A Warning for AI from the Age of Enlightenment

In March 1811, stocking frame knitters in Arnold, Nottinghamshire smashed over 200 machines in three weeks. Within a year, the British government deployed 12,000 troops to suppress the movement, Parliament made machine-breaking a capital crime, and seventeen men were hanged at York. We’ve been calling them technophobes ever since.

Modern scholarship — E.P. Thompson, Eric Hobsbawm, Kevin Binfield’s collection of actual Luddite writings — tells a different story. The Luddites weren’t opposed to technology. They were opposed to how the gains from technology were distributed. Skilled framework knitters, croppers, and weavers watched employers use new machinery to bypass apprenticeship standards, replace trained workers with unskilled labor at a fraction of the wage, and concentrate the productivity gains entirely among owners. The Combination Acts of 1799 had made collective bargaining illegal, so machine-breaking became what Hobsbawm called “collective bargaining by riot,” the only pressure mechanism left.

When a recent study from Berkeley Haas found that AI doesn’t reduce work but consistently intensifies it—employees working faster, expanding into more tasks, losing recovery time, and burning out without being asked to do so—we recognized the pattern immediately. The technology is different. The organizational failure is the same.

The Pattern We Keep Repeating

The textile revolution didn’t destroy workers overnight. It ground them down across a generation. Handloom weaver earnings fell from 25-30 shillings per week in the 1790s to 5-6 shillings by the late 1820s, a roughly 75% wage decline that exceeded the span of a working career. The generation that bore the costs never lived to see the benefits. Yorkshire croppers, the most skilled finishers in the wool trade, wielded 40-pound hand shears four feet long. By 1817, over a third were unemployed, and another 43% only partially employed. Within a decade, the trade was extinct. One unskilled worker operating a shearing frame could replace eight of them.

The speed of AI adoption makes this compression worse, not better. The textile transition from initial mechanization to workforce stabilization took roughly 70-90 years. Generative AI has reached 40% workplace adoption in approximately two years. We’re compressing the same displacement dynamics into a fraction of the timeline, which means the stress response hits harder, recovery windows shrink faster, and the adaptive capacity we need most gets impaired first.

The Ranganathan and Ye study captured this in real time. Workers managed multiple AI threads simultaneously, revived deferred tasks because AI made them feel manageable, and gradually eliminated the natural pauses that prevent cognitive overload. They felt productive. They weren’t less busy. Output was up. Judgment was quietly eroding underneath it.

What Robert Owen Proved in 1800

The most important fact from the Luddite era isn’t the destruction. It’s the alternative that existed and worked.

Robert Owen took over the New Lanark cotton mills on January 1, 1800, and did something no one expected: he invested in workers during the transition and turned a profit. He prohibited child labor under age 10, built schools and healthcare, improved housing, and paid full wages during a four-month production shutdown caused by a US cotton embargo. New Lanark was profitable for the entire 25 years Owen ran it, generating returns above 5% to investors while treating labor as a long-term asset rather than a disposable input.

Over 20,000 people visited New Lanark to see the model in operation. Owen proved that humane transition management and commercial viability were complementary. The mills that fortified against workers (like William Cartwright’s Rawfolds Mill, which opened fire on Luddite attackers in 1812, killing two) won the immediate battle and lost the community. The mill owner who invested in people won the generation.

Most employers chose Cartwright’s model. The results were Manchester, where 57% of working-class children died before their fifth birthday, and Peterloo, where cavalry charged 60,000 protesters — many of them displaced textile workers — killing over fifteen.

Being Owen, Not Cartwright

Owen’s insight wasn’t sentimental. It was operational. He understood that workers who felt invested were workers who could adapt, and that adaptation during technological disruption is worth more than raw compliance. The question for executives managing AI adoption is whether they’re willing to build that same logic into how they deploy the technology today.

In practice, we see three things that separate organizations navigating an AI transition well from those running their people into the ground.

They measure what actually predicts sustainability, not just what looks good on a dashboard. Output metrics will tell you AI is working right up until your best people start leaving. Connection quality between managers and direct reports, the frequency of genuine developmental conversations, and decision quality under time pressure are the leading indicators that predict whether a team is building capacity or consuming it. If a team’s throughput doubles but the manager hasn’t had a real one-on-one in three weeks, that’s not a productivity story. It’s a turnover forecast.

They make boundaries specific and structural, not aspirational. “Take intentional pauses” is the AI equivalent of telling someone to manage their stress. It sounds reasonable, and it fails under pressure every time. What works: concrete behavioral commitments. A fixed time the AI tools close and don’t reopen. A standing weekly review where teams assess what AI-generated work actually moved the business forward versus what was just more activity. A rule that no AI-assisted deliverable ships without a human review checkpoint. Not because the AI can’t produce it, but because the review is where your people maintain the judgment that makes the output worth anything. These aren’t guidelines. They’re operational decisions an executive makes and then stands firm on, even when quarterly pressure says otherwise.

They account for who actually bears the cost. The Luddite movement had an equity structure that maps directly onto today’s data. Skilled croppers in Yorkshire had more economic buffer than framework knitters in Nottinghamshire. Senior leaders in the Berkeley Haas study had more control over their workflows and more organizational permission to set boundaries than junior employees, who reported burnout at significantly higher rates. Any AI governance framework that ignores power differentials in who can actually enforce work boundaries will reproduce this 200-year-old pattern. Audit AI workload impact by level and role, not just by team. Ask who’s absorbing the intensification and who has the authority to push back. If those are different groups — and they almost always are — redesign the boundaries so they protect the people with the least power to protect themselves.

Owen didn’t manage New Lanark with inspirational speeches about worker wellbeing. He changed the operating conditions: who worked, for how long, under what terms, with what support. The decisions were structural, and the results were both humane and profitable for a quarter century.

The Choice Is the Same One It Was in 1812

The Luddites were right about the problem. Owen proved the economics of investing in people during disruption. Two hundred years later, we have better tools, better data, and no excuse for choosing Cartwright’s model—fortify, suppress, and win the quarter—when Owen’s model is sitting right there in the historical record, profitable and proven.

The question isn’t whether AI will transform work. It’s whether the gains will be shared with the people doing the adapting, or hoarded by the people making the deployment decisions.

Two hundred years of evidence say the answer matters more than the technology.

This work synthesizes research from E.P. Thompson (The Making of the English Working Class, 1963), Eric Hobsbawm (”The Machine Breakers,” Past & Present, 1952), Kevin Binfield (Writings of the Luddites, 2004), Daron Acemoglu and Simon Johnson (MIT, 2024), G.H. Wood’s 1910 wage series, and Ranganathan and Ye (Harvard Business Review, 2026).

—