和工业革命一样，总量层面的故事会是凯旋的。但个体层面的故事常常不是。兰开夏的手织工没有被再培训成火车司机。他们中很多人就是输了，而且损失横跨一代人。经济史学家 Robert Allen 给这段时期起了个名字：Engels 停顿。1780 到 1840 年间，英国的人均产出上升了 46%。实际工资只上升了 12%。
The best-written bullish history of AI: we outsourced our muscles to build the modern world, and now we are outsourcing our minds.
Indigo's conclusion
His optimism rests on the analogy holding, and the biggest difference he names himself, speed, is exactly where the analogy may break.
How to read this The author is a Sequoia partner investing heavily in AI, so “turbulent but better in the end” suits him. To his credit he doesn't dodge the bad side: the Engels pause, a lost generation, speed as the real risk. Use the historical frame; check the step to “the whole will surely be better” on its own.
What to remember
Physical work took two hundred years to go from 99% to 99.9% machine-done; cognitive work is replaying it, bigger and faster.
The killer app that defines this revolution hasn't appeared. Three candidates: science at machine speed, a chief-of-staff agent for each person, new ways for people to connect and coordinate.
The Engels pause: from 1780 to 1840 British output per person rose 46% while wages rose only 12%. The Luddites were right about their own lives and wrong about their grandchildren's.
Breakdown · 5 steps
01
Thinking work is retracing the path physical work took
Over two centuries physical work went from 99% done by people and animals to 99.9% done by machines. A century ago 99% of cognitive work was done by people; soon 99.9% will be done by machines. Read this part →
02
The playbook: inputs get cheap, demand explodes
Power plus compute are the new coal and iron; the Transformer is the new steelmaking process. Jevons saw it long ago: more efficient engines multiplied Britain's coal use. Read this part →
03
Coding agents are the spinning jenny; the car hasn't been built
The 1764 spinning jenny moved people from doing the work to supervising machines. Textiles started the Industrial Revolution, but the car defined it. Read this part →
04
Jobs, education and medicine all replay
In 1800, 75% of American workers farmed; today about 1%, and mass unemployment never came. The dark side is the Engels pause. Education and medicine were always short of people; that limit is lifting. Read this part →
05
What stays with people was never cognitive work
Wanting something, choosing between options, answering for the choice, being trusted by others. A machine can draft the treaty; someone still has to sign it. Read this part →
What would change my mind
the hiring data he promised for software engineers, law and finance is published and really does show an upturn.
How to read this
The author is a Sequoia partner investing heavily in AI, so “turbulent but better in the end” suits him. To his credit he doesn't dodge the bad side: the Engels pause, a lost generation, speed as the real risk. Use the historical frame; check the step to “the whole will surely be better” on its own.
Thinking work is retracing the path physical work took
Over two centuries physical work went from 99% done by people and animals to 99.9% done by machines. A century ago 99% of cognitive work was done by people; soon 99.9% will be done by machines.
We externalized our muscles and built the modern world. Now we are externalizing our minds.
Most mornings I ride to work in a Waymo.
The car is doing physical work: five thousand pounds of metal and glass moving across the San Francisco Bay Area. The car is also doing cognitive work: reading the road and making predictions that result in 17x fewer serious-injury crashes than human drivers.
The experience asks nothing of my muscles and nothing of my mind. Both have been externalized.
This one ride encapsulates two revolutions. The physical one is 200 years old. The cognitive one has barely started.
It seems the mood around that second revolution is a cocktail of optimism, skepticism and anxiety. Talk of a singularity. Recursive self-improvement. Trillion-dollar clusters. Superintelligence. Unemployment. Social disruption. A race between superpowers. It’s dizzying.
My view is far from dizzying. I think we have already undergone a transition that rhymes with the AI revolution in the coming decades: the Industrial Revolution.
Two Kinds of Work
We can bifurcate work into physical work and cognitive work. Most valuable tasks necessitate both.
Agriculture is one of the oldest examples. Working the land took muscle, ours and our animals', for ten thousand years. It also took insight: reading the seasons, observing which seed produced which yield, etc.
The two kinds of work are different. Physical work is force times distance: moving mass through space. Cognitive work is thinking. But both are scarce, have a price, and get applied to a task somebody wants done. In both cases when price collapses, supply floods in. In most cases, demand expands to meet supply.
What Happened to Physical Work
For most of history, nearly all physical work done for humans was done by muscle, ours and animals. The chart above goes flat before the 1700s, and the flat part runs back as far as we do.
In the late 1700s and early 1800s things started to change. First steam, then combustion and the electric motor. Each wave took a larger share of the world's physical work. Over the course of about two centuries, physical work went from 99% biological to 99.9% machine.
Today the results are everywhere. The screen you’re reading this on. Nearly every article of clothing you’re wearing. The plane that carried you on your last trip. The lattice of ships, trains and trucks behind every ordinary product in modern life. Human muscle is a rounding error in the global energy budget.
History Rhymes: What’s Happening to Cognitive Work
For most of history, essentially all cognitive work was done by humans (plus a small assist from animals like the sheepdog). On top of that sat a thin mechanical sliver: tools that computed a little, like the clock or Pascal's mechanical calculator.
Then came electronic computation. Inside a century, it spread from simple calculations and spreadsheets to trillions then quadrillions of operations running 24 hours per day. Navigating your commute, processing payroll, pricing your insurance…this is all cognitive work done by machines.
The next wave is neural networks. This wave drastically expands the types of cognitive work machines can do. Just as combustion engines magnified the scope of the Industrial Revolution, neural networks are set to accelerate the Cognitive Revolution.
A century ago, 99% of cognitive work was done by humans. In the near future, 99.9% will be done by machines. Not because humans will think less, but because machines will compute much much more.
The parallel in the two curves is stark. The Cognitive Revolution will look a lot like the Industrial Revolution. But it will be bigger and much faster. Divide the global economy by the nature of work, and cognitive and physical work could split a ~$120 trillion world economy roughly equally. The physical half has been mechanizing for 200 years. The cognitive half has barely started.
So what actually happens in a revolution like this? Let’s run it back.
02
The playbook: inputs get cheap, demand explodes
Power plus compute are the new coal and iron; the Transformer is the new steelmaking process. Jevons saw it long ago: more efficient engines multiplied Britain's coal use.
Inputs Scale Massively and Get Cheap
The industrial age required massively scaling up input commodities: first coal and iron, then oil and steel. Iron was made by a process known as puddling. Steel was made by a process called the Bessemer process.
The parallel is clear: electricity and compute. The “process” here are algorithms like the Transformer.
The providers of those components were the titans of their era. Carnegie and U.S. Steel, Rockefeller and Standard Oil. Process designers like Henry Bessemer. The modern equivalents are Jensen Huang's NVIDIA, Morris Chang’s TSMC, authors of the Transformer paper, founders of the great Model Labs and the hyperscalers pouring trillions into data centers. Nations that once measured power in refining capacity will measure it in grid capacity and intelligence output.
A unit of mechanical work in 1900 cost a small fraction of what it cost in 1800. That collapse in price made mass production possible, and made the revolution universal rather than a curiosity for the rich.
The cost of cognition is collapsing faster. Intelligence per Watt has been falling more than 10x per year.
With price decline, customers will buy wildly, absurdly more intelligence.
Demand Explodes to Meet Supply
Nobody in 1800 flew to Tokyo, shipped strawberries across a hemisphere in January, or air-conditioned Phoenix. All these are physical processes that either were impossible or impractically expensive.
Demand for physical work exploded once new technologies were invented and price collapsed. William Jevons identified the phenomenon in 1865: more efficient engines did not reduce Britain's coal consumption. They multiplied it.
Cognition will follow the same pattern, and here the latent demand is almost unbounded in scale. Most problems on Earth currently go un-thought-about, because thinking is expensive. A doctor has limited time and doesn’t review the latest literature. A radiologist studies a scan for minutes instead of hours. A small business owner has no budget for a financial analyst.
When cognition costs approach zero, every curiosity can get a research team.
The amount of thinking done by machines will grow by orders of magnitude. Which raises the question, what will all that thinking be spent on?
03
Coding agents are the spinning jenny; the car hasn't been built
The 1764 spinning jenny moved people from doing the work to supervising machines. Textiles started the Industrial Revolution, but the car defined it.
The Application Era
The primal application of the Industrial Revolution was spinning textiles.
Spinning and weaving were skilled work. Textile makers were respected in your society: years of apprenticeship, a craft guild, a trade the family had held for generations. In 1764, James Hargreaves built the Spinning Jenny, which let one person spin eight threads at once, and later eighty. Clothing prices fell rapidly and global clothing consumption exploded.
The big change was the human moving from performing the task to supervising it. One person tending a frame produced what a room of hand-spinners once had.
The primal skilled application of this era is AI coding. Writing software is our era's elite craft: highly skilled, highly paid, a decade to master. And it is the first cognitive task where the pattern has fully flipped. The machine now completes most of the work, and the engineer's job is to direct, review and correct it. It’s no accident that this is where the Cognitive Revolution found its first commercial traction.
While textiles kicked off the industrial revolution, they did not define it. The ultimate innovation of the industrial revolution was the automobile: a civilization grew around it, including suburbs, highways, supply chains, and the entire geography of modern life.
Coding agents are our Spinning Jennies: the first act, automating a skilled cognitive task. The automobile of this revolution has probably not been built yet. It will be something that was impossible while thinking was scarce.
My candidates: science conducted at machine speed, the personal agent that serves your life the way a chief of staff organizes a CEO's, or something in how humans connect and coordinate. In 1800, the automobile would have been hard to predict.
04
Jobs, education and medicine all replay
In 1800, 75% of American workers farmed; today about 1%, and mass unemployment never came. The dark side is the Engels pause. Education and medicine were always short of people; that limit is lifting.
Employment: Where the Anxiety Lives
In 1800, over 75% of American workers worked the land. Machines came for that work. Today, farming is about 1% of U.S. employment. By the pessimists' arithmetic, over 74% of the country should be unemployed.
Instead we invented work no farmer in 1800 could have named. Radiologist. Software engineer. Podcast producer. Flight attendant. The machines did not end labor. They eliminated categories of work and expanded the total amount of it. There are more people employed today than at any point in human history, and they are vastly richer than their ancestors. All of my ancestors worked in agriculture 100 years ago. I knew all four of my grandparents, all of whom were born to farming families. None of them finished their lives in agriculture. All of them considered the non-agricultural work they did at the end of their lives more pleasant and prosperous than the hard agricultural work their families had done for generations.
The same is taking shape in the Cognitive revolution. Despite the “spinning Jenny” phenomena in AI coding, data supports that demand for software engineers, the most AI exposed occupation was accelerating higher, has inflected. In the coming week, we plan to share data supporting inflections in hiring in other areas being transformed by AI, including Legal Services and Financial Services.
Machines doing cognitive work will not mean humans stop doing cognitive work any more than the tractor meant humans stopped doing physical things. The specific tasks change. The junior analyst who builds the model by hand becomes the person directing a hundred models. Further, categories of work we cannot currently name will grow the way "engineer" became a job following the invention of the engine during the Industrial Revolution.
Like in the industrial revolution, the aggregate story will be triumphant. But, the individual story will frequently not be. The handloom weavers of Lancashire were not retrained into locomotive engineers. Many of them simply lost, and their losses spanned a generation. The economic historian Robert Allen gave the period a name: Engels' pause. Between 1780 and 1840, British output per worker rose 46%. Real wages rose 12%.
Luddites were not wrong about their own lives, but about their children’s and grandchildren's. Between 1840 and 1900, British output per worker rose 90% and real wages rose 123%. Wages caught up to productivity, and then outran it.
Apause is not a foregone conclusion. Mark Zuckerberg put this well recently: “There is no rule that AI must increase automation faster than it increases individuals' capabilities or demand for new skills.” Humanity’s demand for new skills, for better and worse, will be insatiable.
The difference this time is speed. The Industrial Revolution gave a farmhand's son forty years to become a factory hand. This revolution may offer five. Society's shock absorbers, like education, retraining, and safety nets, were designed for the old clock. The central policy question of the next two decades is whether we can help humans transform as fast as the jobs do.
Education
The Industrial Revolution created the modern education system, in at least two ways.
First, it made mass education possible. When most of humanity had to work the fields, children were labor. Machines taking over the physical work is what freed children to sit in classrooms and freed societies to fund them. The modern university exists because society no longer needed every able body in the field, and could redirect a civilization's youth toward cognitive skill-building instead.
Second, and less flatteringly, it shaped what education became. The factory needed workers who could read instructions, do arithmetic, show up on time and perform standardized tasks in synchronized shifts. So we built schools that look suspiciously like factories: Bells. Rows. Fixed schedules. Batch processing by age. Standardized outputs. Mass education was, in large part, industrial job training.
The Cognitive Revolution upends both of these dynamics. If machines do the cognition, the factory-model school is the equivalent of training for hand-spinning. But the deeper change is on the delivery side. For all of history, the binding constraint on education was the ratio of teachers to students. Benjamin Bloom quantified the cost of that constraint in 1984: the average student tutored one-to-one with mastery learning outperformed 98% of students in a conventional classroom.
That constraint is dissolving. The marginal cost of an infinitely patient, endlessly knowledgeable tutor is approaching zero. Hybrid human-AI tutors are found to be more effective and less costly than traditional education. Further, new hybrid AI approaches could help spread education in parts of the world where public education has been prohibitively expensive.
Health
Modern medicine was also built by the industrial revolution. Mass production gave us penicillin at scale, refrigeration that made vaccines deliverable, and the chemical industry that spawned the pharmaceutical industry.
The Cognitive Revolution pushes this much further. Drug discovery is a search across a space too large for humans to comprehend, which is exactly the work that scales best when cognition gets cheap. By predicting structures for virtually every known protein, AlphaFold solved a challenge where even minor breakthroughs previously demanded entire lifelong careers.Thousands of diseases currently have no treatment, not because they are unsolvable but because the cost of investigating each was never justified by the number of patients. When that cost collapses, the long tail becomes addressable the way every other latent demand does.
Curing disease has been the ambition of every generation of physicians. This is the first generation where the binding constraint is no longer how many of them there are.
Then there are the doctors in clinical practice. We face a massive shortage of physicians both in the United States and globally, with demand far exceeding supply. This issue is deeply personal to me, as I have long seen healthcare workers in my family stretched thin attempting to meet patient needs. Cheaper, accessible cognitive tools will not lead to doctors taking time off; instead, they will allow more patients to receive higher-quality care. Given the virtually limitless demand for better healthcare, scalable intelligence can help shift our medical model from reactive treatment to proactive care.
The Shape of Daily Life
Beyond work and school, the Industrial Revolution changed the texture of ordinary life so thoroughly that we now mistake its artifacts for human nature. The Cognitive Revolution will change them once again.
Clock time. Before factories, rural work followed seasons and sunlight. But a factory owner needed hundreds of people to start and stop simultaneously. The machine ran on a schedule, so the humans had to. Rigid shifts, the workday, the work week, modern timekeeping itself: these are industrial inventions, barely two centuries old. The Cognitive Revolution loosens their grip. When your machine colleagues work continuously and asynchronously, there is no reason for a hundred humans to think in unison from nine to five. Work drifts back toward task rhythms rather than clock rhythms.
Home and Work. Work happened in and around the home for most of human history: the farm, the cottage workshop, the family trade. Industrialization pulled the workplace out of domestic space because the work had to happen where the machine was. The daily commute is the residue of that separation. The Cognitive Revolution reverses it, because the machine now lives everywhere and one of the most powerful tools in history fits in your pocket. Home and work are recombining.
Global trade and power. Industrial nations imported raw materials and exported finished goods, an asymmetry that drove imperialism. The new raw materials are energy, compute and algorithms. The finished good is intelligence. The last time this asymmetry appeared, it defined a century of international order and disorder: the same industrial base that laid rail and poured steel also eventually created atomic bombs. Intelligence will be dual-use in the same way, and faster. Nations that miss the Cognitive Revolution will be disadvantaged and suffer for many generations.
05
What stays with people was never cognitive work
Wanting something, choosing between options, answering for the choice, being trusted by others. A machine can draft the treaty; someone still has to sign it.
What Remains Human
Follow the thread and you eventually arrive at the real question underneath the anxiety about AI: If the machines do the thinking, what are we for?
The last revolution already answered the physical version. At the end of the Industrial Revolution, humans did not stop moving their bodies. We remained embodied creatures.
Proportionally, fewer of us work intense physical jobs. However, far more of us move for sport and purpose: we run marathons, climb mountains, and pay for the privilege of holding yoga poses in uncomfortably hot rooms. Along the way, we created sports-leagues and even the modern Olympics were the result of the Industrial Revolution.
Cognition will make the same migration. Chess is a clean example. Deep Blue beat Garry Kasparov in 1997 and Kasparov later half-jokingly described himself as the first knowledge worker whose job was threatened by a machine. Chess did not die. More people play it today than at any point in history. When AlphaGo beat Lee Sedol, researchers found that the novelty of moves played by human Go professionals increased significantly afterward. Superhuman machines made human play more creative, not less. The “centaur” era of machine-human collaboration was brief, but the human era never ended.
Critics of this essay will argue that “this time is different.” Specifically, they will point out that humans were the smartest animals on earth and that this is the first time something uniquely human is displaced by a machine. That is incorrect. It is incorrect for at least two reasons. First, thinking is not uniquely human. Animals think. Even nature follows complex and “intelligent” patterns. Second, there are many uniquely human skills that machines exceed already. For example, making fire is strictly a human skill…and now your stove does it with the press of a button. Same with writing and transferring information (the internet). Same with making tools (us and chimpanzees), but now machines make most of our tools. And then Einstein reinvented it completely-- humans making a comeback -- with a new human thought experiment that was needed to accurately plot the orbit of Mercury! Flight was a super-human task until airplanes made it accessible. This time is different, but not that different.
What stays economically human the longest is what was never really cognition to begin with: wanting things, choosing between them, being accountable for the choice, and being trusted by other people. The machine can draft the treaty. Someone still has to sign it. Over 2500 years ago, Greek philosopher Protagaros wrote "Man is the measure of all things." Indeed, value as we experience it is in what we offer to other people.
Conclusion: We Have Made This Trip Before
Stand far enough back and this is all one story. For all of history, humans did the work: with our muscles and with our minds. Twice, we built machines that took the work and externalized it to machines.
The first time, we were terrified, and the fear was short-term justified and long-term wrong. The world after the machines took most of the physical work was profoundly more abundant, healthy and humane.
The same trade is on the table now, at greater scale and higher speed. The Cognitive Revolution will be turbulent, unevenly distributed, and deeply uncomfortable . And, it will end the way the last one did: with the world unrecognizably better, its results so completely woven into daily life that our grandchildren will live inside them.
We have made this trip before.This time, we can navigate it with our eyes open: shaping the transition calmly, sharing the prosperity, and remembering that the goal of externalizing our minds is ultimately to elevate our humanity.
11:07 PM · Sep 2, 2026·560.3K
Where Indigo landsFurther
Indigo's conclusion
His optimism rests on the analogy holding, and the biggest difference he names himself, speed, is exactly where the analogy may break.
What to remember
Physical work took two hundred years to go from 99% to 99.9% machine-done; cognitive work is replaying it, bigger and faster.
The killer app that defines this revolution hasn't appeared. Three candidates: science at machine speed, a chief-of-staff agent for each person, new ways for people to connect and coordinate.
The Engels pause: from 1780 to 1840 British output per person rose 46% while wages rose only 12%. The Luddites were right about their own lives and wrong about their grandchildren's.
Claims you can check later
Claim
Who
When we will know
How firm
Demand for software engineers (the job most exposed to AI) has already turned up; data on hiring turns in law and finance comes next week
Konstantine
Around mid-September 2026
First-hand; promised, not shown; to be checked
The cost of cognition falls more than 10x a year
Konstantine
Ongoing
First-hand citation; needs checking
Soon 99.9% of cognitive work will be done by machines
Konstantine
Decades
First-hand; extrapolated from history
This revolution's “car”, the application that defines it, hasn't been built yet
Konstantine
Future
First-hand; a call on direction
Speed of transition is the real risk (40 years shrinking to 5); the core policy question is whether people can move as fast as the jobs do
Konstantine
Next twenty years
First-hand; a call on direction
Back on the long-running theses
confirms + adds to
The AI jobs apocalypse: how the view has evolved The bullish history version: better overall, with an Engels pause, and this time the difference is speed. Consistent with AI replacing work task by task.
confirms
You don't own the model: value moves to what can't be rented “Wanting, choosing, answering for it, being trusted” is almost this view word for word. He thinks it is enough; this view treats it as a moat.
adds to
The shape of demand: bounded vs. unbounded “When thinking costs almost nothing, demand explodes” gives the demand thread a historical reference point.
conflicts
Gregory Conti, “The AI apocalypse is already here” Two poles on one subject: one says we have been here before, the other says this time is different in kind.
confirms
Sarah Guo (Conviction) Sarah also bets on the Jevons paradox and “we will all be employed more”; this is the long historical lens on the same call.