This series is adapted, with permission, from a Chinese-language series by our friend Marcus Ji, who looks at historical events through an economic lens. We have condensed and adapted the original for an international audience.
The debate over whether AI will destroy jobs usually reaches for the same reassuring precedent: technology has always killed some occupations and created others, and it has always worked out. The horse-drawn carriage driver became a taxi driver; the telegraph operator found something else.
But coachmen and telegraph operators were never the real story. The occupations that actually disappeared at scale – the ones that once fed the majority of society – died so quietly that history barely recorded them. Three of them were the giants of their eras: hand spinning, domestic service and farming were each, at their peak, the single largest occupation of their age. How the giants ended is far more instructive than the folklore, and considerably less reassuring.
Giant #1: Hand spinners – the largest technological unemployment on record
Around 1770, hand spinning employed roughly 8% of England’s entire population – not the workforce, the population, elderly and infants included. Among women and children, the share reached about 20% (per research published in Oxford’s Past & Present and by the British Academy). For millions of households, spinning was the main source of cash income.
Then came the spinning jenny (1764), the water frame (1769) and steam-powered spinning (1785). Within about half a century, the occupation was erased – economic historians class it as the earliest mass technological unemployment, and measured as a share of total population, plausibly the largest single technology shock ever.
Two facts from this episode deserve more attention than they get.
First, the new jobs did not absorb the old workers. Factory spinning positions were far fewer than the jobs destroyed, and went mostly to men. The displaced women and children did not “transition” — they simply vanished from the economic record. The endlessly cited rule that “machines destroy old jobs but create equivalent new ones” fails at the exact moment it is tested against the largest case on record.
Second, almost nobody documented it at the time. The workers were dispersed across ordinary households, too unremarkable to notice. Historians only reconstructed the scale of the event in recent decades.
Giant #2: Domestic servants – the largest job category of the 19th century
In 1891, roughly 41% of Britain’s female workforce worked as servants in other people’s homes; in 1870s America, the figure was about half of all employed women (per census-based estimates). If a woman worked outside her own home, the single most likely place she worked was somebody else’s.
No single machine killed this occupation. A swarm of cheap appliances did – washing machines, refrigerators, vacuum cleaners – combined with factories and offices offering women alternative employment, and middle-class households no longer able to afford staff. By 1950, the share of British women in domestic service had fallen from 41% to 8.4%. In developed economies today, the live-in servant has essentially ceased to exist as a formal occupation.
The reverse test confirms the mechanism: where appliances haven’t penetrated and women’s employment options remain narrow, the occupation never died. The ILO estimates roughly 75 million domestic workers globally today, overwhelmingly in developing economies – a figure readers in Southeast Asia and the Gulf will not find abstract.
Giant #3: Farmers – history’s only successful mass reallocation
For nearly ten millennia after the agricultural revolution, farmers made up 80–90% of the global workforce. Then: England’s male agricultural share fell from ~78% in 1381 to 11% by 1911; the US went from ~90% in 1790 to under 2% today. Fertilizer, machinery and breeding did it – one modern farmer produces roughly what 200 farmers did in the early 1800s.
The historical weight of this case comes down to one sentence: this is the only time in human history that roughly 90% of the workforce was successfully moved into entirely new categories of work – and it took about 200 years.
The mainstream expectation for AI-driven labor reallocation is that a comparable migration happens in about 20 years. (A forecast, not a fact – but it is the forecast the current debate rests on.) A 10x compression of the only precedent we have is the backdrop for everything that follows.
One more observation falls out of the three giants together: the more people an occupation fed, and the more ordinary its practitioners, the more quietly it died. The episodes history remembers loudly – strikes, machine-breaking – were the sideshows.
Three trend lines, two thousand years
Stretch the occupational data across two millennia and three lines emerge.
Variety is exploding – and only recently. Roman inscriptions record about 268 identifiable occupations in 0 AD. The US census counted 323 in 1850. Today the official US classification lists 867. The striking part is not the growth but its timing: almost nothing happened for 1,800 years, then the count nearly tripled in 170. The “doctor” of 1900 has splintered into hundreds of specialty codes; the “engineer” into dozens of disciplines.
The largest single occupation keeps shrinking. Farming was ~95% of all work in 8000 BC, ~85% in 0 AD, ~43% in 1850s America. Today the largest US category – retail and food service – is 5-6%. Ancient societies had almost everyone doing the same thing; today no single occupation reaches even a tenth of the workforce.
There is, however, a counter-signal worth stating plainly: while occupations have dispersed, employers have concentrated. US employer concentration rose roughly 46% between 2002 and 2023 (per BLS-based estimates), and Amazon alone employs close to 1% of the American workforce. Job titles have never been more varied; bosses have rarely been fewer.
Manual work flipped to knowledge work in one lifetime. In 1900, about three-quarters of workers earned a living with their bodies; today about three-quarters earn it with knowledge and services. The farm-to-factory transition took two centuries; the muscle-to-brain transition took eighty years. The transitions themselves are speeding up.
A caution before drawing the optimistic conclusion (“freed from manual work, people will create”): freed labor does not automatically become innovative labor. Song dynasty China – the most commercialized, urbanized economy of its era, as the previous installment of this series described – did not convert its enormous non-farm population into sustained technological progress. Depression-era America converted displaced factory workers into poverty, not invention. India’s IT boom moved millions from fields into offices, but largely into standardized data entry and call centers – knowledge work, not creative work, and now itself squarely in AI’s path. Freed cognitive capacity converts into innovation only where education, institutional incentives and judgment-demanding work exist to receive it.
The asymmetry that actually hurts
Across every technology revolution, one asymmetry does the damage: the destruction of old jobs is immediate; the creation of new ones lags.
The lag has been shrinking – roughly 70 years from the steam engine to mass new employment; about 40 for electricity; about 15 for the PC and the internet; and for generative AI, extrapolations suggest perhaps 6-11 years. But shrinking is not zero, and the replacement happens now. One door closes before the other opens; that gap is where the concentrated unemployment of every previous revolution lived.
This round adds a structural novelty. Every prior wave respected a boundary: routine physical work was exposed, cognitive work was safe. AI cuts that line vertically – anything definable as rules or patterns is in range, physical or mental. And what it covered first is precisely what humans spend the most years learning: reading, searching, summarising, drafting, analysing.
The timing lesson from the handloom weavers
Among England’s handloom weavers – the spinners’ successors in obsolescence – the ones who successfully crossed over were those who learned machine operation in the 1810s, while their incomes were still intact and the crisis had not yet arrived. Those who waited until wages collapsed in the 1830s entered a labor market already flooded with people exactly like them.
The window is before the crisis, not during it. That was true in 1820, and it is the single most transferable finding in this entire history. If the 200-year precedent really is about to be compressed into 20, the one variable an individual – or a company, or a government – actually controls is which side of the window they start moving on.
Previously in this series: Why was food delivery already popular in imperial China almost 1,000 years ago?
Next in this series: Why your job might be replaced in the next ten years?
