Economic history is a history of substitution: superior technology replaces the prime mover before it, and the jobs built around the old one vanish with it. No case shows the speed and totality of that shock more vividly than the working horse — the reference point economists have reached for since 1983 when thinking about what artificial intelligence might do to human labour. Here is the story, the pattern it shares with four other collapses, and what it does and doesn't tell us about the scale of AI's disruption.
A common misreading assumes the steam engine and railroad displaced the horse. They did the opposite. Rail moved bulk goods cheaply between railheads but couldn't solve the last mile; horses were the connective tissue that moved everything from the depot to the door. Demand for horsepower climbed for the entire nineteenth century.
Early steam engines were also less efficient than a horse. A horse converted feed to traction at 15–20% efficiency — more than triple an early coal-fired machine — and, as one 1894 engineer noted, it was self-feeding, self-maintaining, and self-reproducing. Driven by rail, farming, and urbanisation, the U.S. horse and mule population grew nearly six-fold between the 1840s and World War I.
The horse sustained hundreds of thousands of livelihoods — farriers, blacksmiths, harness and whip makers, carriage builders, hay farmers, livery stables, street sweepers. In 1900 the single mid-sized city of Pittsburgh supported 23 carriage makers, 56 harness makers, 81 farriers, 40 blacksmiths, and 46 livery stables. Cities ran on horses: by 1890 New Yorkers averaged 297 horsecar rides each per year. When the Great Epizootic — an equine flu — swept the Northeast in 1872, streetcars stopped, freight piled up at the wharves, and households went without groceries. It was a total economic paralysis, a grid failure in animal form.
| Year | U.S. horses & mules | People per horse |
|---|---|---|
| 1840 | 4,300,000 | ~4.0 |
| 1900 | 21,500,000 | ~3.5 |
| 1910 | 24,000,000 | ~3.8 |
| 1920 | 25,200,000 — peak | ~4.2 |
| 1930 | 18,900,000 | ~6.5 |
| 1950 | 7,600,000 | ~20 |
| 1960 | 3,090,000 | ~58 |
From roughly one horse per four people to one per fifty-eight — in forty years.
The horse didn't fall to invention alone. The urban horse economy had already hit an environmental limit — and once a cheaper substitute crossed the line of economic viability, the decline was almost total within a single working lifetime.
A city horse produced 15–35 pounds of manure a day. In New York and Brooklyn, with 150,000-plus horses, that meant millions of pounds on the streets daily — piling in vacant lots, breeding the flies that spread typhoid, and killing people in traffic at nearly seven times the rate of modern car accidents. A streetcar horse lasted two to three years; New York cleared some 15,000 carcasses a year. The apocryphal "Great Horse Manure Crisis of 1894" — London supposedly buried under nine feet of dung — was a later fabrication, but it endured because it captured a real and intractable despair.
The internal combustion engine and electric streetcar were sold as sanitation as much as transport. But displacement wasn't instant: through the 1910s, horse-drawn wagons stayed cheapest for short-haul urban delivery. The inflection came in the 1920s — Ford's mass production collapsed the cost of the car, and the Fordson tractor mechanised the farm. Ford cut the Fordson from $625 to $395 in 1922; plowing an acre fell from 90 minutes with a five-horse team to 30 minutes with a tractor. Once the math tipped, the curve went vertical.
This was textbook creative destruction. Ancillary trades — carriage and harness makers, farriers, whip makers — evaporated; a few reskilled into auto bodies and mechanics, but the decentralised horse economy was replaced by centralised oil and automotive giants. Freeing 93 million acres of feed-crop land flooded grain markets and helped depress prices into the Great Depression. Farms consolidated as one operator could work vast tracts alone, pushing labour into the cities. And the swap traded visible, local pollution — manure, flies, carcasses — for the invisible, systemic kind: fossil carbon and the suburban sprawl the car made possible.
The horse is not an anomaly. Look at four other transitions and the shape repeats: a new technology incubates slowly and even looks complementary, crosses a threshold of economic viability, then adoption explodes — and the workforce tied to the old technology becomes obsolete fast.
The "frozen water trade" employed 90,000 workers and 25,000 horses harvesting ice from northern rivers. Artificial ice overtook it by 1914 (26M vs 24M tons) and the industry collapsed between the wars, leaving only abandoned icehouses.
Steam crossed the Atlantic in 1838 but sail held bulk and long routes for decades until compound and triple-expansion engines won. Speed — not fuel — closed it, and the sailor's specialised craft was displaced with it.
Edison's 1879 bulb ended the lamplighter's trade entirely. Its odd legacy: the term "gaslighting," from a 1938 play about a husband dimming the gas lamps and denying it.
In 1920 operating was the largest occupation for young American women; AT&T was the nation's biggest employer. Mechanical switching cut operator jobs 50–80% per city almost overnight — yet new clerical work later absorbed the next cohort.
Displacement is legible, immediate, and concentrated. Job creation is diffuse, gradual, and often unimaginable in advance. That asymmetry is why every transition feels like the end of work — and why it usually isn't.
The switchboard lessonIn 1983 the Nobel laureate Wassily Leontief argued that computing would make human labour redundant just as the tractor made the horse redundant. But the analogy has a flaw: unlike a horse, a human can invent new tasks and move up the value chain. Economists frame the difference as seven mechanisms. Every one resolved against the horse. For humans facing AI, most are still open.
| Mechanism | Horses | Humans vs AI |
|---|---|---|
| Frontier hazard Is the machine simply cheaper at the tasks that matter? | Yes — tractor won | Contested |
| Reinstatement Do new tasks appear that need the displaced actor? | No | Yes |
| Team complementarity Do actor and machine produce more together? | No | Yes & growing |
| Skill replenishment Can the worker retrain into a new advantage? | No — biologically fixed | Yes, with friction |
| Reliability stakes Do mistakes force a human in the loop? | Low | High |
| Expenditure migration Where does the money saved by automation flow? | Away from horses | Open |
| Scale vs. substitution Do people buy far more when the price drops? | No — inelastic | Varies |
The horse lost because it was rigid: it couldn't invent a new task, team with the tractor, or retrain. Human labour is adaptable — as machines absorb routine work, human advantage tends to migrate toward high-stakes reliability, social intelligence, and complex problem-solving. That's the analytical gap in the horse metaphor: labour is more likely to move than vanish.
The history is reassuring about the destination and alarming about the journey. The same S-curve governs AI adoption — but compressed harder than any transition before it, which concentrates the pain on whoever holds an automatable role right now.
Digital network effects let AI bypass the physical constraints that slowed steam and the tractor, producing the most compressed adoption-and-investment cycle in technological history. The macroeconomy will very likely generate new roles — but the individuals in today's most automatable cognitive jobs (routine coding, data processing, paralegal work) face the same immediate, concentrated shock the 1920s switchboard operator did. Technology transitions are unsympathetic to legacy skill sets.
The lesson of the horse economy is not that human labour is doomed. It is that surviving a phase-shift depends entirely on the capacity to adapt to the infrastructure that follows — the one thing the horse could never do, and the very thing that makes humans different.
The takeawayFor leaders and policymakers the implication is concrete: the severity of the AI transition will be set by the velocity of reskilling — how fast a workforce can move into roles that work with AI rather than against it, using the team-complementarity the horse never had.
Understanding the pattern is one thing; deciding where your organisation and your people sit on the curve — and what to reskill toward first — is the practical work. That's what we do with the teams and leaders we coach: turning a model on a screen into a deliberate next move.
Talk to us about the shiftThe figures and framing above are drawn from economic-history and labour-economics research. Start here to go deeper — or to check the numbers.