Suppose a model called Sable appears tomorrow morning: not a machine god, but a system better than the best human teams at science, software, legal reasoning, engineering design, diagnosis and research. The important economic question would not be whether intelligence had become abundant. It would be where scarcity, and therefore power, moved next.
Intelligence arrives before execution
The economy does not run on intelligence alone. It runs on electricity, factories, laboratories, contracts, permissions, building sites, courts, legacy software and people willing to accept liability. Sable could make cognition abundant without making action abundant.
A bridge design might take seconds while permits, concrete, land and political agreement take years. Medical diagnosis could become excellent while nursing time and operating theatres remain scarce. A machine could propose thousands of useful experiments without creating clean rooms, recruiting clinical cohorts, obtaining ethics approval, growing samples or securing beam time.
Research would shift from an idea bottleneck toward an execution bottleneck. The scarce contribution would increasingly be access to distinctive evidence, the capacity to run decisive tests and the ability to make the result verifiable to another institution. Cheap analysis can increase the value of the systems needed to establish that something happened in the physical world.
Physical work gains time, not immunity
An engineer commissioning a crane works at that boundary. Sable could automate calculations, simulation, standards research, documentation and failure analysis. The crane would still need steel, inspected welds and a safe foundation. Components would still arrive late. Soil, weather and tolerances would remain uncooperative. Somebody would still have to examine the actual installation and sign for it.
The same is true inside an old building. A model can diagnose a plumbing failure from video, prepare a quote, order parts and guide a worker. It cannot instantly reach a pipe behind an undocumented renovation. Cheap design may initially increase demand for retrofits, grid upgrades, factories, housing and robotic installations by lowering the cost of deciding what to build.
Germany’s 2025 skilled-worker analysis identified shortages in 157 occupations, including pressure in nursing, construction and skilled trades.1 Existing shortages give parts of the physical economy room to absorb labour while cognitive work compresses.
That is a delay, not an exemption. Sable would also improve robotics, redesign products for automated maintenance and replace irregular components with machine-friendly systems. The first robotic plumber may be a human-led van carrying scanners, cutting tools, lifting devices and specialized manipulators. Each generation would handle more of the physical work. Old infrastructure preserves ambiguity; new infrastructure will be designed to remove it.
Scarcity moves into ownership and permission
Markets would reprice themselves around these bottlenecks. Businesses that sell expensive human cognition would lose part of their advantage. Owners of compute, electricity, proprietary data, robot fleets, laboratories, distribution, trusted brands, land and legal authority would gain it.
Permission matters because a good answer is not yet an economically recognized action. Sable could recommend that a company release funds or sign a contract. Another system must still establish who may act for the company, which evidence is admissible, which rule applies and who carries liability. The institutional mechanism that turns an output into an accepted event is a separate control problem. Here, its economic significance is that organizations able to grant recognition can ration what cheap cognition is allowed to do.
Regulation can shape this boundary. The EU AI Act uses a risk-based framework and assigns different duties according to the system and its use.2 Such rules can require records, tests and accountable actors around sensitive deployments. They cannot restore the old price of intelligence or make a high-cost incumbent competitive merely by preserving its headcount. A durable bargain would connect permission to automate with inspectable systems, worker participation, transition support and a claim on the productivity gain.
Abundance does not distribute itself
If firms produce more with fewer salaries, higher productivity does not automatically place purchasing power in the hands of the people whose work lost its market value. Technical abundance can coexist with weak mass demand and extreme asset inequality. Whoever owns Sable and its physical and institutional complements can collect rents before tax and labour institutions adapt.
The eventual bargain could include reduced working hours, wage insurance, public investment, negotiated employment guarantees or broader ownership of productive assets. The appropriate mix is political. The underlying constraint is economic: distribution cannot be treated as a consequence that technology will settle by itself.
The labour-market path into this change may first be visible in vacancies that never open, while the loss of junior work creates a separate problem for the reproduction of professional judgement. Neither changes the destination of value. If cognition becomes cheap, returns move toward the owners of scarce execution and the institutions that decide which machine actions count.
Sable would not abolish scarcity. It would reveal which forms of scarcity had been hidden behind expensive cognition. Human worth would not be reduced by that repricing, but the distribution of the resulting abundance would remain institutional and political.
Footnotes
-
Fachkräfteengpassanalyse 2025, Federal Employment Agency. ↩
-
AI Act overview, European Commission. ↩