Walk into an American factory in 1900 and you walk into a machine. One steam engine sits at the heart of the building. It turns a steel driveshaft that runs the length of the plant, sometimes through the ceiling to the floors above. Belts and pulleys hang off the shaft and drop power to every lathe, press, and loom.
Everything follows from the shaft. Machines cluster close to it, because leather belts bleed power with distance. Buildings grow tall and narrow, because a vertical plant keeps more machines near the shaft. Workers stand where the belts are. When the engine runs, everything runs. When it stops, nothing does.
By 1900, Edison’s power stations had been selling electricity for eighteen years. The electric motor was proven and available. Yet less than five percent of mechanical drive power in American factories came from electric motors. An industrial economy looked at a transformative technology for two decades and passed.
One big motor
The owners who did adopt electricity did the obvious thing. They unbolted the steam engine and bolted a large electric motor in its place. Same shaft. Same belts. Same building. Same jobs.
The result: not much. The motor was cleaner and a little cheaper to run. But the shaft still spun all day whether one machine drew power or fifty. The layout was still hostage to power transmission. The owners had bought a revolutionary technology and installed it as an incremental one.
Paul David documented what happened next in “The Dynamo and the Computer” (1990). The payoff from electrification arrived in the 1920s, forty years after the technology worked. It arrived when factory owners stopped asking where the motor goes and started asking what a factory looks like when power travels through a wire.
The answer looked nothing like the old building. A small motor on every machine. Single-story plants on cheap land, because nothing needed to hug a shaft. Machines arranged in the sequence of the work. Windows, because the ceiling no longer carried belts. Each station started and stopped on its own, which changed what a worker was: less muscle tending a belt, more judgment running a process. Hiring changed. Pay changed. Supervision changed.
When the redesign spread, manufacturing productivity grew through the 1920s at multiples of its historical rate. The dynamo did not do that. The reorganization did.
The constraint that became furniture
The driveshaft began as a solution to a real constraint. Steam power is expensive to move, so it must be generated centrally and distributed mechanically. Every downstream choice, the tall buildings, the clustered machines, the all-or-nothing operation, was a rational response.
Then the constraint disappeared and the responses stayed. The shaft stopped being a solution and became furniture. Factory design, job design, and management practice kept encoding a limitation that no longer existed. Not because owners were stupid. Because the structure no longer looked like a response to anything. It looked like how factories are.
This is the pattern with general-purpose technologies. The technology is the cheap part. The expensive part is finding which of your structures are fossilized constraints, then paying to rebuild around their absence. Adoption is a budget line. Reorganization is an identity crisis: job definitions, status, floor plans, the org chart itself.
David wrote his paper to answer a puzzle of his own era. Robert Solow quipped in 1987 that you could see the computer age everywhere but in the productivity statistics. David’s answer: wait. The dynamo took forty years because the reorganization had to come first, and computers would pay off the same way. US productivity surged in the mid-1990s, once firms rebuilt processes instead of just wiring desks.
What forced it
The 1920s redesign was not triggered by vision. It was triggered by a labor shock. The First World War, then immigration restrictions, cut off the flow of cheap workers to American factories. Wages rose fast. When labor is expensive, you redesign work to get more from each worker. The design that did that was the one the electric motor had enabled for decades.
The capability sat on the shelf for forty years. The reorganization happened when the labor market made the old structure unaffordable. Scarcity, not technology, was the forcing function.
The driveshaft in the org chart
Now look at how organizations are adopting AI in 2026.
The dominant pattern is one big motor. Each department gets its tool. Legal gets a contract copilot. Support gets a chatbot. Marketing gets a drafting assistant. Engineering gets code completion. Each function optimizes its own area of focus and reports single-digit gains that are real and unimpressive. The workflow between functions stays untouched. The handoffs, the review layers, the human-sized batches of work: preserved. Same shaft, new motor.
The structure being preserved was built around a constraint that is dissolving. The modern org chart encodes one assumption everywhere: cognition is scarce. It arrives in human-sized units, works eight hours, communicates through meetings, and needs layers of supervision because attention is expensive. Departments, spans of control, approval chains, quarterly planning. Driveshaft, all of it.
AI changes the constraint. Cognition becomes cheap, parallel, and continuous. A structure designed to ration scarce human attention is the wrong structure for allocating abundant machine attention, the same way a building designed to hug a shaft is the wrong building for wires.
The question is not which area of focus gets the tool. It is the 1920s question: what does the work look like organized around the capability? What is the shape of a claims process, a compliance review, a client onboarding flow, when work no longer has to be chopped into human-sized pieces? Some processes collapse into continuous flows. Some roles invert, from doing the work to specifying and verifying it. Some management layers exist only to compensate for expensive coordination, and cheap coordination deletes them. That is the difference between buying AI tools and fielding an AI workforce. One slots into the org chart. The other forces you to redraw it.
The forcing function is arriving on schedule. Services labor is getting scarcer and more expensive in the industries most exposed: healthcare, financial services, professional services. Baumol guarantees the wage pressure. Demographics guarantee the scarcity. As in 1919, the redesign will not be driven by technology demos. It will be driven by operators who can no longer afford the old structure.
Who wins
The gains did not go to the plants that installed motors first. Early adoption on the old layout conferred almost nothing. The gains went to plants built around the motor, and new buildings had the advantage: no shaft to defend. When the required change is structural, incumbency inverts. Your installed base of process, hierarchy, and habit becomes the liability.
The dynamo’s lag was forty years. The computer’s was about ten. The lag compresses, but it never disappears, because it was never technical. It is the time it takes an organization to see its own driveshaft.
The question is not where to add AI. The question is what your organization was built around, and whether that constraint is still real. Organize the work around the capability, or keep polishing the shaft.
References
Baumol, William J. “Macroeconomics of Unbalanced Growth: The Anatomy of Urban Crisis.” American Economic Review 57, no. 3 (1967): 415-26.
David, Paul A. “The Dynamo and the Computer: A Historical Perspective on the Modern Productivity Paradox.” American Economic Review 80, no. 2 (1990): 355-61.
Devine, Warren D. Jr. “From Shafts to Wires: Historical Perspective on Electrification.” Journal of Economic History 43, no. 2 (1983): 347-72.
Harford, Tim. “Why didn’t electricity immediately change manufacturing?” BBC News, 2017.
Solow, Robert M. “We’d Better Watch Out.” New York Times Book Review, July 12, 1987.