The Abundance Thesis

Economists hear the word “deflation” and reach for the panic button. Which is understandable, because the deflation they usually encounter is the demand-side kind: prices fall because nobody is buying anything, because the economy is collapsing. That kind of deflation is genuinely terrible.

But supply-side deflation — prices falling because things become dramatically cheaper to produce — is a completely different machine. It is, in fact, the engine of almost all human economic progress. The price of light has fallen by a factor of roughly one thousand over two centuries. Computing power has halved in cost every eighteen months for five decades. Sequencing a human genome went from three billion dollars in 2003 to under two hundred today. Every one of these was a deflationary event. Every one of them made us, unambiguously, richer.

AI is producing supply-side deflation across a category of human activity that has never been significantly deflationary before: knowledge work. A task that required a team of ten can now be done by one person with the right tools. A project that took three months can take three weeks. A capability that was only available to large businesses is now available to anyone with a laptop and a monthly subscription.

But AI-driven deflation does not stop at the borders of the knowledge economy. China’s so-called dark factories — automated facilities that operate with minimal human presence, lights off, no heating required for a workforce that isn’t there — are not a curiosity. They are the opening act of a wholesale transformation in how physical goods are made. A robotic manufacturing line does not take breaks, does not require benefits, does not need a canteen. It runs continuously, at consistent output quality, and the cost of operating it falls as the technology improves.

Labour, across most categories of manufacturing, is the dominant cost. Remove it by an order of magnitude, and the price of manufactured goods does not nudge downward. It falls through the floor.

For thirty years, offshoring kept goods prices flat while services kept climbing. AI is the first technology in history that attacks services deflation directly.

This is the first time in history that deflation hits both sides simultaneously. AI compresses the cost of services embedded in every product; robotics compresses the cost of making the product itself. For the first time, both input categories are deflationary at once.

The knowledge economy is built on a specific logic: certain types of intellectual work require enough specialisation and collaboration that they’re best performed by large groups of people under a single roof. The law firm, the consulting firm, the investment bank — these are factories for producing intellectual output. I realise nobody in a law firm wants to hear that, but it is.

AI dissolves the constraints that made those factories necessary. One person with capable AI tools can now research, synthesise, model, design, write, and build at a level that previously required a team. The constraint was never human intelligence. It was human time. AI extends human time.

What replaces the large knowledge-work employer is not unemployment. It is a proliferation of small, highly capable independent operators. Platforms like Upwork, which currently sit at the margins of serious professional work, are positioned to become central to how the economy allocates talent in the 2030s. The two problems that historically prevented solo professionals from operating at scale were finding clients and establishing trust. Platforms dissolve both. A verified track record, a portfolio of rated engagements — these are trust signals that transfer. Freelance will not merely survive the AI transition. It will emerge as one of the defining work arrangements of the decade.

The US alone formed 5.5 million new businesses in 2023 — a record high, 48% above 2019 levels. Each annual cohort of startups, at current survival rates, generates approximately 24 million jobs by year five. As AI improves both the survival odds and the growth rate of small firms, that figure rises.

One person with AI tools produces what ten people produced before. That is not a threat to employment — it is a transformation of what employment means. The output doesn’t disappear. It gets redirected. The person who once managed a team of ten now operates alone, at equivalent output, and the nine others find new work in the expanding economy that cheaper services create.

AI cannot lay a floor. It cannot wire a house, fit a kitchen, service a boiler, fix a pipe, build a wall, or pour a concrete foundation. Your robot vacuum still gets stuck under the sofa. We are not close.

But this framing misses the actually important point. The blue collar trades are not merely safe from AI displacement. They are positioned to benefit enormously from it. And the mechanism is straightforward.

Consider what happens when AI makes the rest of the economy run significantly faster. Businesses grow faster. Capital deployment speeds up. Consumer spending power increases. The economy, in aggregate, moves faster. A faster economy demands more of everything. More offices, more warehouses, more restaurants, more shops — all of which need to be built, fitted out, maintained, and periodically renovated.

Now apply basic economics to physical labour in this environment. Demand for blue collar work increases. Supply is constrained — you cannot produce a skilled plumber in three months. Inelastic supply plus rising demand produces one outcome: rising prices. Wages go up. Significantly. Not as a matter of ideology, but as a matter of market mechanics.

The trajectory has been visible for several years in the UK. Electricians, plumbers, and heating engineers already command rates that embarrass many professional salaries.

The evidence is already visible. UK electricians and plumbers now command rates that embarrass many graduate starting salaries. The best apprenticeships in desirable trades are genuinely competitive. A young person entering a skilled trade today is, by most financial measures, making a better decision than a peer entering a generic university degree. AI will not reverse this trend. It will dramatically accelerate it.

The 82 million professional driving jobs that AI threatens are a useful example of how to read this more carefully. 36 million of those drivers retire naturally by 2040 — the industry was already facing a driver shortage before a single autonomous vehicle hit the road commercially. The honest displacement figure is 46 million jobs over 15 years. Spread globally, in the economies most likely to adopt AV fastest — China, Europe, Japan, all of which face shrinking working-age populations — autonomous vehicles are not punching a hole in the labour market. They are filling one that demographics had already started digging.

There is also a question the displacement narrative conveniently ignores: who monitors the machines? The original version of this thesis assumed a remote-operator ratio of roughly 1 to 250 — one human overseeing 250 autonomous vehicles. That was wrong. Waymo's own operation reveals the real number: 70 remote assistance agents supervising a fleet of 3,000 vehicles across four centres in Arizona, Michigan, and the Philippines. That is one operator for every 41 cars — six times more human-intensive than the optimistic projection. And these are not passive observers. They handle blocked lanes, unexpected construction, collisions, and regulatory interactions that the AI cannot safely navigate alone. Even with 127 million autonomous miles and a 90 per cent reduction in serious-injury crashes, the system still leans on people for the messy, unpredictable edges of driving that algorithms have not solved. Scale that ratio across 2 billion vehicles globally and even at a conservative 1 in 100 — allowing for significant efficiency gains as the technology matures — autonomous vehicles create roughly 20 million new remote-operator, fleet-management, and maintenance jobs. The displacement is real. But so is the creation, and it is far larger than anyone initially assumed.

Professional Services Become a Consumer Good. Legal advice, financial modelling, architectural drafting, medical second opinions — all of these become accessible at a fraction of their current cost. The professional class does not vanish. But the barrier to accessing professional-quality output drops to near zero. What was once a luxury becomes a utility.

The Solo Operator Economy Matures. By 2030, a significant share of the professional workforce operates independently. The platforms that connect them to clients have matured. Trust signals are robust. Payment infrastructure is reliable. The solo operator is not a freelancer on the margins — they are a serious economic actor, often earning more than their salaried peers, with more autonomy and lower overhead.

Apprenticeships Become Competitive with Degrees. As the premium on physical work rises and the premium on generic knowledge work falls, the financial calculus of education shifts. A young person who enters a skilled trade at 18 — earning immediately, accumulating no debt, building equity through work — finds themselves financially ahead of their university-educated peers well into middle age.

Energy Becomes the Binding Constraint. AI’s growth is physically dependent on the workers it supposedly threatens. The data centres that power AI require enormous amounts of electricity. The infrastructure to generate, transmit, and distribute that electricity requires enormous amounts of physical labour. Every new data centre is a construction project. Every solar farm, every wind installation, every grid upgrade — these are blue collar jobs, created by the very technology that was supposed to eliminate them.

The Transition Period Remains Uneven. None of this means the transition is painless. It will not be. Some regions will adapt faster than others. Some individuals will find the shift wrenching. Policy matters — education systems, retraining programmes, social safety nets all need to evolve. But the direction of travel, for a society willing to navigate the transition intelligently, is toward more abundance, not less.