When the pie grows but the slice shrinks: what black cabs and handloom weavers know that AI evangelists don't

It's Tuesday evening in Marylebone. I'm in the back of a black cab. The driver is sixty-something, second-generation cabbie, did the Knowledge in 1987. He's telling me he now drives twelve-hour shifts to clear what eight hours used to bring him a decade ago. "My boy is driving an Uber now. Pays the same." He looks at me in the mirror. "Maybe less, after the rent on the car."

I give him a tip and walk the last couple of minutes home thinking about handloom weavers.

The promise

Here's the perspective I want to share with you: the Industrial Revolution did make almost everyone richer in the end, but the people who lived through the transition mostly didn't see it, and the mechanism by which the gains eventually got shared had very little to do with technology and a lot to do with politics. Most popular comparisons to today's AI boom collapse a 150-year process into a single triumphant arc. The actual arc has phases. Some of those phases are brutal.

Unfortunately, almost everyone in the AI debate has picked a tribe. The optimists treat “the Industrial Revolution turned out fine” as a closing argument. The doomers treat the Luddites as visionaries. Both are doing history badly. The Luddites were not wrong because they feared machines. They were wrong because smashing machines could not solve the political economy of who owned them.

There's a middle position worth holding: this transition will probably make society richer, it will also crush specific occupations for decades, and whether ordinary workers share in the gains depends on choices we haven't made yet. This is harder to defend than either tribe, but closer to true.

The thesis I want to defend is this: AI is unlikely to cause mass unemployment, but it is very likely to cause a long period in which productivity rises and wages don't, and in which displaced workers cannot easily find equivalent work. We have a name for this period from the last revolution. It's called Engels' Pause. Black cab drivers are living through their own miniature version of it right now. With that, let's dive in.

The numbers nobody quotes

Pop history teaches you that the Industrial Revolution made everyone better off. The economic data say something more uncomfortable.

British per capita GDP grew at an average annual rate of 0.13 per cent between 1086 and 1700, then accelerated to 0.48 per cent between 1700 and 1870. Still not particularly fast. By modern standards this is barely growth at all. The point of the Industrial Revolution wasn't a sudden burst of wealth. It was that growth became sustained for the first time in human history. Every previous period of expansion had eventually plateaued. This one didn't.

But the gains did not start arriving for the people doing the work. They arrived for the people who owned the machines.

The economic historian Robert Allen, building on a phrase from Engels, named the phenomenon. Between 1780 and 1840, output per worker in Britain rose by 46 per cent. Real wages rose by 12 per cent. The productivity gains went almost entirely to capital. In the first half of the 19th century, the real wage stagnated while output per worker expanded. The profit rate doubled and the share of profits in national income expanded at the expense of labour and land.

Two generations of stagnation. And then, around 1840 to 1850, the pattern flipped. Between 1840 and 1900, output per worker increased by 90 per cent and real wage growth increased by 123 per cent. Productivity and wages began moving together for the first time since industrialisation began.

So if you were a Lancashire mill worker born in 1790, you spent your entire working life producing more and earning roughly the same. Your grandchildren would do better. You would not.

The handloom weavers

The single sharpest case study sits inside that period, and it's the one Daron Acemoglu and Simon Johnson keep returning to in their work on AI. The story goes like this.

The mechanisation of spinning in the late 1700s was, on net, good for weavers. More yarn meant more demand for cloth, and weavers were the bottleneck. Real wages for handloom weavers rose. This is the story the techno-optimists love: automation here, more jobs there.

Then came the power loom. A single power loom could produce more cotton cloth than 10 to 20 handweavers working from home, and the machines were so large they had to be housed in factory buildings, taking cottage industry weaving off the table. As factory weaving eclipsed home weaving, the displaced workers had no place to go because power looms created relatively few new jobs in the factories.

The numbers are grim. Handloom workers in the English cotton industry averaged 240 pence per week in 1806, but by 1820 they were making less than 100 pence weekly. A 60 per cent collapse over fourteen years. The British economy overall did not create enough other well-paying new jobs, at least not until railways took off in the 1830s. Hundreds of thousands of weavers stayed in the trade because there was nowhere else to go. They watched their families' incomes halve and then halve again.

The Luddites sit inside this story, but not in the way they are usually used. They were not idiots smashing machines because they hated progress. Many were skilled textile workers attacking specific machines that were being used to destroy wage rates, bypass apprenticeship norms, and degrade the quality of work. Their target was not technology in the abstract. It was a new bargain between capital and labour in which employers captured the productivity gain while workers absorbed the loss.

That distinction matters for AI. “Luddite” is now used as an insult, a synonym for irrational resistance to technology. Historically, that is too lazy. The Luddites were wrong that machine-breaking could preserve their world, but they were not wrong about the distributional question. They understood, earlier than most economists, that a machine can be socially productive and still ruin the people whose skill it replaces.

David Ricardo, the great classical economist, had once argued that machinery would not reduce the demand for labour. In 1819, he said as much in the House of Commons. But by 1821, in the third edition of Principles, he added a new chapter, “On Machinery,” and admitted that his views had “undergone a considerable change.” He was 49 years old, wealthy, well-connected, and at the top of his profession. He could have ignored the data. He chose to revise his theories instead. That's worth remembering.

The London cabbies

Now jump 200 years and stand in a rank in Bloomsbury. The pattern rhymes, just compressed.

Uber arrived in London in 2012. TfL data show licensed taxi vehicles falling from 22,810 in 2013/14 to 14,800 in 2023/24, while licensed taxi drivers fell from 25,538 to 17,416. Private-hire driver licences rose from 65,656 in 2013/14 to a peak of 117,712 in 2016/17, and still stood at 106,267 in 2023/24.

The total addressable market expanded enormously. People who would never have hailed a cab now took an Uber to the corner shop. By any measure of consumer welfare, this was a win: cheaper rides, faster pickup, more transparent pricing.

But here's the wage story. The trade press now openly describes black cab earnings having fallen by as much as £15,000 annually for many drivers, largely because of rising operating costs and intense competition from app-based services. Black cab drivers, who'd built a credentialled profession around the Knowledge (two to four years of memorising 25,000 streets), are working longer hours for less money. Their craft was a moat. The moat got drained.

The Uber side of the ledger is more complicated, and I want to be careful here. Carl Frey and Thor Berger studied London Uber drivers in 2017-18 and found that median driver earnings exceeded the London Living Wage at the time. But aggregate hourly earnings before vehicle costs sit in a £15-£25 range today, and after fuel, insurance, vehicle rental and platform commission, many drivers operate close to the floor that minimum-wage legislation now provides. Uber was forced by a 2021 Supreme Court ruling to provide statutory minimum wage rights and paid annual leave, which tells you what the prior equilibrium looked like.

This is Engels' Pause playing out in real time. The pie grew. The TAM expanded. Ride volumes are higher than ever. And the per-driver economics compressed toward the wage floor for everyone in the market, black cab and PHV alike. The capital owner here is the platform. The productivity gain (better matching, better routing, better utilisation) accrued to consumers and shareholders. The labour share of value compressed.

What this isn't, and what it is

A note on terminology. You will no doubt have heard people call the Uber phenomenon "Jevons paradox" and that's only half right. Jevons describes a real thing: efficiency gains in the use of a resource often increase total consumption rather than decreasing it, which is why ride volumes exploded. But Jevons doesn't tell you anything about how the surplus gets distributed between capital and labour. For that you need Allen, you need Acemoglu, and frankly you need Marx.

The relevant concept isn't a paradox. It's the displacement effect. In Acemoglu and Restrepo's model, new technologies (steam-powered looms, industrial robots, AI) automate tasks workers used to do, which reduces the share of labour in national income and decouples wages from productivity. Whether this matters for ordinary workers depends entirely on whether the technology also produces a reinstatement effect: new tasks where humans have comparative advantage. Spinning mechanisation reinstated weavers. Weaving mechanisation reinstated nobody, until the railways came along a generation later.

The honest question for AI is which kind of revolution it turns out to be. And the honest answer is: nobody knows yet, but the early evidence on knowledge work is not yet “weaving” in the strong historical sense. But it is enough to worry that AI may raise output before it raises wages, especially where it strips scarcity value from professional judgement, credentials, or routine cognitive expertise.

The piece nobody mentions: politics

Here is the part of the historical parallel I find most under-discussed. The Engels' Pause didn't end because the technology matured. It ended because the political settlement changed.

Look at the timeline of British labour protections. The Health and Morals of Apprentices Act of 1802 had no enforcement mechanism and was ignored. The 1819 Act banned under-9s from cotton mills and was barely enforced. Althorp's Factory Act of 1833 was the first one with teeth: it prohibited workers under 9, restricted the working day in textile mills to 12 hours for ages 13–17 and 8 hours for ages 9–12, mandated 2 hours of school per day for child workers, and crucially established a four-member inspectorate to enforce the law. Then the 1842 Mines Act banned women and under-10s from underground work. The 1844 Factory Act required dangerous machinery to be fenced. The 1847 Ten Hours Act finally won the campaign that had begun in the 1830s.

That's 45 years between the first attempt at regulation and the win that the campaigners actually wanted. It was contested at every step by factory owners and by the dominant intellectual framework of the day. Adam Smith and Ricardo's free-market economics held that markets should determine who was employed and under what conditions, and these views gained widespread acceptance in England in the first half of the nineteenth century.

The 1833 Act mattered because it created a professional inspectorate. Earlier factory laws were not merely advisory, but their enforcement machinery was weak enough that they were often ignored. After 1833, the state had a mechanism to check compliance. This is a deeply unfashionable thing to say in 2026, but the lesson of British industrial history is that productivity gains do not naturally trickle down to workers. They get redistributed by political and institutional pressure, or they don't get redistributed at all.

What this means for anyone working in AI's path

I've spent the last decade in operational risk in banking. The work involves codifying judgement into structured decisions, and AI does this kind of work very well. I am, in the technical sense, a power loom.

This sharpens a few things for me, and probably for anyone reading this whose work overlaps with what large language models do well:

The "they'll find new jobs" argument is true on a long enough timeline and false on the timeline that matters to a 45-year-old. Handloom weavers' grandchildren did fine. Handloom weavers did not. If you're mid-career and your craft is being mechanised, the question that matters isn't whether the economy will eventually reabsorb your skill set. It's whether it will reabsorb you, given your age, your geography, and your mortgage.

The Knowledge mattered for one reason: it was the moat that supported black cab incomes. App-based dispatch and turn-by-turn routing didn't make the Knowledge obsolete in the technical sense. People still pay a premium for it on certain trips. But it stripped out the structural scarcity that gave it economic value. AI is doing this to a lot of professional credentials right now. The CFA, the actuarial exams, the bar exam, accountancy qualifications. None of these are about to become useless. All of them may become less valuable, in the sense of supporting the income they currently support.

The political settlement is contested now and will be contested harder. Whether AI gains accrue to labour or only to capital is not a question the technology answers. It's a question the institutional framework answers. UBI debates, AI safety regulation, antitrust on platform monopsony, the legal status of synthetic content: these are the modern equivalents of the Factory Acts. The people doing this work today are the modern equivalents of Lord Ashley. They will probably be derided by economists for the next twenty years. They will probably also be right.

The bottom line

The old framing of the AI debate has two camps. The boosters say everything will be fine because it always has been. The catastrophists say this time is different and we're all about to be unemployed. Both are arguing about the wrong question.

The right question is the one Engels asked in 1845 and Acemoglu is asking now: who captures the gains, on what timeline, and through what institutions? The historical record is unambiguous. Productivity gains do not flow to workers automatically. They flow when the political settlement compels them to flow. And the gap between the productivity gain arriving and the political settlement catching up is measured in decades, not quarters.

If you are a black cab driver, that gap is your career. If you are a knowledge worker in the path of generative AI, it might be yours too.

Until next time, you'll find me in the back of a London black cab, making sure I tip, and wondering which industry pies are growing while the professional slices are shrinking.