Scarcity Was the Operating System
On what happens when the foundational assumption of every institution evaporates
Not the kind of operating system that runs on a computer. The kind that runs on a civilization. The underlying set of assumptions — so deeply embedded they are invisible — that determine how every institution is structured, how every incentive is calibrated, and how every person understands their place in the economic order.
For the entirety of modern history, the operating system has been scarcity. Specifically, the scarcity of human labor — both physical and cognitive. Every institution we have was designed for, and calibrated to, a world in which human work was the scarce input.
The tax code assumes value is created by humans working and being paid for it — individual income tax and payroll taxes account for approximately 85% of federal revenue. The mortgage market assumes humans will remain employed at roughly current income for thirty years — the U.S. residential mortgage market carries approximately $13 trillion in outstanding debt, every dollar underwritten against the durability of human earning power. The education system assumes a return to human skill acquisition. The insurance industry assumes human behavior is the primary source of risk and that it is statistically predictable at population scale. Labor law assumes that the fundamental economic relationship is between a human worker and an employer.
All of these institutions work because they are calibrated to the same underlying reality: human labor is scarce, valuable, and the primary input to economic production. What happens when that assumption evaporates?
The question is in the air. Viral macro memos imagine cascading unemployment and mortgage market collapse. Economists have begun modeling the conditions under which automation could trigger demand collapse — Alex Imas at Chicago formalized the scenario and stress-tested it, concluding the extreme conditions are unlikely but the demand-side forces are real. The anxiety is not confined to speculation — recent evidence suggests AI is already affecting junior-level employment in specific professions, and the timeline debates have shifted from “whether” to “when.”
Some of this discourse is overwrought. But the underlying concern — that institutions calibrated to labor scarcity may fail when that scarcity disappears — is serious enough to deserve a serious answer. Doom scenarios are easy to write and, paradoxically, not very useful for governance. What is useful is understanding mechanism — the specific institutional dependencies that make the transition dangerous — because the mechanism is what you can intervene on.
The most sophisticated counterargument to displacement anxiety came from Frank Flight at Citadel Securities, who marshaled real data against the doom scenarios circulating earlier this year.
Flight makes several strong points. AI adoption, measured by the St. Louis Fed’s Real Time Population Survey, shows no inflection in daily workplace use — the trend is stable, not exponential. Software engineer job postings are up 11% year-over-year. Construction hiring is booming. New business formation is expanding. Flight invokes Keynes’s 1930 prediction of the fifteen-hour workweek: productivity gains don’t produce leisure — they produce SUVs, iPhones, streaming subscriptions, and new consumption categories that didn’t previously exist.
Flight is probably right that the displacement cliff is not imminent. The data does not show one. AI capex is creating jobs even as AI capabilities threaten them elsewhere. His analysis may hold for the rest of the decade.
But Flight's analysis has a characteristic blind spot, and the blind spot is structural, not personal. Financial analysis operates on the timescales that financial instruments care about — quarters, years, the near-to-medium term where tradeable positions exist. Flight's evidence — software engineer postings, AI adoption surveys, construction hiring — is entirely about the digital economy over the next two to three years. His implicit model is one in which displaced white-collar workers shift into physical and service work, new business formation absorbs slack, and fiscal transfers smooth the transition. Physical labor is the absorptive buffer. The escape valve.
Flight is not alone in this assessment. Oren Cass, writing in the American Compass collection For Whom the Machine Toils, arrives at the same near-term conclusion from the opposite direction. Where Flight argues that displacement is not happening, Cass argues that the real problem is too little automation — the United States lags Germany, Japan, Korea, and now China in robot density, productivity growth has stalled, and workers suffer not from machines taking their jobs but from insufficient investment in the machines that would make their labor more valuable.
Visit a Schaeffler auto-parts plant in Cheraw, South Carolina — a town of five thousand that already lost its textile industry to NAFTA — and you can see what Cass means. A humanoid robot carries baskets of bearing components eight hours a day. The worker who had the job was moved to inspection. A “Now Hiring” sign still hangs outside. This is not displacement. It is a company solving a staffing problem with a technology that is, for now, more expensive and less capable than the person it replaced.
Cass is right that the United States needs more of this.
But the diagnosis contains its own warning. Countries that automated gradually — Japan over four decades, Germany over three — built institutional capacity alongside deployment. Safety practices, workforce pipelines, liability frameworks evolved incrementally as the machines arrived. The United States has not done this. And the timeline is compressing. In Shanghai, a factory automation CEO who supplies General Motors told the Guardian’s Chang Che that he expects deep learning to make full automation of final assembly achievable by the mid-2030s — a goal that defeated Tesla’s Alien Dreadnought attempt less than a decade ago. When Che pressed him on what happens to displaced workers, the CEO said higher-skilled employees could train the next generation of robots. He did not say what would happen to the lower-skilled ones.
Cass frames the technology question as distributional — his analysis reduces every automation decision to cui bono, who captures the gain. That is a real question. But distribution depends on governance, and governance is where the body forces the issue.
Both Flight and Cass model a world in which physical labor is the escape valve. Displaced product managers drive for Uber; displaced consultants work in warehouses; the service sector absorbs the cognitive-labor surplus, albeit at lower wages. This assumption is the load-bearing wall of their analyses. Without it, displacement has no floor.
Robotics removes the floor.
But the convergence thesis I previously wrote about is about capability, not deployment. The labs are building both minds and bodies. Physical deployment, however, faces frictions that cognitive AI does not — robots must navigate spaces not designed for them, clear liability questions that software avoids, and move through supply chains that bits don't need. Cognitive displacement will arrive first. The escape valve exists.
But it is finite. Years, not months — but not forever.
Eventually, displaced analysts will not be able to become warehouse workers — warehouse robots will be doing the picking. Eventually, displaced lawyers will not be able to become delivery drivers. The question is whether “eventually” is ten years away or twenty, and whether the institutions that need to adapt will have used the window or squandered it.
The strongest objection to this thesis is genuinely powerful: the next sector cannot be named because it does not exist yet. People in 1900 could not have named “software engineer” or “social media manager” as future jobs. New work categories have always emerged that were unimaginable before the transition that created them. This objection has centuries of evidence behind it.
But every historical transition that created new work categories required a physical substrate — a domain where human bodies could do things machines could not. The transition from agriculture to manufacturing absorbed workers because factories needed hands. The transition from manufacturing to services absorbed workers because offices, stores, hospitals, and restaurants needed human presence and human dexterity. The new categories were unimaginable in advance, but they shared a common feature: they required human bodies to be somewhere, doing something physical that machines of the era could not do.
If cognitive AI displaces cognitive labor and physical robotics eventually displaces physical labor, that common feature disappears. The question is not “what new jobs will emerge?” but “what will those jobs require that neither a digital mind nor a physical body made of metal can provide?” The people who invoke the historical pattern cannot answer this either.
Mass physical labor displacement is not imminent — not in 2028 or even 2030. Current robotics systems are not yet reliable or general enough. But the trajectory is accelerating. The IFR projects global robot installations will surpass 700,000 units per year by 2028. At some point during this progression — I would guess between 2032 and 2038, with wide uncertainty bounds — robotics systems will be capable and cheap enough to displace a significant fraction of physical labor in advanced economies.
The mortgage market dependency illustrates why the timing matters. Every mortgage is a bet that the borrower will remain employed at roughly current income for thirty years. Underwriting models encode this through credit scores, debt-to-income ratios, and employment verification, all calibrated to a world where job loss is cyclical — people get laid off in recessions and re-hired in recoveries.
AI-driven displacement is structural, not cyclical. A borrower who loses their job to automation is not going to be re-hired when the economy recovers, because the job no longer exists. The closest analogue — the localized devastation of single-industry towns when the industry leaves — has always been geographically contained. AI displacement is geographically distributed, concentrated in the income brackets that hold the largest mortgages. The underwriting assumptions embedded in $13 trillion of outstanding debt are quietly becoming less reliable, and nobody in the system has a framework for assessing how unreliable they have become.
And even when robots become more economically efficient than humans for a given task, economic efficiency will not settle the question. Communities that watched the last transition — and remember the promises that were broken — will demand more than efficiency gains. They will demand a stake.
The institutional mechanisms matter, but the political economy may matter more.
Every previous wave of technological displacement produced backlash, from the original Luddites through the anti-globalization movements of the 2000s. Robotics may very well produce backlash that is more visceral, more localized, and more actionable than anything provoked by software automation, because robots are visible. A language model that displaces a paralegal is invisible to the public. A robot that replaces a warehouse worker is a physical object that people can see, resent, and — as the historical record confirms — destroy.
The available regulatory obstacles are enormous. Zoning laws, occupational licensing, liability standards, workplace safety regulations — every one of these can be weaponized to slow robotics deployment, and the political incentives to weaponize them will be intense. Datacenters have already faced aggressive local opposition. Robots will face opposition that is orders of magnitude more personal, because the displacement is direct and visible.
Unless the people affected are presented with a social contract that makes the mass rapid adoption of robotics a benefit for them personally — not in the abstract long-term sense, but in the concrete sense of income, security, and dignity — the backlash will strangle deployment through exactly the regulatory channels that make the United States slow to adopt physical technology. And the deployment paradox means that strangled deployment produces not safety, but strategic disadvantage.
This is the political economy trap: the countries that can suppress backlash (because the state can override it) will deploy fastest and accumulate the most data. The countries that cannot suppress backlash (because the democratic process empowers the displaced) will deploy slowest, unless they solve the social contract problem. A credible social contract for robotics deployment is not a progressive policy preference. It is a strategic necessity.
Operating systems do not crash all at once. They degrade. Features that used to be reliable become intermittent. Processes that ran smoothly start producing errors. Each individual failure looks like a bug, not a system-level problem. And by the time you realize the operating system itself is failing, the migration is already an emergency.
The question is not whether the migration will happen. It is whether we will use the window — the years between cognitive displacement and physical displacement — to build the institutions that the transition requires. The countries that use the window will deploy robotics with legitimacy. The countries that squander it will face a choice between backlash and strategic irrelevance.
The window is open. It will not stay open.