The Ladder Is Being Eaten

For two hundred years, one piece of advice survived every wave of disruption: climb higher. The Industrial Revolution ate muscle, and cognitive work became the refuge. The digital revolution ate clerical work, and knowledge work became the answer. Study harder. Credential up. The next generation will be richer than this one. For a very long time this advice was not aspiration. It was empirically correct.

Artificial intelligence breaks the advice at its root. It does not displace one kind of labour while raising the value of another. It substitutes for cognitive labour itself: analysis, judgment, creativity, translation. These are the very rungs that displaced workers were always told to climb toward. When the rung above is the thing being automated, the instruction to climb stops meaning anything.

The mechanism that always made automation politically tolerable stops working when the ladder is what is being eaten.

Two Hundred Years of Climbing the Skill Ladder

The skill-climb bargain was real, and it was durable. Each major automation displaced a category of work and, in the same motion, created a higher-value category for displaced workers to move into. Looms displaced weavers, but the factory economy demanded clerks, managers, and engineers. The computer displaced the typing pool, but it rewarded analysts, programmers, and the credentialed professional class. The pattern held because the new work sat above the old work on a ladder of skill, and human beings could climb that ladder faster than machines could follow.

Goldin and Katz framed this as a race between education and technology. When education won the race, the gains from automation spread widely and inequality compressed. When technology pulled ahead, the gains concentrated and inequality widened. The implicit promise of the whole arrangement was that education could, in principle, keep pace. Workers could acquire the next tier of skill before automation reached the tier they were standing on. That promise rested on a quiet assumption about timing, and it is the timing that has now changed.

The assumption was that a serious cognitive skill, once acquired, would retain its value long enough to repay the years invested in learning it. For most of the industrial and digital eras, that held. A skill could be acquired faster than it depreciated. The ladder was stable enough to climb. What follows is the arithmetic of why that stability has ended.

The Depreciation-to-Acquisition Ratio

Begin with a simple question. How long does it take a person to acquire a serious cognitive skill? Roughly five years: an undergraduate degree plus the early-career experience that turns a credential into competence. Now ask the inverse. How long does it take for AI to substantially devalue that same skill? Two to four years per capability generation, per domain.

Put those two numbers together and you get a ratio of depreciation to acquisition of approximately 1.7 times. Skills now lose their value faster than people can build them. Crucially, this ratio crossed 1.0 somewhere between 2020 and 2024. In the Goldin and Katz era, the same ratio sat comfortably below 1, which is precisely why the climb-higher bargain worked. Skills could be acquired faster than they depreciated, so investment in human capital paid off. That arithmetic has now reversed.

A ratio above 1 does not announce itself in any single career. It works statistically, across cohorts, as the expected return on years of cognitive training quietly turns negative for an expanding set of domains. The macroeconomic signature shows up in the labour share of national income. In the United States, labour share of GDP stands at 58 percent today. In 1980 it was 67 percent. On current trajectories it is projected to fall toward roughly 40 percent by 2050. That is not a rounding adjustment. It is a structural redistribution of national income away from labour and toward capital, and it is the most reliable single number for watching this transition unfold.

Five Trajectories

The interesting question is not whether this transition happens. The arithmetic above suggests it already has begun. The interesting question is how societies respond, because the technology underdetermines the outcome. There are five plausible end-states, and they look very different from one another. Each carries a rough probability weight, and the weights matter more than any single forecast.

The most likely current trajectory is B with elements of D, evolving under political pressure toward A or C, with E as a persistent tail risk. The point of the table is not to pick a winner. It is to make explicit that the binding constraint is not the technology. The binding constraint is the political and institutional response, and the spread between the best and worst weighted outcomes is enormous.

This matters for anyone allocating capital or planning a workforce, because most portfolios and most plans implicitly bet that Trajectory B simply continues. That is a concentrated bet on a single column of a five-column table. The disciplined posture is to diversify against it: to hold positions that survive A and to hedge E explicitly, rather than assuming the current path extends indefinitely.

What to Track Now

The transition announces itself in data well before it reaches headlines, which is the practical reason to watch a small number of series closely. Four are worth quarterly attention. Audit workforce planning against AI substitution timelines by function rather than by industry, because the granularity is where the signal lives. Structurally overweight the capabilities AI cannot easily substitute: embodied skill, judgment under genuine uncertainty, and relationship capital. For capital allocators, diversify away from the implicit Trajectory B bet. And track the macro series that move first: labour share of national income, the corporate profit share of GDP, and prime-age employment.

The mechanisms for Trajectory A are not speculative. They are documented and reproducible. The Nordic countries sustain a labour share of 60 to 64 percent through coordinated wage bargaining, active labour market policy, and high taxation of capital. What that arrangement requires is not a technological breakthrough. It is political will. History suggests that this kind of will tends to arrive during crises rather than before them, and the arithmetic suggests the crisis arrives in the 2030s if nothing changes. The window to act ahead of it is still open, but it is finite.

The binding constraint is not the technology. It is the political and institutional response.

For a young, fast-developing economy, the lesson is less about defending an existing labour share than about deciding what kind of one to build. An economy still composing its institutions has a degree of freedom that mature economies have already spent. It can design active labour market policy, capital taxation, and human-capital investment with the depreciation-to-acquisition ratio already in view, rather than retrofitting them after a crisis. The two series most worth watching at home are the same two that move first elsewhere: the labour share of national income, and prime-age employment. Held qualitatively, the question for the Gulf is whether a generation entering the workforce now climbs a ladder that is still standing, or one that is being eaten from the top. That outcome is not decided by the technology. It is decided by the institutions built around it.