The motor was already there
In 1899, electric motors made up less than 5% of the mechanical power in American factories. It took roughly another twenty years to reach half.
They made no mark on manufacturing productivity growth before the early 1920s. By then, four decades had gone by since the first central power station opened.
The economist Paul David told this story in “The Dynamo and the Computer”, in the American Economic Review of May 1990. It was the era of the line credited to Robert Solow: we see computers everywhere except in the productivity statistics. David’s answer: in 1900, you could have said the same about the dynamo.
A new motor in an old factory
A factory usually had one source of power, at its centre: steam or water. It turned long shafts hung from the ceiling, and belts carried the power down to the machines.
The first electric motor simply took the place of the steam engine. The shafts, belts and pulleys all stayed. The only thing that changed was what turned the shaft.
Then came a halfway stage: one motor for each group of machines. In existing factories it was typically added to the plant, and the old equipment stayed where it was.
Owners back then weren’t getting their sums wrong. Scrapping a factory that still worked simply didn’t pay.
When the factory changed
The real shift came with a motor on every machine. With no shafts overhead, buildings could be lighter and single-storey. Machines could follow the order of the work rather than the line of the shaft. Changing one department no longer meant shutting down the whole power system.
Warren Devine, in a 1983 study, quotes an engineer from a motor manufacturer: many factories took up electricity to cut their coal bills, and found savings other than the ones promised. The fuel for generating electricity was a small cost, usually between about 0.5% and 3% of the total cost of a unit of output. The gain lay in how the work flowed. The US Government Printing Office made room for forty more presses without adding any floor area.
In American manufacturing, output per hour worked grew by an average of 1.3% a year before 1919. After that, up to the early 1950s, by 3.1%. Motors alone can’t take the credit. Electricity became cheaper after 1914-17, the 1920s brought a wave of investment, and architects and engineers learned the trade one site at a time. David looks at a different measure, total factor productivity, which counts labour and capital together. He estimates that about half of its acceleration in the 1920s, compared with the decade before, is accounted for, statistically, by the growth in electric motors.
And artificial intelligence?
In 2025, according to ISTAT, Italy’s national statistics office, 16.4% of Italian businesses with at least 10 employees say they use at least one AI technology. Among small and medium-sized firms it’s 15.7%; among large ones, 53.1%. Of those using it, the most common areas are marketing and sales (33.1%), administrative processes (25.7%) and research, development or innovation (20.0%). Production processes stop at 17.5%, logistics at 6.1%.
Those figures can’t tell you whether anyone has changed the way they work. Our reading, and it’s only a reading, is that AI today is bolted onto yesterday’s processes, as the first electric motor was bolted to the steam shaft. An email written faster doesn’t change the route an order takes.
In Denmark, Anders Humlum and Emilie Vestergaard studied chatbots in exposed occupations. Workers report productivity benefits. Yet two years after ChatGPT launched, official records show no effect on earnings or hours, and the authors rule out anything above 2%. Employers, meanwhile, are reorganising tasks.
Erik Brynjolfsson, Daniel Rock and Chad Syverson describe the mechanism. Technologies like this need complementary investment, often intangible: new processes, new skills. National accounts measure it badly. Their model shows how productivity growth can then look understated early on, and overstated later.
What this story doesn’t tell you
Go carefully with the comparison. David himself warns that computers are not dynamos, and that the analogy has many limitations if taken too literally. The story doesn’t say AI will go the same way, or when. It warns against two mistakes: too much optimism and too little patience.
There is one thing David adds, though, and it concerns you. A building ages, and sooner or later you get the chance to rebuild it. The way a business gathers and moves its information doesn’t wear out. David says so plainly: you can’t rely on the passing of time to create the occasion to redesign it.
So, as we see it, the question isn’t where to add AI. It’s what your central shaft is: what, in your work, sits where it does only because there was once no other way.
That’s the question we start from: agents connected to email, management software and machines, sensors we bring where the data is missing, a measurement before and one after, and a person who decides.
Sources
- Paul A. David, The Dynamo and the Computer: An Historical Perspective on the Modern Productivity Paradox, American Economic Review (AEA Papers and Proceedings), vol. 80, no. 2, pp. 355-361, May 1990 (copy read). www.gwern.net
- Warren D. Devine Jr., From Shafts to Wires: Historical Perspective on Electrification, The Journal of Economic History, vol. 43, no. 2, pp. 347-372, June 1983 (copy read). gwern.net
- Warren D. Devine Jr., From Shafts to Wires: Historical Perspective on Electrification, The Journal of Economic History, vol. 43, no. 2, pp. 347-372, June 1983 (publisher’s page). doi.org
- ISTAT, Imprese e ICT, anno 2025, press release, 15 December 2025. www.istat.it
- ISTAT, Imprese e ICT, anno 2025, full text and methodological note, 15 December 2025. www.istat.it
- Anders Humlum, Emilie Vestergaard, Still Waters, Rapid Currents: Early Labor Market Transformation under Generative AI, National Bureau of Economic Research, Working Paper 33777, May 2025, revised March 2026. www.nber.org
- Erik Brynjolfsson, Daniel Rock, Chad Syverson, The Productivity J-Curve: How Intangibles Complement General Purpose Technologies, American Economic Journal: Macroeconomics, vol. 13, no. 1, pp. 333-372, January 2021. www.aeaweb.org
- Erik Brynjolfsson, Daniel Rock, Chad Syverson, The Productivity J-Curve (working paper version, abstract), National Bureau of Economic Research, Working Paper 25148, October 2018, revised January 2020. www.nber.org