Losing the Bottleneck: Why AI Threatens More Than Our Jobs
The advice everyone gives right now is simple: if you want to keep your job, use more AI. It’s hard to argue with. The colleague who uses it looks faster and cheaper than the one who doesn’t.
But follow the logic one step further, and the advice falls apart. The strategy that protects you individually speeds up the outcome you’re trying to avoid. Everyone stands up at the concert to see better, and nobody ends up with a better view.
AI is not something you just use
This is the point most of the debate misses. We talk about AI as a tool, like a hammer, a spreadsheet or a search engine. It isn’t one, and the difference is everything.
A tool extends you. AI substitutes for you. A hammer makes your arm stronger, but the hammer still needs the arm. AI doesn’t just make you faster at the task. It can do the task. Every improvement in a tool used to make the worker more valuable. Every improvement in AI makes the worker less necessary.
A tool doesn’t learn from you. AI does. A hammer doesn’t get better at carpentry by watching you swing it. With AI, every prompt, every correction and every “no, do it like this” turns your know-how into something the system can repeat. Using it is a form of training. You aren’t just doing your job with it. You’re teaching it your job.
A tool belongs to you. AI belongs to someone else. When you buy a hammer, what it produces is yours. When you use AI, you’re working through a system owned by a company, usually far away. The skill you transfer doesn’t stay with you. It ends up with the owner, who can sell it to your employer without you.
A tool waits for instructions. AI makes decisions. A hammer never decides what to build. AI breaks goals into steps, chooses between options and proposes what you should want. The more you rely on it, the more of the deciding it does.
A tool can be put down. AI becomes infrastructure. You can stop using a hammer tomorrow. Once a workplace, and then a whole society, runs on AI, nobody can step away from it, and that includes the people who are supposed to regulate it.
So “just use AI” isn’t neutral advice. Using it means taking part in your own replacement and handing over the knowledge that made you needed to someone who doesn’t need you. That doesn’t mean refusing it. Refusing it alone only gets you replaced sooner. It means seeing clearly what is happening when you use it.
“Judgment” is just the next prompt
The usual comfort is that AI handles the routine and humans keep the judgment. I don’t buy it. Judgment is often just the next rung on an abstraction ladder. First AI writes the email. Then it decides which emails to write. Then it decides the communication strategy. Every time we step up to “the part AI can’t do,” that level gets captured too. Eventually the instruction is just: do it.
Even goals aren’t safely human. Ask an AI to pick targets for your business and it will give you options. If people increasingly choose from menus written by AI, the “human decides” step stays on paper while the real shaping of goals moves into the system.
What is clearly different is that the AI loses nothing when the wrong goal is picked. Someone else takes the downside. Arrangements where one side collects the upside and the other carries the risk have never been stable for long.
Replacement is fine. Losing my share is not.
I can live with being replaced. Plenty of work disappearing would be good news. The problem is what goes with the job: income, pension, insurance, a place in society.
When AI makes my work unnecessary, the freed resource is me: my time and my labor. On the company’s books that’s called efficiency. In my life it’s called unemployment. It’s the same event, labeled by whose ledger you look at. Freed time could be leisure, but only if part of the gain followed it. Today the mechanism that returns part of the gain to the worker is the wage, and the wage is exactly what disappears.
An economy that shrinks while getting more efficient
Now scale it up. All companies automate at the same time, so none of them loses ground to competitors. Production costs fall across the board. But unemployed people don’t buy much, so demand falls too. Companies become more efficient, and selling becomes harder.
The economy’s capacity grows. The realized economy, what is actually produced and sold, shrinks. Abundance that can’t find buyers.
There is no robot company 2.0
Earlier technologies displaced workers too, but the labor had somewhere to go: from farms to factories, from factories to services. The new sector absorbed the people the old one released.
With AI that loop breaks. Imagine a robot that does everything. Start a new company and it’s just another set of robots. Robot company 2.0 doesn’t hire you either, and starting it takes capital, which is exactly what the displaced person doesn’t have. What’s left is work where being human is the product itself: care, craft, performance. It’s real, but small, and it still needs customers with incomes.
The exits that don’t work
“Politics will fix it.” The EU began pushing for a common phone charger around 2009. The law took effect at the end of 2024. Fifteen years for a cable. AI changes in months. Politics can move fast, but mostly in a crisis, which means the correction comes after the damage, and the first people displaced pay for the wait.
“Just unplug it.” You can’t switch off something everything depends on: hospitals, payments, logistics, and the politicians’ own offices. The leverage has to be on the owners, not the machine. That only works while the state is still stronger than the companies it depends on.
“Own a piece of it.” Right now a handful of US companies control frontier AI. For someone in Sweden, the gains land in American tax bases and with American shareholders, while the displacement happens here. And even owning AI shares may not capture the gains. Providers compete prices down and pay much of their revenue to chipmakers, data centers and energy producers. The savings mostly go to the companies that deploy AI. A broad index doesn’t save you either, if the whole economy’s demand is shrinking.
“Things will just get cheaper.” Maybe. But central banks treat falling prices as a danger and respond with easier money. New money flows first through financial markets and lifts asset prices. Once again the gain goes to people who already own things.
Every exit runs into the same wall: any mechanism that goes through markets or finance ends up rewarding ownership.
The bottleneck
Here is what I think it comes down to. Value always gathers where something is scarce. For most of modern history, human labor was one of those bottlenecks. Nothing got made without people, and that scarcity was the source of workers’ bargaining power. Unions, wages, the welfare state were all built on it.
AI removes labor as a bottleneck. But bottlenecks don’t disappear. They move to what AI can’t make abundant: energy, land, chips, minerals, data center capacity, housing in the right places, and access to the models themselves. Whoever controls those captures the value. Anyone with only labor to offer stands on the wrong side of every chokepoint.
That’s also where the conflicts will come. When AI makes everything else cheap, the pressure piles up on the few things that stay scarce. And workers enter those fights having just lost the one source of leverage they used to have.
So yes, losing the job matters. But losing the scarcity that gave us a say matters more. The real question isn’t how to stay employable for a few more years. It’s what gives ordinary people a claim on the economy once their labor no longer does, and whether we answer it while we still have the leverage to.