From Luddites to LLMs: How Fast Does Technology Replace Workers?
A data-driven comparison of six historical technology transitions — and what they predict about the pace of AI displacement.
Every time a transformative technology appears, the same question surfaces: how fast will it replace human work? The answers tend to cluster around two poles — utopian abundance or mass unemployment — with little historical grounding in between.
This essay takes a different approach. Instead of predicting the future from first principles, it examines six well-documented technology transitions, measures how long each actually took, and identifies the structural patterns that governed their speed. The goal is not prophecy but calibration: given what we know about past transitions, what should we actually expect from AI?
From the moment a technology is capable of replacing human labor to the moment that replacement is largely complete — how long does the transition take? And is AI likely to be faster or slower than historical precedents?
I. The Historical Record
1. Textile Machines & the Luddite Rebellion (1764–1816)
The Spinning Jenny appeared in 1764. The power loom followed in 1785. Yet the Luddite uprising didn't erupt until March 11, 1811 — a full 47 years after the first machine.
Why the delay? Because for decades, expanding export markets meant machines added to human output rather than replacing it. The tipping point came when three shocks hit simultaneously: the Napoleonic Wars collapsed export markets, successive crop failures drove food prices up, and factory owners pivoted from using machines alongside workers to using machines instead of workers.
The Luddite movement itself was intense but brief — 5 years (1811–1816), with the peak lasting only 2 years. Parliament made machine-breaking a capital offense. Seventeen men were publicly hanged in York. The movement was crushed by military force.
| Year | Event | Gap to Luddite Uprising |
|---|---|---|
| 1764 | Spinning Jenny invented | 47 years |
| 1785 | Power Loom invented | 26 years |
| 1790s | Mechanized spinning scales massively | ~15–20 years |
| 1811 | Luddite movement begins | — |
| 1816 | Movement ends | — |
2. Automobiles Replacing Horses (1886–1950s)
The Benz Patent-Motorwagen appeared in 1886. Cars outnumbered horses in New York City by 1912. Yet the US horse population kept growing until 1920, peaking at 25 million — because rural markets were still expanding.
The transition in cities took roughly 25–30 years (1900–1930s). Including rural areas, the full arc stretched to 50+ years. The auto economy created approximately 12× more jobs than it destroyed (7.5 million created vs. 623,000 lost).
There was organized resistance — the Farmers' Anti-Automobile Society proposed that drivers encountering horses must "take the machine apart and conceal the parts in the bushes" — but it was ultimately overwhelmed by economic reality.
Governments initially resisted (Britain's Red Flag Acts required a person walking ahead of every car), then pivoted to accelerating adoption through highway programs and zoning laws that banned stables.
3. Automated Switching Replacing Telephone Operators (1892–1978)
This case is perhaps the most instructive for the AI era.
The automatic telephone exchange was invented in 1892. AT&T didn't begin serious deployment until 1919 — and only after a strike by 12,000 operators convinced management to reduce dependence on human labor. The full phaseout wasn't complete until 1978.
Total transition: 86 years. Intensive displacement phase: ~40 years (1920–1960).
Operator employment grew for 30 years after automation began (1920–1950), peaking at ~350,000 telephone company operators, because the telephone network was expanding faster than automation could eliminate positions. The absolute decline only began when network growth slowed.
When a city finally "cut over" to mechanical switching, 50–80% of young women employed as operators immediately lost those jobs (Feigenbaum & Gross, QJE 2024). Displaced operators had permanent earnings losses. Many left the workforce entirely.
By 2024, only 1,460 telephone operators remained in the entire United States — a 99.6% decline from peak.
4. Containerization Replacing Dockworkers (1956–1975)
This is the fastest and most complete displacement in the historical record. Malcolm McLean's first container ship sailed in 1956. Within 20 years, major ports had lost 90% of their dockworkers. The Port of New York went from 35,000 workers to 3,500.
Productivity leapt from 1.5 tons/hour/worker to 37.5 — a 25× improvement. Ship loading time dropped from 60 hours to 4.
Unlike other transitions, union resistance was intense and resulted in remarkable buyout agreements: the ILA negotiated a Guaranteed Annual Income — displaced longshoremen received a full salary even if no work was available. A Container Royalty Fund, paid per-ton by shipping lines, still exists today.
5. Computers Replacing Typists & Secretaries (1964–2010s)
The "boiling frog" displacement. IBM's first word processor appeared in 1964. At peak, secretary was the most common job in 31 US states. Office/admin support peaked at 12.7% of all US workers in 1980.
By 2022, it had fallen to 6.8% — a ~47% decline over 40 years. But no single crisis moment occurred. No strikes. No legislation. The work didn't vanish; it was redistributed. Every manager became their own secretary.
6. Agricultural Mechanization (1900–1970)
The largest displacement by absolute numbers. US farm labor went from 11.5 million workers (41% of the workforce) in 1907 to 3.5 million (4%) by 1970. The transition was absorbed because it coincided with manufacturing growth, WWII mobilization, and the GI Bill.
Government policy both caused and failed to mitigate the displacement: the Agricultural Adjustment Act's subsidies to landowners inadvertently funded tractor purchases that displaced sharecroppers.
II. The Master Comparison
| Transition | Honeymoon | Displacement | Total Span | Peak Job Loss |
|---|---|---|---|---|
| Textile machines | ~30 yr | ~5 yr | ~50 yr | Thousands |
| Auto → horse | 15–20 yr | ~20 yr | 50 yr | 623K |
| Auto-switch → operators | 30 yr | ~40 yr | 86 yr | ~1.35M |
| Containers → dockers | 5–10 yr | ~15 yr | 20 yr | ~90K |
| Computers → typists | ~25 yr | ~20 yr | 40 yr | ~3.9M |
| Farm mechanization | ~40 yr | ~30 yr | 70 yr | ~8M |
| AI → ? | ? yr | ? | 3 yr in | TBD |
Technology maturity → mass displacement begins → displacement complete
Farm mechanization ████████████████████████████████████████░░░░░░░░░ 70 yr
Telephone operators ██████████████████████████████░░░░░░░░░░░░░░░░░░ 86 yr (dense: 40)
Automobile / horse █████████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░ 50 yr
Computers / typists ████████████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 40 yr
Containers / dockers ████████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 20 yr
Textile / Luddites ████░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 5 yr (movement)
AI → ? ██░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░ 3 yr and counting
▲ We are here
III. The Universal Honeymoon Pattern
The single most important finding from the historical record is this: every technology transition has a honeymoon period during which the new technology coexists with — and even increases — the employment it will eventually destroy.
- Horses kept growing for 20 years after cars appeared (peaked 1920)
- Telephone operators grew for 30 years after automation began (peaked 1950)
- Bank tellers grew for 30 years after ATMs (peaked 2000) — because ATMs made branches cheaper, so banks opened more branches
- Farm workers held steady for 40 years before the post-WWII cliff
What ends the honeymoon?
Not the technology maturing. The honeymoon ends when the market stops expanding or employers shift from augmentation to replacement.
- Textile workers: Napoleon's Continental System collapsed export markets
- Telephone operators: network growth slowed as saturation approached
- Bank tellers: smartphones eliminated the need for branches entirely
- Farm workers: manufacturing absorbed surplus labor; mechanization reached a threshold where one tractor replaced dozens of hands
Electric power existed for decades before it transformed industry. The disruption came not when the technology appeared, but when Henry Ford redesigned the entire factory around it (Highland Park, 1913–14). Productivity surged — but annual worker turnover hit 370%. The parallel to AI: the disruption won't come from plugging ChatGPT into existing workflows, but from companies that redesign their operations around AI from scratch.
IV. Where Does AI Fit? Three Camps
Camp 1: "This Time Is Different — Faster"
| Source | Core Argument |
|---|---|
| Dario Amodei (Anthropic CEO) | 50% of entry-level white-collar jobs eliminated within 5 years |
| Kai-Fu Lee | 40% of world's workers displaced in 15 years — "no revolution has arrived this fast" |
| EY-Parthenon | Steam engine = decades; computers = 10 years; AI productivity boost = 3–5 years |
| Cambridge Associates | ChatGPT 3.5-level inference cost dropped 280× in 2 years — faster than mobile or cloud |
| Eduardo Levy Yeyati (CEPR) | Speed alone can cause permanent labor-force exit when retraining is slower than displacement |
The strongest version of the "faster" argument: AI is the first technology wave that targets cognitive work. Every prior transition primarily displaced manual labor. Knowledge workers — traditionally the most secure employment category — are now squarely in the line of fire.
Camp 2: "Same Pattern, Bigger Scale"
| Source | Core Argument |
|---|---|
| WEF 2025 | By 2030: 92M jobs displaced, 170M created. Net +78 million |
| Goldman Sachs | 300M jobs affected globally; displacement effects fade within 2 years |
| McKinsey | 60% of today's US jobs didn't exist in 1940 — technology creates more than it destroys |
| ITIF | In 2024: ~120K direct AI jobs created vs. ~100K AI-related losses |
Camp 3: "Slower Than the Hype"
| Source | Core Argument |
|---|---|
| Daron Acemoglu (MIT, 2024 Nobel) | Only 4.6% of tasks automatable in 10 years; GDP impact: 1.1–1.6% |
| Yale Budget Lab | 33 months after ChatGPT — labor market data shows "stability, not major disruption" |
| Dallas Fed | "Very little evidence of AI taking away jobs on a large scale to date" |
| Penn Wharton | AI's 2025 productivity impact: 0.01 percentage points. Peak effect not until 2032 |
V. Synthesis: What the Pattern Predicts
If we take the historical record seriously, several predictions follow:
Prediction 1: We are in the honeymoon period
90% AI is currently augmenting workers, not replacing them. Only 23% of US workers use AI tools weekly (St. Louis Fed, late 2024). The labor market shows no statistically significant displacement effects 33 months after ChatGPT (Yale Budget Lab). This is exactly the pattern we see in every historical transition.
Prediction 2: The honeymoon will be shorter than historical precedents
75% Historical honeymoons lasted 15–40 years. AI's will likely be 5–10 years (2022–2030), because: (a) AI diffusion has no geographic friction (software vs. physical machines), (b) inference costs are falling 280× in 2 years (no historical parallel), and (c) the technology improves itself — AI accelerates AI development, a feedback loop that did not exist for looms or tractors.
Prediction 3: The trigger will be organizational redesign, not AI capability
80% Just as electrification's productivity surge came when Ford redesigned the factory (not when electricity was invented), AI displacement will accelerate when companies redesign workflows around AI rather than bolting AI onto existing processes. The companies doing this today are still the exception.
Prediction 4: Net job creation will likely exceed destruction
70% This has been true for every technology transition in the historical record, without exception. 60% of today's US jobs didn't exist in 1940. However — and this is critical — "net positive" does not mean "painless." The new jobs require different skills, appear in different locations, and take years to materialize. The individuals displaced rarely benefit directly.
Prediction 5: Speed can cause permanent damage even if the destination is good
75% Levy Yeyati's model is sobering: when the inflow of displaced workers exceeds retraining capacity, wait times grow, expected wages fall, and rational workers permanently exit the labor force. The telephone operator case confirms this — displaced operators suffered permanent earnings losses even though the broader economy adjusted. If AI's honeymoon is 5–10 years instead of 30–40, the adjustment pain concentrates into a much shorter window.
VI. The One Thing That Would Change This Assessment
If labor market data from 2026–2027 shows statistically significant employment declines in AI-exposed occupations (customer service, data entry, junior programming, content moderation), the honeymoon may already be ending. That would move our timeline estimate from "displacement accelerates ~2028–2030" to "already underway," and raise the urgency of retraining investment from "important" to "emergency."
Until then, the historical record suggests we have a window — probably measured in years, not decades, but a window nonetheless. The question is whether we use it.
References
Feigenbaum, J. & Gross, D. (2024). "Answering the Call of Automation." Quarterly Journal of Economics. NBER Working Paper 28061.
Bessen, J. (2015). "Toil and Technology." IMF Finance & Development.
Acemoglu, D. (2024). "The Simple Macroeconomics of AI." NBER Working Paper 32487.
Levy Yeyati, E. (2026). "Too Fast to Adjust." CEPR VoxEU.
World Economic Forum (2025). The Future of Jobs Report 2025.
Goldman Sachs (2024). "How Will AI Affect the Global Workforce."
McKinsey Global Institute (2023). "Driving Impact at Scale from Automation and AI."
EY-Parthenon (2024). "Tech Disruptions Can Inform the Economic Impact of AI."
Cambridge Associates (2025). "Navigating the AI Revolution."
Yale Budget Lab (2025–2026). "Evaluating the Impact of AI on the Labor Market."
Dallas Federal Reserve (2025). "Will AI Replace Your Job? Perhaps Not in the Next Decade."
Penn Wharton Budget Model (2025). "Projected Impact of Generative AI on Future Productivity Growth."
Merchant, B. (2023). "What the Luddites Can Teach Us About Artificial Intelligence." TIME.
The Breakthrough Institute. "Revolutionary Engines."
Richmond Federal Reserve (2019). "Goodbye, Operator." Econ Focus.