Throughout human economic history, the difference in earning potential between two people has always been determined by some dominant arbitrage — a scarce human advantage that one person could leverage over another. Technology has repeatedly destroyed these arbitrages, one by one. Each time, the axis of competition shifted to something new.
The frame is useful because it explains both the anxiety of the AI moment and the opportunity in it: we are watching the most powerful arbitrage of the modern era — cognitive ability — get commoditized. What comes next is partly predictable.
A history of arbitrage decay
Strength
For most of civilization, physical strength was the primary determinant of economic value. The stronger you were, the more land you could farm, the more goods you could haul, the more buildings you could raise. Armies were won by the side with more capable bodies.
Killed by: Machines. Once a steam engine, a tractor, or a crane could outproduce the strongest human, the arbitrage collapsed. Today, being able to deadlift 300 kg is a hobby, not an economic strategy.
Knowledge
With machines handling labor, knowledge became the differentiator. But knowledge was gatekept — to learn astronomy you needed an astronomer; to learn medicine, an apprenticeship. Geography and social class determined access. A brilliant mind born in a remote village had almost no way to realize that potential.
Killed by: The printing press, then libraries, then universal education, then the internet. Knowledge became abundant. You no longer needed a master; you needed a library card.
Mental computation
Even with widespread knowledge, some people had genuine advantages in mental arithmetic, spatial reasoning, and rapid calculation. Banks, trading floors, engineering firms rewarded computational speed.
Killed by: The calculator, then the computer. Once a $5 device outperformed the fastest human at arithmetic, there was no premium for it.
Information access (memory and research)
Doctors who recalled rare diagnoses, lawyers who remembered case precedents, consultants who pulled the right framework from memory — these people commanded premiums for superior information retrieval.
Killed by: The search engine. The premium for encyclopedic memory collapsed. A junior with Google was nearly as informed as a 30-year expert.
Geographic reach
A craftsman in a remote village could only sell to those within walking distance. A restaurant on a busy main street outperformed an identical one in a back alley. Location was destiny.
Killed by: Globalisation and the internet, in waves. Shipping expanded reach. The internet obliterated geographic constraints for digital goods. Uber Eats and DoorDash killed the location arbitrage for restaurants — a back-alley restaurant paying low rent could now reach the same customers as the main-street one. Cloud kitchens disrupted that. Dark stores disrupted supermarkets.
Pattern recognition and synthesis
The current arbitrage. Two people with the same computer, internet, and search engine produced wildly different outcomes based on their ability to recognise patterns, connect dots, and synthesize. The analyst who spots the anomaly. The investor who reads the same earnings report but sees something others miss. This is the world of the last 15–20 years.
Killed by: Large language models. Pattern recognition, data interpretation, logical reasoning, even strategic analysis are increasingly being performed at superhuman levels. The premium is collapsing on a timeline of months.
Why each common “but what about…” objection fails
”Taste will be the moat”
Argument: AI can execute, but someone has to decide what’s worth building. The Steve Jobs function.
Why it fails: Taste was only valuable because each iteration was expensive. In the 1960s, an ad campaign cost millions and months, so legendary copywriters mattered. Today, A/B testing replaces visionary copywriting — produce 10 variants, scale the winner. AI accelerates this to the limit. When iteration cost drops to near zero, vision is replaced by volume and measurement.
”Product visionaries will still matter”
Argument: Someone has to know what people want.
Why it fails: The product-visionary role exists because building is expensive — you have to guess right because you can’t afford to build 50 versions. When building becomes radically cheap, people will hyper-personalize their own tools rather than buying one that meets 80% of their needs. The role of the product visionary was to aggregate demand across millions. When each user has their own version, there is no demand to aggregate.
”Capital will be the durable advantage”
Argument: Whoever owns the AI, compute, and robots captures the returns.
Why it fails historically: Capital has not persisted across technological waves. It has been recreated by new entrants each time. Kings → merchants → bankers → industrialists, and so on. None of today’s billionaires (Ambani, Musk, Gates, Zuckerberg, Page, Brin) inherited the previous era’s wealth. Each wave creates new capital holders; it does not enrich the old ones.
Two kinds of people who make money
Re-examining the historical pattern reveals two distinct economic roles:
The arbitrage winner. The stronger person, the more knowledgeable doctor, the better-located restaurant, the analyst with sharper synthesis. They optimise for being the best at the current game.
The arbitrage killer. The person who made the machines, the printing press, the search engine, the Uber Eats, WordPress, YouTube, GitHub Pages. They make the current game irrelevant.
The arbitrage killers historically captured more value than the winners. The strongest worker in 1900 made decent money; the person who built the machine that replaced him built an industry.
This matters because the second path is actionable right now, while the first depends on guessing the next arbitrage correctly. You don’t need to predict the future to build a tool that kills a current arbitrage. The current cognitive arbitrage has hundreds of small subarbitrages embedded in it (legal research, financial modelling, content writing, code review). Killing any one of them is a defensible bet.
Can we predict the next arbitrage?
A common defeatist conclusion is “we can’t know what’s next.” That’s only partly true.
A Roman soldier in 100 AD genuinely could not have predicted that future arbitrages would involve transistors and electricity — the prerequisite concepts didn’t exist for him. But the visionaries at each transition often did see it: Edison and Tesla saw electricity. Gates said “a computer on every desk.” Bezos saw e-commerce. They were positioned at the frontier and could read the signals.
The frontier-reading strategy works because being close to where capabilities are developing makes the next layer visible months or years before it’s visible to the general public. See Spatial Intelligence as Next Arbitrage for one specific candidate that emerges from this kind of frontier-watching.
Two strategies that survive every wave
If you take the framework seriously, two strategies hold up across every historical transition:
- Build the tool that kills the current arbitrage. Most current “AI knowledge work” tasks have a current human arbitrage worth killing. Whoever ships first usually wins outsized returns.
- Position at the frontier of the next capability. See Spatial Intelligence as Next Arbitrage for the leading specific bet.
What does not work: optimising deeper into the arbitrage that’s about to die. Spending years sharpening “data analysis” or “decision-making frameworks” right now is the equivalent of perfecting sword-fighting in 1900 — admirable but economically irrelevant.