The Scaling Wall: A 1960s Economist Explains Why Every Industry That Matters Is Running Out of People

A few weeks ago I wrote about riding the sandworm. The big picture: AI is an abundance engine, not a job killer, and the industries hitting capacity ceilings are the ones with the most to gain. The response blew me away. But the question I kept hearing was some version of: "OK, I buy that there's a wall. But why is it there? And why hasn't anything fixed it before?"

Fair question. Let's go deeper.

Quick note on terminology before we dive in: if you've spent any time in the AI world, you may have heard "scaling wall" used to describe the point of diminishing returns when you increase model size, dataset volume, or raw compute power. The models stop getting smarter just because you pour more resources in. I'm borrowing the phrase here because the same phenomenon plays out in service industries, but with people instead of parameters. You can't just throw more humans at the problem and expect performance to scale. There's a ceiling, and it has a name.

Rationed Air

If you've watched The Expanse (and if you haven't, stop reading this and go watch it), you know the Belters. They live in the asteroid belt, millions of people crammed into stations and hollowed-out rocks, surviving on recycled air, rationed water, and whatever scraps the inner planets let trickle out. Earth has oceans, forests, an entire biosphere. The Belt has just enough. There's a moment in the show that captures it perfectly: a Belter character explains that you learn, as a child, not to fully exhale. Every breath is calculated. You don't waste oxygen on a deep sigh. The scarcity isn't just a resource problem; it shapes how people think, how they negotiate, how they raise their kids. They know abundance exists somewhere. They just can't get to it.

That's not a metaphor. That's a description.

Right now, roughly 27 million Americans had a diagnosable mental illness last year and received no treatment at all.¹ Not "chose not to seek help." Tried, or wanted to try, and hit a wall. If you gathered them in one place, they'd outnumber the entire population of Texas. Nearly half of all Americans with a diagnosable mental illness receive no treatment in any given year, and the reason is brutally simple: there aren't enough people to provide it. One hundred and sixty million Americans live in areas the federal government has officially designated as Mental Health Professional Shortage Areas.²

The U.S. is projected to be short 187,000 physicians by 2034.³ There are about one million practicing physicians today, so that shortfall represents nearly 20% of the current workforce. And it can't be fixed quickly because training a physician takes a minimum of eleven years after high school. The constraint isn't funding. It's time and human capacity.

Then there's elder care, and this one is worth pausing on. Harvard Public Health projects 4.6 million unfilled home care positions by 2032.⁴ To fill that gap, 53 million Americans are currently serving as unpaid family caregivers, and this isn't kids taking their mom to a doctor's appointment once a month. The average family caregiver provides 24 hours of care per week. One in four provides more than 41 hours; a full-time job on top of whatever else they're doing with their life. The tasks are clinical: medication management, wound care, physical therapy exercises, help with mobility and bathing. Together, these 53 million people provide an estimated $870 billion per year in labor that the professional system simply cannot staff.⁵ That's larger than the entire federal defense budget. It's not volunteer work. It's what happens when a system runs out of people and the burden quietly shifts to families who have no other option.

In legal services, 92% of civil legal needs for low-income Americans go entirely unserved.⁶ Not inadequately served. Unserved. There are roughly 1.3 million licensed attorneys in the United States, and yet for tens of millions of Americans, the legal system functions as if those attorneys don't exist at all. Justice is expensive. So people go without.

More than 400,000 teaching positions sit vacant across public K-12 schools, concentrated in the subjects and communities where outcomes are already worst.⁷ In drug discovery, 80% of clinical trials fail to meet enrollment timelines; not because the science is insufficient, but because the human infrastructure can't keep pace with what researchers already know.⁸

Every one of these industries shares the same condition. Enormous demand. Growing demand. Deeply personal demand. And the supply of qualified human beings has hit a ceiling that wages, policy, and good intentions haven't been able to raise.

I call this the scaling wall. It's not a market failure. It's not a political failure (though politics sometimes makes it worse). It's a human capacity ceiling, structural and economic in nature, and in most cases decades in the making.

The Belters live behind it. So do millions of Americans right now. The wall is the barrier keeping them from breathing freely.

The String Quartet Problem

So why does the wall exist? Why can't these industries just... grow?

There's an answer, and it comes from a 1960s economist named William Baumol. Stay with me, because once you see this, you can't unsee it.

Baumol noticed something deceptively simple: in industries where the labor is the product, productivity can't improve the way it does in manufacturing.⁹ A car factory can build more cars per worker by adding automation, redesigning the assembly line, improving materials. But a string quartet can't play Beethoven faster without turning it into something other than Beethoven. The labor is the output.

Think about that for a second. A therapist listening to a patient. A teacher working with a struggling student. A doctor reading a patient's affect during an exam. In these industries, the human interaction isn't overhead. It's the product. You can't speed it up without destroying it.

Here's where it gets structural. Wages in service industries rise alongside wages in the broader economy because labor markets are interconnected. If tech companies and finance firms are paying more, hospitals and schools have to raise wages too, or lose people. But output per worker doesn't rise, because the work resists mechanization. So the cost of delivering healthcare, education, therapy, and legal counsel climbs year after year while the capacity to deliver them stays flat. Demand grows with population and prosperity. Supply doesn't. The wall gets taller.

Go back to the string quartet. Ten years after Baumol made this observation, those four musicians are earning more; wages in the broader economy rose, and they had to raise their rates to keep up. But they're still playing four instruments. Still taking seventy minutes to perform the Beethoven. The cost of the concert went up. The output didn't. That is the cost disease, expressed in four people on a stage.

Baumol called this "the cost disease." Economists have been watching it play out for sixty years.

Healthcare is the clearest example. American physicians earn an average of $352,000 per year, compared to roughly €52,000 to €280,000 (approximately $57,000 to $305,000) in European nations.¹⁰ The instinctive reaction is to think American doctors are overpaid, but that misreads the situation. U.S. physician compensation has actually been falling in real terms even as total healthcare spending keeps rising.¹¹ The cost isn't going to doctors. It's going to the system built around doctors. Administrative overhead consumes approximately 25% of U.S. hospital expenditures, double the rate of other developed nations, totaling an estimated $504 billion per year in excess administrative costs.¹²

Read that again. Half a trillion dollars a year. Not on medicine. Not on doctors. On paperwork.

We didn't just hit Baumol's wall. We built a second wall on top of it, made entirely of documentation, billing codes, prior authorizations, and compliance tracking.

The Gerbil Wheel

Abstract economics only gets you so far. Let's look at an example most people have experienced.

If you've ever waited weeks for an MRI, or wondered why it takes so long to get imaging results, you've experienced the wall personally. The average wait for an outpatient MRI in the United States is two to three weeks (I recently waited almost three months for an appointment, further delayed by my hand surgeon's schedule), and in academic medical systems, nearly half of all MRI appointments are delayed beyond the originally scheduled date.¹³

There are approximately 41,000 radiologists in the United States; roughly 13 for every 100,000 people.¹⁴ Imaging volume is growing at about 4.6% per year. The radiologist workforce is growing at about 1% per year. Only 29 new first-year residency positions have been added nationally in the last four years.¹⁵ The two lines don't meet.

A research team at the Mayo Clinic analyzed over 1.5 million CT and MRI exams across a ten-year span and found that the average CT exam had grown from 82 images to 679, and the average MRI from 164 to 570.¹⁶ To keep pace, a radiologist now has to interpret one image every three to four seconds, eight hours a day, 255 working days a year. That's not a sustainable profession. It's a gerbil wheel. Radiologist burnout has risen from 36% to nearly 50% in just four years, and radiologists have been leaving the profession 50% faster than pre-pandemic levels since 2020.¹⁷ The shortage creates burnout. The burnout accelerates the shortage. It feeds itself.

Now here's the turn.

A study across Northwestern Medicine's 11-hospital system analyzed nearly 24,000 radiology reports over five months and found that AI-assisted reporting improved efficiency by an average of 15.5%, with some radiologists seeing gains as high as 40%.¹⁸ The system generates a report that's 95% complete, personalized to each radiologist's own reporting style, which the radiologist reviews, edits, and finalizes. The diagnostic judgment stays entirely human. The labor of composing, structuring, and formatting the report doesn't.

Think about what that means for a practice of ten radiologists, each reading roughly 50 exams per day. At 4.6% annual growth, within seven years that practice needs to handle about 680 exams per day. Under the old model, that means four additional radiologists recruited from a labor market that produced 29 new positions nationally in four years. Those four radiologists don't exist. With technology absorbing report generation, those same ten radiologists handle the equivalent of twelve to fourteen. They absorb seven years of volume growth without recruiting a single person who was never going to be available anyway. Patients don't wait three weeks. Radiologists spend more time on the cases that actually need them: the ambiguous scan, the complex finding where years of training and a human eye make the difference between a caught diagnosis and a missed one.

Now flip it. What if that technology didn't exist? You don't get a staffing shortage. You get a collapse; wait times measured in months, radiologists burning out faster than they're replaced, and diagnostic delays that cost patients not just time but outcomes. The wall doesn't just hold. It gets taller every year.

That's the scaling wall made real, inside a single specialty, for patients currently sitting in waiting rooms wondering why it takes so long.

The Ring Gates

Here's where the Belters come back in.

In The Expanse, the discovery of the protomolecule opens the ring gates: portals to over 1,300 habitable star systems. Overnight, the resource ceiling that defined Belter existence dissolves. The air is no longer rationed. The constraint that shaped an entire civilization disappears, and what floods through isn't chaos. It's demand that was always there, finally meeting supply that can serve it. And that moment of stepping through something powerful and unknown, not knowing what's on the other side? That feeling is familiar. It's what a lot of people feel about AI right now. Every time a new tool lowers the human scaling wall, we're stepping through our own ring gate.

That pattern isn't science fiction. It's played out every time a capacity constraint was removed in human history.

Before Gutenberg's printing press, Europe had perhaps tens of thousands of books, all hand-copied by scribes. Fifty years after the press, an estimated 15 to 20 million volumes were in circulation.¹⁹ The constraint was human copying speed. When it was removed, the demand for written knowledge didn't shrink. It exploded. Literacy rates began climbing for the first time in centuries.

In 1930, producing 100 bushels of corn took 15 to 20 hours of human labor. Today it takes just over three.²⁰ Malcolm McLean's standardized shipping container, introduced in 1956, dropped port loading costs from $5.86 per ton to $0.16; a 97% reduction.²¹ The bottleneck was longshoremen loading cargo by hand, piece by piece. When it was removed, global trade expanded until containers carried 95% of the world's cargo.²²

The pattern is identical every time. A human capacity constraint conceals massive latent demand. Technology removes the constraint. Demand floods in. The total amount of meaningful human work expands rather than contracts.

But here's what makes this moment different from every previous technological revolution. Every prior wave broke walls in the production of goods: food, books, manufactured products, shipped cargo. Service industries; the ones that touch human lives most directly; remained behind Baumol's wall because the human judgment at their core couldn't be mechanized. That's what's changing. Not because AI replaces the judgment. It can't. But because AI can absorb the non-judgment work that surrounds it. The documentation. The administrative processing. The information retrieval. The prior authorizations. Work that required language comprehension and contextual pattern-matching; capabilities that until very recently only humans possessed. Those capabilities aren't what make these professionals valuable. They're the overhead that prevents these professionals from reaching the people who need them.

For the first time, the wall in service industries can come down.

What This Means for You

The scaling wall isn't just an abstract concept for policymakers and economists. It touches you every time you wait weeks for a doctor's appointment. Every time you can't afford a lawyer. Every time your kid's class has a substitute for the third week in a row. Every time your aging parent needs care and the only option is for your family to provide it.

When the wall comes down, those things change. Not overnight. Not perfectly. But meaningfully. More people get care. More people get justice. More students get the attention they need. The humans doing that work aren't replaced; they're finally free to do the real work instead of drowning in everything around it.

As a pragmatic person, I think there are probably a few ways all this technology goes badly wrong for humans. But I'm not a doomer (OK, maybe less than 5% doomer). What I do believe is that building a future of genuine abundance seems very achievable; but only if we build intentionally. If we build blindly, there will be painful outcomes. I'm passionate about this because I want my kids to understand how work is changing and how to thrive in it. I want to see my neighbors; the physical therapists, kitchen designers, contractors, and doctors; thrive too.

The Belters lived behind the wall for generations. They adapted to rationed air because that was all they'd ever known. When the ring gates opened, the ones who thrived weren't the ones who clung to the old stations. They were the ones who understood what the new abundance made possible and moved toward it. Pioneers go through ring gates into the unknown because they recognize the opportunity on the other side.

The wall is real. It's been here for decades. And for the first time, the tools to lower it exist.

Hope this helps. Reach out through the contact page if you'd like to talk about what this looks like in your industry.