At microagi, we will deploy a billion robots. When that happens, the economic and societal consequences will be profound.
We believe that it will lead to a world of abundance, as robots increase productivity and output to such an extent that material scarcity, which forms the basic assumption of most economic models, will cease to exist. Almost every economic model is built on the principle of supply and demand, with the foundational assumption of resource scarcity. Lionel Robbins captured this in 1932, defining economics "as a relationship between ends and scarce means which have alternative uses". We believe robots will remove this scarcity. In such a world, the standard of living for all will far exceed what anybody enjoys today, and the possibilities of what humanity can achieve will be infinite.
However, we must also acknowledge that in the world of abundance created by widespread robotics, one consequence will be the removal of the need for much of the human physical labour we know of today. In the past, when machines increased supply and brought prices down, demand subsequently increased and employment rose in order to meet this. This time however, general-purpose robotics will be capable of fulfilling that demand on their own. If our economic system stays as it is, this would mean the loss of millions of current jobs, and could rob people of their income, their status, and for many, their purpose.
We are faced with a dilemma, which the majority of the robotics community has so far failed to acknowledge publicly, let alone address. The technology necessary to materially improve all of our lives will inevitably alter them enormously. Critically, if we fail to provide an adequate solution to this problem, we will never reach such a world of abundance. The majority of people will unite against the diffusion of intelligence into the physical world, and under that pressure, governments will stop robot deployments.
This essay is our attempt to look at that dilemma. We describe where the technology stands, why the prevailing economic models of displacement are insufficient for what is coming, and why the question of who people are once they no longer need to work will decide whether abundance arrives at all. We then turn to Europe, partly because we are a European company, but mainly because Europe is extremely exposed to this dilemma. European welfare states are funded largely by taxes on work, European industry still employs millions in the kinds of jobs robots are learning to do, and Europe builds few of those robots itself. We describe three potential futures for our continent. Two of them we must avoid, and one of them we intend to build.
The Race for Robotics’ GPT Moment
For all of human history, material comfort has depended on someone’s labour. Food had to be grown, houses had to be built, and goods had to be carried from one place to another. Robotics promises to automate this physical labour, driving down the cost of goods and enabling an expansion of output that results in material abundance. This future is within reach, and it is approaching rapidly.
To give an illustration of just how rapidly, consider the last eighteen months alone.
Figure AI’s 02 humanoids completed an eleven-month deployment at BMW's plant in Spartanburg, South Carolina. By the tenth month the robots were running full ten hour shifts, lifting sheet metal from bins and placing it on welding fixtures to a five millimetre tolerance. Over the programme they loaded more than 90,000 parts and contributed to the production of more than 30,000 BMW X3s . Although retired in November 2025, this is an indicator of the direction of progress.
In April 2026, Generalist reported that its GEN-1 model raised average success on selected physical tasks from 64% to 99%, working roughly three times faster than prior approaches on box folding and phone packing, with about one hour of robot data per task. Its demonstrations included folding cardboard boxes 200 times in a row without human intervention. These remain company evaluations, but reliability and speed are precisely the qualities that turn a technical achievement into useful output: a robot that succeeds 64% of the time is sufficient for a demo, but a robot that succeeds 99% of the time could replace an employee.
On July 30th this year, Google DeepMind released Gemini Robotics 2, bringing a humanoid's legs, torso, arms and five-fingered hands under a single trained policy, with an on-device version and the ability to adapt to an entirely new embodiment in hours. On August 25th, Skild AI announced S1, a model that watches a single video of a person doing an unfamiliar task and then performs it, with tasks running up to ten minutes and dozens of steps long. On its internal benchmark, Skild reports 66% success on unseen tasks against 9% for language prompted models trained at the same scale.
All of these companies are competing to create the “GPT Moment in robotics”, which would mean that a foundational robotics model could be asked to perform almost any physical task and do it well enough to be useful. ChatGPT did this for language in November 2022, turning a research curiosity into something hundreds of millions of people used every day.
The big difference to language model capability is that in the physical world, tasks are much more verifiable; this is particularly true for those in industrial settings or with short time horizons (e.g. packing a box or manufacturing a piece of equipment). Verifiable tasks can be automated, so the impact of AI systems will be much more visible in the physical world.
The most rigorous attempt so far to measure what this means for work was published by Anthropic's economists on September 30th. Using O*NET's catalogue of roughly 19,000 job tasks, Russell Legate Yang and Maxim Massenkoff find that robots that exist today can perform 74% of physical tasks in the United States, equal to 34% of all working hours, although mostly in purpose built or structured environments. Combined with large language models, around 80% of job tasks by working time are exposed to one technology or the other. Their backtest across fifty years shows that jobs more exposed to the robots of their day saw wages and employment fall in later decades, and that each year robots become able to do about 2% of the physical work they previously could not.
Public discourse fails to acknowledge the risk of robotics
Greater productivity does not, by itself, guarantee material security or contentment. On this subject, the people building artificial intelligence have begun to speak with unusual candour, and it is worth listening carefully to what they say, and what they leave out.
In May 2025, Dario Amodei warned that AI could eliminate half of all entry level white collar jobs and push unemployment to between 10% and 20% within one to five years, describing a world in which "cancer is cured, the economy grows at 10% a year, the budget is balanced – and 20% of people don't have jobs". Three months later, Stanford economists using payroll records from the largest payroll processor in the United States found that workers aged 22 to 25 in the most AI exposed occupations had experienced a 13% relative decline in employment since the spread of generative AI, while older colleagues in the same jobs were unaffected.
This leaves out the potential impact of physical AI. Anthropic's September Scenarios for our Economic Future set out three models for labour market impact. In the substantial scenario, AI can do half of all knowledge work and knowledge workers' wages are essentially flat. In the extreme scenario, US GDP rises by roughly a third, the labour share of income falls to about 45%, knowledge workers' wages fall by more than 10%, and unemployment rises "beyond typical recessionary levels".
Finance has reached similar conclusions from a different direction. In August 2026, Bridgewater's chief executive Nir Bar Dea and chief investment officer Greg Jensen argued in The New York Times that because only labour is taxed, the tax code actively rewards replacing people with machines. They proposed a consumption tax on AI tokens to lower taxes on work and to fund stakes in AI companies that could be shared "with all citizens". By late September, Bar Dea was warning publicly of "societal breakdown" if the transition were mishandled.
Whilst it is encouraging to see foundation model labs and the world's largest hedge fund at least acknowledging the potential for broad social impact, we believe their projections fundamentally underestimate the size of the labour market shock that is coming.
Even their most extreme scenario "does not include" the possibility that "humanity develops hyper capable robots." And the alternative for displaced knowledge workers is explicitly physical: they must "switch to jobs like electrician and nurse, which are less exposed to AI," a transition the authors acknowledge will be hard and slow for many people. Every model of cognitive displacement rests, somewhere, on this same quiet assumption: that displaced cognitive labour will eventually transition to physical roles.
However, once the widespread deployment of robotics becomes feasible, this assumption is critically undermined. The consultant will not spend years retraining as an electrician if she or he expects a robot to be wiring the building soon after she qualifies. The paralegal will not move into care if she expects a machine to be lifting the patient. The economists' escape route from cognitive automation runs directly into physical automation coming the other way.
When will Abundance arrive
Although robots may currently be able to complete 74% of physical tasks in at least some settings, as Legate Yang & Massenkoff’s paper states, they are cost competitive with human labour for just 0.3%. If robot prices keep falling at their historical rate of about 3% per year, it would take around 40 years for that share to reach 10%. Under their fast scenario, robots become cost competitive for half of today's physical work only by 2050. However, this measures the past rate of change and projects it forward, and its own authors concede that "AI powered robots could leapfrog our scale."
We think they will. Our argument proceeds as follows:
1. By 2027, AI systems will be doing serious robotics research and the development cycle will collapse
In June 2025, OpenAI's chief executive Sam Altman predicted that 2026 would bring systems able to produce novel insights, and that 2027 might bring "robots that can do tasks in the real world". By September 2026, OpenAI said it had built an "automated research intern", a system able to carry out research tasks that would take a skilled human researcher a few days, and it is now aiming for a fully automated AI researcher by March 2028. Independent measurement points the same way. METR found that the length of software tasks frontier AI agents can complete with 50% reliability has doubled roughly every seven months for six years. Robotics is, among other things, a software problem, a simulation problem and a data problem. Those are exactly the problems these systems are getting better at fastest.
Applied to robotics, that will collapse the development cycle. Hardware has always been slow because every design had to be built, broken and rebuilt, and every new skill had to be taught on a real robot, one teleoperated demonstration at a time. Both bottlenecks are giving way. AI can now propose an actuator, write the control policy and read the failure logs by morning, so iteration moves at something closer to software speed.
Data sources are now significantly more varied than traditional teleoperation. In February, NVIDIA pretrained a robot model on 20,854 hours of first-person video of people using their hands, with just four hours of robot teleoperation, less than 0.1% of the training mix. It found that performance improved predictably as the hours of human video grew. NVIDIA and its academic collaborators trained the DreamDojo world model on more than 44,000 hours of the same kind of footage (NVIDIA, DreamDojo, 2026). Confidence in the role of egocentric data is reflected in the more than $700 million raised in the past eighteen months for robot training data startups. Skild's single video learning and DeepMind's cross body transfer also indicate that the expensive step of collecting data for every new task, on every new body, is dissolving.
2. By 2029, robots will be technically capable of completing all economically important tasks
For this to occur, we do not necessarily require a single, complete system robot that can accomplish everything. Instead, it is the combination of a general robotic intelligence able to operate across many bodies, humanoid systems for suitable environments and specialised embodiments for specific, valuable industrial tasks.
In February, Physical Intelligence showed its π0.6 foundation model, fine-tuned for each job, running other companies' robots in live commercial settings: folding customers' laundry in a San Francisco laundromat, and packing real e-commerce orders in a warehouse for a full shift at 96.4% autonomy. Where the workplace can be redesigned around the machine, it will be, as it was for the assembly line. Where it cannot, the machine is learning to fit the workplace: the warehouse robots in that trial were built to slot into existing workstations and start working within hours.
3. Social effects will arrive with capability, not with adoption
This is the crux of our argument. Diffusion curves describe when robots will be everywhere, but they do not describe when people will believe robots will be everywhere, and make decisions accordingly. A worker will not wait for a robot to displace them before they understand what such a development may mean for their children’s future. In Europe, people already anticipate the worst: in the Commission's own survey, 66% believe AI will destroy more jobs than it creates. That figure was recorded in the spring of 2024, before any humanoid had worked a full production shift anywhere in the world. The social consequences we are concerned with do not require diffusion immediately, only the belief that it is coming.
If the dominant story is that the economy is learning to function without people, then politics will act on that long before it is necessarily reflected in economic statistics.
This is why we have called this piece Abundance 2029. Of course, not every warehouse in Europe will be automated by then; the cognitive displacement we described in the previous section may take 5-10 years, physical displacement perhaps even longer. But by 2029, the capability of physical AI will be very clear, the trajectory, sparks of Abundance so to say, will be visible to everyone, and the decisions that determine which future awaits Europe will already have been taken.
How People Earned Their Place in Society
To understand how people will react to the capability for abundance, we have to go back in history.
Seventy thousand years ago, as Yuval Noah Harari writes in Sapiens, humans loved, played, formed friendships and competed for status and power, but so did chimpanzees, baboons and elephants. "There was nothing special about them (humans)". Yet the desire to matter to one another was already there.
Physical strength offered one route to status: the ability to win a contest, defend the group or impose one's will. But as physically unremarkable primates, with no claws and little chance against a lion, we developed other advantages instead. We learnt to sharpen a stick, to coordinate a hunt through language, to make fire. We outcompeted everything above us in the food chain because we could pass down useful knowledge, allowing each generation to begin where the previous one had left off. Every generation, however, first has to learn what the last one knew and thus the development of new knowledge and innovation is more difficult and requires more specialisation. Benjamin Jones referred to this dynamic as the “burden of knowledge” for to stand on the shoulders of giants, “one must first climb up their backs.” Knowledge that takes years to acquire is held by few, and is thus scarce and therefore deemed to be valuable.
In small bands, skill earned respect. Someone who could make a better tool or guide others through a difficult winter gave their peers a reason to trust their judgement. But respect within a band is not the same as a social order built on intelligence, and for most of European history no such order existed.
The advent of Enlightenment philosophy altered this existing system of social hierarchy. Kant's motto for the age, sapere aude, dare to know, was an instruction to think for yourself, and it carried a new idea of what made a person worthy of respect: reason, rather than birth or piety. Intelligence increasingly became the characteristic to which monetary wealth accrued and thus to which social status was conferred.
Today, the quant trader's salary and the lawyer's prestige are modern expressions of that dynamic. The modern economy treats intelligence as scarce, and therefore valuable, and rewards those deemed to possess it with both material and social status. For those whose work was physical, there was a parallel ladder, older than the Enlightenment and rooted in the guilds: the master craftsman, the Meister, the foreman who knew the line better than the engineers, the welder whose seams never failed. Skill in the hands was its own kind of intelligence, and it too was honoured.
Work, in short, is the institution through which most people earn not only their living but their place in society. It provides a sense of meaning and of purpose, a defence against the existential questions that otherwise arrive at three in the morning. So what happens when the capabilities people have built their livelihoods and standing around are no longer scarce, and therefore no longer valuable? What happens when the machine is smarter than the lawyer and steadier than the welder, at the same time?
A likely coalition
Advances in the automation of cognitive work are already provoking debate, and rightly so. But for most people, the prospect of McKinsey needing fewer consultants understandably triggers little sympathy. The threat becomes much harder to dismiss when a colleague loses their job and a very tangible, physical machine takes their place the following morning.
Automation has the potential to bring people with very different incomes, education and political loyalties to the same conclusion: that the economy is beginning to function without them. A consultant and an industrial worker do not need to agree about how society should be organised to agree that its direction threatens their place in it. Such a coalition would not require mass unemployment to have already arrived. The expectation of exclusion could be enough to create a shared political interest. And when groups with different grievances come to identify a common opponent, history suggests that dramatic change often follows shortly after.
In France in 1789, parts of the upper middle class wanted political power, urban workers wanted affordable bread, and peasants wanted relief from feudal burdens. Their grievances differed, but they converged against the same order. The price of bread in Paris peaked in the very week the Bastille fell. Marie Antoinette became the hated symbol of that order's privilege, her extravagance exaggerated into an explanation for the country's distress. In Russia in 1917, opposition to the Tsar likewise extended from liberal intellectuals to workers and soldiers. Lenin's promise of "peace, land and bread" connected different grievances to a single programme, and a regime that had survived three centuries fell in a matter of months.
The nearer history of technology itself is just as instructive. Between 1811 and 1816, the Luddites of the English Midlands and North smashed the wide stocking frames and shearing machines that threatened skilled textile work, and Parliament responded with the Frame Breaking Act of 1812, which made the destruction of machinery a capital offence. The Luddites are remembered as enemies of progress. In fact, they were skilled workers who saw clearly that the gains of a new technology would flow to others while its costs fell on them, and within their own lifetimes they were right.
In Britain, output per worker rose for roughly half a century while real wages barely moved, a stretch the economic historian Robert Allen calls "Engels' pause". The men who broke the frames did not live to see the gains, and the work that replaced theirs was often worse. Adam Smith, no enemy of industry, feared that a lifetime of repeating a few simple operations would make a worker "as stupid and ignorant as it is possible for a human creature to become.”
Where the Luddites were wrong was in the long run. Breaking machines did not stop the Industrial Revolution, and it went on to make ordinary people richer than any generation before them, as we expect robotics to do. And through all of it, the work was still there.
What separates the current AI and robotics revolution is the speed and extent of progress. Bill Gates drew a similar distinction in August: earlier transitions, such as the move from farms to office jobs, played out over generations and created new jobs that required human cognition. However, with AI exceeding our capabilities in most domains, and at a far greater pace, "many jobs will disappear forever," he wrote. The Industrial Revolution's gains reached ordinary people through work: through the jobs it created and the wages they paid. If robotics creates abundance without creating work, those gains will need another route to people, and people will need another source of standing.
Until that route exists, people across the political and economic spectrum will have an incentive to resist this technological change, and the convergence of 1789 and 1917 may become possible again. If governments and technology firms fail to show how these advances benefit working people, the material abundance that would improve everyone’s standard of living will not be achieved, even where it is technically feasible. And if governments and the technology sector do fail, who can blame people for reaching such a conclusion?
The coalition is already forming
These dynamics can already be seen in the fight over data centres. In the United States, Data Center Watch counted $18 billion of projects blocked and $46 billion delayed by local opposition over roughly two years, across red and blue states alike, driven by electricity bills, water, noise and land. In the first quarter of 2026 alone, it recorded at least 75 projects worth around $130 billion disrupted, the largest quarterly figure on record. In Ireland, where data centres consumed 23% of all metered electricity in 2025, the grid operator EirGrid stopped offering new data centre connections in the Dublin region until 2028.
The failure of governments and model developers to show how their technology would make ordinary lives better has produced an alliance against data centres that spans the left and the right. This matters directly for robotics. Every general purpose robot depends on compute for training and, increasingly, for inference. A continent that cannot build data centres cannot run physical AI at scale.
On June 8th 2025, protesters in downtown Los Angeles set at least five Waymo robotaxis on fire, and the company suspended service in the area. The protest was about immigration raids, not automation. When a society is angry, the machine in the street becomes a symbol of whatever it is angry about. In October 2024, 45,000 dockworkers of the International Longshoremen's Association shut 36 ports from Maine to Texas, and protection against automation was one of their central demands.
It is incumbent upon every robotics company not to repeat the mistakes of the data centre builders: to acknowledge the social impacts that will occur, and to ensure that the transition to abundance benefits everyone. The more abruptly people experience automation as the removal of income and status without a credible replacement and a stake in the future, the stronger the demand to slow the system creating the disruption. A society can have the technology to build abundant intelligence and still refuse to deploy it.
A credible answer to how people retain status, power and purpose must therefore precede full deployment.
Europe’s Achilles heel
Who gives that answer depends on who owns the machines. As a European company, our concern is whether Europe will be able to give it. Much of what follows applies elsewhere, but Europe's combination of ageing demographics, strong unions, generous welfare states, deep industrial knowledge and weak positions in both AI and robotics manufacturing makes it the most relevant case to examine.
Physical AI has two halves: the intelligence that lets a robot see, reason and act, and the body that carries it out. The geography of both is already set, and Europe holds neither.
The United States leads the intelligence layer. The foundation models that let a robot see, reason and act are overwhelmingly being built in American labs and startups: Google DeepMind's Gemini Robotics, Physical Intelligence, Skild, Generalist, OpenAI Robotics, NVIDIA's Cosmos and GR00T platform. These are the "brains", and they are improving on accelerated software timescales.
China leads the body, and the capacity to manufacture it at scale. In 2024 China installed 295,000 industrial robots, a record and more than half of all global installations, against 67,800 across the entire European Union and 26,982 in Germany. Chinese suppliers now hold 57% of their home market, up from roughly 28% a decade earlier. In humanoids the asymmetry is starker still. Of 13,318 humanoid robots shipped worldwide in 2025, AgiBot shipped more than 5,100 and Unitree about 4,200, with Chinese firms filling most of the top ten. Unitree listed on Shanghai's STAR Market in August 2026 at a valuation of about $9 billion after its humanoid revenue overtook its quadrupeds for the first time.
In November 2023, China's Ministry of Industry and Information Technology issued its Guiding Opinions on the Innovation and Development of Humanoid Robots, which set out to build a robot's "brain, cerebellum and limbs" by 2025 and to integrate humanoids deeply into the real economy by 2027. It treated humanoids as a technology on the scale of the computer and the smartphone, and then behaved accordingly.
America is teaching the machine to think. China is learning to build it cheaper than anyone else. Europe is in danger of owning nothing in this future, and yet is in need of this technology more than either of them. A European Commission analysis finds that, with current population projections and unchanged educational attainment, the EU workforce could shrink by up to 18.8 million people by 2050.
Robotics could solve Europe's most predictable crisis, or be the cause of its next one.Three Futures
Future 1: Europe stops deployments
Europe fails to understand either the scale of the long term displacement risk due to lack of automation, or the shape of the possible solutions. Labour unions, local campaigns and politicians across the spectrum come to see robotics the way many already see data centres: as something that takes power, water, land and jobs, and gives little back to the people who live beside it. Grid connections are refused, as they already are in Dublin. Planning permissions stall. Safety rules written for caged industrial arms are applied to mobile, learning machines in ways that make deployment practically impossible. Worker fears that have long slowed traditional automation in European plants solidify into a robust political veto, and as such, no politician risks running on a platform that does not explicitly oppose accelerating robot deployments.
We believe this will be the default trajectory if nothing changes. Two thirds of Europeans already believe AI will destroy more jobs than it creates, and nobody has yet offered them a reason to believe otherwise. As a consequence, Western democratic societies block robot deployments while their competitors proceed. China, without the same democratic constraints, continues driving down the cost of production. Its factories are already installing more than four industrial robots for every one installed in the entire European Union. Production moves abroad, taking investment, supplier networks and practical knowledge with it.
We can already see the outline of this future in the present. German industry shed around 124,000 jobs in 2025 alone, nearly double the loss of the year before, and about 266,000 since 2019, with the automotive sector losing 111,000 jobs, or 13% of its workforce, in six years. In November 2025, BASF began production at its new Verbund site in Zhanjiang, southern China, an investment of around €8.7 billion, the largest single project in the company's history. Ludwigshafen, the company's home for 160 years, has meanwhile been closing plants. This is the rational response to unsustainable domestic economic conditions.
The Mittelstand, the backbone of German industry, loses its customers. Its genius has always been tacit knowledge: the engineer who knows exactly how a particular alloy behaves at a particular temperature. As production moves abroad, European companies first outsource manufacturing, then find their intellectual property increasingly conditional on manufacturing abroad, and finally discover that the knowledge itself has vanished.
Industrial workers lose employment. At the same time, professionals face cognitive automation from American AI systems that Europe does not regulate away because it cannot afford to. The two waves of displacement arrive together.
In 2024, taxes on labour, including social contributions, made up 51.5% of all tax revenue across the EU. The European welfare state is, at its foundation, financed by people having jobs. Meanwhile the old age dependency ratio is projected to rise from 36% in 2022 to 59% by 2070, from roughly three people of working age for every person over 65 to fewer than two. Mario Draghi called Europe's demographic and productivity position an "existential challenge".
In Future 1, those trends meet. Welfare needs rise, because more people are out of work and more are retired. The tax base shrinks, because European firms are not making or selling enough to pay for it. Governments face three options: let services deteriorate, raise taxes on a shrinking base, or allow debt to accumulate even further. All are undesirable or unsustainable. In the extreme, the welfare state becomes unable to sustain its promises. A policy intended to protect workers ends up protecting neither their jobs nor the institutions built to support them.
Europe can stop robotics within its own borders. However, it cannot stop robotics in Shenzhen or San Francisco, and it cannot realistically stop their products arriving at its ports at prices no European factory can match. To block robotic deployment in Europe would mean entering a new chapter of technological progress and missing out on the benefits entirely.
Future 2: Europe settles
In this second future, Europe does not block the widespread deployment of robotics and thus substantive improvements in everyone’s material living conditions do occur. However, the future still goes wrong.
Firms facing foreign competitors with fully automated “dark factories” respond in kind. Instead of redesigning work around people, the rational economic decision comes to be simply removing people from the equation altogether. Acemoglu and Restrepo found that in American local labour markets, each additional robot per thousand workers reduced the employment-to-population ratio by approximately 0.2% and wages by about 0.42%. Their study was conducted in reference to the robots of the 1990s and 2000s: caged, single-purpose and blind. The consequences of general purpose robotics of the kind we envisage by 2029 will be far more significant.
A secondary problem is ownership. Europe invented much of the science behind the digital economy and then watched others build the companies: Amazon, Microsoft and Google now hold around 70% of the European cloud market, while all European providers together hold about 15%, down from roughly 29% in 2017. The frontier models Europeans use every day are trained in California, on American chips, under American law. Robotics may follow a similar trajectory. A failure to invest in sufficient compute and hardware manufacturing, combined with an environment that is unfavourable to European companies, leads to aggressive market penetration by the United States and China. American intelligence is installed in Chinese bodies, and both are installed in European factories.
Europe has already experienced the effects of such geopolitical dependency. In April 2025, China imposed export licensing on rare earth magnets, which most high performance motors, including those in every robot’s joint, depend on. Within two months several European production lines had shut down, and some applications for Chinese technology face a requirement to disclose IP sensitive information. A Europe whose factories, hospitals, farms and warehouses run on foreign robots could have their bodies withheld by Beijing and their minds switched off from California.
In such a scenario the gains from automation flow to the owners of foreign models and foreign hardware, while the costs stay at home. Mass unemployment arrives without the required tax base to absorb it as value accrues in other jurisdictions, and socioeconomic divides worsen. Governments will try to close that gap with whatever they are still able to tax. That may manifest as a consumption tax on AI tokens of the kind Bridgewater proposes, public stakes in whatever automation governments can buy into, and taxes on the land, energy and compute resources that may remain scarce for a time. If redistributed as a form of universal basic income, this may be able to stave off material poverty as purchasing power is maintained. But crucially, the people being paid are still out of work.
In The Human Condition, Hannah Arendt warned that automation might deliver us into “a society of laborers without labor,” a society that had glorified work as the source of all value and would then be deprived of the only activity left to it. “Surely,” she wrote, “nothing could be worse.” Keynes, writing in 1930, thought his grandchildren would face exactly this, and predicted “a general ‘nervous breakdown’” among those who had been trained for generations to strive.
As previously mentioned, for many of us work forms an important part of who we are, and a transfer payment is insufficient to replace that. Status and our desire for it does not evaporate, but rather it will move onto whatever can still confer it: wealth, attention, grievance or group membership along arbitrary lines. This shift will be destabilising in itself regardless of the state of material conditions.
The core problem of Future 2 is societal upheaval. Companies race toward dark factories, producing economic disparity, and that disparity produces upheaval. Even if a degree of redistribution softens the economics, it will leave the human problem unsolved.
Upheaval of that kind can destroy the political conditions on which abundance depends: the consent to keep building data centres, deploying machines and maintaining the robots. A Europe with abundance but no purpose may vote, in its unhappiness, to dismantle the systems that feed it. At best, Future 2 is a bad future. At worst, it is merely the long way round to Future 1.
Future 3: Europe adopts robotics and builds a new economy
In the third future, Europe fully adopts robotics and builds the capacity to shape it. We develop the models, manufacture the robots and build the infrastructure that powers them. European industry remains competitive, with more of the knowledge, ownership and returns retained here. International cooperation continues, but Europe has the technological strength to negotiate its terms and the productive capacity to sustain its public institutions.
Europe has an opportunity to build this future from strengths it already possesses. It has world-class research in robot learning, a manufacturing base that still employs millions, and an engineering tradition that can turn a general robotic intelligence into thousands of specific applications. It has begun to fund the compute: in February 2025 the Commission announced InvestAI, aiming to mobilise €200 billion for AI including €20 billion for AI “gigafactories”. What it has lacked is the conviction that physical AI is the place where Europe can still lead, because physical AI rewards precisely what Europe is good at: precision manufacturing, deep domain knowledge, and the patience to make machines reliable.
Moreover, the returns from robotics should be taxed in Europe. Alaska has paid every eligible resident an annual dividend from its oil wealth since 1982, and Norway’s sovereign fund, built from North Sea oil, now owns on average 1.5% of all listed companies on behalf of its citizens. A European robotics dividend, financed by the machines themselves, would follow the same model. But as Future 2 shows, income alone does not give people a role, and if our ambitions stay where they are, there will be very few roles left to give.
Robots that can do almost everything we can will, sooner or later, do everything we currently want done, leaving us with material abundance and no part in producing it. Yet humans will remain human. We will still want our efforts recognised, opportunities to advance and communities in which our absence is noticed; none of that survives in a world where our work is needed for nothing we care about.
Our answer is a principle we call Interstellar. In essence, the scope of humanity’s ambition must always be greater than our productive capacity. If the robots we have can deliver a thousand projects, we should be attempting a thousand and one. With every machine already occupied, the remaining work, however small a share of the total, falls to people. Robots will build more robots and close the gap, so each time our machines catch up with our ambitions, our ambitions have to move again. On that condition, people remain necessary even in a world where machines are better than us at almost everything.
Whether or not we ever reach the stars, the name points to a goal that always lies beyond what we can currently do. Robert Browning expressed this idea in 1855: “a man’s reach should exceed his grasp… Or what's a heaven for?”. The nearest precedent is Apollo. At Rice University in 1962, John F. Kennedy said America would go to the Moon and “do the other things, not because they are easy, but because they are hard”. At its peak, the programme engaged some 400,000 people across government, industry and universities, supported by more than 20,000 firms. Engineers, welders, mathematicians and the seamstresses who stitched the spacesuits found themselves part of the most visible human achievement of their century.
Few of them had trained for spaceflight, but they brought the skills and expertise they had to a goal that needed them. Europe’s larger goals will ask the same of the people whose roles robots take over, and most will need help to make that move. This is the role we envision for shift: a platform for continuing education, reskilling and meaningful work, built in partnership with governments, employers and local institutions. The goal is that shift will help train people for the necessary work at the margin and connect them to it.
Whilst a private company can improve access to learning and create opportunities, the rights to income, public funding and fair access must still rest with democratic institutions, because they are of greater importance than the fortunes or priorities of any individual company. Managing the transition may span from income protection before displacement to a share of the gains from automation (using some of the mechanisms described above), and funded pathways through shift into the work the larger goal creates.
Nor can any one company be solely responsible for setting ambition on this scale. It needs a settlement in which companies, workers and public authorities agree on what Europe, and humanity more generally, sets out to do with its abundance.
We cannot responsibly deploy at scale without such a settlement, and it is needed soon. We cannot yet say what the work at the margin will be, or whether this can give people the sense of contribution that a job gives them today. But the alternative is a society whose ambitions have been met entirely by its machines, and Future 2 has shown where that leads.
There is no foreseeable end to what humanity could choose to undertake. We could find new ways to heal illness, restore the ecosystems we have damaged, explore the oceans, establish life beyond Earth and, in time, reach other stars. Each achievement would reveal further possibilities, and each would ask more of us than the machines of its day could supply on their own. A civilization that keeps setting goals of that size never runs out of work for its people, because every goal it reaches shows it a larger one. That, more than any destination, is what we mean by Interstellar.
However capable the robots become, the scale of what we are attempting would keep a place for people in it, and with that place comes the recognition, standing and sense of contribution that work has given people for generations.
Sources @ bercankilic.com
