Loading...
Patterson Consulting Contact Us

Demand

By: Josh Patterson
October 2026

Key Idea:

Firms hire labor because people buy what labor produces, so labor demand is derived from human needs and wants. Body needs such as food level off as they are met. The needs for stimulation and status, and the wants people learn from experience and from each other, keep growing. When automation makes an output cheaper, the work of producing it grows only if the savings reach buyers and buyers want much more at the lower price. The income freed by cheaper goods goes to new goods, and new goods need new work. Automation has expanded total work so far because the wants behind demand have kept growing.

By 2000, lighting in the United Kingdom cost less than one three-thousandth of what it cost in 1800. Over the same two centuries, the amount of light used per person rose about 6,500-fold and total use about 25,000-fold, while income per person rose about 15-fold.[1]

William Nordhaus showed that standard price indexes missed most of this decline, because they tracked the prices of candles and lamps instead of the price of light itself.[2]

Cloth followed the same path. In the United States, labor productivity in cotton cloth production rose about 30-fold over the nineteenth century, consumption rose about 100-fold, and employment in the industry grew.[3] People bought so much more of these goods as they got cheaper that the work of producing them expanded.

What Demand Is

Demand is the quantity of something buyers are willing and able to buy at each price. It depends on the price, on buyers' incomes, and on their preferences.[4]

People mostly want the service a good provides, such as light, warmth, transport, or information. Kelvin Lancaster modeled goods as bundles of characteristics, which helps explain why a need persists while the goods that meet it change, as candles gave way to gas and then electric light.[5]

Labor demand is derived demand. Firms hire labor because people buy what labor produces, so the demand for labor depends on demand for the product, on labor productivity, and on the price of substitutes such as machines.[4]

Demand

The quantity buyers are willing and able to buy at each price, given their incomes and preferences.

Derived demand

Demand for labor that follows from demand for what labor produces.

"Labor Automation" described two effects that pull in opposite directions: machines take over tasks from people, and lower costs raise output and the demand for labor in the tasks that remain. Demand decides which one wins. When machines get cheaper, the demand for labor rises if buyers respond to lower prices more strongly than firms can swap machines for workers, and it falls if they respond less.[6]

Where Demand Comes From

Demand starts with biology. Ulrich Witt lists the needs people share as part of their genetic makeup: body needs such as water, food, sleep, sex, shelter, and clothing, and other needs such as affection, stimulation, a positive self-image, and status.[7] Arthur Robson argues that evolution shaped preferences like these, and that human evolutionary history "logically must be the foundation of economic behavior."[8]

Body needs level off. In 1857, Ernst Engel found that the poorer a family is, the larger the share of its spending that goes to food. A century later, Hendrik Houthakker found the same pattern in about 40 household surveys from some 30 countries.[9] In the United States, food took 42.5 percent of household spending in 1901 and 13.1 percent in 2002–03.[10]

Shares of US household spending. 1901: food 42.5%, clothing 14.0%, housing 23.3%, everything else 20.2%; 1950: food 29.7%, clothing 11.5%, housing 27.2%, everything else 31.6%; 2002–03: food 13.1%, clothing 4.2%, housing 32.8%, everything else 49.9%. 1901 Food 42.5% (1901) 42.5% Clothing 14.0% (1901) 14.0% Housing 23.3% (1901) 23.3% Everything else 20.2% (1901) 20.2% 1950 Food 29.7% (1950) 29.7% Clothing 11.5% (1950) 11.5% Housing 27.2% (1950) 27.2% Everything else 31.6% (1950) 31.6% 2002–03 Food 13.1% (2002–03) 13.1% Clothing 4.2% (2002–03) Housing 32.8% (2002–03) 32.8% Everything else 49.9% (2002–03) 49.9%
Food Clothing Housing Everything else
YearFoodClothingHousingOther
190142.5%14.0%23.3%20.2%
195029.7%11.5%27.2%31.6%
2002–0313.1%4.2%32.8%49.9%
Shares of US household spending, from the Bureau of Labor Statistics report 100 Years of U.S. Consumer Spending (2006). Food and clothing fell from more than half of spending to under a fifth, while housing and everything else grew. The early surveys covered wage-earner and urban households, so the comparison is approximate.

Spending moved to other goods as incomes rose. Diego Comin, Danial Lashkari, and Martí Mestieri find that rising income, more than changes in relative prices, accounts for most of the shift in spending from agriculture to manufacturing and services within countries.[11] Part 1's "Structural Transformation" traced the same shift in where people work, and "Labor Automation" showed what it meant for farm employment.

Other needs do not level off. Witt calls this differential satiation: food satisfies quickly, while the needs for stimulation and status do not, because people adapt to what they have and compare themselves with others.[7] Tibor Scitovsky drew a similar line between comfort and stimulation, writing that "the pleasures of stimulation, unlike those of want satisfaction, are not eliminated by their too persistent and too continuous pursuit."[12]

Most of what people buy meets these needs through wants they have learned. In Witt's account, learned wants grow out of "a limited number of innate wants," and they multiply as people discover new goods and new ways to meet the same needs.[13] "The Evolution of Work" takes up the need for status and why it keeps rising.

Innate need

A need people share as part of their genetic makeup, such as food, sleep, affection, stimulation, or status. Body needs level off as they are met; stimulation and status do not.

Learned want

A desire for particular goods that meet a need, learned from experience and from other people. Learned wants multiply as new goods appear.

Wants for new goods have drawn people into more work before. Jan de Vries argues that in the seventeenth and eighteenth centuries, households in northwestern Europe and North America worked more for the market and sold more of what they produced so they could buy new consumer goods such as sugar, tea, tobacco, and cotton cloth. He calls this the industrious revolution.[14]

The evidence is debated. Robert Allen and Jacob Weisdorf find that English farm laborers did work longer, but they attribute the longer hours to hardship, while leaving room for an industrious revolution among urban workers.[15]

Demand also grows with income, as productivity raises what an hour of work earns ("What Is Work?"), and with population. The first large expansion came with farming, when surplus supported denser settlements that needed record-keepers and administrators, as Part 4's "The Neolithic Age" and "What Is Work?" described.

How Automation Expands Demand

Automation cuts the labor needed for each unit of output. Three things decide whether that turns into more work or less:

  • whether the savings reach buyers as lower prices
  • how much more buyers want when the price falls
  • where the income freed by cheaper goods is spent

The savings have to reach buyers. On average they do. Across 19 developed economies, David Autor and Anna Salomons find that rising productivity in an industry lowers its prices.[16] AT&T, as "Labor Automation" noted, did not pass its savings from dial switching on to callers: its rates typically rose after a city converted.[17] How much of a cost cut reaches prices depends on competition and on the shape of demand, and a firm with market power can keep more of it as margin.[18]

Buyers have to want more. The key quantity is the price elasticity of demand, which measures how strongly the quantity bought responds to the price. When elasticity is above 1, a price cut raises total spending on the output, and with it the work of producing it. When elasticity is below 1, total spending falls.

Illustrative example: price and labor per unit both fall 40 percent. With elastic demand (elasticity 2), quantity bought rises about 2.8 times and labor hours rise about 67 percent. With inelastic demand (elasticity 0.5), quantity rises about 29 percent and labor hours fall about 23 percent. Elastic demand elasticity 2 After: quantity 2.78, spending and labor hours 1.67 (+67%) Before: quantity 1.00, spending and labor hours 1.00 1.0 0.6 1.0 2.8 Price Quantity bought before Labor hours +67% Inelastic demand elasticity 0.5 After: quantity 1.29, spending and labor hours 0.77 (−23%) Before: quantity 1.00, spending and labor hours 1.00 1.0 0.6 1.0 1.3 Price Quantity bought before Labor hours −23%
Spending before the price cut Spending after
Elastic (2)Inelastic (0.5)
Price and labor per unit−40%−40%
Quantity bought×2.8+29%
Spending and labor hours+67%−23%
An illustrative case: automation cuts the labor needed per unit by 40 percent, and the price falls by the same share. Each box is price times quantity, which here is proportional to labor hours. With elastic demand the work grows; with inelastic demand it shrinks.

James Bessen found that automation raised employment in textiles, steel, and automobiles for decades while demand for their products was elastic. Once demand was largely satisfied, further productivity growth reduced employment in those industries.[3] Automation that barely lowers costs, the "so-so technologies" described in "Labor Automation," gives buyers little reason to buy more.

Freed income goes somewhere. Energy economists call the extra demand that follows an efficiency gain the rebound effect. William Stanley Jevons made the original observation in 1865, when he argued that more efficient steam engines would increase Britain's total use of coal.[19] The rebound comes through three channels:[20]

  • direct: more of the same output is bought
  • indirect: the savings are spent on other goods
  • economy-wide: growth raises demand across the economy

Autor and Salomons measured these channels for employment, across 28 industries from 1970 to 2007. Productivity growth reduced employment in the industries where it occurred. Employment grew in the industries that bought their output and through added final demand, and the gains more than made up the loss.[16] "The Value of Human Labor" covers the other half of their finding, the fall in labor's share of income.

Much of the freed income goes to services. William Baumol showed that when demand for services with slow productivity growth, such as education and the performing arts, holds up as their relative costs rise, those services take a growing share of spending and employment.[21] Baumol's mechanism helps explain why in-person services kept growing, as "Physical Labor and Cognitive Labor" showed.

How Demand Creates New Work

New goods bring new work. As cars replaced horse travel in the 1920s, roadside motels and fast-food restaurants grew up to serve the "motoring public."[22]

Most of today's jobs carry titles that did not exist a few generations ago, as "What Is Work?" noted. David Autor and his coauthors trace this new work to two sources: augmentation, which "Labor Automation" described, and rising demand for what an occupation produces. They find that "the flow of new work responds elastically to shifts in demand for occupational output."[23]

Demand works in both directions. Occupations hit by competition from Chinese imports created fewer new job titles, and occupations that gained from demographic shifts created more, especially in personal and care services.[23] In a follow-up study, workers in counties that received large industrial plants during World War II were more likely to hold new job titles by 1950, and most of that new work was linked to new technology.[24]

New work usually starts informally, as new tasks inside existing jobs. When demand grows enough, firms formalize a role, the title enters the Census list of occupations, and training programs follow. The wage premium for the new expertise is highest while the title is new and fades over the following decades, as "The Value of Human Labor" described.

Demand for Cognitive Work

LLMs are lowering the cost of steps in knowledge work quickly. Epoch AI estimates that the cost of reaching a given score on benchmark tasks has fallen about 13-fold a year since 2023, counting the extra text that reasoning models generate.[25]

Use has grown faster. Google processed about 9.7 trillion tokens a month across its products in 2024 and more than 3.2 quadrillion a month by May 2026, more than 300 times as many.[26] Machine-generated text is following the path lighting took: as its price falls, people use far more of it.

Whether that expands human cognitive work depends on how the machine output relates to what people do. Where it substitutes for human output, demand for the human work has fallen so far. On a large freelancing platform, demand for writing and translation dropped 20 to 50 percent below its prior trend after ChatGPT's release, while demand for machine-learning programming rose 24 percent and demand for building AI chatbots nearly tripled.[27] Another study found 21 percent fewer postings for writing and coding jobs that were easy to automate, compared with jobs that needed manual skills, and the jobs that remained were more complex and better paid.[28]

Demand for the substituted output did not grow enough to offset the substitution, and new work appeared around the new tools. The early labor-market evidence in "Labor Automation" points the same way: jobs that LLMs augment have held up so far, and entry-level hiring has slowed where AI substitutes for workers.

Over time, the result depends on the outputs firms sell. Some, such as software or tailored analysis, may find many more buyers at a lower price. Others are fixed by rules or by the calendar: a firm files one annual report no matter how cheap reports become.

For a business, the practical question in any automation plan is where the savings go: into lower prices in a market that will buy more, or into margin in a market that will not.

Where the Demand Channel Breaks

Demand does not always absorb displaced labor:

  • Some demand is fixed, and some markets saturate, as Bessen's industries eventually did.
  • New work arrives later than displacement. The incumbent telephone operators in "Labor Automation" were more likely to end up in lower-paying work or out of work, and it was later cohorts who found the new jobs.
  • If most of the gains go to owners of capital, as "The Value of Human Labor" discussed, demand can lag, because households with high incomes save a larger share of what they earn.[29]

Wassily Leontief compared human workers to horses, whose role in farming shrank and then disappeared once tractors arrived.[30] The comparison breaks down on demand. Horses do not buy anything, while people are both the workers and the source of demand. "Engines of Creation" takes up the harder version of the question: what happens if machines can do every task people can.

Every channel in this article depends on wants that keep growing. Spending on food stopped keeping pace with income long ago, and the needs and wants that took its place keep rising. "The Evolution of Work," the next article in Part 5, explains why.

References

  1. Fouquet, R., & Pearson, P. J. G. (2006). Seven centuries of energy services: The price and use of light in the United Kingdom (1300–2000). Energy Journal, 27(1), 139–177. doi:10.5547/ISSN0195-6574-EJ-Vol27-No1-8
  2. Nordhaus, W. D. (1997). Do real-output and real-wage measures capture reality? The history of lighting suggests not. In T. F. Bresnahan & R. J. Gordon (Eds.), The Economics of New Goods (pp. 29–66). University of Chicago Press.
  3. Bessen, J. (2019). Automation and jobs: When technology boosts employment. Economic Policy, 34(100), 589–626. doi:10.1093/epolic/eiaa001
  4. Marshall, A. (1890). Principles of Economics. Macmillan. [derived demand, Book V, ch. VI]
  5. Lancaster, K. J. (1966). A new approach to consumer theory. Journal of Political Economy, 74(2), 132–157. doi:10.1086/259131
  6. Hicks, J. R. (1932). The Theory of Wages. Macmillan. For the modern statement, see Hamermesh, D. S. (1993). Labor Demand. Princeton University Press. [The cross-price elasticity of labor demand with respect to the price of capital is sK(σ − η), where sK is capital's cost share, σ the elasticity of substitution, and η the price elasticity of product demand. When capital gets cheaper, labor demand rises if η > σ.]
  7. Witt, U. (2017). The evolution of consumption and its welfare effects. Journal of Evolutionary Economics, 27(2), 273–293. doi:10.1007/s00191-016-0459-3 [innate needs as "human universals," Section 2; the differential satiation hypothesis, Section 3]
  8. Robson, A. J. (2001). The biological basis of economic behavior. Journal of Economic Literature, 39(1), 11–33, at p. 11. doi:10.1257/jel.39.1.11
  9. Engel, E. (1857). Die Productions- und Consumtionsverhältnisse des Königreichs Sachsen. Zeitschrift des Statistischen Bureaus des Königlich Sächsischen Ministeriums des Innern, Nos. 8–9. Reprinted in Engel, E. (1895), Die Lebenskosten belgischer Arbeiter-Familien früher und jetzt, Anlage I [the law, pp. 28–29 of the reprint]. Houthakker, H. S. (1957). An international comparison of household expenditure patterns, commemorating the centenary of Engel's law. Econometrica, 25(4), 532–551. doi:10.2307/1905382
  10. U.S. Bureau of Labor Statistics. (2006). 100 Years of U.S. Consumer Spending: Data for the Nation, New York City, and Boston (Report 991; M. L. Dolfman & D. M. McSweeney). bls.gov/opub/100-years-of-u-s-consumer-spending.pdf [1901: p. 6, Table 5; 1950: p. 25, Table 14; 2002–03: p. 63, Table 29. The report cautions that comparisons across surveys "should be interpreted with caution"; the early figures come from surveys of wage-earner and urban households.]
  11. Comin, D., Lashkari, D., & Mestieri, M. (2021). Structural change with long-run income and price effects. Econometrica, 89(1), 311–374. doi:10.3982/ECTA16317
  12. Scitovsky, T. (1976). The Joyless Economy: An Inquiry into Human Satisfaction and Consumer Dissatisfaction. Oxford University Press, pp. 77–78.
  13. Witt, U. (2001). Learning to consume: A theory of wants and the growth of demand. Journal of Evolutionary Economics, 11(1), 23–36. doi:10.1007/PL00003851 For a review of the research that followed, see Chai, A. (2017). Tackling Keynes' question: A look back on 15 years of Learning To Consume. Journal of Evolutionary Economics, 27(2), 251–271. doi:10.1007/s00191-016-0455-7
  14. de Vries, J. (1994). The Industrial Revolution and the industrious revolution. Journal of Economic History, 54(2), 249–270. doi:10.1017/S0022050700014467 See also de Vries, J. (2008). The Industrious Revolution: Consumer Behavior and the Household Economy, 1650 to the Present. Cambridge University Press. doi:10.1017/CBO9780511818196
  15. Allen, R. C., & Weisdorf, J. L. (2011). Was there an "industrious revolution" before the industrial revolution? An empirical exercise for England, c. 1300–1830. Economic History Review, 64(3), 715–729. doi:10.1111/j.1468-0289.2010.00566.x See also Clark, G., & Van Der Werf, Y. (1998). Work in progress? The industrious revolution. Journal of Economic History, 58(3), 830–843. doi:10.1017/S0022050700021197
  16. Autor, D., & Salomons, A. (2018). Is automation labor share–displacing? Productivity growth, employment, and the labor share. Brookings Papers on Economic Activity, Spring 2018, 1–87. doi:10.1353/eca.2018.0000 [19 countries, 28 industries, 1970–2007. Annual employment effects in log points: own industry −0.222; industries downstream of the productivity gain +0.353; upstream +0.051; final demand +0.301; net +0.482.]
  17. Feigenbaum, J., & Gross, D. P. (2024). Answering the call of automation: How the labor market adjusted to mechanizing telephone operation. Quarterly Journal of Economics, 139(3), 1879–1939. doi:10.1093/qje/qjae005 [Dial cutovers "were typically accompanied by telephone rate increases, rather than decreases": Appendix Figure A.4.]
  18. Weyl, E. G., & Fabinger, M. (2013). Pass-through as an economic tool: Principles of incidence under imperfect competition. Journal of Political Economy, 121(3), 528–583. doi:10.1086/670401
  19. Jevons, W. S. (1865). The Coal Question: An Inquiry Concerning the Progress of the Nation, and the Probable Exhaustion of Our Coal-Mines. Macmillan. The modern rebound literature begins with Khazzoom, J. D. (1980). Economic implications of mandated efficiency standards for household appliances. Energy Journal, 1(4).
  20. Sorrell, S., & Dimitropoulos, J. (2008). The rebound effect: Microeconomic definitions, limitations and extensions. Ecological Economics, 65(3), 636–649. doi:10.1016/j.ecolecon.2007.08.013
  21. Baumol, W. J. (1967). Macroeconomics of unbalanced growth: The anatomy of urban crisis. American Economic Review, 57(3), 415–426.
  22. Autor, D. H. (2015). Why are there still so many jobs? The history and future of workplace automation. Journal of Economic Perspectives, 29(3), 3–30, at p. 7. doi:10.1257/jep.29.3.3 [citing Jackson, K. T. (1993), Crabgrass Frontier]
  23. Autor, D., Chin, C., Salomons, A., & Seegmiller, B. (2024). New frontiers: The origins and content of new work, 1940–2018. Quarterly Journal of Economics, 139(3), 1399–1465. doi:10.1093/qje/qjae008
  24. Autor, D., Chin, C., Salomons, A., & Seegmiller, B. (2026). What Makes New Work Different from More Work? NBER Working Paper 34986. doi:10.3386/w34986 [WWII plants and new work by 1950; 85 to 90 percent of that new work was technology-linked]
  25. Emberson, L., & Roodman, D. (2026, September 22). The plunging price of thought. Epoch AI. [about 47 percent a quarter, or 13-fold a year, since 2023, for the cost of reaching a fixed benchmark score] An earlier Epoch AI analysis of per-token prices found declines of 9- to 900-fold a year, with a median of 50-fold: Cottier, B., Snodin, B., Owen, D., & Adamczewski, T. (2025, March 12). LLM inference prices have fallen rapidly but unequally across tasks. Epoch AI.
  26. Pichai, S. (2026, May 19). I/O 2026: Welcome to the agentic Gemini era. Google. blog.google/innovation-and-ai/sundar-pichai-io-2026 [9.7 trillion tokens a month two years earlier; more than 3.2 quadrillion a month in May 2026]
  27. Teutloff, O., Einsiedler, J., Kässi, O., Braesemann, F., Mishkin, P., & del Rio-Chanona, R. M. (2025). Winners and losers of generative AI: Early evidence of shifts in freelancer demand. Journal of Economic Behavior & Organization, 235, 106845. doi:10.1016/j.jebo.2024.106845
  28. Demirci, O., Hannane, J., & Zhu, X. (2025). Who is AI replacing? The impact of generative AI on online freelancing platforms. Management Science, 71(10), 8097–8108. doi:10.1287/mnsc.2024.05420 [21 percent fewer posts for automation-prone writing and coding jobs within eight months of ChatGPT's release, relative to jobs requiring manual skills]
  29. Keynes, J. M. (1936). The General Theory of Employment, Interest and Money. Macmillan, Book III, ch. 8, pp. 96–97. [Kuznets's long-run data later showed no rise in the share of income saved as incomes grew, which led to the permanent-income and life-cycle theories. The point here concerns differences between households at a point in time.]
  30. Leontief, W. (1983). Technological advance, economic growth, and the distribution of income. Population and Development Review, 9(3), 403–410. doi:10.2307/1973315