Note: This article is somewhat speculative, and I hope you have the patience to read it all.

Author: Xiao Bing

On July 26, Sam Altman stood on the stage of the JPMorgan Center in San Francisco and said, “The idea that if you don’t enter the frontend labs, you will fall into the permanent bottom is absurd.”

“Permanent bottom,” this trendy term has entered the spotlight once again.

It gradually became popular in the tech circle X in the summer of 2025, with theoretical foundations from “The Curse of Intelligence,” as well as the paper “Capital in the Twenty-Second Century” by Trammell and Patel.

This statement is not merely predicting that “AI will cause many people to lose their jobs”; it describes a more extreme future: when AI can perform the vast majority of cognitive and physical labor, companies will no longer need to hire so many people to create wealth. The importance of wages in the economy will decline, while profits will increasingly flow to those who control the models, computing power, energy, and data centers.

It’s not just a group of people temporarily becoming poor, but rather that their labor, which they relied upon to move up, loses its value; it’s not just the rich getting richer than before, but that those without capital cannot become capital owners through work, which is what is referred to as the “permanent bottom.”

In the absence of large-scale wealth redistribution, public ownership, or global progressive capital taxes, this theory ultimately derives a disturbing conclusion:

Almost all assets may gradually flow to that group of people who were already the richest when the AI transformation occurred.

Of course, this is still just a deduction, but it reminded me of a small matter.

Over the past year, writer Jasmine Sun conducted a nearly awkward field survey. She met with researchers from leading AI labs one by one, each time talking for an hour, without recording or keeping notes, probing like an anthropologist into their genuine views on the future of AI.

There was one question she asked almost every time:

Assuming you have an average 17-year-old American kid standing in front of you. He’s not a genius programmer, has a B average, and usually doesn’t care much about AI. What would you suggest he do to prepare for the future?

Almost no one could provide an answer.

These researchers had varied political views, with assessments of AI ranging from optimistic to pessimistic, yet their answers were surprisingly consistent:

I don’t know. The current situation is alarming; the jobs left for him are probably few, and he happens to be stuck in the most painful transition period...

But the real trouble that the 17-year-old kid faces may not be whether he will become poor.

Permanent is not poverty

Every era has poor people, and there are always lower classes, but the “lower class” of the past is not necessarily a permanent identity.

A person can sell labor for wages, then convert those wages into education, housing, and assets; workers can also organize, strike, and collectively bargain, demanding capital to convert a portion of productivity growth into higher wages and better benefits.

This mechanism is not fair, but at least it retains a pathway upward.

The truly piercing aspect of the “permanent bottom” is that it believes this pathway may disappear.

Looking back at labor history, unions, minimum wages, and weekends were all established on the same fact: Capital and labor cannot completely replace each other.

Factory owners need workers, tech companies need engineers, so both sides must sit down and negotiate.

Marx once predicted the high concentration of capital, and he also provided a solution. Because in his world, workers still held something that capital needs but cannot create out of thin air: labor.

Strikes are effective because workers can temporarily withhold it. If workers don’t enter the factories, machines cannot operate; if engineers don’t write code, products cannot go live.

But if computing power can buy all labor a person can provide, the things that workers can withhold will approach zero.

At that time, the issue is not just some people’s wages declining, but capital will no longer need to negotiate with the majority. Workers will find it difficult to accumulate assets through wages and hard to demand a redistribution of gains by stopping work.

What makes the bottom “permanent” is that the machinery that once allowed the poor to return to the middle class, the labor, bargaining power, and asset accumulation, have been dismantled.

Evidence that this machinery is loosening has moved from theory to payrolls.

The Stanford Digital Economy Lab, using microdata covering 4.6 million laborers from ADP, has found that after the popularization of generative AI, in the occupations most exposed to AI, the employment of workers aged 22 to 25 has seen about a 16% relative decline, the unemployment rate for recent graduates in the U.S. has reached 5.6%, an increase of 1.6 percentage points compared to three years ago; the recruitment scale for recent graduates at large tech companies has dropped by 25% over two years.

The elevator is still operating normally; it’s just that the button for the first floor has been removed.

Those who can’t answer are buying themselves insurance

Back to Sun’s interviews.

Those researchers who could not answer “What should the 17-year-old kid do” did not stop their work because of that.

Sun continued to press: Since you think the future is so dangerous, why continue to build it?

Answers roughly fall into three categories.

The first group sincerely believes that as long as they endure the transition period, AI can ultimately cure diseases and eliminate scarcity.

The second group believes in technological determinism: even if they don’t do it, others will.

The third group is the most honest and also the most uncomfortable:

If great upheavals are really coming, at least secure a position for yourself in the future.

This impulse to “secure a position” is changing the talent flow in the entire AI industry.

A friend of Sun’s, pursuing a PhD in AI at Berkeley, said that about half of the cohort decided to graduate early, with some even choosing their dissertation topics based on “which research direction is most likely to get an AI lab offer.”

Many independent writers and policy researchers around Sun have also abandoned their previous positions to join AI labs.

These statements may sound like a joke, but they genuinely shape a generation’s life choices, continuously drawing talent away from independent research and public policy ecosystems.

And those being drawn away are precisely the group most likely to build the negotiating table.

The factory automation of the 20th century did not universally evolve into intense conflict, one important reason being that before machines entered the factories, companies usually had to negotiate first with unions: automation had to be explained as a safety upgrade, and wage increases needed to be linked to productivity gains.

There is a negotiating table between capital and labor.

Today’s white-collar workers do not have that table.

More subtly, those most capable of building negotiating tables—scholars, independent researchers, and policy talents—are being lured into labs by the narrative that “labor is about to lose its value.” Because only by entering labs can they possibly acquire equity, standing on the side of capital in advance.

A closed loop is thus formed:

The more people believe that the bargaining power of labor is about to disappear, the more people will give up building that bargaining power and instead compete for equity; and the fewer builders there are, the faster the bargaining power of labor will disappear.

No one is absolutely safe

That 17-year-old kid is not entirely without advice.

In the face of the uncertainties brought about by AI, people have provided many answers, resulting in two direct reactions.

One is to learn AI as quickly as possible, striving to become the last one to be replaced; the other is to acquire AI assets as quickly as possible, aiming to align with capital before labor loses its value.

The latter impulse is particularly evident within the wealthy classes of Silicon Valley and even China.

Some are willing to go to any lengths to obtain equity in Anthropic, OpenAI, or other leading AI companies; the theory is simple: If the singularity really comes, labor may rapidly depreciate, and what will determine a person’s situation then is not what he can do, but what he has in advance.

According to this logic, as long as one exchanges labor income for equity in AI companies before the window closes, there is a chance to shift from being replaced by technology to owning technology.

This is also the most alluring aspect of the “permanent bottom” narrative; it not only creates fear but also suggests an escape route:

Since capital may replace labor, then quickly shift from being a laborer to a capital owner.

The problem is that this path does not belong to the majority of people in the first place.

OpenAI and Anthropic are not companies that ordinary people can easily purchase on the open market. The ones who really have the opportunity to obtain shares are usually lab employees, early investors, and the wealthy who can enter the private equity market.

Thus, this is almost a circular argument:

To avoid falling into the bottom due to lack of capital, one must first have the capital that can only be obtained by the upper class.

Even more importantly, even if a person successfully acquires equity, this insurance may not be permanently valid.

Fernando Borretti, the author of the programming language Austral, points out that there is a contradiction within the “permanent bottom” theory.

If AI can indeed complete almost all cognitive and physical labor at a lower cost, then while ordinary workers may lose their economic value, those who hold shares in AI companies in advance may not necessarily become the “permanent upper class.”

The reason is simple: Wealth is not inscribed in natural laws.

A person has ownership of a company, land, or computing power not only because their name is on the contract but also because courts, police, and governments are willing to recognize and protect that property right.

But following the extrapolation of “super AI replacing everything,” in the end, AI may even be able to handle production, management, governance, and even warfare. At that point, today’s rich neither provide labor nor possess actual power outside of machines. So why would future nations or superintelligent beings have to forever recognize their ownership acquired in the old era?

Buying equity in AI companies may help a person get through the transition period, but it does not guarantee that they will permanently stay in the upper class.

If owning AI assets is not a path available to everyone, a more universal answer is to learn AI.

This path presents another paradox:

The better a person is at using AI to improve productivity, the more they may help companies reduce the need for other workers.

They might be able to temporarily stay in the elevator but are also participating in dismantling the buttons for other floors.

This does not mean that people shouldn’t learn AI. For individuals, being proficient in AI may still be the most reasonable choice right now. The issue is that when everyone tries to avoid being replaced by enhancing their own replaceability efficiency, the end result may be that the workers needed by companies diminish.

Individual rational choices, when aggregated, may actually accelerate the overall decline of labor's bargaining power.

Another answer is to leave the fields most easily replicable by AI and turn toward jobs that require personal presence.

If the supply of digital products approaches infinity, then physical operations, on-site responsibilities, and interpersonal trust that cannot be remotely replicated may indeed command a higher premium.

Data centers need electricians, an aging society needs caregivers, and people are still willing to pay for the real presence of doctors, teachers, and service personnel, but this is also not a fallback that everyone can take.

The above seem like three different paths, but in reality, they all answer the same question:

How to delay falling down?

But what that 17-year-old kid really needs an answer to may be: Why must a person's existence depend on whether they can still be needed by capital?

The reason researchers cannot answer is that this question ultimately requires not just a personal plan but a new distribution system.

When labor is still irreplaceable, wages, unions, and strikes together form the distribution mechanism; if labor really begins to lose its scarcity, then society must find another way to distribute the wealth created by machines to those no longer needed by machines.

This cannot be resolved by everyone working harder to learn AI, nor can it be achieved by everyone purchasing a few stocks in advance.

What individuals can buy is just buffer time; what is genuinely lacking is still that negotiating table.

A negotiating table

Building a negotiating table is not something that any individual effort can replace.

China unexpectedly provides a contrasting example.

In December 2025, a court in Beijing ruled that being replaced by AI does not directly constitute a legal reason for dismissal. Some state-owned enterprise employees also reported that the AI tools they use could approximately accomplish the work of two employees, but the company promised not to lay off workers due to AI.

After spending two weeks in China, Sun summarized, the social attitude there is not simple technological optimism, but rather a kind of pragmatism: “If resistance is not realistic, then get on board first.”

But at least there are people trying to catch those falling during technological transformation with a multitude of fragmentary rules and legislation.

What the U.S. faces is closer to an institutional vacuum.

In April 2026, someone threw a Molotov cocktail at Altman’s residence in San Francisco; in the same month, a city council member supporting data center projects was shot at home. During graduation ceremonies, jeering directed at executives from AI companies began to emerge.

If emotions do not find institutionalized outlets, they will seek exits themselves.

Meanwhile, data from the St. Louis Federal Reserve shows that 39% of GDP growth in the U.S. in 2025 came from data centers and AI-related investments.

The entire country is betting on a technology that most citizens presently feel no benefits from for growth.

The U.S. will hold midterm elections in 2026, and the primary elections in 2028 are also bound to be crowded, with Mark Kelly, Ro Khanna, and Josh Hawley having released AI action agendas respectively.

Therefore, the only indicator worth observing in the next two years may be:

Can that negotiating table be set up before the anger erupts?

It could be a law stipulating that companies cannot simply convert all productivity increases into layoffs; it could be a new form of union that allows employees of the same company, regardless of their profession, to jointly participate in negotiations over AI deployment; it may also be a check bearing the names of AI companies that redistributes the wealth generated by technology to those who bear the costs of transformation through taxes, dividends, or public funds.

The specific form has yet to be determined, but the core issue can no longer be avoided:

When companies no longer need to employ the majority of people to create wealth, how can the majority share in that wealth?

As for that 17-year-old kid with only a B average, he will probably never know:

When the smartest group in an entire industry is asked “What should be done with him,” the answer they give is “There is no answer.”

Then they each go back and continue to buy insurance for themselves.