Ai is eating the white-collar jobs we swore were safe
Silicon Valley keeps promising that artificial intelligence will “augment” human labour. Tell that to the 19th-century hand-loom weavers who watched their wages collapse by 50 % in fourteen years once the mechanical loom arrived. The same script is rolling again, only this time the victims wear Patagonia vests, carry Moleskines and bill £400 an hour.
The last guild to fall
Between 1806 and 1820, British weavers went from earning double the factory wage to 25 % below it. No quarterly earnings call announced the drop; villagers simply noticed that piece-work dried up and the cloth merchant suddenly owned a smoke-belching mill. Historians call it the first great deskilling. Economists call it a “sectoral shock.” The weavers themselves called it hunger.
What terrifies labour historians is how closely today’s prompt engineers, paralegals and junior analysts resemble those artisans. They are educated, urban, politically articulate—and, crucially, expensive. Precisely the demographic whose cost structure makes CFOs salivate when a subscription to GPT-4 Enterprise lands on their desk for $60 a month.
Google’s latest ad copy boasts that its Vertexai now drafts 60 % of the code committed inside Alphabet. Goldman Sachs quietly retired half its first-year analyst intake this spring, replacing them with internal large-language-model tools that read IPO prospectuses overnight. No picket lines, no smashed servers—just a Slack message: “Your access has been deactivated.”
Why this wave is faster than steam
The spinning jenney needed bricks, boilers and canals. Neural nets need only electricity and a credit card. Adoption curves that once stretched across two lifetimes now fit inside a single budgeting cycle. When the UK Parliament repealed the Calico Acts in 1774, it took forty years for mechanised cotton to dominate output. ChatGPT reached 100 million users in sixty days, each prompt a tiny knife-cut to somebody’s billable hours.
Mediocre coders were first. Mid-tier copywriters followed. Now the frontier is creeping into radiology, where Stanford’s CheXagent flags lung nodules faster than any fellow, and into M&A due-diligence, where Harvey ai spits out 200-page risk memos before the junior lawyer has finished her flat-white. The common denominator: tasks once deemed too fuzzy, too linguistic, too “human” for machines.
What the novels tried to warn us
Elizabeth Gaskell’s North and South is marketed as a romance, but its real tension is economic: can a southern magistrate’s daughter stomach the moral cost of a northern mill owner’s profits? The answer, delivered amid strikes and cholera, is that technology does not politely stop at the garden gate of the gentry. It rewrites class, courtship, even the air you breathe.
Dickens’ Scrooge is less a miser than an early venture capitalist, outsourcing risk to the workhouse while he counts compound interest. His epiphany is not spiritual; it is reputational. Once the ghosts reveal that the city will toast his death, he realises generosity is simply better PR. Likewise, today’s CEOs discovering that “responsible ai” audits insulate share price against regulatory backlash are not enlightened; they are hedging.
The metric that matters
The Office for National Statistics will not publish a “prompt engineer redundancy index.” Instead, watch the payroll-to-revenue ratio inside S&P 500 firms. It has fallen for nine straight quarters, even as top-line growth stagnates. The money saved is being funnelled straight into Nvidia purchase orders, a circular trade that enriches the hardware layer while hollowing out the human one.
History offers no comfort that new, better jobs will magically appear. The weavers’ grandchildren did find work—inside the same mills that destroyed their elders—only at half the wage and triple the tuberculosis risk. The lag between productivity gains and wage gains stretched seventy years. Most adults alive during the transition were dead before the rebound.
How to stay employable—for now
Specialise in the hand-grenade tasks ai still fumbles: cross-examining a hostile witness, calming a terrified patient, negotiating a truce between two founders who want to kill each other. These are not “soft” skills; they are entropy-rich problems where context, emotion and legal liability collide. The moment they become pattern-recognition exercises, they too will be eaten.
Second, rent your brain to the machine instead of competing with it. The top-earning radiologists this year are those who annotate training data for DeepMind, earning more labelling tumours than diagnosing them. A gruesome irony: the better you teach the model, the faster you automate yourself. But the pay cheque cashes today.
Finally, read the novels. Not for nostalgia, but for the recognition that every generation believes its own disruption is unprecedented. The weavers also assumed their craft was too intricate for gears. Their mistake was trusting the mill owner’s promise to “augment” them rather than replace them. The page turns; the loom clatters; the office lights stay on, but the desks are empty.