AI Labor Tracker / Foundations

Bigger than the Industrial Revolution? Testing the claim

Generative AI is spreading faster than electricity, the car or the internet, but history says the payoff and the pain both arrive on a delay, and the delay is where workers get hurt.

Here is a claim worth testing: AI will turn out to be a bigger deal than the Industrial Revolution. I happen to believe it. But an opinion that strong should have to survive contact with the numbers.

So run the test. Take the biggest technologies people have ever adopted, from fire to steam, electricity, the car, flight and the internet. Score AI against them on three things: how fast it spread, how many jobs it destroyed and created, and how long the transition hurt before the benefits reached ordinary workers.

The short answer: AI wins on speed by a mile. On jobs and pain, the history is a warning, not a comfort.

Speed: nothing has spread like this

Start with the American household, where the long-run data are best. Economists Diego Comin and Bart Hobijn assembled adoption series for dozens of technologies, published by Our World in Data.

Electric power reached 10% of US households in 1908 and 52% in 1925, about 17 years. The automobile went from 10% in 1915 to 49% in 1925, about 10 years. The internet went from 10% in 1993 to 52% in 2002, about nine years.

Generative AI did not even start at 10%. By the third quarter of 2024, about two years after ChatGPT launched, 44.6% of working-age American adults were already using it, according to the Real-Time Population Survey run by economists Alexander Bick, Adam Blandin and David Deming and published on FRED.

It passed half in the second quarter of 2025 and stood at 60.8% in the third quarter of 2026. In their NBER paper, the same economists found that, measured from each technology’s first mass-market product, overall adoption has been faster than either the personal computer or the internet.

Now the asterisk. A household that wired itself for electricity or bought a car spent real money. An adult who tried a free chatbot once has spent nothing.

The same NBER paper found generative AI assisted with only 1% to 5% of all work hours in late 2024. Trying is not the same as depending.

Even so, the speed is real, and it matters. Across 15 technologies and 166 countries, Comin and Hobijn found that poorer countries adopted new technologies an average of 45 years after the leaders, and that the lag has shrunk for newer technologies (American Economic Review, 2010). A tool that ships over the internet, on phones people already own, skips most of that wait.

On speed, my claim holds. Score one for AI.

Fire: the original time-saving technology

It is worth going back further than steam, because the biggest technology in human history was arguably the first.

Harvard anthropologist Richard Wrangham argues cooking reshaped our bodies. In an interview with Scientific American, he noted that a primate our size would be expected to spend about half its day chewing, while modern humans spend less than an hour.

“So you’ve got four or five hours a day freed by the fact that you’re eating relatively soft food,” he said. He believes Homo erectus had mastered cooking by 1.8 million years ago, a date his critics dispute.

Fire did not eliminate work. It freed hours, and early humans filled them with other work, mostly hunting, according to Wrangham.

That is the oldest data point we have on what people do with time a technology gives back. They found something else to do.

Steam: the long fuse

Steam is the technology people mean when they say “Industrial Revolution,” and its most surprising feature is how slowly it paid off.

James Watt patented his improved engine in 1769. Yet steam only reached parity with water as a power source in Britain in 1830, according to economic historian Nicholas Crafts (LSE working paper).

His published results “indicate that steam contributed little to growth before 1830 and had its peak impact about a hundred years after Watt’s famous invention” (Economic Journal, 2004). Only with high-pressure steam after 1850 did it take off.

A hundred years. That is the first thing anyone predicting an instant AI productivity boom should sit with.

The long wait for wages

The worst part of the Industrial Revolution was not the machines. It was the decades when output grew and workers’ pay did not.

Economic historian Robert Allen calls it “Engels’ pause,” after Friedrich Engels, who described the misery of Manchester’s workers in the 1840s. Using standard estimates, Allen shows that British output per worker rose 46% between 1780 and 1840, while real wages rose 12% (Allen, 2009).

Labour’s share of national income fell from 50% in 1801 to 45% in 1841. Profits took the gains.

Then it flipped. From 1840 to 1900, output per worker rose 90% and real wages rose 123%, by Allen’s figures.

The gains did reach workers. It took roughly two generations.

Here is why that matters today. American labor’s share of business output fell to 52.8% in the second quarter of 2026, the lowest since 1947, according to the Bureau of Labor Statistics. Productivity rose 2.2% over the year to mid-2026 while real hourly compensation slipped 0.1% (BLS).

One year is not a pause. The labor share was falling long before AI. But the shape is familiar enough that I want us watching it every quarter, and we will.

The second worry is who pays. Allen’s numbers are averages.

Inside them were handloom weavers whose trade disappeared. That pattern repeats in every transition that follows.

Electricity: everywhere but in the productivity statistics

Electricity is the best match for AI, because it is a general-purpose technology that factories had to reorganize around.

In 1899 electric lighting reached only 3% of American homes, and electric motors provided under 5% of factory mechanical drive, according to Stanford economist Paul David’s classic 1990 paper (American Economic Review). He wrote that in 1900 “contemporary observers well might have remarked that the electric dynamos were to be seen ‘everywhere but in the productivity statistics!’” (full text).

Factory electrification did not lift productivity until the early 1920s, which David notes was four decades after the first central power station opened. The reason was organizational.

Factories built around one big steam shaft had to be torn up and redesigned around small motors at each machine. Once they were, the gains were large: David attributes about half of the jump in manufacturing productivity growth in the 1920s to the spread of those motors.

So where is AI on that curve? US productivity has grown 2.1% a year since late 2019, compared with 1.5% from 2007 to 2019, according to the BLS. That is better, but it is not yet the 1920s.

Most companies are still bolting AI onto old workflows, the equivalent of putting a motor where the steam engine used to be. The real gains come when the work itself is redesigned. So does the real disruption.

The car: a whole species of work, retired

The automobile shows what full replacement looks like, because it happened to an entire workforce, just not a human one.

Horses and mules on American farms peaked at 25.2 million in the 1920 census. By 1959 there were under 3 million, while tractors went from about 246,000 to almost 4.5 million, according to the USDA.

The people did better than the horses, but not painlessly. Farm workers made up 41% of the US workforce in 1900 and 1.9% of employed people by 2000, according to the USDA Economic Research Service. Most of that shift took place across generations, as children left the farm for factories and offices their parents never worked in.

This is the comparison people reach for when they ask whether humans could go the way of the horse. The honest answer is that horses had nowhere else to go. People did, because the same economy that replaced them invented new kinds of work.

Flight and the internet: new work, fast

Air travel shows how quickly a technology can scale once it is cheap enough. Airlines carried 310 million passengers worldwide in 1970 and 4.46 billion in 2019, according to the World Bank. We could not find a clean ledger of jobs it eliminated, so we will not claim one.

The internet gives us better job data. Internet users rose from 15.6% of the world’s population in 2005 to 73.6% in 2025 (World Bank). It wiped out jobs and created others, and the best recent research on translators shows how that works in practice.

Oxford’s Carl Benedikt Frey and Pedro Llanos-Paredes studied 695 US local labor markets and found that each percentage-point increase in use of Google Translate cut translator employment growth by 0.71 points. They estimate about 28,000 fewer translator jobs were created than otherwise would have been (Oxford Martin School).

The jobs were not fired away. They were never created.

An older case makes the same point. Between 1920 and 1940, AT&T replaced telephone operators with mechanical switching in over half the US network.

Economists James Feigenbaum and Daniel Gross found it did not reduce employment for the next generation of young women, who moved into clerical and service work. But incumbent operators were, a decade later, “more likely to be in lower-paying occupations or no longer working” (NBER).

That is the pattern in one sentence. The economy recovers. The specific people displaced often don’t.

And over the long run, the new work wins. MIT’s David Autor and colleagues found that “the majority of current employment is in new job specialties introduced after 1940” (NBER).

The scorecard

Speed: AI wins easily. No technology has reached a majority of adults this fast.

Jobs destroyed and created: too early to score. History says creation outruns destruction over decades. The translator and telephone studies say it does not feel that way to the person whose job went first.

Early AI data look similar. Stanford economists found workers aged 22 to 25 in the most AI-exposed jobs had employment 19% below trend, while experienced workers did not (Stanford Digital Economy Lab).

Length of the pain: steam took a century to peak, Engels’ pause lasted about 60 years, electricity needed four decades to show up in productivity. My bet is that AI’s transition will be shorter, because the adoption lag that slowed every earlier technology is close to zero.

But shorter is not painless. A transition that takes 15 years instead of 60 still swallows a career.

So is it bigger than the Industrial Revolution? On reach and speed, I think the answer is yes. On the part that matters to workers, whether the gains are shared, we don’t know yet.

The Industrial Revolution eventually raised everyone’s living standards. It also made two generations wait. For a business owner, the useful question is the same one on a smaller scale: how long does it take for a new tool’s gains to reach the people using it?

Predicting timelines is not my strong suit, which is exactly why this publication runs on data instead of my hunches.

What to do with the delay

The lesson of every one of these transitions is that the lag is where people get hurt, and the lag is also where people get ahead.

For employers, the advice is not to wait for AI to show up in the productivity statistics. Redesign one workflow end to end, the way the 1920s factories rebuilt around small motors, and train the people who do that work first.

If you are a worker, take the lesson from the telephone operators: the next generation did fine because it moved toward the new work early. Learn the tools on your current job, and look hard at the fields that are hiring now, including skilled trades you can enter without a degree.

The people who came out of every past transition well were the ones who put the new tool in their own hands before someone else used it to replace them. Pick one tool this month, use it on real work, and make sure the gains show up in your team’s pay or skills, not just the quarterly numbers.

Sources

1. Comin and Hobijn and others, “Technology adoption by households in the United States,” Our World in Data, accessed Nov 1, 2026. https://ourworldindata.org/grapher/technology-adoption-by-households-in-the-united-states
2. FRED, Federal Reserve Bank of St. Louis, “Generative AI adoption rate, overall, working-age adults” (RPSGENAIUSAGESHAREALL), Real-Time Population Survey, accessed Nov 1, 2026. https://fred.stlouisfed.org/series/RPSGENAIUSAGESHAREALL
3. Bick, Blandin and Deming, “The Rapid Adoption of Generative AI,” NBER Working Paper 32966. https://www.nber.org/papers/w32966
4. Comin and Hobijn, “An Exploration of Technology Diffusion,” American Economic Review 100(5), 2010. https://www.aeaweb.org/articles?id=10.1257/aer.100.5.2031
5. Kate Wong, “Case for (Very) Early Cooking Heats Up,” Scientific American, Sep 1, 2013. https://www.scientificamerican.com/article/case-for-very-early-cooking-heats-up/
6. Nicholas Crafts, “Steam as a General Purpose Technology: A Growth Accounting Perspective,” Economic Journal 114(495), 2004. https://ideas.repec.org/a/ecj/econjl/v114y2004i495p338-351.html
7. Nicholas Crafts, same title, LSE Economic History Working Paper 75/03, May 2003. https://www.lse.ac.uk/Economic-History/Assets/Documents/Research/LSTC/wp7503.pdf
8. Robert C. Allen, “Engels’ pause,” Explorations in Economic History 46(4), 2009. https://www.nuffield.ox.ac.uk/Users/Allen/engelspause.pdf
9. Paul A. David, “The Dynamo and the Computer,” American Economic Review 80(2), 1990. https://ideas.repec.org/a/aea/aecrev/v80y1990i2p355-61.html
10. Paul A. David, 1990, full-text PDF mirror. https://gwern.net/doc/economics/automation/1990-david.pdf
11. USDA NASS, “Trends in U.S. Agriculture: Farm Mechanization, 1900-1997.” https://www.nass.usda.gov/Publications/Trends_in_U.S._Agriculture/Mechanization/mechanization.csv
12. Dimitri, Effland and Conklin, “The 20th Century Transformation of U.S. Agriculture and Farm Policy,” USDA ERS EIB-3, June 2005. https://ers.usda.gov/media/8254/eib-3.pdf
13. World Bank, “Air transport, passengers carried,” updated Oct 8, 2026. https://data.worldbank.org/indicator/IS.AIR.PSGR
14. World Bank, “Individuals using the Internet (% of population).” https://data.worldbank.org/indicator/IT.NET.USER.ZS
15. Feigenbaum and Gross, “Answering the Call of Automation,” NBER Working Paper 28061, revised Feb 2024. https://www.nber.org/papers/w28061
16. Frey and Llanos-Paredes, “Lost in Translation: Artificial Intelligence and the Demise of Foreign Language Translators,” Oxford Martin School, Mar 2025. https://oms-www.files.svdcdn.com/production/downloads/academic/Frey_LlanosParedes_2025_LostInTranslation.pdf
17. Autor, Chin, Salomons and Seegmiller, “New Frontiers: The Origins and Content of New Work, 1940-2018,” NBER Working Paper 30389. https://www.nber.org/papers/w30389
18. U.S. Bureau of Labor Statistics, “Productivity and Costs, Second Quarter 2026, Revised,” Sep 3, 2026. https://www.bls.gov/news.release/prod2.htm
19. U.S. Bureau of Labor Statistics, “Labor share at its lowest level, 52.8 percent, in second quarter 2026,” Sep 15, 2026. https://www.bls.gov/opub/ted/2026/labor-share-at-its-lowest-level-52-8-percent-in-second-quarter-2026.htm
20. Brynjolfsson, Chandar and Chen, “Canaries in the Coal Mine?,” Stanford Digital Economy Lab, revised Aug 12, 2026. https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/