AI Isn't Killing Jobs. It's Shifting the Bottleneck.

To understand what AI is about to do to your work, you have to go back about two hundred years, to a world where every piece of clothing on Earth was made by hand.

Before the Industrial Revolution, cloth was woven one thread at a time. A weaver sat at a wooden loom, usually in his own cottage, passing a shuttle of thread back and forth, row after row. Hours of that work produced a single yard of fabric. Cloth was so expensive that an ordinary family owned a few garments, wore them for years, and patched them until they fell apart. A new shirt was a serious purchase. Weaving was skilled, respected, slow work, and it employed enormous numbers of people.

Then came the power loom. It was a machine, driven first by water and then by steam, that did the weaver's physical motions automatically: throwing the shuttle, packing the thread, advancing the cloth. One operator could now tend several machines at once instead of one artisan working one loom. Over the nineteenth century, machines like this automated 98% of the labor it took to weave a yard of cloth.

Ninety-eight percent. If you'd stood in a Manchester mill in 1830 and announced that number, every weaver in the room would've predicted the end of their trade. Plenty of them did predict exactly that. Some smashed the machines. Investors, meanwhile, poured fortunes into anything with a smokestack. Half the country was convinced work itself was ending, and the other half was convinced they were about to get rich.

Here's what actually happened: the number of weaving jobs went up, and it kept going up for the rest of the century.

The mechanism is the whole story. Automation made cloth radically cheaper. Cheaper cloth meant ordinary people could suddenly afford more than one shirt, and the world's appetite for fabric turned out to be enormous. Demand exploded, mills multiplied, and they needed more weavers to tend the machines. And the skills the machines couldn't touch became more valuable, not less: a weaver who could coordinate work across multiple looms earned wages that rose sharply against everyone else's. The economist James Bessen documented all of this, and his conclusion fits in one sentence: when a job is partially automated, employment in it can grow.

Three thousand years before any of that, King Solomon famously recorded in Kohelet (Ecclesiastes) 1:9: "there is nothing new under the sun."

People always believe the development in front of them is unprecedented. It's happening right now: the certainty that AI breaks history, that no generation has ever watched its work change this fast. But what we saw with the Industrial Revolution is exactly what we're seeing with AI today. A new capability arrives that does the core task better than people ever could. Everyone splits into panic on one side and mania on the other. And underneath all the noise, the same quiet pattern plays out: costs fall, demand explodes, and the work regroups around a new constraint.

Here's what I've come to believe, watching it from the inside: AI isn't eliminating the need for human contribution. It's exposing where the next human bottleneck lives.

When one part of a workflow becomes dramatically faster, the constraint doesn't vanish. It moves. I see this every single week in my own work building AI systems for businesses. We automate a company's content production, and suddenly the bottleneck isn't writing anymore, it's judgment: deciding what's worth saying, what sounds like the brand, what ships and what dies. We automate the research, and the bottleneck moves to the person who has to act on it. We automate the reporting, and the constraint becomes the meeting where somebody actually makes the decision. The queue never disappears. It reforms somewhere new, usually somewhere more interesting.

And the companies expecting AI to shrink their teams keep discovering the opposite. The people freed from the automated step get pulled straight into the new constraint, because that's where the value AI just unlocked is sitting, uncaptured. The work didn't end. The work moved, and it brought friends.

Marc Andreessen put the entire two-hundred-year mechanism into one line this week:

Marc Andreessen 🇺🇸@pmarca

Technology increases productivity → cost of output falls → demand for output rises → more total output gets built → more jobs (and at higher wages).

Marc Andreessen 🇺🇸@pmarcaTotally predictable!10:02 PM · Jul 10, 2026

1.5KReplyCopy link

Read 89 replies

That arrow chain is the power loom's biography. It's also the ATM's. Cash machines were supposed to erase bank tellers, but branches got cheaper to run, so banks opened more of them. Teller employment grew for decades while the job transformed from counting cash into handling relationships. The machine took the mechanical part and handed humans the judgment part. Every generation gets its own version of this story, and every generation is sure that this time it's different.

I don't think it's different. I think the appetite is the constant. The world's demand for cloth wasn't fixed in 1830, and the world's demand for intelligence isn't fixed now. Make thinking cheaper and the world will consume unimaginably more of it. More analysis, more software, more content, more experiments, more decisions that used to be too expensive to make carefully. Somebody has to stand at the new bottleneck of all that expanded work. That somebody gets paid.

So I'd suggest retiring the question everyone keeps asking, which is "which jobs will AI eliminate?" The loom-smashers were asking a version of that one, and they were standing next to the wrong problem. The better question, for your company and for your own career, is "where does the constraint move next?" Find the step that becomes scarce when everything around it gets fast: the judgment, the taste, the accountability, the human being who decides. That's where the wages rise. It's the same lesson that had me betting on the rush instead of guessing where the gold lands. You don't have to predict the technology to position yourself around what it makes scarce.

None of this means the transition is painless. It wasn't painless in 1830 either, and pretending otherwise is how you lose people's trust. Individual weavers got hurt while weaving thrived. The honest version is that the trade grew while the tasks transformed. The people who did best moved toward the new constraint instead of guarding the old one.

Every generation convinces itself that its moment has no precedent, and the pattern outlasts every one of them. The sun's the same. The panic's the same. The mania's the same. The loom didn't end the weavers, the ATM didn't end the tellers, and the model won't end the workers. The bottleneck just shifts. It always has.

Your job is to be standing where it lands.