle
LG UltraGear 32GX850A-B 32-inch 4K UHD OLED Gaming Monitor
As an affiliate, we earn on qualifying purchases.
From Quartet to Cost Disease
William Baumol and William Bowen examined the finances of performing arts in their 1966 book, Performing Arts: The Economic Dilemma. Their example of a Schubert quartet captured a basic productivity difference: a live performance requires a set amount of musicians’ time, while manufacturing has repeatedly found ways to reduce labor needed per product. Baumol developed the broader argument in his 1967 paper, “Macroeconomics of Unbalanced Growth: The Anatomy of Urban Crisis.”
Baumol described activities where labor is mainly an input that technology can reduce, and activities where labor is closely tied to the service being delivered. He treated this as a difference of degree, not a strict division: retail, for example, has adopted self-service, while technology can make some forms of teaching or communication reach more people. The report’s central point is that service employment was already growing well before the US manufacturing job peak in June 1979, when manufacturing employed almost 20 million people, and continued to grow afterward.
The source is an episode in the series “Work After Work,” published by ThorstenMeyerAI.com on November 9, 2026. Its numbers combine cited public data, the author’s calculation from BLS payroll statistics and a referenced customer-support study. The distinctions between those sources matter: the employment series uses different worker populations, while the price and AI figures measure separate outcomes.
““When someone purchases an air conditioner he neither knows nor cares how much labor went into it.””
— William Baumol, quoted in the report from his 1967 paper
What AI Gains Could Mean
The report does not establish whether AI will reduce total service-sector employment, change the kinds of jobs available, or mainly help existing workers handle more tasks. The 15% result comes from one study of customer-support agents; the source does not provide enough detail here to assess how representative the participants or tasks are, or whether the gain lasts over time.
It is also unclear how far the cited productivity result can extend to services where human contact is central. The report does not claim that AI has resolved the cost pressures in health care or education, nor does it show that lower labor costs would translate into lower prices. Service costs reflect more than staffing, and the United States-Germany spending comparison does not, by itself, identify the causes of the difference.
The employment shares also use distinct definitions. The BLS percentages refer to payroll jobs outside farming and exclude the self-employed; the ILO figure is an estimate of all workers. They should not be read as interchangeable measures. The source does not provide a forecast for future employment shares or a date by which AI might substantially change them.
The Test for Service Productivity
The next development to watch is whether AI productivity findings hold across more employers, service occupations and time periods. Further studies would need to distinguish faster task completion from better outcomes for customers and workers. For customer support, that means examining not only how many issues are resolved per hour but also whether issues are resolved accurately and whether customers need to return for help.
For employment, future data can show whether service jobs continue to grow, how roles change and whether measured output per worker rises. The report sets up those questions but does not give a timeline or predict a particular outcome. Its immediate evidence is a long rise in services and a productivity result in one customer-support study; the scale of AI’s wider effect remains open.
Key Questions
What share of US workers were employed in services in 2025?
The report cites an International Labour Organization estimate of 79.6% of all US workers in 2025. Its separate BLS-based calculation puts service-providing industries at 86% of nonfarm payroll jobs that year; that measure excludes farm workers and the self-employed.
What is Baumol’s cost disease?
It is an economic explanation for why labor-intensive services can become more expensive as wages rise across the economy, even when it is difficult to increase output per worker. A live quartet, for example, still requires musicians to perform the music.
Did AI reduce customer-support employment in the study?
The result cited in the report does not answer that question. It says an AI assistant raised issues resolved per hour by 15% on average among 5,172 agents. The source does not report whether the study found fewer jobs.
Does the report say AI will make services cheaper?
No. It describes a measured productivity gain in one customer-support study, but does not establish that the gain will lower prices or apply across services such as health care and education.
Source: ThorstenMeyerAI.com
