The standard story about AI and employment may have the causality backwards. Technology becomes capable of performing work, companies discover that machines can substitute for people, and workers are displaced whether they wanted that future or not. Automation acts first; the labour market reacts.
There is another path into the same future. Societies can reduce the amount of human time devoted to paid work, grow older, run short of skilled workers or accumulate financial obligations that require more production to sustain. Automation then arrives from the opposite direction.
The machine is not initially pushing the human out of production; the economy is creating a gap that something else has to fill.
Sweden offers one version of this mechanism. Miljöpartiet now advocates reducing normal working time to 35 hours as a step toward a four-day week, and shorter working time has become part of the wider Swedish labour-policy debate. The case for such reforms does not have to be dismissed for their economic constraint to be real. Leisure has value. Better recovery may reduce sickness and allow people to remain in working life longer. But unless those effects completely compensate for the hours removed, the economy receives less human production time. If society wants to preserve output at the same time, productivity has to make up the difference.
The United States is approaching the same problem from almost the opposite direction. Its immediate constraint is not a preference for fewer working hours but an accumulation of claims on future production. The Congressional Budget Office's 2026 outlook projects federal debt held by the public rising from 101 percent of GDP in 2026 to 120 percent in 2036. Net interest expenditure rises from 3.3 to 4.6 percent of GDP over the same period, while the annual deficit remains unusually large even without a severe recession.
This is where Niall Ferguson's proposed "Ferguson's Law" becomes useful. Ferguson argues that a great power enters dangerous territory when debt service exceeds defense spending, because resources committed by the past begin constraining what the state can do in the present. He calls the crossing point the Ferguson limit and argues that the United States crossed it in 2024. It is better understood as a historical warning light than as a literal economic law: there is nothing magical about defense spending as a threshold. The important point is that servicing accumulated obligations can eventually consume enough current production to narrow a country's room to manoeuvre.
That makes productivity unusually valuable. A heavily indebted state can improve its position through higher taxes, lower spending or some mixture of the two, but faster economic growth changes the denominator itself. The same stock of debt becomes smaller relative to the economy supporting it, while taxable incomes, profits and consumption can expand. When debt is already large, even a modest persistent increase in growth compounds into a very different fiscal trajectory.
AI therefore enters the story at a scale far beyond whether a programmer, accountant or administrator loses a particular task. A 2026 Brookings analysis by Ben Harris, Neil Mehrotra and William Overcash models what a large AI-driven productivity shock could do to American public finances. In their conventional productivity scenario, the federal deficit in 2036 falls by roughly five percentage points of GDP relative to baseline. Their more disruptive AI scenarios recover substantially less of that improvement, but the scale is still remarkable. At that magnitude, AI productivity is no longer merely a technology-sector effect. It is large enough to alter sovereign arithmetic.
The Swedish labour argument and the American debt argument therefore converge on the same requirement. Sweden can choose to make human time less available to production. The United States can inherit financial commitments that require more future production. Aging societies can do both at once, reducing workforce growth while increasing spending on pensions and healthcare. Companies experience their own version through skill shortages, rising labour costs and pressure to defend margins. Different causes increasingly produce the same demand: more economic value from each unit of human time.
There are only a few broad ways to satisfy that demand. An economy can add workers, ask existing workers to work more, consume less, reduce public commitments, increase the amount extracted through taxation or produce more from the labour and capital it already has. Countries will choose different combinations, but where more human labour is difficult to obtain, politically unwanted or simply unavailable, productivity has to carry a larger share of the adjustment.
This is where the usual automation narrative begins to invert. If six people assisted by machine execution can produce what ten produced before, it sounds immediately like four jobs have been destroyed. That can certainly happen. But if an economy has already lost equivalent capacity through shorter working time, retirement, demographic change or persistent skill shortages, the machine is also replacing work that the human labour system no longer supplies on the previous terms. Automation is being pulled into an empty space rather than merely forcing its way into an occupied one.
Generative and agentic AI are unusually suited to that role because the available frontier is not confined to industrial machinery. Previous waves of automation transformed physical production and particular clerical processes. AI reaches into analysis, administration, software production, documentation, customer handling, planning, coordination and parts of operational decision-making. These activities occupy enormous amounts of human time in advanced service economies, and much of their cost exists not because the underlying decisions are inherently difficult but because information has to be moved, transformed, checked and coordinated by people.
The American fiscal version makes the reversal especially stark. The government does not need AI because AI is fashionable, nor does it need automation because replacing workers is an objective in itself. It needs some combination of fiscal adjustment and greater economic output because its existing balance sheet demands it. If machine-assisted production can raise output materially faster than the workforce grows, AI becomes one possible way of keeping the economy above a financial waterline established long before today's models existed.
The apparent solution contains a complication of its own. Modern tax systems are unusually good at taxing an economy in which production flows through human employment. Wages generate income and payroll taxes, workers become consumers, and household income circulates back through the economy. The Brookings analysis therefore models a shift from labour toward more lightly taxed capital income as one of the mechanisms that can reduce AI's fiscal benefit. If machines increase output while labour receives a smaller share of it, GDP can grow even as the tax architecture built around human work becomes less effective.
That produces two very different versions of the same productivity boom. In one, machines amplify people: fewer human hours produce much more value, while employment, wages, profits and tax receipts remain broadly connected to the expansion. In the other, machines substitute heavily for labour and an increasing share of economic output flows directly to capital. Both can produce exceptional productivity numbers, but only the first fits comfortably inside institutions designed around employment income. The second eventually forces a further transition in taxation, redistribution and the way households receive claims on machine-produced output.
Nor does a country with healthier public finances escape the pressure. If American companies reorganise around substantially higher output per employee, their Swedish, German and Japanese competitors inherit the new productivity benchmark. One country may enter the transition through debt, another through demographics, another through working-time reform and another through industrial competition. Once one part of the global economy learns to produce materially more value with less human coordination, everyone else competes with the result.
The strange possibility is therefore not simply that AI will take our jobs. Developed societies may themselves make human labour scarcer while simultaneously preserving, or even enlarging, the economic promises that labour once supported. Each individual choice can be rational: more leisure, longer retirement, better social protection, greater public spending or borrowing against future prosperity. Taken together, they increase the value of any technology capable of separating economic output from the number of human hours available to produce it.
For years we worried that machines would force humans out of work. The stranger possibility is that humans increasingly choose to withdraw their labour first -- and then discover that they need the machines to preserve the economy they expected to keep.