The Three Economies That Built the Modern World and the Fourth That Will Replace Them

This is a companion piece to our AE Manifesto, exploring the historical arc behind our Autonomous Economy thesis.

Every few generations, the fundamental constraint on economic life changes.

When the constraint was land, we built an agricultural economy. When it was machinery, we built an industrial economy. When it was human expertise and time, we built a services economy.

Each transition followed the same pattern: it destroyed an entire way of life, created vastly more prosperity than what came before, and moved so fast that the people living through it could barely believe what was happening.

We are living through the fourth transition right now. The constraint is shifting from human time to compute, energy, and data. The result is the Autonomous Economy.

In our AE Manifesto, I described this transition and the investment thesis behind it. This piece goes deeper into the historical pattern, because understanding where we have been is the only way to see clearly where we are going.

The First Economy: Land

In 1800, roughly 83% of the American workforce worked in agriculture. Farming was not an occupation. It was the economy. If you were alive in 1800, you almost certainly grew food, or you directly served someone who did.

The constraint was land. Wealth meant acres. Power meant controlling fertile ground. The entire social, political, and economic structure of the country was organized around who owned the land and who worked it.

Then a series of inventions broke that constraint. The steel plow in 1837. The mechanical reaper in 1831, which let one farmer do the work of five. Tractors, which replaced horses on farms between the 1920s and the 1950s. Each innovation multiplied output per worker, and each one made fewer workers necessary. Between 1930 and 2000, US agricultural output roughly quadrupled while the number of inputs remained essentially unchanged.

Today, agriculture employs 1.2% of the US workforce. Direct farm output represents 0.8% of GDP. Americans spend just 12.9% of household income on food, down from over 22% in 1950.

Let those numbers sink in. An activity that consumed 83% of all human labor now requires 1.2%. And we produce more food than ever.

Where Did All the Farmers Go?

This is the question everyone asks, and the answer is not what most people expect. A landmark NBER study found that 63% of the decrease in agricultural employment happened because rural counties industrialized internally, not because workers migrated to big cities. In the most agricultural counties, 240 new towns were incorporated between 1880 and 1940. Small factory towns absorbed displaced farm workers locally. The share of the population living in urban areas within these formerly agricultural counties rose from 5% to 30%.

Manufacturing expanded from 2.5 million workers to 10 million between 1880 and 1920. The Great Migration sent 6 million African Americans from the rural South to Northern and Western cities between 1910 and 1970, driven significantly by agricultural mechanization.

The agricultural workforce did not disappear into unemployment. It was absorbed into an entirely new economic paradigm that did not exist a generation earlier. Nobody in 1850 was planning a career in automobile manufacturing or telephone installation, because those industries had not been invented yet.

The Horses Tell the Story Best

I show the 1900 and 1913 Fifth Avenue photographs at every presentation I give. One car among thousands of horses. Thirteen years later, one horse among thousands of cars. But the horse population data is even more striking.

The US horse and mule population peaked at approximately 27.5 million around 1910. By 1960, it had fallen to around 3 to 4.5 million. An 85% decline. Horses were never retrained. They were replaced.

Nobel economist Wassily Leontief saw this and wrote in 1983: the role of humans as the most important factor of production is bound to diminish in the same way that the role of horses was first diminished and then eliminated by the introduction of tractors.

That quote is 43 years old. It hits differently now.

The Second Economy: Machinery

The industrial economy shifted the constraint from land to capital. Wealth no longer meant acres. It meant factories, machines, and the capital to build them. The entire social structure reorganized around who controlled the means of production.

Manufacturing employment rose steadily through the late 1800s and peaked during World War II, when nearly 39% of all US employees worked in manufacturing (November 1943). The peacetime headcount peak came later: 19.6 million workers in June 1979, representing 22% of total nonfarm employment.

Today, manufacturing employs about 12.8 million workers, roughly 8% of the workforce.

But here is the twist that most people miss, and it is the single most important data point for understanding economic transitions:

Since 1990, the US has lost 30% of its manufacturing jobs while manufacturing output grew 148%.

Real manufacturing output hit an all-time high in 2024. America makes more stuff than ever. It just does not need as many people to do it.

The Pattern Repeats

This is the same pattern as agriculture. Employment collapses. Output soars. The steel industry illustrates it most dramatically: US steel output remained roughly flat between 1962 and 2005, but the industry shed 75% of its workforce, approximately 400,000 employees. Fewer people. Same output. Better technology.

And the cause is overwhelmingly technology, not trade. A Ball State University study found that approximately 88% of manufacturing job losses between 2000 and 2010 were attributable to productivity growth (automation, technology improvements), while only about 13% were due to trade. The long-term trend is clear: machines replace human labor, output increases, and the economy transforms.

But the Human Cost Was Real

I want to be honest about this, because I think too many technology optimists skip over the pain.

The communities that depended on manufacturing were devastated. The Rust Belt did not recover on its own. Manufacturing jobs offered something most service replacements do not: a path to middle-class life without a four-year degree. Over 60% of US adults lack a bachelor’s degree. The replacement jobs in healthcare, retail, and food service paid less, offered fewer benefits, and carried less dignity.

Research has linked manufacturing decline to the opioid epidemic, finding that factory closures predicted up to 92,000 overdose deaths between 1999 and 2017. These regional shocks were, as the researchers put it, highly persistent because of limited labor mobility.

The lesson: economic transitions create enormous aggregate wealth, but they distribute the pain unevenly. Knowing this does not mean we should resist the transition. It means we should invest in building the infrastructure that makes it work for more people.

The Third Economy: Human Time

The services economy shifted the constraint once more: from machinery to human expertise and time. Wealth now meant skilled labor hours. A doctor’s time. A lawyer’s judgment. An engineer’s design work. A teacher’s attention.

Today, services represent approximately 80% of US GDP and 79% of employment. The United States is the largest services economy in the world, at roughly $18 trillion.

And it is hitting a wall.

Baumol’s Cost Disease: The Structural Flaw Nobody Fixed

In 1966, economist William Baumol identified a devastating pattern. Labor-intensive services with low productivity growth see their costs rise inexorably faster than inflation. The mechanism is simple: wages in low productivity sectors must rise to compete with wages in high productivity sectors, but since these services cannot proportionally increase output per worker, unit costs climb endlessly.

Baumol called this the cost disease of the services economy. And the data now, sixty years later, proves he was exactly right.

Healthcare: In 1960, healthcare was 5% of GDP. In 2024, it reached $5.3 trillion, or 18% of GDP, growing at 7.2% annually, well above GDP growth. It is projected to reach 20% of GDP by 2033. The US spends more per capita on healthcare than any other developed country and achieves worse outcomes on many measures.

Education: Since 1980, college tuition and fees are up 1,200%, while overall consumer prices rose only 236%. College tuition has grown at 5.8% annually since 1983, nearly twice as fast as medical costs and four times faster than home prices. The result: $1.8 trillion in student loan debt.

Meanwhile, the cost of manufactured goods has plummeted. A television that cost several weeks’ wages in the 1960s costs a few hours today. The pattern is consistent: goods get cheaper because they benefit from automation and scale. Services get more expensive because they remain trapped by the cost of human time.

That is not a policy failure. It is a structural constraint of the services economy itself.

The Time Constraint Is Already Binding

The US currently has 6.9 million unfilled job openings (February 2026), concentrated in healthcare, professional services, and skilled trades. Healthcare alone is projected to need 2 million additional workers over the next decade. There are not enough humans to fill these positions. Research on knowledge worker productivity consistently finds that workers accomplish only 2 to 3 hours of focused, productive work per day despite 8+ hour workdays, with the rest consumed by meetings, email, and administrative overhead.

The Bureau of Labor Statistics projects total US employment growth of just 3.1% over the next decade. The economy is running out of workers to throw at problems. That is the binding constraint. And it is what makes the fourth economy not just possible, but inevitable.

The Fourth Economy: Compute

Every economic transition followed the same script. Employment collapsed in the old paradigm. Output in that sector actually increased. And the speed of each transition accelerated.

The agricultural transition took about 150 years (1760 to 1910). The industrial to services transition took about 60 years (1920 to 1980). The autonomous transition may take 10 to 20 years.

The adoption curves confirm this acceleration. The telephone took 50 years to reach 50 million users. Radio took 38 years. Television took 13 years. The internet took 4 to 7 years. ChatGPT reached 100 million users in two months. It now has over 800 million weekly active users.

The Constraint Shift

Here is the framework I use to explain this to our LPs:

• The Agricultural Economy was constrained by land. Wealth meant acres.

• The Industrial Economy was constrained by capital and machinery. Wealth meant factory access.

• The Services Economy is constrained by human expertise and time. Wealth means skilled labor hours.

• The Autonomous Economy is constrained by compute, energy, and data. Wealth means access to intelligence infrastructure.

When the constraint changes, everything built on top of it changes too. That is what is happening right now.

The Infrastructure Investment Is Historically Unprecedented

In 2015, the four major hyperscalers spent $23.8 billion in capex. In 2025, the Big Five (Amazon, Alphabet, Meta, Microsoft, Oracle) spent approximately $443 billion. In 2026, the projection is $600 to $690 billion, with roughly 75% targeting AI infrastructure specifically.

That is a 13-fold increase in a decade. Amazon alone plans approximately $200 billion in 2026 capex. Capital intensity has reached 45 to 57% of revenue at these companies, levels that resemble utilities, not technology firms.

To put this in perspective: five companies are investing approximately 2.2% of US GDP in AI infrastructure this year. That is how the physical foundation of a new economy gets built.

It Is Not Theoretical

Waymo operates 3,000 robotaxis across 10 US cities, completing 500,000 paid rides per week. They served 14 million trips in 2025 alone. Safety data shows 90% fewer serious injury crashes versus human drivers across 127 million miles. Amazon crossed 1 million robots in its warehouses in 2025, up from 1,000 in 2013, a thousandfold increase in 12 years. Aurora launched driverless trucking on US public highways. Zipline surpassed 2 million autonomous deliveries.

The METR benchmarking organization tracks how long AI agents can work autonomously on real tasks. In early 2024, the answer was about 4 minutes. By mid 2025, it was nearly 5 hours. By February 2026, it reached 14.5 hours, a full workday. The doubling time is approximately 90 to 120 days. At current rates, week-long autonomous tasks are projected by late 2026.

What History Actually Tells Us

When I present this thesis, I always get the same two reactions. Optimists say the new jobs will be better. Pessimists say this time is different. I think both are partially right, and the history helps us see why.

The Automobile Created More Jobs Than the Horse Lost

The automobile destroyed the horse economy: farriers, stable hands, hay farmers, carriage makers, street sweepers (someone had to clean up after 27 million horses). But it created an entirely new economic ecosystem. Today the US automotive industry supports over 7.25 million jobs, generating $830 billion in annual paychecks and driving $1.5 trillion into the economy. Each job in auto manufacturing creates approximately 10.4 other positions across the supply chain.

In the 1920s alone, entirely new occupations appeared that had never existed: gas station attendants, auto mechanics, insurance adjusters, driving instructors, highway engineers, motel operators, suburban real estate developers. Nobody planned these careers. They emerged from the new economic paradigm.

The ATM Paradox: Sometimes Automation Creates More Jobs

ATMs were first installed in 1969 and were expected to eliminate bank tellers. The opposite happened: teller employment grew from 268,300 in 1970 to 608,000 in 2006. ATMs reduced tellers per branch from about 21 to 13, but this made branches cheaper to operate, so banks opened 43% more of them. Total employment went up, not down.

But here is the critical nuance. Teller employment started declining after 2010. Not because of ATMs, but because mobile banking automated nearly all the remaining tasks. The lesson: technology that automates some tasks within a role can increase employment.

Technology that automates nearly all tasks eliminates it. The question for any given job is: which scenario are we in?

The Productivity J Curve: We Are in the Trough

In 1990, economist Paul David showed that the electric dynamo was invented in the 1880s but did not show up in productivity statistics until the 1920s, a 40 year lag. The reason: factories initially just swapped steam engines for electric motors without redesigning workflows. The productivity gains only came when they adopted unit drive, individual motors for each machine, which required completely new factory layouts.

MIT’s Erik Brynjolfsson extended this into the Productivity J Curve: general-purpose technologies require massive investment in process redesign, retraining, and organizational restructuring before they pay off. In the early years, measured productivity looks flat or even declines because firms are investing in intangibles. Later, when those investments are harvested, productivity surges.

Robert Solow wrote in 1987 that you could see the computer age everywhere but in the productivity statistics. The same observation is being made about AI today. But if the J Curve framework is correct, we are in the investment phase right now, and the payoff is coming.

The Engels’ Pause: A Warning About Distribution

Economic historian Robert Allen documented what he called Engels’ pause: the period from 1790 to 1840 in Britain when working-class wages stagnated while per capita GDP expanded rapidly. Between 1780 and 1840, output per worker rose 46%, but real wages rose only 12%. It took roughly 50 years before the gains of the Industrial Revolution reached ordinary workers.

We may be entering a similar period. Block’s February 2026 announcement, cutting 40% of its workforce while citing AI as the reason, targeting $2 million in gross profit per person versus $500,000 historically, may be the first major signal. The gains are real. The question is who captures them.

This is why I care so much about the infrastructure layer.

The companies that build the governance, safety, and management systems for autonomous agents do not just enable the transition. They shape how the benefits get distributed.

What Happens When the Cost of Expertise Approaches Zero

The services economy has one defining feature: the things humans need most (healthcare, education, legal counsel, financial advice) keep getting more expensive because they are bottlenecked by human time. That is Baumol’s cost disease.

But what if AI does for cognitive services what mechanization did for physical goods?

A shirt that cost a month’s wages in 1800 costs minutes of labor today. What if a medical consultation, a legal review, a personalized education plan, and an hour with a financial advisor followed the same curve?

Peter Diamandis calls the coming period the Abundance Inflection Point: the moment when AI leverages accelerating technological progress to solve foundational human challenges at scale. He argues that AI will dematerialize, demonetize, and democratize services, dramatically improving the quality of life for 8 billion people.

Vinod Khosla has argued that AI will make high-quality expertise in medicine, law, education, and finance available to everyone at near-zero marginal cost. When labor is essentially free, the cost of goods and services drops dramatically. Education becomes free. Healthcare outside of interventional procedures becomes free.

I do not know if they are right about the timeline. But the directional logic is powerful. If healthcare is 18% of GDP and rising because it is constrained by human time, and AI can break that constraint the way tractors broke agriculture’s dependence on human labor, the implications for cost, access, and quality are almost incomprehensible.

The Pattern Is Clear. The Speed Is New.

Three economies built the modern world. Each one was defined by its binding constraint. Each one was transformed by technology that broke that constraint. Each transition was faster than the last.

Agriculture: 83% to 1.2% of employment. Output quadrupled.

Manufacturing: 39% to 8% of employment. Output hit all time highs.

Services: 80% of GDP and rising. Constrained by human time. And AI is coming.

The autonomous economy is not a prediction about what might happen in 2040. Waymo is completing 500,000 rides a week. Amazon has a million robots in its warehouses. AI agents can now work autonomously for a full business day. The hyperscalers are investing $600 billion this year to build the infrastructure.

We have seen this movie before. We know how it ends. The question is not whether the fourth economy is coming. It is what we build to make it work.

At Untapped Ventures, we are investing in the infrastructure layer of the Autonomous Economy: the trust systems, the governance frameworks, the workforce management tools, and the autonomous applications that will define how this transition unfolds.

If you are building for this future, we want to meet you. 

Pitch us

Read the full AE Manifesto

George Bandarian II

Founder & General Partner, Untapped Ventures

untapped.ventures