Where the jobs went, and where they go next

I went to a concert of sea shanties last week, and shanties are work songs, written to be sung by crews hauling on ropes, and the crews they were written for mostly don't exist any more, which got me reading about where they went, and then about where everyone else went each time a machine took their work.

As far as I can tell every big shift in where people work in the last two hundred years has gone the same way, a machine does the work that needed a lot of hands, the hands go somewhere else over a generation or two, and the place they went becomes the next normal. What changes each time is the speed, and what stays the same is the direction, because the work keeps moving up the list of things people need, from food to goods to services.

Starting with the sea

The American whaling fleet peaked in 1846 at 736 ships, each with a crew of about 35, and by 1876 only 39 ships set out (New Bedford sent out its last whaler in 1927) ^[1]. The merchant crews went the same way, just a bit later. In 1938 there were 131,885 British seamen in the first crews of sea-trading ships ^[2], and in 2025 the Department for Transport counted 24,550 UK seafarers active at sea, which is about where the number sat every year from 2012 to 2020, apart from a dip in 2021 and 2022 ^[3]. That's a fall of about four fifths in under a century, while the cargo went the other way.

The Cutty Sark sailed with a crew of around 26 ^[4], and on her first trip home from Shanghai she brought back 1.3 million pounds of tea, about 590 tonnes ^[5]. The Ever Given (the one that got stuck in the Suez Canal in 2021) sails with a crew of 25 ^[6] and carries just over 20,000 containers ^[7], and even if you only put ten tonnes in each of them that's something like 300 Cutty Sarks' worth of cargo in one trip, with one fewer person on board. Nobody planned it, the engines got better and the ships got bigger and the crews got smaller, one shipping company at a time.

The farm and the factory

The land had emptied long before the ships did. In 1801 about 32% of Britain's workers were on the land, 36% in industry and the rest in services, by 1851 the land was down to 24% and industry up to 46%, and by 1871 the land was at 17% ^[8]. The 1851 census had been designed to measure that move from the countryside to the towns, and it confirmed a significant shift ^[9]. Today agriculture employs about 290,000 people in the UK ^[10], out of 34.6 million in work ^[11], which is under 1%.

The United States ran the same change later and faster. In 1900, 41% of the American labour force was in agriculture, and it was 21.5% in 1930, 16% in 1945, 4% in 1970 and 1.9% in 2000 ^[12]. That's seventy years to lose nine tenths of the people, and the tractor, the combine and fertiliser did it, because one person with machines could feed the people who'd moved to the factory, and that's what made the cities possible.

Around 20% of the world lived in cities in 1950, the UN called the world majority urban in 2007, and around 80% of people now live in cities, towns and suburbs ^[13]. Nobody planned that either, the machinery freed the hands and the hands went where the work was, which for most of them meant the factory, until the factory went too.

UK manufacturing employment peaked in 1966 at over eight million jobs and was a little over 2.5 million by 2009 ^[14], and between June 1979 and June 2022 the sector lost nearly 4.1 million jobs, the largest fall of any ^[15]. In the US the peak was June 1979, 19.6 million people and 22% of non-farm jobs, and by June 2019 it was 12.8 million and 9% ^[16].

Forty years, and the children of the people on the line went to the office, because that's where the work had moved. Some of that was automation and some was the work moving to other countries, and from the point of view of the town with the factory the difference doesn't matter much.

Faster each time

The example that looks most like what's happening now is smaller and sharper, and it's the telephone exchange. Telephone operation in the US peaked in 1930 at 180,000 operators, and for young, white, American-born women it was about 4% of all employment, one of the biggest jobs a young woman could get. Between 1920 and 1940 mechanical switching replaced operators in the exchanges serving over half the country, and the number of young women working as operators fell by 50 to 80% where it arrived. Feigenbaum and Gross followed the people, and the young women who would have become operators were not worse off, they found other work, but the ones who were already operators were, ten years on they were much less likely to be operators and the ones still working had often moved to lower paid jobs ^[17]. Full diffusion through the Bell network took sixty years, but the people in the exchanges felt it in twenty.

And the office has data now. Brynjolfsson, Chandar and Chen went through ADP payroll records and found no economy-wide displacement, but for workers aged 22 to 25 in the most AI-exposed occupations they found employment about 19% below where it would be if it had kept pace with their peers in less exposed work, a gap that grew from 15% to 19% in the year to June 2026 and that started from nothing in November 2022 ^[18]. That's the operator pattern again, the entry job going first, in under four years instead of twenty.

So it took about a century for the sea, seventy years for the farm, forty for the factory, twenty for the exchange, and four so far for the office. The exact years depend on where you start counting (and you could argue about every one of them), but each move is faster than the one before, because the tool arrives everywhere at once, a tractor had to be bought, delivered and learned on each farm, and a model is a download.

Where the jobs went

If you put the moves next to each other the direction is the same every time. Maslow published his list of what people need in 1943, in the order he thought they come, the body first (food, warmth, sleep), then safety, then love and belonging, then esteem, then what he called self-actualisation, the need to become what you're capable of ^[19]. Tay and Diener's work on 123 countries in 2011 found that the needs are real and matter to people everywhere, and that the order matters a lot less than Maslow thought ^[20], so I'm using the list as a rough map of what people pay for, not as psychology.

The farm was the bottom rung, food. When machines took it, the hands went to the factory, which made the things one step up, clothes, housing, heat, tools, the stuff of safety and comfort. When machines and cheaper countries took that, the hands went to the office, which sells the next rung, insurance and banking and law and admin, the services that keep the comfortable life safe, plus the information and coordination that run everything else. Each time the machine took the rung below and people climbed one (the sailors fit as well, because the sea was how the goods travelled, and the engine took that job too).

The models are taking the office rung now, reading, deciding and writing, which is most of what the middle of the ladder is made of, so the hands move up again, and the rungs above are love and belonging, esteem and self-actualisation. In plain words that's care, teaching, feeding and entertaining people in person, the things people make for people they know, and the people who help you get good at something. Health priority occupations employed 1,487,000 people across the UK in 2024, adult social care employed 1.47 million in England alone, and demand for the priority care occupations is projected to grow 28% by 2035 before you even count replacing the people who leave ^[21]. The shanty concert was the same thing from the other end, a room full of people paying to hear other people sing work songs for work that no longer exists.

So my guess is that the next big employer is each other, care for the old, which is the one job we know grows no matter what, teaching in person, the trades that fix the house you live in, live music and live anything, and the small businesses run by someone you can name, who makes money from the product and not from your attention. The models can't do that rung, because the thing being paid for is that a person did it.

There are two catches though. The first is that the top rungs pay worse than the middle one does today, a care worker earns a lot less than the admin job the models are taking, and the bills that were built on the admin wage don't shrink as fast as the job does. The second is that the top rungs are the easiest to fake, a model will tell you it cares and a company will sell you belonging by the month, and the only defence is knowing the difference, which is what the LocalGhost post on manipulation patterns is about ^[22]. The exchange operators of 1930 found other work because the rungs above them were still empty, and ours are full of things people want that nobody has worked out how to pay enough people for yet.

Is the learning curve steeper?

Yes, and there's research behind it, because Benjamin Jones looked at US patents in a 2009 paper and found that the age at which inventors make their first invention rose by about two thirds of a year per decade, that inventors specialise more narrowly as time goes on, about 6% more per decade, and that the average team on a patent grew from 1.73 people in 1975 to 2.33 by the end of the 1990s, a 35% increase ^[23]. He called it the burden of knowledge. The frontier moves out, there's more to learn before you can add to it, so people learn it later, learn a narrower slice, and work in bigger groups, and it's the same thing you feel trying to hold a modern system in your head, you can't, so you hold one part of it.

The models go the other way. Every earlier tool raised the burden, because the tractor needed a mechanic and the switch needed an engineer, and a model is the first tool that carries the knowledge for you, so one person can use the parts they don't understand. The complexity is still there, inside the model now, and nobody knows yet what goes wrong when a whole society runs on tools nobody fully understands. But it's why one person with good tools now does what a team did.

We'll figure it out

Every one of these moves looked like the end of something to the people inside it, and every one of them ended with more people in work than before, doing things that the people who lost the old work couldn't have described. The whalers' grandchildren didn't go back to sea, they went to the factory, and the factory workers' grandchildren went to the office. We'll figure this one out too, faster than the last ones because the tools are in everyone's hands, and the people who do well will be the ones who start exploring now, while they still have the wage, and pick the one small thing they care about and go deeper into it than a company can. If you're building tools for agents, build for that person, the one who just lost the ladder and still has the skill. The directory on this site is there so they can find what they need without having to hold all of it in their head.

References

Read on 2 November 2026. Quotes are as published.

1. [1]Wikipedia, "Whaling in the United States". Source for the fleet peaking in 1846 at 736 registered vessels, a whaleship typically carrying around 35 crew, only 39 American ships setting out in 1876, and New Bedford sending out its last whaler in 1927. en.wikipedia.org
2. [2]Sir William P. Elderton, 1947, "Merchant Seamen during the War", Journal of the Institute of Actuaries 73, Table 3, "Approximate total of first crews of sea-trading vessels actually employed during year", from the Board of Trade's pre-war return. Source for 131,885 British seamen in 1938. The same row's total is 192,375, with 9,790 foreign seamen and 50,700 lascars, so the British row is the one that compares with the Department for Transport's count of UK seafarers. actuaries.org.uk
3. [3]Department for Transport, 25 February 2026, "Seafarers in the UK shipping industry 2025". Source for an estimated 24,550 UK seafarers active at sea in 2025, and the number staying broadly between 24,000 and 25,000 a year from 2012 to 2020 before falling in 2021 and staying low in 2022. gov.uk
4. [4]Royal Museums Greenwich, "Crew on Cutty Sark". Source for the crew varying by voyage but usually numbering around 26. rmg.co.uk
5. [5]Royal Museums Greenwich, "Ten things you should know" about Cutty Sark. Source for her coming back from her first trip to Shanghai "loaded with 1.3 million pounds of tea", about 590 tonnes. rmg.co.uk
6. [6]TradeWinds, 29 April 2021, "Three crew members to switch out from Ever Given amid detention". Source for the Ever Given being run by a crew of 25. tradewindsnews.com
7. [7]Wikipedia, "Evergreen G-class container ship". Source for the class carrying 20,124 to 20,160 containers (twenty-foot equivalent units), and 20,124 for the ships retrofitted with scrubbers, the Ever Given among them. Taking a container at ten tonnes of goods, our assumption and well under what a full one weighs, that's about 200,000 tonnes, a few hundred times the 590 tonnes of tea in [5]. en.wikipedia.org
8. [8]Stephen Broadberry, Bruce M. S. Campbell and Bas van Leeuwen, 2013, "When did Britain industrialise? The sectoral distribution of the labour force and labour productivity in Britain, 1381 to 1851", Explorations in Economic History 50(1), Table 7B and Table 6C. Source for agriculture at 31.7%, industry 36.4% and services 31.9% in 1801, 23.5%, 45.6% and 30.9% in 1851, and agriculture at 16.9% in 1871. pureadmin.qub.ac.uk
9. [9]Wikipedia, "1851 United Kingdom census", citing Christian and Annal, "Census, The Expert Guide", The National Archives, 2008. Source for the census being designed to measure migration from rural to urban areas and confirming "a significant shift from the countryside to the towns". en.wikipedia.org
10. [10]OECD Main Economic Indicators via FRED, "Employment, Agriculture, Forestry and Fishing, Total for United Kingdom". Source for 290,000 people in 2025. fred.stlouisfed.org
11. [11]Office for National Statistics, 20 October 2026, "Employment in the UK". Source for 34.577 million people aged 16 and over in employment in June to August 2026. ons.gov.uk
12. [12]Carolyn Dimitri, Anne Effland and Neilson Conklin, June 2005, "The 20th Century Transformation of U.S. Agriculture and Farm Policy", USDA Economic Research Service, EIB-3, box "Farming's changing role in the Nation's economy". Source for the share of the employed labour force in agriculture, 41% in 1900, 21.5% in 1930, 16% in 1945, 4% in 1970 and 1.9% in 2000, which the chart shows per 1,000 people in work. ers.usda.gov
13. [13]Our World in Data, "Urbanization", citing the UN World Urbanization Prospects 2025 and the UN national-definition series. Source for around 20% of the world's population living in cities in 1950, the UN announcing the world became majority urban in 2007, and around 80% living in cities, towns and suburbs today. ourworldindata.org
14. [14]Stephen Nickell, 25 August 2010, "Number cruncher, manufacturing a decline", Prospect. Source for UK manufacturing employment peaking in 1966 at over eight million jobs and a little over 2.5 million in 2009, which the chart shows as 8,000 and 2,500 thousand. prospectmagazine.co.uk
15. [15]Office for National Statistics, Freedom of Information response, "Employment loss by sector 1979 to 2022". Source for manufacturing showing the largest fall between June 1979 and June 2022, nearly 4.1 million jobs. ons.gov.uk
16. [16]US Bureau of Labor Statistics, November 2020, "Forty years of falling manufacturing employment", Beyond the Numbers. Source for the June 1979 peak of 19.6 million and 22% of non-farm employment, and 12.8 million and 9% in June 2019. bls.gov
17. [17]James Feigenbaum and Daniel P. Gross, November 2020, revised February 2024, "Answering the Call of Automation, How the Labor Market Adjusted to Mechanizing Telephone Operation", NBER Working Paper 28061. Source for operators peaking in 1930 at 180,000, about 4% of employment for young, white, American-born women, mechanical switching reaching exchanges serving over half the country between 1920 and 1940, the 50 to 80% fall in young women working as operators, later cohorts not being worse off, incumbents being less likely to be operators ten years on and moving to lower paid work, and full diffusion taking until 1978. nber.org
18. [18]Erik Brynjolfsson, Bharat Chandar and Ruyu Chen, 12 August 2026, "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence", Stanford Digital Economy Lab, on ADP payroll data. Source for employment of workers aged 22 to 25 in highly exposed occupations standing about 19% below its expected level as of June 2026, up from 15% at the July 2025 data vintage, levels measured from November 2022, and "We do not see widespread, economy-wide job displacement associated with AI." digitaleconomy.stanford.edu
19. [19]A. H. Maslow, 1943, "A Theory of Human Motivation", Psychological Review 50, pages 370 to 396. Source for the five needs in the order he gave them, the physiological drives, the safety needs, the love and belongingness needs, the esteem needs and the need for self-actualisation. psychclassics.yorku.ca
20. [20]Louis Tay and Ed Diener, 2011, "Needs and Subjective Well-Being Around the World", Journal of Personality and Social Psychology, on Gallup World Poll data from 123 countries, as reported by the University of Illinois on 29 June 2011. Source for the needs being universal and important to happiness everywhere, and for the order in which they are met having "little bearing" on life satisfaction. news.illinois.edu
21. [21]Skills England, 4 August 2026, updated 6 October 2026, "Sector skills needs assessment, health and adult social care". Source for health priority occupations employing 1,487,000 people across the UK in 2024, adult social care employing 1.47 million people across England in 2024/25, and demand for the priority adult social care occupations projected to grow by 281,000, 28%, from 2025 to 2035 in the central scenario, with total demand including replacement at about 685,000. gov.uk
22. [22]Vlad Cealicu, 10 May 2026, "A Ghost Should Not Possess", LocalGhost Hard Truths. The catalogue of manipulation patterns, from the slot machine to the social feed, now aimed at people through their AI. localghost.ai
23. [23]Benjamin F. Jones, 2009, "The Burden of Knowledge and the 'Death of the Renaissance Man', Is Innovation Getting Harder?", Review of Economic Studies 76(1). Source for age at first invention rising 0.66 years per decade, specialisation rising about 6% per decade, and average team size rising from 1.73 in 1975 to 2.33, a 35% increase. kellogg.northwestern.edu