For about 200 years the advice never changed.
A machine takes your job, so you climb to harder work. The loom takes the weaving, you run the factory. The spreadsheet takes the arithmetic, you do the strategy. Every time, there was a higher rung to climb to.
This is the first time that might not be true.
The ladder we always climbed
Here’s the pattern, every time we automated something.
We started at the bottom. Muscle. Lifting, hauling, weaving by hand. Then machines took the muscle, and we climbed a rung. We did routine work instead. Adding up columns, filing, the same task over and over.
Then computers took a lot of the routine. So we climbed again, to the top rung. Judgment. Deciding what to do. Reading a room. Taste. The work that needed a human brain, not just a human back.
Reid Hoffman has a good name for what’s happening now. He calls it the cognitive industrial revolution. The steam engine, he says, gave us physical superpowers. This is the same thing, but for the mind. For anything we do with language: reasoning, analysis, selling, marketing, support.
That’s a useful frame. The steam engine was for muscle. This is for thinking.
And it points at exactly one thing. The top rung.
The part nobody likes to say
Every machine before this one pushed us up the ladder. This one is climbing it with us.
It’s reaching for the judgment rung. The one we always backed up to when the machines took everything below it. There’s nowhere above it we’ve ever needed to go.
Yuval Noah Harari put his finger on why. He wrote that language is the stuff almost all human culture is made of. Our stories, our laws, our deals, our religions, all of it runs on language. And AI just got very, very good at language. He asked a question that stuck with me: what happens to history when AI starts producing the stories, the melodies, the laws?
For a founder, that question isn’t abstract. I’ll come back to why in a minute.
Here’s the second thing people skip. Even when the machine age works out, it can be brutal first.
We tell the industrial revolution like a happy ending. Machines came, life got better, the end. That leaves out the middle. The economist Robert Allen documented it: in Britain, roughly from 1780 to 1840, output per worker shot up about 46 percent. Working-class wages? They barely moved, up around 12 percent over the same stretch. Profits doubled. The gains went to the owners. Workers only really started catching up after 1840.
That’s about 50 years. A whole generation lived and worked and died inside the gap, before the payoff arrived. So “it worked out last time” is survivorship talk. It worked out, eventually, for the people who survived the wait.
Now the genuinely scary part, and I want to be honest that it’s an open question, not my prediction.
There may be no higher rung left.
Every prior wave, we climbed up. Farm to factory to office. This is the first technology aimed straight at the top. Serious people think there may be nowhere above it to go. Erik Brynjolfsson and Andrew McAfee raised it years ago in The Second Machine Age. Daniel Susskind built a whole book around it, A World Without Work. I’m not saying they’re right. Nobody honest knows. The two people most likely to build this thing can’t even agree, within a factor of five, how fast that top rung disappears, and they admit they haven’t found the edges of what today’s models can already do.
So we’re left with a real question, not a tidy answer. And I’d rather sit in the honest version of it than pretend I know how it ends.
Why this is the wrong argument
Almost everyone writing about AI and your job is having the wrong argument.
One camp says it’s the end of work. The other says relax, just learn to use AI, new jobs always show up. Both are arguing about the same thing: whether there’s a higher rung. Will a new kind of work appear above this one or not.
That’s the wrong question for an operator. Here’s the one I think actually matters.
Stop trying to find a higher rung to stand on. Own the ladder.
Decide what gets built. Own who already trusts you. Own the outcome. The few things words can’t fake. That’s the move, and it’s the same move whether the doomers or the cheerleaders turn out to be right.
I’ll be fair to the optimists, because their case is strong. There’s a real economic idea behind it called Jevons paradox. Make something cheap and people use way more of it, not less. Make thinking cheap and maybe we just want a lot more thinking. New work appears. That’s genuinely possible.
But two things have to be true for “it works out” to mean “it works out for you.” A higher rung has to exist. And you have to survive the gap before the new work shows up. Being right about 2075 doesn’t pay your rent in 2027.
So the macro argument doesn’t change your move. It just removes your excuse to wait for it to settle.
A startup is a machine made of words
This is the part that made it click for me.
A startup is a machine made of words.
Look at your actual week. The pitch deck is words. The cold email is words. The landing page, the strategy memo, the job spec, the standup update, the product spec, the support reply. Words. More and more, even the code is words.
When Harari says language is the stuff culture is made of, that’s not philosophy for a founder. That’s a parts list. It’s most of what’s inside your company.
And the words are going free. Copy, decks, content, memos, a lot of code. The stuff that used to take a person a day now takes a model a minute. So the value isn’t in the parts anymore. The parts are cheap.
The value is in the few things that aren’t words.
What to build. That’s taste, and a model has none of its own. Who already trusts you. That’s distribution and relationships, and you built those with real time. Who’s accountable when it’s wrong. That’s judgment and skin in the game, and AI has no skin in anything. Plus what’s actually yours: your customers, your data, the stuff you’ve shipped.
Harari has a darker note here, and I think he’s right. AI can now mass-produce intimacy. It can sound warm and personal to millions of people at once. So even being persuasive gets cheap. What gets rare is being a real human someone actually wants to deal with.
That’s the diagnosis. Now the move.
How a founder actually moves
Four things. In order. Boring on purpose, because the boring version is the one that works.
1. Be the operator, not the customer
Jensen Huang said the line that everyone quotes now: most people won’t lose their job to AI, they’ll lose it to someone who uses AI. He’s right. But I think it’s sharper than that, and I learned it the slow way.
For a while I thought I was AI-native because I paid for the tools. I had the ChatGPT subscription. I had the Claude tab open. I felt ahead. I wasn’t. I was a customer. I was renting someone else’s leverage and feeling clever about it.
The shift was when I stopped subscribing and started building. I built Kirro, an A/B testing tool. Then I co-founded Olavi, which helps teams see how they show up inside AI search. That’s when it changed. Not using AI. Wielding it at the system level. Building it into how the work runs, instead of just typing into a box.
I wrote up the full version of that mistake separately, the year I treated AI like a vending machine before I built it into a system instead of just prompting it. That post is the worked example. This is the principle.
Being the customer is table stakes now. It’s the floor, not the move. The move is to run AI as a system you own.
2. Build the moat AI can’t copy
Once everyone has the same models, the model isn’t your edge. Aaron Levie put it well: when your competitor has the exact same AI you do, the win comes from your own data, your own knowledge, the workflows only you have.
So build on the things a competitor can’t download. Your taste. Your audience. The trust you’ve earned. The data only you sit on.
Distribution is the big one, and it’s why I bet so hard on owning a channel instead of renting one. I made that whole bet in the open: a channel you own that compounds instead of ads you rent. An audience that already trusts you is a moat AI can’t copy, because it took real time to build and a model can’t fake the relationship.
3. Stay small and sharp
This is the fun part, and it still kind of blows my mind.
Garry Tan at Y Combinator put real numbers on it. Companies with 10 or 20 people are getting to 10 or 20 million dollars a year in 10 or 20 months. That literally never happened in software before. Sam Altman expects billion-dollar companies run by two or three people.
The pattern: let AI be the headcount. You be the judgment. Don’t grow the team to grow the output. Point the cheap intelligence at the work, and keep the small group of humans on the decisions only humans should make.
So you don’t need a big team anymore. You need a small one that owns the few things AI can’t.
If you’d rather have this built into how your team works than figure it out solo at 2am, that’s the work I do. You can see how here.
4. Move now
We’re in the cognitive version of that 50-year gap. Engels’ Pause, but for the mind. The edge is real right now because most people are still standing on a rung, waiting to see what happens.
Going first compounds while going first is still rare. I’d bet most people are using about 5 percent of what AI can already do. The 5 percent gap is the opening. It closes a little every month.
So move now. Not because you’ve got it all figured out. Nobody does. Move because the wait has a cost, and the cost is the head start.
What to do this week
Here’s the reframe I want you to leave with.
Stop asking “how do I use AI to do my work faster.” Ask “which parts of my work were only ever words, and what’s left when those go free.”
And here’s something to actually do this week. Write down everything you did at work last week. The real list. Every task.
Now cross out every line that’s just words a model could now produce. The draft. The summary. The first version of the deck. Cross them all out.
What’s left is your actual job.
If the page is nearly blank, that’s not a failure. That’s the wake-up call, while you’ve still got time to do something about it. If a lot is left, that’s your moat. Protect it, and point all that cheap intelligence at the rest.
Either way, you’ve stopped guessing whether there’s a higher rung. You’ve found the part of the ladder that’s yours.
The future isn’t decided. It’s authored. So pick the rung you stand on, or better, own the ladder.