The Speed of Forgetting
In 1900, New York City had a manure problem.
The city’s 100,000 horses produced roughly 2.5 million pounds of manure every day. It piled in the streets. It drew flies. The Times of London predicted that by 1950, every street in the city would be buried under nine feet of horse manure.
By 1912, the last horse-drawn streetcar made its final run. By 1920, horses had essentially vanished from American cities.
The speed shocked everyone. Not because they couldn’t see it coming — the automobile had been around for years — but because they couldn’t feel it accelerating. The thing about exponential change is that it arrives slowly, then all at once. And the human mind, evolved to notice the rustling of leaves and the approach of predators, is spectacularly bad at intuiting what “doubling every few months” actually means in practice.
I wrote recently that the biggest companies will capture AI’s value. I still believe that. The incumbents with distribution will absorb AI functionality and expand their reach. But there’s a second question I didn’t answer: what happens to the people inside those companies? What happens to the workers when coordination becomes cheaper and the optimal workforce size shrinks?
The answer, if history is a guide, is the same thing that always happens. They get displaced faster than they expect.
My sense is that most people tried an AI tool, found it mildly clever but obviously limited, and moved on with their day.
This is exactly what Western Union did in 1876 when they turned down Alexander Graham Bell’s offer to sell them the telephone patent for $100,000. An internal memo described the device as “hardly more than a toy.”
They weren’t idiots. The telephone was limited. The sound quality was poor. It only worked over short distances. And Western Union already had a perfectly good system — the telegraph — that had been printing money for decades.
What they couldn’t see was the slope of the curve.
The problem with exponential improvement is that the early days look linear. You try the tool, it’s 70% as good as a human, and your brain does the math: if it improves 10% a year, I’ve got a decade before I need to worry.
But it doesn’t improve 10% a year. It improves 10% every few months. Then faster.
This is what people get wrong about exponential growth. It’s not that things get a little better each year. It’s that the rate of improvement itself accelerates. The jump from last January to last June was significant. The jump from June to December was larger. The jump from December to now was larger still. Each step forward makes the next step bigger.
We understand this intellectually. We do not understand it intuitively. The human brain expects linear progression — steady, predictable, manageable. Exponential progression feels like nothing is happening, and then everything is happening, and there was no moment in between where you could have braced yourself.
The Luddites have been mythologized as backwards fools who smashed machines because they feared progress.
The reality was more specific. They were highly skilled textile workers — weavers, croppers, spinners — who had spent years learning their craft. And they were right. The power looms really did destroy their livelihoods. Within a generation, the wages of handloom weavers in Britain fell by 75%. Many starved.
History remembers them as the punchline. The cautionary tale of what happens when you stand in the way of innovation.
But here’s the thing: being right about what’s coming doesn’t mean you can do anything about it. The Luddites saw the future clearly. They just couldn’t stop it.
The mill owners, meanwhile, got richer. They owned the distribution — the factories, the supply chains, the customer relationships. The technology changed, but the people who controlled the infrastructure kept winning.
It’s the same pattern as Rockefeller. When Standard Oil was broken up, Rockefeller owned the distribution networks and institutional knowledge. The technology shifted from kerosene to gasoline, but he captured the value because he owned the rails and refineries. His workers didn’t own anything. When their specific skills became obsolete, they just became obsolete.
Corporations capturing value and workers losing livelihoods aren’t contradictions. They’re two sides of the same coin.
I think about this when people say, “AI will create new jobs we can’t imagine yet.”
Maybe. Probably, even.
But the Luddites’ children didn’t become loom operators. The jobs that eventually emerged — factory supervisors, mechanical engineers, department store clerks — required different skills, in different places, for different people. The weaver who lost his job in 1815 didn’t retrain as an industrial engineer. He died poor.
The economy adapts. People often don’t.
The AI labs have been clear about what they’re building and how fast it’s arriving. The estimates from the people closest to the technology — not the pundits, not the skeptics, but the engineers and executives actually training the models — range from one to five years before the majority of entry-level white-collar work is fundamentally changed. Not replaced overnight. Changed. Compressed. Restructured around what the machine can’t yet do.
The word “changed” is doing a lot of work there, and I think that’s intentional. Because the messy truth is that this doesn’t happen cleanly. It’s not that one day you have a job and the next day you don’t. It’s that your job slowly becomes a smaller job, then a different job, then a supervising-the-AI job, then maybe no job.
People say, “I tried AI and it wasn’t that useful.”
Of course it wasn’t. You’re successful. You’ve spent years building skills and workflows that already work. The AI is clumsy and requires hand-holding and doesn’t understand context the way a human does.
But the 22-year-old who’s just starting out? They’re learning to use the AI from day one. They’re building their workflows around it. They don’t see it as a threat to their established methods because they don’t have established methods yet.
This is how every transition happens. The old guard doesn’t adopt the new thing. They age out. The new generation grows up native.
There’s a folk song about John Henry, the steel-driving man who raced a steam-powered hammer to prove that human muscle could still beat the machine.
He won the race. Then he died, hammer in hand, his heart gave out from the effort.
The song is sung as a tragedy, but I’ve always thought the real tragedy is what it says about how we frame these contests. We tell ourselves it’s about human dignity. About proving our worth. About not being replaced.
But the steam hammer didn’t care whether John Henry beat it that day. It just kept working. The next day, and the day after that, getting faster and cheaper and more reliable.
The race was never the point. The result was already determined.
The question isn’t whether your job involves technology. It’s whether your job involves patterns.
Reading a contract and flagging risk. Analyzing a spreadsheet and finding the story. Writing a brief from precedent. Diagnosing from symptoms. Building a model from assumptions.
These are pattern-recognition tasks. And pattern recognition is what AI does, relentlessly and at scale. The specific domain doesn’t matter. The underlying cognitive structure does.
If your work is built on patterns, the machine is learning them.
The companies with distribution will absorb this and grow more efficient. That’s the investment thesis. But “more efficient” is a euphemism. It means doing the same work with fewer people.
The institutions win. The individuals inside them face the same question the handloom weavers faced: what do you do when the thing you’re good at stops being scarce?
The Fortune 500 companies aren’t going anywhere. Their distribution moats are real. But the number of people working inside them is a different question entirely.
I don’t know how this ends. Nobody does.
But I know what the pattern looks like, because it’s the same pattern every time.
A new technology arrives. It’s expensive and limited. Experts dismiss it. Then it gets a little better. Then a lot better. Then it’s everywhere, and the world we built around the old technology starts to crack.
The people who see it early feel crazy because nothing has changed yet. The people who see it late feel blindsided because everything changed at once.
The truth is that both groups are looking at the same curve. They’re just standing in different places.
The scribes who copied books by hand in medieval Europe were the most educated people on the continent. They knew Latin, Greek, theology, history. A single Bible took months to complete. Their work was sacred. Irreplaceable.
Then Gutenberg’s printing press arrived. By 1500, 50 years later, Europe had printed 20 million books. The scribes didn’t become printers. The skill sets were too different. The entire profession just vanished.
What’s strange is that we remember Gutenberg as a hero. The printing press unlocked the Renaissance, the Reformation, the Scientific Revolution. It was obviously good.
But if you were a scribe in 1450, watching your life’s work become obsolete, “obviously good” is not how it felt.
I think that’s the thing about exponential change that’s hardest to hold in your head. It can be both true that something is good for the world and devastating for you. Both true that the future is abundant and your skills are worthless. Both true that progress is real and you’re being left behind.
The curve doesn’t care about your mortgage. It doesn’t care that you did everything right. It just moves.
The speed of forgetting is faster than the speed of change.
In 1900, horses were everywhere. By 1920, people had to be reminded they ever existed.
We think of transitions as slow because we remember them in retrospect, compressed into a paragraph in a history book. But if you lived through it, it felt like the ground disappearing beneath your feet.
My sense is we’re in the early days of one of those transitions now. Not the middle. Not the end. The beginning.
Which means most of what’s coming is still invisible.
The manure is piling up in the streets, and we’re still arguing about whether automobiles are a fad.