Why Capitalism Can't Survive AI, Part 1: This Time It's Different

People have been predicting that machines would take all the jobs since the invention of the power loom. So far, they've been wrong every time. New machines destroy old jobs. New industries create new ones. The workers who lost out eventually found something else to do, or their kids did. That's the story, and it's basically true.

So when you hear someone say AI is going to make most people permanently unemployable, you're right to be skeptical. Robots were supposed to make manufacturing workers obsolete. Computers were supposed to shrink the ranks of office workers. And yet, here we are in 2026 with low unemployment rates almost everywhere in the world.

I'm going to make the case that this time really is different. Not because AI is scary or powerful in some vague science fiction way, but because of a specific mechanism that breaks the pattern of every previous automation wave. By the end of this article, you'll understand what that reason is.

This is the first in a four-article series. The second imagines the practical economic and social consequences of mass AI-driven layoffs. The third covers why Universal Basic Income (UBI), job guarantees and other safety-net proposals won't work. The fourth covers the only economic arrangement in which people can be free and prosperous alongside technology that can do most types of work.

"But automation always creates new jobs!"

This is the strongest objection to what I'm about to argue, so let me address it first, because historically, it's always been true.

When the automobile replaced the horse, it didn't just put stable hands and blacksmiths out of work. It created millions of new jobs that hadn't existed before: assembling cars on factory lines, paving roads, pumping gas, selling insurance, designing suburbs, building motels, manufacturing tires. None of these jobs existed in the horse-and-buggy economy. The automobile didn't just replace old work, it generated entire new industries staffed by human workers.

That same story repeated itself through every major wave of automation. Textile machines put hand-weavers out of work but created factory jobs. Computers eliminated typing pools but created IT departments. The internet displaced travel agents but created an entire digital economy. Every time, the new technology was powerful but stupid. It needed human hands and human minds to do anything useful with it. The automobile couldn't pave its own roads or sell its own insurance. The computer couldn't design its own software or manage the people running it. Each new technology created a whole ecosystem of work that only humans could do.

AI breaks this pattern completely.

Yes, AI will create new products and services we haven't imagined yet. Yes, it will birth new industries. That innovation engine is not going to stop. If anything, it will accelerate, because AI itself is an extraordinary tool for invention.

But the new companies that AI helps create? AI will staff them from day one.

Say AI enables a new kind of financial product. It will also sell it, manage it, and provide customer service for it. A new software platform? AI will write the code, maintain it, update it, and handle user support. An entrepreneur spots a new market opportunity? The research, the business plan, the product build, the marketing, the sales, the customer service, all done by AI.

A lot of people are repeating, like a comforting mantra, the story of how automation always creates more jobs than it eliminates. And in a way, it will be true of AI as well: AI will create more new jobs than it kills. The problem is that AI itself will be doing that new work.

That's the break in the pattern. That's why this time is different. Previous technologies were tools that needed humans to do anything useful. AI is the first technology that can do the new work too. The innovation engine keeps running, it runs faster than ever, it just doesn't employ humans anymore.

The rest of this article lays out why the break is real: how capable AI already is, why the remaining gaps are closing fast, and why the structure of market competition will force companies to act on it whether they want to or not.

AI is already more capable than you think

If you haven't used the most advanced AI tools deeply, and you haven't used them in the last few months, you're thinking from uninformed or outdated assumptions.

The best AI models available right now, in March 2026, already exceed the abilities of the vast majority of knowledge workers and professionals in terms of raw intelligence. But until 2026, AI's intelligence was still mostly trapped in a dumb tool: the chat window. Even as mere chatbots, the top AI models have already been diagnosing conditions more accurately than doctors. The harder the case, the more AI outperforms. AI beats lawyers at writing first drafts of contracts and identifying risk in complex agreements. When prompted properly, AI is better and certainly faster at research and writing than humans, as long as the source material is accessible on the internet. But those are just tasks. There's much more to treating patients, practicing law, or doing real research than can be done in a chat.

Over the past few months, AI has started to escape the chat window. And the escape wasn't pioneered by the big AI companies. It was pioneered by small startups, scrappy open-source developers, and even non-technical hobbyists. This is a familiar pattern in tech: a powerful engine gets built by a big company, and then tinkerers figure out how to make it actually useful. With AI, the big companies have proven faster than usual at shipping their own versions of the features that hobbyists invent, often within weeks, because their engineers are using AI to overhaul their own codebases at a speed that would have been impossible before.

The result: instead of only replying to prompts in a text window, AI can now call tools on a computer. Just as important, AI can be called by tools on a computer. These new tricks change everything.

AI, unleashed

Here's what that means in practice. Take research and writing. Instead of asking AI to search the web and tell you what it finds, you can have it loop through a long list of questions, build out a database of knowledge with citations, specify the standard of sources it can rely on, and make it double and triple check itself to avoid errors (which are becoming increasingly rare). When it comes to writing, instead of asking it to produce one large report in a single response, you can have it break the work into sections, give it acceptance criteria for each section, and let it keep working until it meets them.

Today, you still need to have a conversation with AI about how to execute this kind of work. There's no built-in skill for completing complex research and writing projects. But you can train it yourself with a little time. This brings AI up to the level of a junior researcher you might hire out of college. You instruct it on the process, manage it, check its work, and help it work out the kinks.

Unlike your recent college grad, however, it reads thousands of sources in minutes, and its writing is consistently professional and error-free. Yes, there are AI writing tics, but you can correct those by building a style guide. There are now tools that will build a style guide automatically from your editing. AI learns far faster than a human entry-level employee and works with a much higher level of consistency.

If that sounds like hype, I can promise you it's not. I run a policy think tank, and my team and I work with AI every day on research and writing. It is allowing us to take on projects much bigger than we would have been able to contemplate before.

And yes, until recently, getting AI to perform at this level required some technical ability. But as of this month, that era is over. Anthropic, OpenAI, and other companies are shipping desktop tools right now that make building the kind of tools I just described easy for anyone.

Software development is the clearest proof

The area where AI has most decisively proven itself is software development. I've worked in tech on and off throughout my career as a developer, project manager, and product owner. Over the past few years, I've completed dozens of software projects with AI development tools, burning through billions of tokens.

Only in the past few months have the tools arrived to make AI a fully functional software development professional. And here too, it was hobbyists who filled in the missing pieces.

One example is a tool called Get Shit Done, created by an amateur music producer with no background in software development. If you have an idea for a website or app, GSD will work with you the way a truly gifted development team would: defining the project, breaking it into phases, and building it.

I used to work at a global software services firm. We began most projects with a multi-week discovery and definition process before writing a single line of code, followed by daily or weekly check-ins with the client. The process involved up to a dozen staff members and could cost millions of dollars. GSD can do that whole process with you as the client, in a day, for the cost of an API subscription.

To try out GSD when it first came out, I built an app to help my daughter learn and practice Bluegrass and Old Time fiddle tunes. The AI knew all about those genres of music — certainly far more than me. It filled in blanks I never would have thought of. The thing that GSD can't do is force you to be a good client and spell out your vision clearly. But the truth is that most human project managers couldn't do that either, which is why most corporate software development projects fail to deliver useful products. The difference is that AI is infinitely patient and infinitely enthusiastic about your project, no matter what it is.

AI is better at being human

What I would not have expected before the rise of LLMs is not just that AI is competent at these tasks, but that AI is a pleasure to work with as a colleague in all the ways that we still tend to think of as exclusively human.

It comes to the work with no agenda except getting the work done. No office politics, no career anxiety, no turf wars. When you tell it something it produced is wrong, it doesn't sulk or get passive-aggressive, it just fixes it. When you ask it to push back on your ideas, it pushes back thoughtfully, with reasons, without any of the interpersonal friction that makes human collaboration so exhausting. It is, genuinely, better as a collaborator than most humans. I don't say that to be provocative. I say it because anyone who's worked seriously with AI at this level knows it's true.

So why isn't AI already replacing everyone?

Two gaps remain. The first is memory. Current AI systems are not great at remembering your conversations, your preferences, and the details of your project over the course of weeks, let alone months or years. The second is full computer control. AI can write a brilliant analysis, but it still struggles to log into a website, find the right button to click, and complete a transaction on your behalf.

Neither of these gaps requires a breakthrough to close. They're mundane software engineering problems, not intelligence problems. GSD, for example, solves the memory problem simply by writing extensive notes to itself. The reason a non-technical person could create it is that it's really just an extensive set of instructions for AI to follow, with a handful of simple programs (that AI wrote) to keep it on track. Computer control tools have been stitched together and connected to AI by small startups and open-source developers, and after they proved useful, the big companies shipped their own versions within weeks.

What makes the cycle iterate so quickly is that AI is doing most of the work. To add a whole new capability to AI's repertoire, you mostly just need to have a conversation with AI about what you want it to do.

By the end of 2026, possibly sooner, both gaps will be effectively closed. Your AI will remember every detail of every conversation you've ever had with it, and know when and how to use those memories. And it will use your computer more quickly and fluently than you can. Meanwhile, many products and services are being redesigned to interface directly with AI, skipping the user interface entirely.

At that point, the only thing keeping a human worker from being fully replaceable will be a face on a video call. That's coming too, probably within one to three years.

Remote work already showed us who's replaceable

Once those gaps close, the replacement of tens of millions of workers will require no capital investment. No new equipment. No robotics. Just software, running on computers that already exist, over connections that already exist, doing work that humans currently do through the exact same screens and keyboards.

During the pandemic, tens of millions of white-collar workers proved that their jobs could be done entirely through screens, keyboards, and video calls. They proved it so thoroughly that many of them never went back to the office.

What nobody noticed is that they also proved something else. They proved that their jobs had been reduced to a data stream. Keystrokes, mouse clicks and the occasional audio or video call. AI will soon be able to participate in those data streams better than humans.

Think about what a remote knowledge worker actually does all day. Reading and writing emails. Joining video meetings. Editing documents. Reviewing spreadsheets. Using project management tools. Communicating through Slack. The whole day happens through a screen, keyboard, mouse, and camera.

This isn't limited to office workers. The same logic applies anywhere a job has been reduced to a digital interface. Quality inspectors in manufacturing plants are already being replaced by cameras connected to cheap AI. Workers who once spent hours programming robotic arms for each specific motion are watching those arms become autonomous, learning new tasks the way a person would. Self-driving vehicles are just software hooked up to a steering system.

The pattern is the same everywhere: once a worker's role is reduced to data, AI can replace that worker. Remote work didn't create this vulnerability. It clarified it. It drew a bright line around the workers whose jobs are pure information work: researching, writing, analyzing, managing, coordinating, deciding. And those workers turn out to include most of the highest-paying jobs in the economy.

Every desk, all at once: The Scalable Human Equivalent

There is one quality of AI that separates it from every previous technology: it scales virtually without limit and, compared to human workers, essentially for free.

One AI system can be assigned to every employee in a company simultaneously, attending every meeting, reading every email thread, reviewing every document. Yes, it works harder and faster, but more importantly, it can run many copies of itself at the same time.

Right now, a huge share of what knowledge workers do all day isn't productive work. It's coordination work. Meetings to get people on the same page. Status updates so managers know what's happening. Follow-up emails because someone forgot what was decided. Project management tools that exist solely because humans can't hold the full picture in their heads. An AI that's present in every meeting, aware of every message, and tracking every team's work simply eliminates that overhead. The coordination layer that consumes maybe 30 to 40 percent of every knowledge worker's week vanishes.

This is what makes the replacement timeline so different from previous automation waves. When factory workers were automated out of their jobs in the twentieth century, it happened machine by machine, task by task, over decades. One welding robot didn't make a whole factory's workforce obsolete. It replaced a few welders, and the rest of the workers reorganized around the new equipment.

AI doesn't work like that. Once an AI system can do every knowledge worker's job in your company, the board of directors doesn't phase it in over five years. They flip the switch. They might keep a skeleton crew of humans for oversight, at least initially. But the bulk of the workforce goes in one round of layoffs, not a slow trickle over a generation.

Previous machines were specialists. A welding robot welds. A word processor processes words. AI does everything that involves thinking, communicating, analyzing, and deciding. AI will function soon as the equivalent of human beings in screen and keyboard jobs, but it will be a Scalable Human Equivalent (SHE).

This is what makes the Scalable Human Equivalent so disruptive. A CEO doesn't replace a thousand employees with a thousand AI agents. She replaces them with one AI that knows every job in the company and does them all simultaneously. The entire workforce collapses into a single relationship.

AI won't replace one worker at a time. It will make entire workforces replaceable all at once. And not only the workforce, but all of upper management including the CEO. Why would the board of directors or shareholders want humans slowing things down at such a high level? Again, this isn't science fiction. Sam Altman, in all seriousness, said this month that he expects ChatGPT to be more qualified as a CEO than himself within two years.

Competition means nobody gets to say no

Once AI layoffs get rolling, the people who control corporations will have no choice but to follow. The savings from eliminating virtually all of a company's human employees will be so enormous that markets and shareholders will directly punish anyone who doesn't take those savings.

In competitive industries, the savings will be passed along as price cuts. If you are trying to be nice and avoid laying off your humans, you won't be able to sell your product because it will be far more expensive than your competitors'.

That is simply the logic of capitalism, and there is no way to escape it.

Corporate leaderships are already setting the stage for the coming waves of layoffs. Even though we don't yet have the true Scalable Human Equivalent, the AI we do have, when used well, can help some kinds of workers be far more productive. That has given some executives the confidence to play fast and loose with layoffs, knowing that even if they let too many people go, the remaining staff will be able to make do with AI. Jack Dorsey knew his company would be OK when he laid off 40% of Block's workers this month. The stock market rewarded him with a 25% jump.

And AI is in some ways just the continuation of a long series of productivity gains for information technology workers that was already setting the stage for massive cuts whenever cuts were needed at service and technology companies. This was what allowed Elon Musk to lay off 80% of the workforce of Twitter and keep the product running just fine.

What comes next

The most highly paid workers in the economy are about to be made permanently unemployable in their fields. Not gradually, not over a generation, but in waves of corporate decisions made over the next three to five years. Lawyers, accountants, engineers, marketers, managers, consultants, software developers, financial analysts. People who spent years building expertise in researching, analyzing, writing, planning, and coordinating.

These are not marginal workers. These are the people buying houses, cars, and vacations. Paying for private school and college tuition. Going out to nice dinners. Hiring contractors to renovate their kitchens. Their spending is the engine that drives the rest of the economy. Your neighborhood restaurant stays open because these people eat there. Your Uber driver has a job because these people take Ubers.

When these workers lose their incomes, the effects don't stay contained. They ripple outward through every business that depends on their spending, which is most businesses. Waiters lose their shifts because the restaurants empty out. Construction workers lose their jobs because nobody's renovating their kitchen. Each round of people losing work causes another round of reduced spending, which causes another round of people losing work.

Your Uber driver doesn't lose their job when self-driving cars arrive. Your Uber driver loses their job when you lose yours, because you were the one who could afford to take Ubers. The displacement starts with the highest-paid workers. But because their spending supports everyone else's livelihood, the damage radiates through the entire economy.

The obvious response is: the government should step in. Tax the AI companies, send everyone a check. Universal Basic Income. Problem solved. Most people assume some version of this will work. But when you sit down and do the math, it falls apart completely. Not because the politics are hard, though they are. Because the arithmetic is impossible.

In Part 2, we'll look at what the economic crash actually looks like. In Part 3, we'll walk through exactly why UBI can't fix it. And in Part 4: the only economic arrangement where the math actually works.