By Austin VornhagenSomeone told you the middle class is dying.
The picture that sentence paints is a country splitting in two: a few people pulling away in their gated communities, and everyone else sliding down into the same struggle. No more middle. Just rich and poor.
So here’s a strange fact to start with. Since 1971, the share of Americans in the middle really has shrunk. But more of the people who left climbed out the top than fell out the bottom. And the families who stayed in the middle got richer, by about 60% after inflation.
So what, exactly, is dying?
The middle class was never a group of people. It was a deal: an ordinary job buys a secure life.
That deal was not a law of nature. It was built, piece by piece, over about a century. It has been under strain since the 1970s. And now we’re building machines that might, someday, do every job a person can do, which raises the only question that really matters here: what does the deal look like when the economy no longer needs your job to run?
To answer that, we need to know how the house got built in the first place.
“Middle” started as an address, not an income
For most of European history, society had a top and a bottom and very little in between. Nobles and clergy held land and power by birth. Peasants worked the land. Your position was mostly decided before you could walk.
Then towns grew, and trade grew with them. Imagine a merchant who owns a thriving shop. He has money, property, and a measure of independence. He has no title and no estate. He doesn’t fit at the top or the bottom. He’s in the middle.
That’s the original meaning, and it explains a confusion that’s still with us. Today, working class usually describes how you earn a living, and middle class describes your income or lifestyle. A union electrician with a paid-off house is both, and there’s nothing contradictory about it.
The bigger point is what the merchant had that the peasant didn’t: something of his own that kept producing. Remember that. It comes back at the end.
How America built the house, one layer at a time
The mass middle class that people mean when they say “dying” is the postwar American version. It took a foundation, a floor, walls, a door, and a roof. Scroll through the drawing.
First, a place to stand that isn’t a castle
In medieval Europe, your rank came from birth, land, and legal status. Then towns and trade grew, and a new kind of person appeared: the merchant with a shop, money, and some independence, but no noble title. The word bourgeois originally just meant a town dweller. “Middle” described a position between the lords and the peasants. It did not mean an average income.
Then a foundation of factories, and a trap
Industrialization, starting in Britain, created engineers, managers, accountants, clerks, and lawyers, and the middle classes grew. It also created a vast industrial working class with dangerous jobs, long hours, and no cushion. More wealth arrived. Security for most people did not. Hold onto that: a growing economy and a broad middle class are two separate achievements.
The Depression proved hard work wasn’t enough. So a floor went in.
You could work hard, save, and still lose everything in a nationwide collapse. Social Security began in 1935. The Fair Labor Standards Act of 1938 set a federal minimum wage, overtime pay, and child labor limits for covered workers. Coverage started incomplete and widened later. None of it made anyone middle class. It made a bad year less likely to become a ruined life.
Walls that grew as fast as the economy
Here is the heart of the deal. Economists Frank Levy and Peter Temin describe labor productivity and median family income each roughly doubling between 1947 and 1973. Growing businesses and collective bargaining pushed wages and benefits up, and unions influenced pay even at competing employers. When the country got more productive, the typical family felt it.
A door you could walk through
The GI Bill helped millions of veterans pay for college and job training. Expanding education let more people qualify for the better-paid jobs that industry was creating. A door is the right image: it didn’t carry you in. It let you in.
A roof, and equity under it
Mortgage programs and a housing boom put homeownership within reach of ordinary families. A paid-down house became a savings account you lived inside, the asset that let a household absorb shocks and pass something on.
Then the lights came on
“Middle class” changed meaning. It stopped describing a rank and started describing a promise: a person with an ordinary job could own a home, raise kids, take some time off, and retire. Not because they were exceptional. Because the house was built that way.
Some of the doors were locked
The promise was real, and it was never universal. Black Americans faced severe discrimination in lending, housing, and hiring, and Black veterans often could not get the same practical benefits from the GI Bill as white veterans. The house had lit windows that many families were not allowed to stand behind.
Then the walls stopped growing with the foundation
There was no single switch. Technology and trade changed which work paid. Union power weakened. Business practices and public policy changed. Economists still argue about the weights, but the research points to both economic forces and the institutions that set workers’ bargaining power. Starting in the 1970s, typical pay stopped keeping pace with productivity, and the link between a growing country and a secure family loosened.
Notice what the house was made of. Not money. Rules.
A minimum wage. A retirement floor. Bargaining power. Cheap mortgages. Tuition. Every one of those was a decision, and every one of them could be built differently, or not at all. The economy grew enormously in the 1800s too, and most workers didn’t get a house like this out of it.
So when people say the middle class is dying, the honest first question is a measuring question. If it’s dying, where did everyone go?
If the middle is dying, where did everyone go?
Here is America as one street of 100 households, using Pew Research Center’s 2024 analysis of government data. Amber houses are middle income. Blue are upper income. Gray are lower income.
Start with a street of 100 households
Pew Research Center sorts households by income compared to the national median, adjusted for household size. “Middle income” means two-thirds to double the median. In 1971, 61 of every 100 Americans lived in middle-income households, 27 in lower-income ones, and 11 in upper-income ones. Pew’s rounded shares add to 99, so one house here stays unlit.
Fifty-two years later, ten houses leave the middle
By 2023, the middle was down to 51. The lower tier grew to 30. The upper tier grew to 19. Watch where the houses actually went. More climbed out the top than fell out the bottom.
And the houses that stayed got bigger
Here’s the part the “dying” story leaves out. Adjusted for inflation, the median middle-income household went from about $66,400 to about $106,100, up roughly 60%. Lower-income households rose about 55%. Upper-income households rose about 78%. Every tier got richer. The top got richer fastest.
But the middle’s slice of the whole street shrank
In 1970, middle-income households took home 62% of all household income in the country. By 2022, it was 43%. Upper-income households went from 29% to 48%. The middle didn’t get poorer. It got smaller and fell further behind the top.
Not a vanishing. A stretching.
So the evidence supports a smaller middle and a widening distance to the top. It does not show America inevitably sorting into only rich and poor. One caution: these are shares of the population, not the same families followed over time. What really eroded is harder to put in a chart: whether an ordinary job is still a dependable route into one of those lit houses.
Having a job and having security are two different things
The deal was never “you will have a job.” It was “your job will be enough.” Enough to afford the basics, absorb a bad month, get ready for retirement, and recover if the job disappears. Those are different tests, and you can pass one while failing another.
The Federal Reserve’s survey of how households were doing at the end of 2025 shows exactly that split.
of adults said they were doing okay financially or living comfortably.
would cover a $400 emergency with cash or its equivalent. So more than a third would not.
of adults not yet retired thought their retirement savings were on track.
Feeling fine today and being protected tomorrow are not the same thing. These numbers cover all adults, not just workers, but the gap is the point.
Then there’s the roof. Two people with the same salary can live in different economic universes depending on when they bought a house. Harvard’s Joint Center for Housing Studies estimates that affording the payment on a median-priced home took about $66,000 a year in household income in early 2020. By late 2025, it took more than $120,000.
National estimate, assuming 3.5% down on a 30-year fixed loan. Local costs vary a lot.
So before a single robot enters this story, the answer to “does an ordinary job still buy a secure life?” is already: it depends. On where you live, what you already own, and whether you got in before the price of the roof nearly doubled.
Now add a machine that makes workers more productive. Surely that helps?
Ten clerks, one AI, four different endings
AI really can make people better at their jobs. In one study of customer-support agents, an AI assistant raised productivity by about 14% on average, and the gains were biggest for the least experienced workers. That’s genuinely good news for ordinary workers’ abilities.
But here’s the trap in “more productive.” Imagine a company where ten people process insurance paperwork. AI lets them handle twice as much. What happens next? Pick an ending.
Demand is there, so the company takes on twice the business with the same ten people. Everyone keeps their job.
Workers keep their seats, but most of the new money flows to the owners unless pay goes up too.
An illustrative scenario, not data. Real companies often do several of these at once.
Becoming more productive does not automatically make a worker more secure. Who captures the gain depends on demand, competition, ownership, and bargaining power.
That’s exactly the lesson of the 1970s, replayed at machine speed. And when you look at where AI is actually showing up in the job data so far, it isn’t where most people expect.
The danger isn’t mass layoffs. It’s a missing bottom rung.
Everyone is braced for the big dramatic version: robots arrive, millions fired. The evidence so far points to something quieter, and in some ways worse.
A career is a ladder you climb by being hired onto the bottom
Nobody is born experienced. You get a first job, someone teaches you, and ten years later you’re the person who teaches. The bottom rungs aren’t the least important part of the ladder. They’re the only way onto it.
The people already on it hold their grip
Stanford researchers analyzing ADP payroll records looked at jobs most exposed to AI. For experienced workers in those jobs, they found no comparable drop in employment. The upper rungs held.
The bottom rungs thin out
For workers aged 22 to 25 in the most AI-exposed occupations, employment by mid-2026 was about 19% below where it would have been had it kept pace with less-exposed jobs. In the original 2025 version of the study, that gap was 13%.
Nobody gets pushed off. Fewer people get on.
The researchers found the adjustment came mainly through reduced hiring, not layoffs. That makes it nearly invisible. No factory closes. No headline. A posting just never goes up.
Read this before you panic
The authors call these descriptive patterns, not proof that AI caused the gap; other forces could explain part of it. And the national picture is sluggish, not collapsing: in September 2026, unemployment was 4.2% and payrolls grew by about 29,000. Ordinary employment is not disappearing.
The quiet way a middle class erodes
Follow the drawing forward. If fewer people are hired at 23, there are fewer experienced people at 33. The comfortable worker keeps their job. The young person never gets the one that would have made them comfortable. That is the failure worth watching, because it doesn’t look like a failure while it’s happening.
So, does an ordinary job still buy a secure life? For many people, yes. As a guarantee, no.
Employment alone has become too thin a foundation to assume that everyone who works steadily will be secure. AI doesn’t create that weakness. It raises the stakes on it. If AI makes workers more capable, demand grows, and workers share the gains, an ordinary job could become a better route than it’s been in decades. If software becomes a cheap substitute for labor, workers could lose bargaining power even while the economy booms. Both paths are possible. Nobody knows their relative size yet.
Which means the real design question isn’t “how do we protect jobs?” It’s harder and more interesting than that.
Now take the hardest version: a machine that can do every job
Let’s stop hedging and run the thought experiment all the way. Suppose a superintelligence exists that can do every economically useful job a human does today, and suppose, a big assumption, that it operates safely under legitimate human control. Economic policy can’t deliver that part. Everything below depends on it.
In that world, the old deal is dead on arrival. You can’t build lifelong security on “find something the machines still need you for,” because by definition there may be nothing. Every middle class in history rested on ordinary people having something the economy needed. The merchant had a shop. The postwar worker had labor to sell. If labor stops being needed, the claim has to move to something else.
Here’s the design principle: as the country gets more capable, an ordinary household should get more independent, not less needed.
What would that feel like? Here’s an imagined, ordinary Tuesday.
A family has a comfortable home, dependable healthcare, and enough money for repairs, recreation, and the occasional trip. They can save. An unexpected illness is a worry, not a threat to everything they own.
One parent coaches softball and runs a small business. The other spends part of the week restoring a local creek and studying architecture. Their interests may change. Their income can change without putting their kids’ security at risk. Part of that income arrives because they own a share of the productive economy: automated businesses, energy infrastructure, and other investments earning on their behalf. And the essentials cost much less at the point of use.
Down the road there’s a town workshop where people design and make things with AI help. Someone is building equipment for people with disabilities. Someone else is making absurdly elaborate furniture. Kids learn next to adults making things they care about.
There are still businesses, arguments, ambitions, and people who get rich. There are still hard decisions about land and money. But an ordinary household can say: We can leave this employer. We can challenge this decision. We can take care of our kid. We can afford to start over.
That is what building a country around the middle class would mean: security, independence, ownership, and a real voice as the normal condition of life. And “middle class” here describes that condition. It doesn’t require a poorer class underneath it.
The house for that world needs four new foundations. Tap through the plans.
Every American owns a piece of the machines.
Today you buy ownership with money left over after the bills. In a world where machines do the work, that becomes a trap: you need wages to buy assets, and the machines are why the wages are disappearing. So give every American, at birth, a permanent share in a set of diversified public investment trusts holding stakes in businesses, energy, and infrastructure. The share itself can’t be sold or pledged to a lender, so it can’t be bought from you in a desperate week. The payments are yours to spend or save.
Pieces already exist. Alaska’s Permanent Fund turns part of the state’s mineral wealth into a fund that pays residents a dividend and supports public services. Norway invests its oil revenue in a diversified fund for future generations. Neither is a full model. Both prove shared wealth can outlast a generation.
How the trusts fill up: when taxpayers finance a commercial project, the public takes fairly priced equity or revenue sharing; part of capital-related tax revenue gets saved instead of spent; workers and communities get financing to buy businesses; and the trusts invest in new productive capacity, taking real investment risk.
Money is useless if there’s nothing to buy with it.
Hand everyone more money while housing stays scarce, and landlords and sellers absorb a lot of it. Give everyone health coverage without enough clinics, and people have insurance and a waiting list. So the second foundation is concrete and steel: more homes where people want to live, enough clinical capacity, abundant electricity and resilient grids, transportation for people who can’t drive, and childcare, eldercare, and lifelong learning that don’t depend on having a job.
There’s an American example close to home. Nebraska is the only state where every home and business gets electricity from public power: public power districts, cooperatives, and municipal utilities, some run by elected boards. It doesn’t settle how future AI infrastructure should work. It does show essential systems can be owned by the people they serve.
Superintelligence could help design, build, and run these systems. They would still need equipment, energy, materials, approvals, and time.
Comfort without capability is just a nicer cage.
A comfortable population could still depend on a handful of companies for every decision that matters. So make capable tools broadly available: an AI assistant that works for you and that you can take to a different provider with your data; shared computing; shared workshops where residents design and build things.
Picture a neighborhood with a drainage problem. Today it waits for the city. In this world, residents investigate it, compare engineering proposals, get an independent review, and commission the repair. Or one person explores an invention without first convincing a wealthy institution it deserves attention.
Projects that affect other people would still need funding and permission. But ordinary people could formulate, test, and pursue an idea.
A machine can be better at predicting. That doesn’t make it the one who decides.
An AI could lay out three plans for a town: one saves farmland, one concentrates housing downtown, one maximizes near-term tax revenue. It should show its assumptions and uncertainties. Then residents choose, through accountable institutions. Being better at estimating consequences doesn’t give anyone, human or machine, the right to decide whose interests count.
The protections need teeth: independent courts and a real human to appeal to, privacy and freedom of association, essential services that can’t be cut off over a political disagreement, multiple providers, and public oversight of big concentrations of economic and technological power.
The trusts themselves are a risk: put everyone’s livelihood under one unaccountable board and you’ve built a giant new single point of failure. So spread management across independent institutions, publish performance and fees, and keep investment separate from paying benefits.
People would still work. Some would chase demanding careers. Others would spend years caring for family, competing in sports, studying, making things, or running their communities. Human work could stay valuable even where a machine does it better. We already know this from sports, music, and friendship. A kid doesn’t need their parent to be the most capable person on Earth. They need their parent.
But a paycheck can’t stay the only recognized proof that someone matters. And none of this works if the money doesn’t add up. So let’s add it up.
The plan has to survive arithmetic
Engineers call this a load calculation: before you build, you check the structure can actually hold what you’re putting on it. Suppose, purely as an illustration, that 250 million adults each received a dividend from the national trusts. Move the sliders.
At 100%, the whole dividend rides on one fund. A 4% payout is an assumption, not a guarantee.
That second number is the point. $20,000 a year for 250 million adults costs $5 trillion a year, and funding all of it from a fund paying out 4% would take $125 trillion in assets. Slide the “share the trusts pay for” down and the fund shrinks fast. That’s why the plan can’t be one mechanism. It needs several working together: accumulated ownership, ongoing tax revenue, affordable public services, and falling production costs. And every dollar has to show up in the budget exactly once. The same earnings can’t pay for a dividend and a service at the same time.
There’s also a twist. If AI makes products wildly cheap through competition, profits in some industries could shrink. The fund’s returns could disappoint even while people are better off through lower prices. A fund is only one tool. In the end, prosperity means real goods and services reliably reaching households.
How to get there: overlapping stages, starting while paychecks still work
No one can honestly promise that superintelligence arrives in a particular year. So commit to a sequence and to measurable milestones instead of dates. The order matters more than the speed: you build the new support while wages still hold most households up, and you only lean on it once it’s proven it can carry the weight.
Pour the foundation and protect people now
Create the trusts, their independent governance, disclosure rules, and first funding. Start negotiating public returns on new subsidies. At the same time, improve unemployment support, keep health coverage continuous across job changes, and pay people to train into new work. Someone losing a job this year can’t wait decades for a fund to mature. This stage costs something real: some consumption deferred, some taxes shifted.
Build a model home people can walk through
Start where the benefits are concrete: community-owned power, housing, shared AI services, financing for workers to buy businesses when owners retire. Measure household costs, service quality, and net public returns. Publish the failures too. A pilot proves an institution can operate. It can’t prove a national system will pay for itself.
Build capacity before promising purchasing power
Expand housing, power, care, and transportation alongside household support, with each dollar tied to something you can see: homes occupied, power delivered, shorter waits, lower bills. This is when families first feel more room in their budgets.
Spread the gains as automation spreads
As machines generate more profit, grow shared ownership and payouts. Where companies still employ people, support bargaining for higher pay, shorter hours, and ownership stakes. Where jobs disappear, keep income steady without pretending retraining always restores an equal job. Shift taxes over time toward profits, capital income, and economic rents, and less onto paychecks alone.
Loosen the tie to wages only when the new support holds
A shorter workweek where productivity and financing allow it, sector by sector. The test is simple: do households keep their purchasing power and services while working less, with enough people still staffing what needs a human? Paid work becomes optional for basic security only after that is proven.
Judge it by ordinary households, not by GDP
Track production and investment, sure. But steer by a scorecard any family could fill out at the kitchen table:
- Can people afford housing?
- Can they get care?
- Can they survive a gap in income?
- Do they have spare money and spare time?
- Can they challenge a decision made about them?
- Are kids growing into independent adults?
The hardest obstacle isn’t technical. It’s political power. People who already control valuable assets will have reasons to resist sharing future returns. Some public institutions will make mistakes. Some politicians will try to raid the funds or steer investments to friends. A credible path needs visible benefits, broad participation, oversight that can bite, and institutions sturdy enough to survive changes in government. It also needs international cooperation: American comfort shouldn’t depend on exporting insecurity, environmental damage, or coercion somewhere else.
None of that makes the transition painless. But it gives it a purpose people can recognize.
The middle class is a house. Every generation decides who it shelters.
Walk back through the drawings. The medieval merchant was “middle” because he owned something that kept producing. The industrial century proved that growth alone doesn’t build a broad middle. The New Deal and the postwar decades built one on purpose, out of rules, and locked many families out of it. Since the 1970s, the walls stopped growing with the foundation. Now AI is quietly removing bottom rungs while everyone watches for layoffs.
Every version of the deal has rested on the same bet: that ordinary people hold something the economy needs. For two centuries that thing was labor. If machines make labor optional, the bet has to move back to where it started, to owning a piece of what produces. The merchant’s shop, scaled to a whole nation.
In the mature version of that future, a parent loses their job because a machine does it better. There’s real disappointment. They’ll miss their coworkers, the routine, the feeling of being good at something.
But their home stays secure. Their kid’s treatment continues. Their ownership income keeps arriving. They have time, tools, and people to help them figure out what comes next.
And the question at the dinner table stops being “How will we survive?” It becomes: what would you like to do now?
This essay began as a conversation with an AI about the history of the middle class, AI and work, and what a middle-class future could look like, then was checked against sources. Key sources: Pew Research Center, The State of the American Middle Class (2024); Levy and Temin, Inequality and Institutions in 20th Century America; the Federal Reserve’s Economic Well-Being of U.S. Households in 2025; Harvard JCHS, The State of the Nation’s Housing 2026; Brynjolfsson, Chandar and Chen, Canaries in the Coal Mine? (August 2026 update); Brynjolfsson, Li and Raymond, Generative AI at Work; and the BLS Employment Situation for September 2026. The ten-clerk scenario and the $20,000 dividend are illustrations, not forecasts. The future described in Sheets 06 through 08 is a proposal, not a prediction. The hero film and drawings are illustrations.
If “own a piece of what produces” stuck with you
The Robot Dividend’s First Check
Why a dividend from the machines would start tiny, and a concrete plan for crossing the years between your last paycheck and one worth having.
Read the essay →Essay · Content PilotsWe Automated the Apprenticeship
The missing bottom rung, up close: how entry-level work was also the training system, and what replaces it.
Read the essay →Primary source · Pew Research CenterThe State of the American Middle Class
The full 50-year data behind the street of 100 houses, including how each group has fared.
Read the report →The merchant’s shop is still the best foundation there is.
Owning something that keeps producing is the oldest route into the middle class, and a small business is still the most direct version of it. I build fast websites for those businesses at Content Pilots, with scheduling and payments wired in, so the shop keeps working while you’re not at the counter. And if you think I got the history or the economics wrong somewhere, email me. I’d genuinely like the correction.