Silicon Valley argues about whether AI will create a permanent underclass. It already has a measured population of 2.1 billion, it was not caused by automation, and the exit is narrower and more open than either side claims.
This in-depth guide covers everything you need to know about escape the permanent underclass: what the data actually shows (2026). Based on verified income data and real-world case studies from our database of 133 side hustle tactics.
The phrase is having a moment in a very specific place. Search it and you will find essays on Medium, Substack and LessWrong, mostly written in the last two years, mostly by people in or adjacent to the technology industry, arguing about whether artificial intelligence is about to automate white-collar work and remove the ladder behind it. The debate is about a future. The participants are, by global standards, comfortable.
Meanwhile the thing they are worried about becoming has an existing population, and it is measured every year.
The ILO's Employment and Social Trends 2026 puts more than 2.1 billion workers, about 57.7 percent of the global workforce, in informal employment. The World Bank's ILO-modelled series for vulnerable employment, meaning own-account and contributing family workers, was 42.4 percent worldwide in 2025, and 71.6 percent in India, 66.5 percent in Nigeria, 55.5 percent in Pakistan. Extreme working poverty covered 7.9 percent of workers, around 284 million people, in 2025, and in low-income countries 68 percent of workers lived in extreme or moderate poverty.
Those are not predictions. That is the current census of people who work full time and cannot accumulate anything, and it existed before anyone had heard of a large language model.
So the honest starting position for this page is that the permanent underclass is not a thing that might happen. It is a thing that has a population of roughly two billion, and the interesting question is not whether it is coming for you. It is what the actual mechanism of escape is, and whether it still works.
What the word "permanent" is claiming
Every version of this argument turns on one word, and it is worth isolating what it asserts.
Poverty is not the claim. Plenty of people are poor for a period. The claim is that the route out has closed: that the mechanism by which someone born at the bottom ends up somewhere else has been removed, so position is now inherited rather than earned.
That is a testable proposition and the evidence on it is uncomfortable in both directions.
The World Bank maintains the Global Database on Intergenerational Mobility, which estimates mobility in education for 153 countries and about 97 percent of the world's population born in the 1980s. Its finding is that mobility is substantially lower in developing economies than in high-income ones, and that progress has stalled for the most recent cohorts it covers. That is the strongest available support for the pessimistic case, and it is real.
What it does not show is closure. Lower mobility and stalled progress describe a ladder that is shorter and harder to climb. They do not describe an absent ladder, and the difference matters enormously to anyone deciding what to do on Monday.
The version of this argument circulating in technology circles goes further and asserts closure, usually on the basis that automation will shortly remove the entry-level rungs. That is a forecast. It may prove right. It has not happened yet, and it is being asserted most confidently by people whose own position is not in the measured population above.
Why the AI framing points at the wrong people
The essays share an implicit subject: a graduate whose analytical job is automated, who then has nowhere to go. That is a real risk and it is not the shape of the existing problem.
The two billion people in informal work are not there because software replaced them. They are there because the economy around them never generated formal jobs at the rate the population grew, because the work available carries no contract, and because the assets that generate the other half of national income are owned by someone else. Automation is not the mechanism. It never was.
This matters practically, because the remedies differ completely. If your model is "AI took the jobs", the response is to find work AI cannot do, which is a race with a moving finish line. If your model is the measured one, which is that most work has never led anywhere and the ownership of productive assets is the actual dividing line, then the response is to acquire a claim on something rather than to out-run a machine.
The second model is older, better evidenced, and considerably more actionable.
The three walls, and which ones are real
Ask what actually keeps someone in the measured population and three constraints come up. They are not equally binding, and treating them as equal is how people spend years on the wrong one.
Capital. You cannot buy an asset without surplus, and there is no surplus. This is the most cited wall and it is the least absolute, because the cost of acquiring certain assets has genuinely collapsed. A person with a phone can now build distribution, publish a product, and accept a payment from another country without capital, permission or premises. That was not true twenty years ago and it is the single largest change in this problem in a generation.
Time. This is the wall people underestimate. Informal work is frequently longer-houred and less predictable than formal employment, and it is combined with unpaid care work and long commutes. Someone working eleven hours and travelling two has no evenings, and every plan that assumes two free hours a day is written for someone else. This wall is real, it binds hardest on women, and no amount of motivation dissolves it.
Information and network. Not knowing that a market exists, what it pays, or how to reach it. This is the wall that has fallen fastest and is still the most commonly binding, because knowing that someone in another country will pay four times your local rate for work you already do is worth more than any productivity technique.
The practical reading is that the capital wall is lower than it was, the information wall is lower than it was, and the time wall is not. Any plan that spends the little time available on the walls that already fell is a waste of the only resource that is genuinely scarce.
The escape mechanism that is still open
There is one, it is unglamorous, and it follows directly from where the money is.
Roughly half of world output does not reach anyone through wages. The UN's SDG indicator 10.4.1 put the global labour income share at 52.6 percent in 2025, and at 43.9 percent in the Philippines. Whoever owns the productive assets receives the rest. Escape, defined properly, means acquiring a claim on that second pool, because a claim on the first pool is what everyone in the measured population already has and it is what left them there.
For someone with no capital, three claims are obtainable, roughly in order of how quickly they arrive.
Sell into a richer market than your own. This is the fastest and the most underrated. Your local rate is set by local purchasing power and by heavy local supply. The identical skill sold to a buyer in a higher-income economy is priced by their market instead. Nothing about you has to improve. This is why remote service work, cross-border marketplaces and digital delivery matter far more in Lagos, Karachi and Manila than the equivalent advice does in London, and it is the single largest income change available without capital.
Build distribution. An audience, a list, a channel, a search ranking, a reputation in a specific niche. It is the only asset that accumulates with time rather than money, which makes it the natural first holding for someone with neither. It also lowers the cost of everything you sell afterwards, which is what makes it an asset rather than an activity.
Convert income into ownership, mechanically. A fixed percentage of every payment, moved on the day it arrives, into something you own. The index investing guide sets out how that actually works. The amount matters less than the automation in the first year. This is the step that actually crosses the line between the two pools, and it is the step that people who earn more but own nothing keep skipping.
None of the three require permission, capital, or a credential. All three are slow. Anyone offering you a fourth option that is fast is selling the fourth option.
What this page will not tell you
It will not tell you that anyone can escape. The time wall is real and for some people it is total. Someone supporting five dependants on eleven-hour days has no surplus of hours or money, and the honest thing to say is that individual effort is not the binding constraint in that situation. Policy is. Telling that person they need better habits is what the course industry does, and it is contemptible.
It will not tell you the odds are good. Most attempts do not work. The reason to attempt it anyway is that the alternative is a fixed position, not that success is likely.
It will not tell you AI is irrelevant. It is plainly changing what generic work is worth, and the categories most exposed are precisely the entry-level, generic, remote tasks that this page recommends as a starting point. The response is to be specific rather than generic as early as possible: a defined buyer, a defined problem, and a reason you are the obvious choice for that narrow thing.
And it will not tell you that reading about this is progress. The people writing essays about the permanent underclass are, almost without exception, not in it, and consuming the argument is a way of feeling engaged with a problem while changing nothing about your position in it.
Where the measured population actually is
The phrase gets used as though it described one condition. The data describes several, and the right response differs in each. These are 2025 World Bank ILO-modelled figures for vulnerable employment, set against unemployment for the same year and source.
India. Vulnerable employment 71.6 percent, unemployment 4.2 percent. Close to three workers in four are own-account or contributing family workers. Almost nobody is idle. If you want a single country that demonstrates that joblessness and the underclass are different phenomena, this is it, and the response that fits is reaching customers outside a labour market with overwhelming domestic supply.
Nigeria. Vulnerable employment 66.5 percent, unemployment 3.1 percent. Two thirds insecure, near-total employment. Here currency is the decisive variable rather than the work itself: the same hour billed in dollars behaves entirely differently from the same hour billed locally, which makes cross-border work less a preference than the whole point.
Pakistan. Vulnerable employment 55.5 percent, unemployment 5.4 percent. Same logic, with payment rails as the practical obstacle. Settling how money will reach you comes before deciding what to sell, because a route you cannot collect from is three months lost.
Indonesia. Vulnerable employment 50.0 percent, unemployment 3.2 percent. Exactly half. A large domestic market makes local-language products viable in a way they are not in smaller economies, so the export-only logic that fits Nigeria applies less strongly.
Philippines. Vulnerable employment 33.6 percent, unemployment 2.2 percent, the lowest unemployment in this set. A third of workers insecure while the headline rate suggests an economy at full employment, which is the clearest single illustration of why that rate should never be used to decide whether people are doing well.
South Africa. Vulnerable employment 11.7 percent, unemployment 32.4 percent. The inversion. Formal work dominates among those who have any and a third of the labour force has none. Here the underclass is defined by exclusion from work rather than by the quality of it, and almost every recommendation on this page has to be read differently.
Six countries, one phrase, and at least three different problems underneath it. Advice that does not distinguish between them is advice written for an average person who does not exist.
What the previous escapes actually looked like
It is worth knowing how people historically got out, because the mechanisms are unglamorous and they are still the ones that work.
Almost none of it was entrepreneurship in the sense the word is used now. It was migration, to a city or to another country, which converted a low local wage into a higher one for the same labour. It was credentialling, meaning a licence or qualification that legally restricted who could compete. It was a formal job with a contract, in an economy generating them fast enough. And it was ownership, usually of land or a small premises, acquired over decades.
Each of those is a version of the same move: acquiring something that limits competition for what you sell. A licence limits it legally. A city job limits it geographically. Ownership limits it absolutely.
The reason the internet matters here is that it created a fifth version, which is reaching a market where your labour is scarce rather than abundant. It is the same mechanism as migration without the migration, and it is available to more people than any of the previous four, because it requires neither capital, nor a visa, nor years of study.
That is the honest case for optimism, and it is narrow. It does not say the ladder is easy. It says one new rung appeared, it is real, and it is being used by a very small fraction of the people who could use it.
If you are reading this from the comfortable side
Some readers arrive here from the technology-industry version of the argument, worried about their own position rather than occupying the measured one. The concern is legitimate and the framing is worth correcting, because it changes what you should do.
The risk to a well-paid analytical worker is not becoming one of the 2.1 billion. Social protection, savings, credentials and property make that improbable within one lifetime. The realistic risk is a compression of the premium your particular skill commands, which shows up as flat pay, fewer senior roles and a longer job search, rather than as destitution.
That is a serious problem and it has the same solution as the one on the rest of this page, which is why the two audiences belong on one page. A salary is a claim on the labour pool. If the premium on your labour compresses, the claim shrinks, and the response is to hold something in the other pool before you need it rather than after. The difference is that you can currently afford to do it easily and the measured population cannot.
The one thing worth avoiding is treating the anxiety as equivalent. It is not, and writing that conflates a compressed premium with informal subsistence work is a large part of why this phrase generates so much discussion and so little useful action.
The industry that sells this word
A phrase describing two billion people in insecure work is a marketing asset, and it is being used as one.
The pattern is consistent. Establish that the situation is hopeless and structural. Establish that ordinary routes are closed. Then present a paid product as the only remaining exit. The despair does the selling, which is why the content that agitates hardest about permanence is frequently attached to the most expensive offers.
Four checks before paying anyone.
Is the price published, or revealed only after a call. A price discovered on a call is set by how desperate you sounded on it.
Does the seller earn from the business they teach, or from teaching it. Where the answer is available it is usually revealing.
Is the income claim about them or about their students. A screenshot of the teacher's earnings tells you that selling courses is profitable.
What is actually free. Almost everything sold as the secret is published somewhere at no cost, which is why the sales page focuses on your position rather than on the contents.
We keep reviews of these sellers with published prices where they exist and documented regulatory findings where they exist. Read the review before paying rather than after. Someone who believes their situation is permanent is the highest-converting customer in this market, and being told so is not evidence that it is true.
What to do about it this month
The instruction is the same as it is everywhere else on this site, and it is short.
Find out what someone outside your local market pays for something you can already do. One hour of research, not a course. If the number is meaningfully higher than your local rate, that gap is your entire strategy and everything else is detail.
Then get one payment from one stranger in that market. Any amount. It proves the chain works end to end: that you can describe something, that someone will decide they want it, and that money can reach you where you are. Payment infrastructure is the step that quietly ends most attempts in several of the countries with the highest informality, so settle it before you sell rather than after.
Then move a fixed share of every payment into something you own, starting with an amount too small to hurt, because the habit is what year one is for and the amount is what year ten is for.
Two billion people work full time inside a structure that does not accumulate. That is the measured reality and it is much larger and much older than the argument currently being had about it. The exit is narrow, slow, and open. It runs through owning a claim on the half of the economy that wages never touch, and the first step costs nothing but an afternoon.
The short version
The permanent underclass is not a forecast. Around 2.1 billion people, 57.7 percent of the global workforce, are in informal employment right now, and 42.4 percent of the world's workers are in the vulnerable categories that carry no contract and no route upward.
Unemployment is not what defines it. India runs 71.6 percent vulnerable employment against 4.2 percent unemployment. Almost everybody works. The work does not accumulate.
The word "permanent" overstates the evidence. Mobility is lower in developing economies and has stalled for recent cohorts, which describes a shorter and harder ladder rather than a missing one.
The dividing line is ownership. Only 52.6 percent of world output reached anyone through wages in 2025, and 43.9 percent in the Philippines. Everything else went to owners, and a wage is a claim on the wrong pool.
Three exits are open without capital: sell into a market richer than your own, build distribution, and convert a fixed share of every payment into something you own. All three are slow. Anything fast is being sold to you.