I currently spend $200/month on Claude Code Max plan, and I’m actively trying to spend more.
Most people aren’t even using AI, nevermind paying $200/m. A 2025 YouGov survey found that 56% of American adults used AI tools, but only 28% use them at least once per week.
Meanwhile there are people like doodlestein, former Wall St. analyst Jeffrey Emanuel, running twenty-two simultaneous Claude Code sessions spending $4,600 per month.
I completely understand the skepticism most people feel regarding the quality of AI-generated work, but let’s assume quality continues to improve exponentially over time.
The people not adopting AI are a little behind those using the free version of ChatGPT or Claude, and those maximizing their usage limits for $20/m will be nudging ahead.
People spending $200/m like me 20x the usage limits and access to the most capable model (Opus 4.6), as well as experimental features like agent swarms. Someone like Emanuel is operating at 440x the output of a standard paying user, and 2,200x what a free user gets.
Sam Altman plans to release a $2,000 a month agent to replace “high-income knowledge workers,” a $10,000 monthly fee for software development agents (productionizing doodlestein’s workflow), and a whopping $20,000 monthly fee for a PhD-level agent, capable of advanced academic work.
Now ask yourself this question: if you’re not at least maxing out a $20/m account, will you learn enough about using AI effectively to justify spending $200/m? If you’re not juggling multiple $200/m accounts, will you be productive enough to afford the $20,000/m plan when it comes out? If your competitors have access to the good AI that can do frontier science, how much further will they get ahead?
Welcome to the permanent underclass theory, a terrifying meme circulating in silicon valley tech circles with increasing intensity. The idea that if you’re not aggressively adopting AI right now, you might be priced out of it in the future. Is this for real? Is there a way out? Let’s discuss.
The Permanent Underclass
“everyone I know believes we have a few years max until the value of labor totally collapses and capital accretes to owners on a runaway loop - basically marx' worst nightmare/fantasy. this is the permanent underclass thing. and everyone I know subscribes to it” – Nic Carter, Partner at Castle Island Ventures
Throughout history, capital and labor have been imperfect substitutes. A factory owner needs workers. A tech CEO needs engineers. This imperfect substitutability is the source of all worker power. Every labor right we take for granted—unions, minimum wage, weekends—exists because capital owners need workers to be motivated, and therefore have to negotiate with them.
AI potentially changes this equation. If you can buy compute that does everything a human worker can do, workers lose their bargaining power entirely. Marx predicted exactly this kind of capital concentration, but he also predicted revolution as the resolution—because workers had leverage. AI removes that mechanism. Workers can't revolt effectively if they have nothing to withhold. You can’t go on strike if ChatGPT can break the picket line. That's what makes the underclass "permanent": not just relative poverty, but the loss of the mechanism to escape it.
Trammell and Patel formalized this in their paper "Capital in the 22nd Century": "Without [a global and highly progressive tax on capital], once AI renders capital a true substitute for labor, approximately everything will eventually belong to those who are wealthiest when the transition occurs, or their heirs." The argument builds on Piketty's famous finding that returns on capital tend to outpace economic growth, which concentrates wealth over time.
So are we all going to be doing gig work for minimum wage or sitting at home on UBI while anyone lucky enough to have Google, OpenAI, or Anthropic equity becomes the new landed gentry? Scott Alexander pointed out that this meme isn't being spread by poor people actually affected—it's preying on neurotic well-off people in Silicon Valley who fret about being merely bourgeois rather than future-oligarch well-off.
Fair point. But maybe these people see the ladder most clearly because they're already on the first or second rung. Their anxiety may be self-interested, but it’s real enough that we shouldn’t dismiss it.
Millionaire Hairdressers
Let me make the strongest case against the theory, because I’m an optimist and I think it's important not to spiral into a doom loop.
Previous technology shifts like the invention of the plow, printing press, steam engine, computer, or the internet didn’t eliminate jobs – even the poorest people in society today live better lives than kings and queens did 200 years ago. Why is that?
There are two economic forces that explain this, and they're both about to matter a lot. The first is Jevons' Paradox: when something gets cheaper, we don't just use the same amount for less money—we use vastly more of it. A transistor cost $1 in 1965. Today it costs a fraction of a millionth of a cent. We didn't pocket the savings. We put computers in thermostats, greeting cards, and disposable shipping tags. The same thing is happening with AI tokens right now—every time inference gets cheaper, someone figures out a new use for it, and demand explodes.
The second force is Baumol's Cost Disease, and it's the one the permanent underclass theory ignores. As Alex Danco recently explained at a16z: when a productivity boom creates tons of well-paying jobs in one sector, wages rise in every other sector too, because everyone competes in the same labor market. If you can make $150 an hour installing HVAC for data centers, you won't accept less for doing home AC repair. And if HVAC pays more, plumbers need raises too, or they'll switch trades. This is how wealth from productivity booms gets spread around—even to people who have nothing to do with the booming industry.
This is what happened with every previous technology wave, and there's no obvious reason it won't happen with AI. If AI makes everything it touches 1000x more efficient, the relative value of human-bottlenecked services goes up, not down. There are reports of electricians being paid $500k salaries to fill demand for new data center buildouts, and we can expect to see more stories like this as AI induces demand in bottlenecked industries.
Tradespeople, hairdressers, therapists, educators, dog walkers—anyone in a role that AI can't fill for regulatory, practical, or preference reasons becomes more expensive, and therefore better paid. The permanent underclass theory assumes human labor becomes worthless. Baumol's suggests the opposite: the things AI can't do become the premium goods.
AGI Trade Partner
But won’t AGI mean AI does everything? Not necessarily. LLMs may have fundamental limits that prevent true labor substitution. Current models are surprisingly poor at generalizing outside their training data. Here's the tell: no scientific breakthrough has come from ChatGPT, despite having access to all the world's published knowledge. If LLMs could extrapolate and discover genuinely new things, they would have by now.
In my "New Taylorism" piece I wrote about Claude getting stuck in Mt. Moon in Pokémon for over 48 hours. The solution required maybe seven moves—something I navigated when I was 12. That gap between strategy and tactics, that messy middle where you have to extrapolate from incomplete data, might require biological intelligence.
Even if a new architecture breakthrough or scaling technique gets us a true general AI agent, there’s the problem of computational irreducibility. There are no free lunches, and any AI agent that can generalize as well as a human might just cost as much as a human. Maybe more, given how efficient millions of years of evolution has made us. Once AI models cost $20,000 and beyond, they might be too expensive to operate on most tasks, reserved for important scientific endeavors or government projects leaving the smaller tasks to us mere mortals.
Ricardo’s law of Comparative Advantage states that even when AI workers are more efficient at producing every single good than human workers, we can both gain from trading with each other thanks to the opportunity cost AI will have when not tasked on their most valuable use cases. This is why the USA and China trades in goods with less economically advantaged countries – any time they spend not producing iPhones or equivalent high tech goods is wasted margin.
Secure Your Legacy
Even if the theory is overstated, the underlying dynamics are real enough to act on just in case there’s a probability it plays out. So what can you realistically do if you want to get ahead of the AI tsunami?
Own assets. If it gets harder to convert labor into capital—which is exactly what the theory predicts—then the amount of capital you have when the transition happens becomes the thing that matters. Own your home if you can. Own shares in companies benefiting from AI, like buying stocks in the MAG7 or working for an AI startup. Start your own AI startup if nobody will hire you, and earn the freedom to experiment.
Actually use the tools. The biggest mistake I see is people who tried ChatGPT six months or a year ago, it didn't do what they wanted, and they wrote the whole thing off. This is like hiring a new employee, giving them one task on day one, watching them get it wrong, and firing them on the spot. Except this employee is getting exponentially smarter. These models are moving so fast that things they couldn't do three months ago, they can do now. Don't try it once, give up, and dismiss it as all hype.
Maximize humanity. When AI can produce any digital artifact—writing, code, images, analysis—the things it can't produce become the premium goods. Physical presence. Human connection. The energy in a room during a brainstorm. Ari Emanuel launched a $2 billion events company in late 2025, explicitly betting on this thesis: as AI handles more of the work, demand for live human experiences goes up. He knows his limits on AI, so he’s leveraging his talents to tap an industry that benefits if AI goes bonkers.
The honest summary: own things, use the tools, and be more human. The first two help you climb the ladder. The third helps you build a life that doesn't depend on it.
