THE ECONOMY AFTER EMPLOYMENT?

Artificial intelligence, the productivity paradox and the future of human economic agency

SUNDAY ESSAYS #01 — 25 October 2026

What happens when machines become increasingly capable of producing economic value, but people still need incomes to participate in the economy?

---

The business model of unemployment?

On 16 October 2026, Australian economist Richard Denniss appeared before a parliamentary inquiry into artificial intelligence and made a rather extraordinary claim. Mass unemployment, he argued, is not simply a possible unintended consequence of artificial intelligence. It is central to the business model of the companies developing it.

His reasoning was straightforward. AI companies are attracting extraordinary valuations because investors expect their technologies to perform economically valuable tasks currently undertaken by human workers. Businesses will buy AI services because doing so allows them to reduce labour costs, and the economic value created by replacing human work will increasingly flow towards the companies providing the automation. If the expectations underlying AI valuations are correct, Denniss argued, the consequences for employment, communities and public finances could be enormous. And if those expectations are wrong, why should governments be in such a hurry to reorganise their economies around them?

It is a provocative argument, perhaps deliberately so. Not all AI applications replace human labour, and the relationship between technological progress and employment is far more complicated than a simple transfer of wages from workers to technology companies. Nevertheless, Denniss raises a question that deserves considerably more attention than it has received: what happens to an economy when the technology that makes businesses more productive also reduces their need to employ the people who buy their products?

To understand the significance of this question, we need to consider something that is often missing from discussions about artificial intelligence: the economic system in which the technology operates. Productivity is not the same thing as prosperity, and producing economic value is not the same thing as distributing it.

The promise we have heard before

Nearly a century ago, John Maynard Keynes made a remarkable prediction. Writing in 1930, in the midst of the Great Depression, he considered the economic possibilities available to his grandchildren. Keynes believed that technological progress and capital accumulation could eventually make humanity vastly more productive. Within approximately a century, he suggested, people in economically advanced societies might be able to satisfy their material needs with a fraction of the labour previously required. He even imagined a fifteen-hour working week.

Keynes was not entirely naive about the transition. He anticipated a phenomenon he called technological unemployment: the possibility that labour-saving innovations could temporarily displace workers faster than societies could find new uses for their labour. But he believed this would ultimately be a transitional problem. Once humanity had become sufficiently productive, the central economic challenge would change. Instead of struggling to produce enough, people would need to learn how to live with abundance.

Almost a hundred years later, Keynes’s prediction looks simultaneously perceptive and strangely unfulfilled. Productivity has increased enormously, entire categories of work have been automated, and technologies that would have seemed miraculous in 1930 are now ordinary parts of everyday life. Yet paid employment remains central to the organisation of modern societies. For most people, the ability to secure housing, food and other necessities still depends heavily on employment income. Rather than approaching the end of economic necessity, many societies continue to experience insecurity, inequality and growing anxiety about the future of work.

There are many reasons why Keynes’s anticipated age of leisure has not materialised. Rising productivity has supported higher consumption and entirely new categories of goods and services. People’s expectations have changed, new industries have created new work, income has been distributed unevenly, and institutions have continued to organise economic life around employment. But Keynes’s central question remains: if technology allows us to produce more with less human labour, what should happen to the relationship between work, income and economic participation? Artificial intelligence makes that question newly urgent.

Workers are also customers

Consider a hypothetical company employing a thousand people. Suppose advances in artificial intelligence allow it to maintain the same production with five hundred employees. From the company’s perspective, the economic argument for automation may be compelling. Labour costs fall, productivity rises, profitability may improve, and competitors may face pressure to adopt similar technologies.

But the five hundred employees who lose their jobs are not merely costs removed from a balance sheet. They are also consumers whose wages pay for housing, groceries, transport, restaurants, cultural experiences and countless other goods and services. One company’s reduction in employment might have limited consequences for the wider economy, but if thousands of companies simultaneously reduce their payrolls, the macroeconomic effects become more complicated.

Income previously distributed as wages may increasingly flow to shareholders, technology providers and owners of computational infrastructure. The money does not simply disappear: it can finance investment, consumption, new businesses and other economic activities. But its distribution matters. Wealthier households typically consume a smaller proportion of additional income than households with limited resources. If automation substantially shifts income from wages towards concentrated capital ownership, the composition of economic demand may change. And if the resulting investment and consumption fail to compensate for weakened household spending, the economy may encounter a demand problem.

In other words, a technological improvement that is profitable for individual companies can create difficulties when adopted across an entire economy. Economists have long recognised variations of this problem. It reflects a fundamental distinction between decisions that make sense for individual economic actors and outcomes that emerge when everyone makes similar decisions simultaneously.

Yet there is no reason to assume that automation must necessarily produce an economic crisis. Lower production costs can reduce prices, increasing consumers’ purchasing power. Technological change can create new industries, products and employment, while investments can generate additional demand. Governments can redistribute income or provide services that reduce households’ dependence on wages. The outcome depends on how the gains from productivity growth move through the economy. The problem is not necessarily that machines produce too much, but that our institutions may fail to distribute the benefits of what they produce.

When demand becomes a technological constraint

This is no longer merely a philosophical concern. In October 2026, the Bank for International Settlements published a working paper titled Transformative AI and Demand Bottlenecks. The researchers examined a possibility that complicates the most optimistic forecasts of AI-driven growth: what if artificial intelligence becomes increasingly capable of transforming production, but the economic conditions required to sustain further innovation begin to weaken?

The mechanism in their model is more subtle than a simple collapse in workers’ purchasing power. As AI automation expands, an increasing proportion of economic output may be directed towards investment rather than consumption. If consumer markets become relatively smaller, the potential returns from developing new consumer products may decline. Since expected commercial returns help determine incentives to invest in innovation, weaker consumer demand can eventually constrain further technological progress.

The researchers identify several possible outcomes. In one scenario, transformative AI generates an extraordinary acceleration in economic growth. In another, growth initially accelerates but subsequently falls below its pre-AI rate as a demand bottleneck reduces incentives for continued innovation. Their calibrated model suggests that the second outcome is empirically plausible. This is not a forecast that such a bottleneck will inevitably emerge; it is a result within a particular economic model, and alternative assumptions produce different outcomes.

Nevertheless, the study illustrates an important possibility: the limits of artificial intelligence may not be exclusively technological. A society can possess increasingly powerful productive capabilities while lacking the commercial incentives needed to develop and deploy them further. That is a rather different perspective from the familiar debate about how much computing power, electricity or data future AI systems will require.

Another recent study addresses a closely related question. In Macroeconomic Policies for AI, economists Luca Fornaro and Martin Wolf investigate how AI-driven automation could simultaneously increase potential output and reduce workers’ share of national income. Because workers tend to spend a relatively large proportion of their earnings, a substantial shift away from labour income could weaken aggregate demand. Under certain conditions, the result could be what the researchers describe as an AI-induced slump: an economy becomes technologically more productive but fails to realise its productive potential because demand is insufficient.

Yet the authors also demonstrate that the outcome depends critically on economic policy. Monetary policy, labour taxation and employment subsidies can influence how the economy responds. The study therefore does not establish that AI inevitably causes a recession. It establishes something arguably more important: the same technological transformation can produce substantially different economic outcomes depending on the institutions and policies surrounding it.

This is worth remembering whenever technological developments are presented as forces operating independently of human choices. Machines may change what is economically possible, but they do not independently determine how the resulting prosperity is distributed.

The jobs that never existed

What does the evidence tell us about AI’s actual effects on employment? So far, considerably less than confident predictions on either side of the debate might suggest.

In August 2026, researchers at Stanford Digital Economy Lab published an updated analysis of US payroll data covering millions of workers. They found no evidence of widespread, economy-wide job displacement attributable to artificial intelligence. However, they identified a striking pattern among younger workers. Employment among people aged 22–25 in occupations highly exposed to AI was approximately 19 percent below where it would have been had it followed the employment trajectory of similarly aged workers in less-exposed occupations.

The researchers found that this divergence appeared primarily through reduced hiring rather than increased layoffs. They also emphasised an important limitation: the findings are descriptive, not definitive causal estimates. The data cannot establish precisely how much of the observed difference was caused by AI. Nevertheless, the pattern suggests a possibility that deserves close attention. Perhaps the first substantial employment effects of artificial intelligence will not arrive as spectacular waves of mass redundancies, but as jobs that are never advertised, vacancies that are never filled and career opportunities that quietly disappear.

Such changes can be difficult to observe. A person who loses an established job is visible in employment statistics. A young graduate who never finds an entry-level position in their chosen profession may simply become part of a broader and less easily explained pattern of economic insecurity. The consequences could extend well beyond immediate employment. Professional expertise is often acquired through experience, so if AI performs an increasing share of entry-level tasks, how will people acquire the practical knowledge required to become experienced professionals? How will industries develop the next generation of specialists?

The economic effects of automation may therefore include not only the displacement of existing workers but also the transformation of the pathways through which people enter working life.

Automation is not the only possible future

There is a danger in treating all of this as an inevitable consequence of technological progress. Artificial intelligence can be designed and deployed in fundamentally different ways. Some applications replace tasks previously performed by humans. Others enhance human capabilities, reduce barriers to expertise or make entirely new forms of work possible.

In February 2026, economists Daron Acemoglu, David Autor and Simon Johnson published Building Pro-Worker Artificial Intelligence. Their analysis distinguishes between several forms of technological change, including automation, labour augmentation, the levelling of expertise differences and the creation of new tasks. They argue that AI’s potential to support human expertise and create new kinds of economically valuable work remains underexploited.

Importantly, not every technology that makes an individual worker more productive necessarily increases the overall demand for labour. A tool can help one employee perform the work of several without creating additional jobs. The researchers therefore emphasise the importance of technologies that generate new tasks requiring human expertise.

This distinction has profound implications. The future economic impact of artificial intelligence depends not only on what the technology can accomplish, but also on what businesses and developers are encouraged to build. An AI system designed to replace a customer service worker has different economic implications from one that enables the worker to provide services that were previously impossible. Both may increase productivity, but they need not have the same consequences for employment, skills or income distribution.

The direction of technological development is therefore itself an economic and institutional question, and the incentives currently shaping that direction deserve scrutiny.

The redistribution question

If AI eventually enables economies to produce substantially more with significantly less human labour, societies will face a difficult question: how should the resulting wealth be distributed?

One possibility is that new industries and new forms of work will absorb much of the labour displaced by automation. Another is that governments will need to redistribute a larger share of income generated through capital-intensive production. Proposals include changes to taxation, universal basic income, public wealth funds, broader ownership of productive assets and new forms of social insurance. These approaches differ considerably in their economic consequences and political implications, but they share a recognition that the relationship between employment and income may need to evolve.

A recent announcement by Anthropic offers a small but revealing example of how this discussion is beginning to influence the technology industry itself. On 18 October 2026, the AI company announced plans to dedicate an initial five percent of its shares to a new public benefit fund, subject to shareholder approval. The fund could eventually support initiatives including direct financial distributions to particular populations. Anthropic explicitly identified the labour-market consequences of AI as one area in which more substantial interventions might be necessary.

This is not a universal basic income system, nor does it come close to resolving the macroeconomic challenges discussed here. But it is an interesting institutional signal. A company developing advanced AI is considering mechanisms through which some of the economic value associated with its own success could be directed towards wider social purposes.

That raises questions extending beyond the size of the proposed fund. Who should decide how the benefits of AI are distributed? Should such decisions be made primarily by governments, private corporations, independent institutions or some combination of these? And how can societies prevent technological dependence from becoming economic and political dependence on a small number of exceptionally powerful organisations?

These questions will become increasingly important if AI fundamentally changes the distribution of economic power. Yet even an effective redistribution system would address only part of the problem, because having an income and having meaningful economic agency are not the same thing.

The economy beyond employment

Modern societies have developed a remarkably strong relationship between employment, income and participation in economic life. For most working-age adults, paid employment remains the principal mechanism for securing a livelihood. But human activity has never been limited to paid work. People care for relatives, maintain communities, exchange knowledge, create art, repair things, organise events and build relationships. Much of this activity produces genuine social value without generating equivalent market income, and economic statistics capture only part of what makes a society function.

If artificial intelligence substantially reduces demand for certain forms of paid labour, this distinction may become increasingly important. People will not suddenly cease to possess skills, interests, ambitions or reasons to collaborate. They may, however, encounter growing difficulties in translating those capabilities into economic security. This leads to a question that conventional discussions about technological unemployment often overlook: what if the future of work is not primarily about the disappearance of human activity, but about the changing relationship between human activity and paid employment?

Perhaps the most significant transformation will not concern what people are capable of doing, but what society is prepared to pay them to do. This is a much broader challenge than unemployment. It concerns the institutions through which people participate in economic life, and institutions can change.

The creator economy as an early experiment

One place where some of these changes are already visible is the contemporary creator economy. Artists, writers, educators, independent professionals and cultural producers increasingly organise their activities through networks rather than relying exclusively on traditional employment relationships. They combine projects, clients, audiences, communities and multiple sources of income, and their economic lives are often structured around relationships, reputation and independent initiative.

It would be a mistake to romanticise this arrangement. For many people, the creator economy is characterised by precarious incomes, intense competition, unequal visibility and dependence on platforms that control access to audiences. Indeed, many of its problems illustrate precisely what can go wrong when individuals are required to function as independent economic actors without adequate institutional support.

Nevertheless, the creator economy provides an interesting experimental environment. It demonstrates how people can organise economically valuable activities through distributed networks of relationships, projects and communities. It also raises the possibility that some of these organisational patterns could become relevant far beyond the cultural industries, not because everyone will become an artist or content creator, but because more people may eventually need to combine different forms of work, collaboration and income outside a single long-term employment relationship.

The creator economy might therefore be understood not only as a specialised sector of contemporary digital capitalism, but also as an imperfect preview of broader economic transformations: an experiment whose successes and failures deserve serious study.

From social networks to economic networks

For much of the digital era, consumer technology companies have organised their business models around capturing and monetising human attention. Social networks have become extraordinarily effective at collecting behavioural data, predicting preferences and distributing content. But social relationships have economic significance far beyond advertising.

People use networks to find collaborators, exchange knowledge, organise projects, establish trust, build communities and discover opportunities. Historically, these activities have always been essential to economic life. Digital technologies simply make it possible to organise them across new distances and scales. Yet the dominant social platforms have generally focused on turning networks into audiences.

What if future social technologies placed greater emphasis on enabling people to act together? What if their purpose extended beyond making content discoverable towards making collaboration, shared resources and independent economic activity easier to organise?

This would not solve the problem of income distribution. A network cannot create purchasing power merely by making people better connected, nor can digital platforms substitute for the public institutions needed to support economic security. But they could become one component of an economic environment in which people possess greater practical freedom to organise their activities and relationships.

Such systems would also face a serious challenge: how could they facilitate independent economic activity without becoming new intermediaries controlling the very relationships they make possible? Questions of ownership, interoperability, portability, governance and trust would become central. The promise would not simply be more efficient markets, but greater economic agency.

A future that remains undecided

None of this establishes that paid employment is approaching its end. Artificial intelligence may create new categories of work, expand human capabilities and generate considerable prosperity. It may also weaken established career structures, concentrate ownership and increase economic insecurity. Different outcomes may occur simultaneously across industries, countries and generations, and the evidence remains incomplete.

What seems increasingly difficult to defend, however, is the assumption that economic institutions developed around the employment structures of the industrial and post-industrial eras will necessarily remain adequate.

Keynes imagined a world in which technological progress might eventually free people from much of the economic necessity that had shaped human existence for centuries. We have not yet arrived in that world, and perhaps we never will. But the emergence of increasingly capable artificial intelligence makes it reasonable to reconsider the questions he raised nearly a century ago. How should the benefits of technological progress be distributed? What is the relationship between economic productivity and human well-being? And what kinds of institutions allow people to participate meaningfully in a society whose productive capacities increasingly depend on machines?

These are not questions that technology companies can answer alone. But technology companies are among the organisations making decisions that will shape the answers. That creates a responsibility to think beyond technological capabilities and immediate commercial opportunities. It also creates a reason to reconsider what digital infrastructure is ultimately for.

Perhaps the most important question about the future of work is not how much work machines will take away from people, but how much freedom technology can give people to organise their lives, relationships and activities on their own terms.

The future of work is not only about what machines will be able to do. It is also about what people will still be able to do for themselves — and for one another.

---

References and further reading

Richard Denniss / Australia Institute (2026)
Parliamentary testimony, 16 October 2026. Reported by Tegan George in The Point, 22 October 2026.
https://thepoint.com.au/new/261022-not-an-unintended-consequence-mass-unemployment-is-the-ai-business-model

John Maynard Keynes (1930)
Economic Possibilities for our Grandchildren.
https://www.hetwebsite.net/het/texts/keynes/keynes1930grandchildren.htm

Bank for International Settlements (2026)
Transformative AI and Demand Bottlenecks. BIS Working Paper No. 1381, 12 October 2026.
https://www.bis.org/publications/working-paper-1381-transformative-ai-and-demand-bottlenecks

Luca Fornaro & Martin Wolf (2026)
Macroeconomic Policies for AI. CEPR Discussion Paper No. 21412, 24 April 2026.
https://cepr.org/publications/dp21412

Erik Brynjolfsson, Bharat Chandar & Ruyu Chen (2026)
Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab, revised 12 August 2026.
https://digitaleconomy.stanford.edu/publications/canaries-in-the-coal-mine/

Daron Acemoglu, David Autor & Simon Johnson (2026)
Building Pro-Worker Artificial Intelligence. NBER Working Paper No. 34854, February 2026.
https://www.nber.org/papers/w34854

Anthropic (2026)
Investing in the Public Benefit. 18 October 2026.
https://www.anthropic.com/news/public-benefit-fund