The Fastest Thing We've Ever Built

I. The Line That Held Through Everything

There is a graph that economists love almost religiously. It shows real income per person in the United States over the past 150 years. On a logarithmic scale, it is nearly a perfect straight line: two percent growth per year, year after year, decade after decade, crisis after crisis.

What makes the graph remarkable isn't the line. It's what happened around it.

The steam engine. Electrification. The internal combustion engine. Antibiotics. Semiconductors. The personal computer. The internet. Each one arrived promising to change everything — and each one did, profoundly, irreversibly. Factories were redesigned. Cities were rewired. Entire categories of work vanished and new ones appeared from nowhere. The world became unrecognizable across a single generation.

And yet: the line held. Two percent. Always, stubbornly, two percent.

Economists have a hypothesis for why. Each transformative technology, however radical, eventually runs into the limits of the tasks it cannot perform. Automate the easy half of a production process, and output is still capped by the hard half. The chain breaks at its weakest link. Progress is real, but it is bounded — absorbed into the long, patient slope of that straight line.

Then came AI.

And for the first time in the history of that graph, serious economists are debating whether the line itself might bend.

II. A Different Kind of Technology

Every previous revolution automated something. Looms automated weaving. Engines automated movement. Computers automated calculation. Each one was extraordinary and each one was, at its core, a specialist.

Artificial intelligence is not a specialist. It automates cognition itself — the capacity to reason, to synthesize, to create, to decide. That is a different category of technology. Not a faster loom. Not a better engine. Something structurally unlike anything that came before.

Consider what this means in practice. When AI raises the productivity of software engineers, those engineers build better AI, which raises productivity further, which builds still better AI. The feedback loop is not metaphorical — it is already measurable. Eighteen months ago, AI could reliably handle tasks equivalent to roughly nineteen minutes of skilled human work. Today that ceiling sits at five hours of human-equivalent work, and it has been doubling every five to seven months. The horizon isn't just moving. It's accelerating away from us.

This is why the debate is no longer about whether AI will have large economic effects. It will. The debate — the genuinely interesting one — is about the shape and speed of those effects. And that debate runs in two directions at once.

III. What Slows It Down

Hypothesis one: The diffusion gap. History is littered with technologies that were revolutionary in the lab and invisible in the statistics for decades. The electric motor was commercially available in the 1880s. It didn't reshape factory productivity until the 1920s, when engineers finally understood that you had to redesign the entire factory around the motor, not just swap it in for the steam engine. The computer was everywhere by 1987. Robert Solow noted that same year that you could see the computer age everywhere but in the productivity statistics. AI is almost certainly subject to the same delay. Institutions, workflows, regulations, and human habits are slow. Technology is fast. The gap between them is where growth goes to wait.

Hypothesis two: Regulation as asymmetric friction. The world does not have a single legal framework for AI. It has dozens, diverging rapidly. The European Union's AI Act imposes compliance layers that slow deployment, particularly for high-stakes applications in finance, healthcare, and hiring. The United States operates under a more permissive, sector-by-sector approach that tolerates more experimentation but also more risk. China moves under state direction — fast in some verticals, constrained in others. Emerging economies often have no specific framework at all, which is simultaneously a freedom and a vulnerability.

This fragmentation matters enormously for growth. A pharmaceutical company developing an AI-assisted drug discovery pipeline faces a different regulatory surface in Frankfurt than in Houston than in São Paulo than in Bangalore. The technology may be identical. The timeline to deployment is not. Countries that develop clear, intelligent, innovation-friendly AI governance frameworks early will attract capital, talent, and the compound gains that follow. Those that regulate by fear or by inertia will import the results — years later, at a premium.

Hypothesis three: The measurement problem. GDP was designed to count what economies produced in the 20th century. It counts cars and steel and hours worked. It struggles with things like the value of a medical diagnosis that didn't happen because an AI flagged the risk six months earlier and the doctors took preventive measures. It has no good instrument for the productivity of a three-person company that, with AI, now produces what previously required forty people. If AI's largest gains are in quality, access, and precision — not just volume — then the most important revolution in economic history might be substantially invisible to the tools we use to measure economic history. The line might be bending already. We might just be looking at the wrong graph.

IV. What Speeds It Up

Hypothesis four: The engine that builds the engine. Every previous general-purpose technology raised productivity in specific sectors. AI raises the productivity of the process by which new ideas are created. Research and development. Scientific discovery. The generation of new knowledge. When DeepMind's AlphaFold mapped the structure of more than 200 million proteins — a problem biologists had spent decades on — it didn't just solve a scientific puzzle. It didn't accelerate pharmaceutical research. It detonated it. That is not productivity growth inside a sector. That is acceleration of the frontier itself. If AI continues to do this across materials science, energy, medicine, and engineering simultaneously, the compounding effects are qualitatively different from anything in the 150-year graph.

Hypothesis five: Leapfrog economies. There is a pattern in economic history that rarely gets the attention it deserves: the advantage of arriving late. Countries without entrenched landline infrastructure adopted mobile phones faster than those with it. Countries without legacy banking systems deployed mobile payments at scale while the developed world was still debating interoperability standards. The same dynamic applies to AI. Economies without legacy institutions — educational systems built around memorization, legal systems built around paperwork, healthcare systems built around geographic proximity — have less to protect and less to dismantle. A smallholder farmer in the Argentine interior who gets agronomist-level crop advice from a phone is experiencing an AI economic effect that will never show up in a model built around knowledge-worker productivity in San Francisco. The gains are real. The map just doesn't include that territory.

Hypothesis six: The democratization of expertise. For most of human history, access to high-quality expert advice — medical, legal, financial, technical — was rationed by geography and wealth. You had to live near the right city and be able to pay the right rates. AI breaks that rationing mechanism. Not perfectly, not instantly, not without new risks. But the direction is clear. When expertise becomes cheap and widely accessible, the economic effects are not linear. An entrepreneur in Lagos who can now draft contracts, analyze competitors, write code, and model financial projections without a team of specialists is not marginally more productive. She is categorically more capable. Multiply that across millions of people who were previously locked out, and the aggregate growth effect is substantial — and largely invisible to traditional measurement.

Hypothesis seven: Energy times intelligence. The physical economy still runs on energy. And AI is beginning to do for energy what it did for protein folding: compress the timeline on discoveries that would otherwise take generations. AI-optimized power grid management, accelerated battery chemistry research, materials discovery for next-generation solar cells, renewed momentum in nuclear fusion — these are not speculative. They are active research programs with measurable results. If the energy constraint on the physical economy loosens within the next fifteen years, every sector that depends on energy — which is all of them — gets a simultaneous multiplier. The thrust doesn't add. It compounds.

V. The Legal Frame Is Not a Detail

It is worth sitting with this longer than most AI commentary does.

The regulatory environment is not a footnote to AI's economic trajectory. It is one of its primary variables. The same technology, operating under different legal frameworks, produces different economic outcomes — not marginally different, but dramatically different over a decade.

The EU's precautionary architecture protects citizens from certain risks. It also exports the first-mover advantage in AI applications to jurisdictions willing to move faster. The US permissive model accelerates deployment but creates liability landscapes that are still being invented in real time. China's state-directed model concentrates AI gains in specific strategic sectors while constraining the distributed innovation that historically generates the most unexpected economic value.

And then there is the rest of the world — Latin America, Sub-Saharan Africa, South and Southeast Asia — where AI governance frameworks are either nascent or absent. This is not simply a gap. It is a choice, being made by default, with consequences that will compound over the next twenty years. Countries that establish intelligent, proportionate, innovation-forward AI regulation now are not just managing risk. They are making an economic bet with generational stakes.

The hypothesis: legal architecture is becoming as important to economic growth as physical infrastructure once was. In the 20th century, you needed ports and roads and power grids. In the 21st, you need regulatory frameworks that let intelligent technology move at intelligent speed.

VI. The Ownership Question Nobody Asks Loudly Enough

Here is a hypothesis that sits at the intersection of economics and ethics, and deserves to be stated plainly.

If AI creates substantial new economic value — and the evidence suggests it will — then the distribution of that value is not a technical question. It is a political and personal one. Historically, the gains from technological transformation have flowed to those who owned the means of production: the capital, the infrastructure, the platforms.

The difference today is that ownership is, in principle, accessible.

Public markets exist. The companies building and deploying AI at scale are, many of them, publicly traded. The infrastructure that runs the AI revolution — semiconductors, data centers, cloud platforms, energy — is investable. This is not a guaranteed path and it carries real risk. But the structure of the opportunity is different from, say, the Industrial Revolution, when factory ownership was not something a working family in Manchester could participate in.

The hypothesis: the most equitable version of the AI economy is one in which the gains are widely distributed through ownership, not just through employment. Not because markets are perfect instruments of justice — they are not — but because the alternative is a world in which the productivity gains flow to a small number of entities while the displacement effects are distributed broadly. Understanding this dynamic, and acting on it deliberately, is not a financial strategy. It is, at its core, an act of conviction.

VII. The Open Verdict

So: will the line bend?

The honest answer is that nobody knows. The forces accelerating AI's economic impact are real and measurable and in several cases already compounding. The forces slowing it down are equally real — institutional, regulatory, structural, human. The race between them is not theoretical. It is happening now, in every sector, in every jurisdiction, at every level of the economy.

What is clear is that the frame matters. If you measure AI's impact with 20th-century instruments, you will undercount it. If you assess it only through the lens of what it cannot yet do, you will miss what it is already doing. If you assume the diffusion timeline from previous technologies applies without modification to a technology that can accelerate its own development, you may be waiting for a lag that never arrives at the expected length.

The two percent line held through everything we built before. It held because every previous technology, however powerful, was ultimately bounded by the tasks it could not perform.

AI is the first technology we have built that is in the business of shrinking that boundary.

Whether it shrinks it fast enough, and broadly enough, and equitably enough — that is not an engineering question. It is the central political and economic question of the next fifty years.

You are already living inside the answer. You just don't know yet which scenario you're in.