Since AI first came out, people have been worried about the same thing:
"Is this going to take my job?"
The standard responses are either breathless optimism about being freed to pursue our passions, or apocalyptic doom about the end of human usefulness. In truth, I'm not really sure myself, and that's because the answer isn't as clear as you might think.
This has happened already, many times before. History is full of labor-saving technologies that automated human effort overnight. Two very different outcomes emerge from those experiences. Sometimes the machine really does destroy the worker. Other times it acts as a rocket ship, lifting the worker into a much better job.
So which one is AI going to be? I wanted to put both cases in front of you, so you can decide.
I will say up front what I think falls out of the history, so you can argue with me as you read it. The tool never decides. Two other things do: whether the machine took the drudgery underneath your expertise or took the expertise itself, and whether there is a higher rung close enough for you to grab. The handloom weaver and the bank teller both watched a machine swallow the core of their daily work. One trade died in poverty. The other is still here, and the job got better. The difference was never the machine.
Let's dive into examples when tech did and did not wipe out jobs.
The Case for Panic: When the Machine Wins
Let's start with the hard truth. The idea that technology always creates more jobs than it destroys in the long run might be true for the aggregate economy, but it is ice-cold comfort to the person standing on the wrong side of the transition. Sometimes, labor-saving technology simply destroys livelihoods, and the people holding those jobs never recover.
Take the English handloom weavers of the early nineteenth century. Around 1820, there were roughly 240,000 to 250,000 handloom weavers in Great Britain. They were the largest single group of skilled craft workers in manufacturing. They worked from home, set their own hours, and enjoyed a golden age of high wages.
Then came the steam-powered loom. It produced cloth three times faster than a human could.
Almost overnight, the artisanal method became unviable for plain cloth. Wages collapsed. A weaver pulling in 240 pence a week in 1806 was making 99 pence by 1820, and 75 pence by 1830. Entire families eventually tried to survive on five to seven shillings a week, well below the ten-shilling subsistence line.
They did not smoothly pivot to become factory mechanics or early software engineers. Older weavers had hyper-specific skills and geographic ties they could not easily sever. They were trapped. Historians describe this cohort as simply having "died off" in poverty. By 1850, there were fewer than 50,000 weavers left in Lancashire. By the 1861 census, the national number was barely 12,000.
This exact despair birthed the Luddites, desperate textile workers who smashed the frames and looms that were starving them. The state response was brutal.
We saw it again a few years later with the Swing Riots of 1830. For rural laborers in England, manual winter threshing (beating grain with flails) was a lifeline that often accounted for a quarter of their annual earnings. When horse-powered threshing machines arrived, they did the work of dozens of men. The laborers smashed the machines and burned the ricks. The government response was merciless: some two thousand people were tried, hundreds jailed, more than five hundred transported to Australia, and nineteen hanged.
Even highly skilled, modern precision workers have faced the same cliff. Consider the London newspaper typesetters of the 1980s. In an incredible twist of recursive irony, the Linotype machine of the 1880s had already destroyed hand compositors (one operator did the work of six). A century later, those very Linotype operators were destroyed by computerized publishing.
In January 1986, Rupert Murdoch secretly built a computerized plant at Wapping. When the unions struck over the transition, he sacked the entire workforce of 5,500 people overnight. The old Fleet Street ecosystem needed 6,800 workers to function. "Fortress Wapping" ran on 670. Despite a year-long strike and over a thousand arrests, the sacked workers were never reinstated.
The Hard Math of Displacement
There is a macro-economic truth that gets tossed around a lot: the US added over 120 million jobs over the last eighty years amid relentless technological change. But that macro truth hides a micro tragedy.
When high-tenure workers are displaced by a sudden automation shock, they rarely leap to a better job. Economists Louis Jacobson, Robert LaLonde, and Daniel Sullivan found that long-tenure displaced workers suffer earnings drops averaging 25 percent per year, and that penalty persists for decades. Other studies show that workers laid off in severe contractions lose roughly 19 percent of their lifetime earnings.
MIT economist David Autor gives this a helpful framework around expertise. When technology automates the simple, routine tasks of a job, the job tends to get more specialized, better paid, and employs fewer people. But when technology automates the expert tasks of a job, the barriers to entry fall away. The labor pool floods, and wages drop.
Think of the taxi industry. GPS and Uber automated the "expert" knowledge of city routes. Ride-hail employment rose sharply, but wages stagnated because suddenly anyone with a smartphone could do the job. Displaced experts rarely become the shiny new thing. An out-of-work typesetter did not become a software developer. He slid into lower-paid service work, or left the labor force entirely.
The Hinge: The Collapse of the Horse Economy
To see how the exact same technology can be a tragedy for one person and a miracle for another, look at the end of the horse economy.
Between 1900 and 1930, the entire ecosystem surrounding the working horse was dismantled by the automobile and the tractor. In 1910, there were roughly 26 million working horses in the US. The 1900 census counted over 200,000 blacksmiths, plus tens of thousands of carriage makers, farriers, and stable hands.
The Columbus Buggy Company was the largest buggy maker in the US in 1900, employing over a thousand workers. By 1913, after Henry Ford introduced the Model T and the assembly line, Columbus Buggy was bankrupt. Most traditional carriage makers lacked the capital and engineering to pivot to autos, and by the early 1920s, they were practically extinct.
But the automobile did not cause mass unemployment. By 1929, the car had created a far bigger economy than the one it destroyed. There were over 330,000 people employed at auto dealerships, 127,000 at service stations, and 104,000 mechanics, plus whole new industries in highway construction, motels, and suburbs.
The hinge was age and mobility. Take the teamsters, the men who drove the horse teams. Between 1920 and 1930, the number of employed teamsters roughly halved. Younger teamsters simply retrained as motor-truck drivers and rode the wave of the new logistics economy. But older workers, carrying decades of specialized animal-husbandry and wagon skills, were trapped in a dying trade as wages fell.
The tool did not dictate the outcome; the worker's ability to grab the new rung on the ladder did.
The Case for Optimism: When the Machine Elevates
This brings us to the second historical pattern. Much of the time, labor-saving technologies move workers up into new, higher-value work rather than out of the workforce entirely.
The most famous example is the ATM paradox, coined by economist James Bessen. When automated teller machines rolled out heavily in the 1970s and 1980s, they automated the core task of the bank teller: dispensing cash and taking deposits. Everyone predicted mass layoffs.
What actually happened was entirely counterintuitive. The number of tellers per branch did fall, from roughly twenty down to thirteen. But because branches suddenly became much cheaper to operate, banks opened far more of them. Urban commercial bank branches rose by 43 percent. The net result? There were more bank tellers employed in 2013 than there were in 1980.
More importantly, the job got better. Tellers moved away from counting paper and toward relationship banking, sales, and loans, the human parts of the work that machines could not touch.
We saw the same dynamic with bar-code scanners in the 1970s. Scanning cut checkout times by roughly 18 percent, which basic logic said should cut cashier jobs by a similar amount. Instead, cashier employment grew through the 1980s. Cheaper, faster checkouts helped supermarkets and big-box retail expand, and that growth required more cashiers, not fewer.
The Spreadsheet and the Paralegal
If you work in software or data, the closest historical parallel to generative AI is the spreadsheet.
When VisiCalc arrived in 1979, followed by Lotus 1-2-3 and Excel, it automated the manual arithmetic of the ledger. This was an existential threat to the bookkeeping clerk, and that specific routine role did decline.
But cheap, instant recalculation sparked an explosion in demand for financial modeling, auditing, and analysis. Accounting and analyst employment grew massively through the 1980s and 1990s. The tool amplified the worker. An analyst who could run fifty complex financial scenarios in an hour was worth far more to a business than a clerk with paper and a pencil. The labor was reinstated into higher roles.
The corporate typing pool underwent a similar metamorphosis. The massive rooms of women retyping drafts on Selectric typewriters were essentially eradicated by word processing and personal computers. Roughly 3.5 million routine clerical jobs were lost. Yet no systemic depression followed. Between 1980 and 2000, US real GDP nearly doubled, and unemployment fell to around 4 percent. The PC era created a massive net gain in knowledge-economy jobs. Former typists became administrative assistants, IT support staff, and database managers.
Or look at legal e-discovery in the early 2000s. Digital data made human-only document review impossible. E-discovery software allowed small teams to search and tag millions of emails instantly. Pundits predicted the death of the paralegal and the junior associate. Instead, e-discovery became a multibillion-dollar field, and paralegal employment grew alongside it. Cheaper-per-document review simply meant far more documents got reviewed. The time saved was redirected to building chronologies, legal strategy, and client advice.
Even on the grandest scales, this pattern holds. At the start of the 20th century, roughly 70 percent of the US workforce was in farming. Today, it is barely 2 percent, and they produce far more food. Mechanized agriculture did not create a permanent 68 percent unemployment rate. That freed labor was reabsorbed into manufacturing and services, creating the modern middle class.
The Machinery of Growth
Why does this happen? Economists have names for the mechanisms that save us from the machines.
The first is the lump-of-labor fallacy, a term coined by D.F. Schloss in 1891. It is the mistaken belief that there is a fixed, finite amount of work to be done in the world. If a machine does half the work, we assume half the workers must be fired. But productivity lowers prices, which creates extra spending power, which creates new demand and new jobs elsewhere.
The second is the Jevons paradox. In 1865, William Stanley Jevons noticed that making steam engines more efficient did not decrease coal consumption; it massively increased it, because cheaper energy made steam power viable for countless new uses. The same applies to cognitive work. Making data analysis, code generation, or document review cheaper can expand total demand for software and analysis much faster than the per-unit labor falls.
Finally, economists Daron Acemoglu and Pascual Restrepo give us the most precise model: displacement versus reinstatement.
They argue that production is just a set of tasks split between machines and humans. Automation naturally causes a displacement effect. But historically, a reinstatement effect follows, creating new, labor-intensive tasks that humans are better at. They found that roughly half of all US employment growth between 1980 and 2015 was in occupations whose tasks and titles were brand new or had fundamentally changed.
Their warning, however, is that reinstatement is not automatic. If companies pursue "so-so automation" (just cutting headcount on a routine task without creating new human-complementary work), displacement dominates. The outcome depends entirely on the direction of innovation.
The Rules of the Machine
So, will AI take your job? The answer lies in synthesizing these two distinct histories.
Rule 1: Technology automates tasks, not occupations. Jobs survive, and often multiply, when the automated task was the drudgery, and a higher-value human task opens up directly above it. The bank teller becomes an advisor. The bookkeeper becomes a financial analyst. The paralegal becomes a strategist.
Rule 2: Jobs collapse when the automated task was the entirety of the expertise. If your entire value to the market is the manual operation of a handloom, or the manual assembly of hot metal type, and no higher rung opens up, the job disappears.
Rule 3: Survival requires mobility. The exact same technological shock that made a young teamster a mechanic left an older teamster stranded. Whether you survive the transition depends heavily on your willingness to drop your attachment to the old tools and climb into the new paradigm.
The honest bottom line is that history shows the aggregate economy usually recovers and often grows, but specific cohorts of workers can be hurt badly, and "the economy adjusts" is no consolation to them. The dividing line is whether the technology pushes you up toward judgment, strategy, and decisions, or simply deletes your task and walks away.
I don't have a crystal ball, but I wish you (and frankly, myself) good luck out there. Here's hoping the optimists got it right.