Work, Labor, and Migration

You find out on a Tuesday. Or a Thursday. It does not matter which day because the day is not the part you remember. What you remember is the tone. Your supervisor's voice has a quality you have never heard before, a kind of practiced gentleness that tells you this conversation was rehearsed. There is language about "restructuring" and "new directions" and "transition support." There is a packet. You are told to read it carefully.

You have been here eleven years. You know where the mops are kept. You know which loading dock sticks in cold weather. You know that the third conveyor on the east wall throws a belt every six weeks and that if you tension it a specific way it holds for eight. Nobody wrote that down. You just know it. And now you are learning that knowing it does not matter.

The thing that replaces you does not look like the robots in movies. It is not chrome. It does not speak. It stands about your height, maybe a little shorter, and it moves with a cautious smoothness that reminds you, unsettlingly, of a person learning a new job. Which is what it is doing. It is learning your job. And it will work the night shift you refused and the holiday weekends you negotiated away. It will never file a workers' comp claim for the shoulder you destroyed in year six.

You are not lazy. You are not stupid. You did nothing wrong. The economy did not crash. Your company is not failing. In fact, your company's stock price will tick up on the announcement, because investors understand what your supervisor's rehearsed gentleness was trying to soften: you are not being replaced because you failed. You are being replaced because you are expensive and fragile, and you eventually ask for more.

What happens next is what this chapter is about.

The History That Repeats — Until It Doesn't#

Every wave of automation in history has provoked the same fear, and every wave has, eventually, proven the fear wrong. That track record is both reassuring and dangerous, because it tempts us into assuming the pattern will hold forever.

Start with farming. In 1900, roughly 41 percent of the American workforce labored on farms. Plowing, planting, weeding, and harvesting were done mostly by hand and animal. Then came the tractor, the combine harvester, chemical fertilizers, hybrid seeds, mechanized irrigation. By 1950, the farm labor force had already dropped from about 11 million to under 6 million. By 2000, agriculture employed less than 2 percent of American workers. A sector that had once absorbed nearly half the country's labor now ran on a skeleton crew. The displacement was staggering in speed and scale: over the second half of the twentieth century, self-employed and family farmworkers fell by 73 percent, from 7.6 million to 2 million.

The workers who left the fields did not vanish. They moved to cities. They found jobs in factories, then in offices, then in services. The economy absorbed them, not painlessly and not immediately, but it absorbed them. Wages rose. Living standards improved. The mechanization of agriculture, in the long run, made almost everyone better off.

This is the story that optimists tell. And they have a second favorite example: the ATM.

When automated teller machines spread through American banks in the 1970s and 1980s, the obvious prediction was that bank tellers would disappear. Machines could dispense cash, accept deposits, and check balances faster and cheaper than any human. By the mid-1990s, over 400,000 ATMs had been installed across the country. The number of tellers per branch dropped from about 20 to 13. And yet, as Boston University economist James Bessen documented, the total number of bank tellers in the United States did not fall. It rose. ATMs made it cheaper to operate a branch, so banks opened more branches, and the new branches needed staff. By 2002, there were 527,000 tellers working alongside 352,000 ATMs. The tellers who remained shifted from routine cash handling to relationship banking: advising customers, selling financial products, solving problems that machines could not.

The ATM story has become something of a parable in automation debates. It illustrates what economists call the "productivity effect": when technology makes a service cheaper, demand for that service can increase enough to offset the labor savings. The tellers were not replaced. Their jobs were reshaped.

But there is another version of this history, and it is less comforting.

The optimistic examples work because they describe partial automation. The tractor replaced the hand plow but still needed a driver. ATMs could handle deposits, yet no machine was advising a small business owner on a loan. In each case, human labor was displaced from some tasks and redirected to others. The technology was complementary: it made certain human skills more valuable, not less.

The trouble starts when automation is not partial but comprehensive. When machines can perform most or all of the tasks within a job, the productivity effect weakens and the displacement effect dominates. There is no residual human role to absorb the displaced worker. This is what happened to telephone switchboard operators, to typesetters, to elevator attendants. These occupations did not transform. They disappeared.

Daron Acemoglu and Pascual Restrepo, economists at MIT and Boston University, have produced the most rigorous research to date on how industrial robots actually affect employment. Their findings are more sobering than the ATM parable suggests. Studying U.S. labor markets between 1990 and 2007, they found that each additional industrial robot per thousand workers reduced the local employment-to-population ratio by 0.2 percentage points and pushed wages down by 0.42 percent. The arrival of one new industrial robot in a local labor market coincided, on average, with an employment drop of 5.6 workers. Critically, the positive effects that should have offset these losses, new jobs in other sectors, increased productivity lifting demand, were small. The displacement effect was winning.

And that was with industrial robots: single-function machines bolted to factory floors, performing welding, painting, and assembly. Machines that could do one thing well in one place.

Humanoid robots are something else. They are designed to operate in human environments, to perform multiple tasks, to adapt. A humanoid robot in a warehouse does not just move pallets; it can unload trucks, sort packages, and handle returns. It competes not with one narrow task but with the full bundle of tasks that constitutes a job. This is the distinction that matters. Previous waves of automation sliced jobs into tasks and automated the routine ones, leaving the non-routine tasks for humans. Humanoid robots threaten to automate the bundle.

If that happens, the historical pattern may finally break. Not because technology has never displaced workers before, it has, repeatedly, but because the escape route that previous generations used may be closing. Farm workers moved to factories. Factory workers moved to services. Each transition took decades and caused real suffering, but there was always somewhere to go. The question that humanoid robots raise is whether the "somewhere" still exists when machines can do physical work and the kind of flexible, adaptive labor that has always been the last refuge of human advantage.

The ATM parable is true. It is also from 1995. The economy it describes, one in which automation reshapes jobs rather than eliminating them, may not be the economy we are entering.

What Goes First#

Rather than catalog every sector that humanoid robots could enter, it is more useful to look closely at three where the conditions for adoption are already aligned: where the work is physically punishing, the workers are scarce, and the economics are straining.

The Warehouse

Amazon operates over 750,000 robots across 300 facilities worldwide, and in 2024 surpassed one million total robotic systems. The company now employs nearly as many robots as humans. But these are overwhelmingly fixed arms and autonomous mobile robots, machines that move inventory along rails and conveyors. The work they cannot do is what Amazon calls "the last meter": picking items from irregular positions, transferring totes between systems, stacking containers in spaces designed for human bodies. This is precisely the work that humanoid robots are being built for.

Agility Robotics' Digit, a bipedal humanoid, passed 100,000 totes moved at a GXO Logistics facility in Georgia by fall 2025, the first measurable commercial deployment of a humanoid in a live warehouse. As of 2025, Agility was preparing its fifth-generation model, with commercial deployments still in early stages. The numbers are still small. But the conditions that make warehousing attractive to humanoid makers are not small at all. Amazon's annual warehouse turnover rate runs at roughly 150 percent, meaning the company replaces the equivalent of its entire hourly workforce every eight months. Seventy percent of new hires leave within 90 days. The company's own internal study, leaked to the press, estimated Amazon could exhaust its available American labor pool if these trends continued. Warehouse work across the industry carries injury rates significantly above the national average, with overexertion and repetitive motion accounting for the largest share of cases requiring time away from work. A humanoid robot that can pick, stack, and transfer totes at $10 to $15 per hour equivalent, running three shifts without injury claims or turnover costs, does not need to be as fast as a human. It just needs to show up.

The Farm

Agriculture faces a different version of the same problem. The average age of American farmers is 58.1, and there are four times as many producers over 65 as under 35. The number of new immigrants arriving to work in U.S. agriculture has fallen by 75 percent in recent years, and even with the H-2A temporary worker program issuing over 315,000 visas in fiscal year 2024, the industry faces a projected deficit of 2.4 million workers. In 2025, farm employment stood at 2.19 million, and labor costs for specialty crop growers were reaching 40 percent of total expenses.

The consequences are concrete: unharvested crops, rising food costs, and the largest agricultural trade deficit in U.S. history in 2024, driven by imports of labor-intensive products like fresh produce. Farm wages averaged $18.12 per hour in 2024, still only 60 percent of the nonfarm average, yet rising fast enough to squeeze margins on crops that already barely pencil out. Immigration enforcement actions in 2025, including raids on California farms, added a new layer of uncertainty. Robotic harvesters and AI-guided tractors are already entering this space, but most current systems are specialized for single crops. A humanoid form factor that could prune, pick, sort, and load across multiple crop types on the same farm would transform the economics of small and mid-size operations.

The Care Home

The most emotionally fraught case is elder care. The U.S. population aged 65 and older grew from 12.4 percent in 2004 to 18 percent in 2024, and the number of Americans living with dementia is projected to double from 7 million to 14 million over the next 35 years. The demand for direct care workers is the largest of any occupation in the country, with a projected 9.7 million job openings between 2024 and 2034. But the average home health aide earns $16.82 per hour, barely above fast food wages, despite work that is physically demanding and emotionally draining. More than 800 nursing homes have closed in the last decade. Approximately 800,000 older Americans needing subsidized care sit on waiting lists because workers simply do not exist.

This is where humanoid robots encounter their most complex terrain. A robot that lifts a patient from bed to wheelchair and monitors medication schedules fills a gap that immigration policy and wage increases have failed to close. On the night shift, when no human aide is available, even a machine presence is better than none. But a robot that replaces the human hand on a dying person's shoulder raises questions that no labor market analysis can answer. These three sectors share the same underlying condition: the work is hard, the workers are leaving, and no policy intervention has reversed the trend.

The Jobs That Change, The Jobs That Emerge#

Every automation wave creates new work. The question is whether the new work is accessible to the people who lost the old work, and whether it pays enough to sustain a life.

Humanoid robots will generate roles that do not currently exist at scale. Fleet operators will manage dozens or hundreds of robots across warehouse floors, monitoring dashboards and intervening when a unit jams or encounters an obstacle its software cannot parse. The machines will need technicians for maintenance and calibration. Companies like Tesla and Physical Intelligence were already hiring teleoperation specialists in 2025 to remotely control humanoid systems and collect training data. Beyond these, the ecosystem will need safety auditors to certify human-robot workflows and integration engineers to configure robots for new environments. According to Lightcast data cited by the Association for Advancing Automation, entry-level robotics technician roles average $66,000 per year, intermediate specialist roles average $105,000, and advanced integrator positions reach $130,000 or more. These are good jobs, and they will multiply as deployments scale.

But there is a brutal arithmetic underneath the optimism. A warehouse that employs 500 pickers and packers might, after humanoid deployment, need 50 fleet operators and technicians. The ratio matters. And the skills required for the new roles bear little resemblance to the skills that defined the old ones. A warehouse picker relies on physical endurance, spatial awareness, and the ability to work at speed under pressure. The fleet operator role demands something completely different: software literacy, diagnostic reasoning, comfort with data interfaces. No weekend workshop bridges that distance.

We have seen this before. When coal production employment fell 42 percent in the Appalachian region between 2011 and 2019, retraining programs proliferated. Federal POWER grants funded initiatives in Kentucky, Ohio, and West Virginia. Politicians from both parties told miners to "learn to code." The results were dismal. Participation rates were low. The jobs that retrained workers found typically paid $12 to $15 per hour, roughly a fifth of what they had earned in the mines. One prominent coding program, Mined Minds, turned out to be riddled with problems: inadequate training, missing pay, participants fired without notice. A Johns Hopkins study concluded that federal and state retraining efforts had "largely failed to attract and retain displaced workers." The programs were described by the communities they targeted as "out of touch." The failure was not primarily a failure of content. It was a failure of context. Workers do not exist in vacuums. They have mortgages, families, aging parents, roots in communities where the new jobs do not exist. Retraining works best when the new industry is in the same place as the old one, when the training is paid, when the transition happens while people still have savings rather than after they have exhausted them.

Humanoid robotics offers one structural advantage that coal-to-code did not: the new jobs are physically co-located with the old ones. If a warehouse deploys humanoid robots, it still needs technicians in that warehouse. If a care facility introduces robotic assistance, it still needs human supervisors on that floor. The retraining pathway is shorter and the geography is the same. But this advantage holds only if companies invest in transition programs before the displacement happens rather than after, and only if the ratio of new jobs to lost jobs is large enough to absorb more than a fraction of the workforce.

The Sanders report captured this tension with a line that deserves attention beyond the political context in which it was delivered: "A factory worker who loses their job cannot be told to learn to code if artificial labor also takes the coding job." Humanoid robots will create work, but they may also close off the escape routes that previous waves of automation left open.

But displacement is not only a domestic story. For decades, wealthy nations have filled their hardest jobs by importing workers from poorer ones. If robots can do that work instead, the consequences reach far beyond any single warehouse or retraining program.

The Migration Equation#

In 1942, the United States and Mexico signed the agreement that created the Bracero Program. American men were leaving for war, and the farms of California, Texas, and Oregon needed hands. Over the next twenty-two years, roughly 4.6 million labor contracts brought Mexican workers north to harvest sugar beets, pick cotton, and lay railroad track. The program was supposed to be temporary. It was not. When it finally ended in 1964, the migration networks it had created kept functioning without it. By the 2000s, over 10 percent of all people born in Mexico lived in the United States.

Across the Atlantic, a parallel story unfolded. West Germany signed a recruitment agreement with Italy in 1955, then with Spain, Greece, Turkey, Morocco, Portugal, Tunisia, and Yugoslavia over the following decade. The Gastarbeiter, guest workers, were supposed to rebuild Germany's infrastructure and go home. Many stayed. Today, Turkish is the second most spoken language in Germany.

The pattern repeated in the Persian Gulf after the 1970s oil boom. South Asian and Southeast Asian workers poured into Saudi Arabia, the UAE, Qatar, Kuwait, Bahrain, and Oman to build cities in the desert. By 2024, migrants accounted for roughly 70 percent of the workforce across the Gulf Cooperation Council states. In Qatar and the UAE, that figure exceeded 90 percent of private-sector employment. The kafala sponsorship system gave employers near-total control over workers' mobility, creating conditions that human rights organizations have compared to forced labor.

Each of these migration waves followed the same logic: wealthy countries needed physical labor that their own populations could not or would not provide. The workers came. The economies grew. The workers stayed, or were replaced by new waves from the same sending countries. Entire national economies came to depend on the cycle. The Philippines alone sent 2.33 million workers abroad in 2023, generating $38.3 billion in remittances, roughly 8.3 percent of the country's GDP. Pakistan dispatched 862,000 workers in the same year, 96 percent of them to the Gulf.

Humanoid robots threaten to break this cycle.

Consider Saudi Arabia. The Kingdom is simultaneously pursuing two goals that have historically been contradictory: reducing dependence on foreign labor through nationalization programs like Saudization, and accelerating economic diversification under Vision 2030. Humanoid robots offer a way to reconcile the contradiction. In October 2025, QSS AI & Robotics signed a non-binding framework agreement with UK-based Humanoid for the potential deployment of up to 10,000 humanoid units over five years, with local assembly at a robotics factory in Riyadh. The deployment targets manufacturing, logistics, energy, and infrastructure, precisely the sectors where migrant workers from South Asia currently dominate.

Saudi Arabia is not alone. Any country that relies heavily on imported labor for construction, agriculture, manufacturing, or domestic care now has a potential alternative. A robot does not require a visa. It does not need housing, healthcare, or legal protection. It does not trigger debates about cultural integration, religious accommodation, or demographic change. For governments that have long struggled with the political costs of migration, this is a powerful proposition.

But the proposition has a hole in it, and the hole is consumption.

A migrant worker is not just a pair of hands. A migrant worker is also a mouth, a wallet, a tenant, a parent. Migrants earn wages, pay taxes, rent apartments, buy groceries, enroll children in schools. They create demand. A construction worker from Kerala who moves to Dubai spends money in Dubai. His remittances support a family in Kerala that spends money there. Both economies benefit from his labor and his consumption.

A humanoid robot produces but does not consume. It fills a position on a warehouse floor or a construction site, but it does not eat at the restaurant down the street. It does not rent a flat. It does not pay income tax. The wealth its labor generates flows to whoever owns the robot, concentrating capital rather than circulating it. For countries already grappling with aging populations and shrinking domestic demand, replacing human workers with machines solves a labor problem while deepening a demographic one.

This matters most in Europe, where the math is brutal. Fertility rates have fallen below replacement in nearly every EU member state. As of 2023, Germany's rate sat at roughly 1.35 children per woman. Italy's at 1.20. Spain's at 1.12. Without immigration, these populations will shrink and age simultaneously, straining pension systems, healthcare, and the tax base that funds both. Robots can fill jobs, but they cannot fill maternity wards or pay into social security or raise the next generation of workers and taxpayers. For Europe, replacing migrants with machines may ease short-term political tensions around integration while accelerating the long-term demographic crisis.

The impact on sending countries could be severe. The Philippines, Pakistan, Bangladesh, India, Nepal, and dozens of smaller nations have built development strategies around labor export. Remittances from overseas workers are not pocket money; they are macroeconomic pillars. For the Philippines, $38 billion a year. For Pakistan, $27 billion. These flows fund housing, education, healthcare, and consumption that would otherwise not exist. If the demand for migrant labor in the Gulf, Europe, and North America contracts because robots can do the same work, the remittance pipelines that sustain these economies will narrow. The consequences would ripple through entire regions.

None of this will happen overnight. Migrant labor is deeply embedded in the political economies of both sending and receiving countries. Employers in the Gulf profit from the kafala system. Recruitment agencies in Manila, Dhaka, and Islamabad employ millions. Governments on both ends depend on the arrangement. But the direction is clear. As humanoid robots become cheaper, more capable, and easier to deploy, the economic case for large-scale labor migration will weaken. Immigration policy will shift from a question of labor supply to one of demographics, humanitarian obligation, and political choice.

Humanoid robots will not end migration. People will still move in search of safety, freedom, and opportunity. But robots could do to labor migration what containerization did to port work or what mechanization did to farming: shrink the workforce required, concentrate the gains among owners of capital, and leave the displaced communities to absorb the cost.

The Political Battlefield#

Humanoid robots are going to scramble political coalitions in ways that no one is fully prepared for.

The existing political grammar divides neatly: the left defends workers, the right defends markets. Automation has always complicated that division, but humanoid robots will blow it apart. When a machine can do the physical work of a warehouse picker and the caregiving work of an eldercare aide, every political faction has to decide what it actually stands for.

The Robot Tax Debate

The most concrete policy proposal to emerge so far is the robot tax: a levy on companies that replace human workers with machines, designed to slow adoption and fund retraining. Bill Gates endorsed the idea in a widely discussed 2017 interview, arguing that if a human doing $50,000 worth of factory work gets taxed on that income, a robot doing the same work should be taxed at a similar level. Mark Cuban and Robert Shiller have voiced support. In October 2025, Senator Bernie Sanders, then the ranking member of the Senate HELP Committee, released a report warning that AI and automation could eliminate nearly 100 million American jobs over the next decade, and called for robot taxes on large corporations to fund worker transition programs, a 32-hour workweek with no loss in pay, mandatory profit-sharing (20 percent of corporate stock to workers), and worker representation on 45 percent of corporate boards.

South Korea became the first country to implement anything resembling a robot tax in August 2017, when President Moon Jae-in's government reduced tax deductions for corporate investment in automation equipment by up to two percentage points. It was modest: not a direct tax on robots, but a reduction in the incentive to deploy them. The European Parliament had rejected a more ambitious robot tax proposal earlier that same year, with EU Commissioner Andrus Ansip warning that any jurisdiction implementing one would become less competitive.

The critics are not wrong about the difficulties. Former U.S. Treasury Secretary Lawrence Summers called Gates's proposal "profoundly misguided," arguing that a sufficiently high robot tax would simply prevent robots from being produced. The definitional problem is real: where does a programmable logic controller end and a robot begin? A robot tax would need to distinguish between a self-checkout kiosk, an automated conveyor system, a collaborative robotic arm, and a full humanoid. Research from Northwestern's Kellogg School found that while a robot tax can reduce inequality, it would need to be extremely high to meaningfully offset the downward pressure of machines on routine workers' wages, and at that level it would distort overall production.

But the most telling critique may be political rather than economic. Capital gains taxes already apply to productive investment. Corporate income taxes already capture automation profits. Property taxes often include machinery. A robot tax stacks another layer on top, and every corporation with a lobbying budget will find ways to reclassify its robots as something else. The policy debate may be less about whether to tax robots and more about whether existing tax systems can adapt fast enough to capture the value that humanoid robots generate while the human tax base erodes.

Strange Bedfellows

The deeper political disruption is not about any single policy. It is about the coalitions that form around the question of who benefits from machines.

Research from Brookings, drawing on studies by Kurer (2020), Im et al. (2022), and Boewein et al. (2024), has documented a consistent pattern: automation exposure increases support for populist right-wing parties across Europe. Workers who lose their jobs to automation tend to disengage from politics entirely or drift toward the left. But workers who keep their jobs while watching automation advance around them, the "survivors" who fear losing their status, move toward the populist right. This helps explain the rise of Alternative for Germany, Fratelli d'Italia, Rassemblement National, and Vox, all of which have surged in regions experiencing economic disruption.

In the United States, the political realignment is equally disorienting. At the 2024 Republican National Convention, Teamsters president Sean O'Brien became the first head of the nation's largest labor union to speak at the event. On the same day, Trump chose J.D. Vance, who had built his political identity around the grievances of displaced blue-collar workers, as his running mate. The Republican Party, historically hostile to organized labor, now includes voices arguing for a reformed vision of worker protection, even as the Trump administration's actual policies have gutted labor protections, paralyzed the National Labor Relations Board, and slashed social programs.

On the left, the response has been uneven. Sanders's HELP Committee report represents the most aggressive congressional intervention to date, but his proposals, which include mandatory worker representation on corporate boards and equity distribution, remain far from legislative reality. The UAW's Shawn Fain has called for an independent workers' political program, noting that among union members across party lines, the top priorities are wages, healthcare, retirement, and control over time, not immigration or culture war issues. A survey by the Center for Working Class Politics found that 57 percent of respondents in Rust Belt states viewed a hypothetical independent workers' party favorably.

Meanwhile, the welfare state itself appears to moderate the political effects of automation. Research shows that unemployment has no statistically significant effect on populist support when unemployment benefits are high and labor market protections are strong. Strip those protections away, and economic shocks translate directly into far-right gains. This finding has direct implications for how humanoid robotics will play politically in different countries. Germany's works councils and sectoral bargaining may absorb the shock. America's threadbare safety net almost certainly will not.

The result is a political landscape where traditional allies become opponents and traditional opponents find common ground. Tech executives who fund Democratic campaigns are building the machines that displace Democratic voters. Republican populists who rail against coastal elites are defending workers whose jobs those elites are automating. Unions that have historically aligned with the left are finding more rhetorical sympathy on the right, even as right-wing governments dismantle the institutional infrastructure that gives unions power. Humanoid robots will not just change the economy. They will rearrange who sits on which side of every political argument about what the economy is for.

Diverging Paths#

The effects of humanoid robotics will not distribute evenly across the world. They will follow fault lines that already exist: between aging societies and young ones, between countries that build robots and countries that supply the labor robots are designed to replace.

The Transatlantic Split

Europe and the United States are approaching humanoid robotics from opposite directions, driven by opposite fears.

The European Union's AI Act, which entered force in August 2024 and began phased enforcement in February 2025, classifies AI systems by risk level and imposes binding compliance requirements on high-risk applications. Robotics in workplaces, healthcare, and critical infrastructure falls squarely into the high-risk category. Violations can trigger fines of up to 35 million euros or 7 percent of global turnover. The EU has also begun incorporating robotics into its liability frameworks, though a proposed AI liability directive was abruptly cancelled in the Commission's 2025 work program, signaling an internal tug-of-war between regulation and competitiveness.

Europe's regulatory instinct is inseparable from its demographic crisis. The EU's total fertility rate hit 1.38 in 2023, the lowest ever recorded, with a 5.4 percent drop in births from the prior year. Italy recorded 1.18 children per woman in 2024, a new record low. Germany fell to 1.35, its lowest since 1994. Spain registered 1.12. France, historically Europe's fertility outlier, dropped to 1.62, its lowest since World War I. Finland hit 1.25, the worst figure in its records going back to 1776. Europe's population peaked in 2021 and is now declining. In Germany, the Federal Statistical Office stated plainly that net immigration was the sole reason for any population growth at all.

This creates a paradox. Europe needs labor. Robots could provide it. But European political culture remains deeply suspicious of unregulated technology deployment, and European labor institutions, from works councils to sectoral bargaining, are designed to protect incumbent workers. The result is likely to be cautious, negotiated adoption: robots introduced through collective agreements, with transition funds and retraining programs attached.

The United States has taken the opposite approach. The Trump administration's 2025 AI Action Plan prioritized deregulation and speed, explicitly rejecting what it characterized as European-style precautionary regulation. At the February 2025 Paris AI Action Summit, Vice President Vance warned Europe to ease tech regulation, and the U.S. declined to sign the EU-backed "Inclusive and Sustainable AI" declaration. Federal AI governance in the U.S. remains a patchwork: no comprehensive national framework exists, though over 500 AI-related bills were introduced at the state level in the first quarter of 2025 alone. The prevailing philosophy is that market forces and voluntary industry standards will allocate robots more efficiently than bureaucratic oversight.

What this means in practice is that American workers may encounter humanoid robots sooner but with fewer institutional buffers. No sectoral bargaining structures exist to negotiate the terms of deployment. No federal transition fund cushions the adjustment. The speed of adoption will be faster, and the disruption will be more concentrated in communities without the political infrastructure to bargain for protection.

The Global South: A Ladder Pulled Up

But the sharpest consequences of humanoid robotics may land far from either Washington or Brussels.

For most of the twentieth century, economic development followed a recognizable path. Poor countries moved workers from farms to factories. Factories produced goods for export. Export earnings funded infrastructure, education, and eventually a transition to services. South Korea did this. Taiwan did it. China did it on a scale unprecedented in human history. The path was not easy and it was not clean, but it was legible: industrialize, export, grow.

Harvard economist Dani Rodrik has documented how this path is narrowing. In a process he calls "premature deindustrialization," developing countries are running out of industrialization opportunities at much lower income levels than the countries that industrialized before them. Latin America and sub-Saharan Africa have been hit hardest, with manufacturing employment peaking earlier and falling faster than it did in the advanced economies. The causes are multiple: trade liberalization, Chinese competition, labor-saving technology. But the consequence is singular. The main channel through which rapid economic growth has historically occurred is being constricted.

Humanoid robots threaten to close it entirely.

Consider Bangladesh. The garment industry is the backbone of the national economy, accounting for roughly 84 percent of total export revenue. In its peak years, the sector employed around 4 to 4.5 million workers, a majority of them women who migrated from rural villages to factory floors in Dhaka and Chittagong. Garment exports reached $39.35 billion in fiscal year 2024-25. This is not an abstract economic statistic. It is the mechanism through which Bangladesh has sustained GDP growth of around 6 percent annually for decades, lifted millions out of poverty, and is on track to graduate from least-developed-country status.

Now imagine that mechanism breaking. Automated sewing, cutting, and finishing systems are already advancing. If buyers in Europe and the United States can source garments from robotic factories closer to home, or from facilities in countries with cheaper energy and no labor costs at all, the comparative advantage that built Bangladesh's economy disappears. The four million workers do not seamlessly transition to service jobs. Bangladesh does not have the educational infrastructure, the capital base, or the domestic consumer market to absorb them. They go back to the villages, or they crowd into urban informal economies where productivity is low and wages are lower.

Vietnam presents a different variant of the same vulnerability. The country has climbed the manufacturing ladder faster and further than Bangladesh, moving from textiles into electronics, with manufacturing exports reaching $356 billion in 2024. Samsung alone accounts for roughly a fifth of Vietnam's total exports, and the electronics sector employs over 1.5 million workers. Vietnam's manufacturing workforce is nearly 12 million people. But the same integration into global supply chains that drove growth also creates exposure. If automation in wealthy countries makes it economical to reshore production, Vietnam's factory workers face the same threat as Bangladesh's garment workers, just at a higher rung on the ladder.

The research is beginning to confirm this mechanism empirically. A 2023 study of Brazil found that automation in advanced export-destination countries reduced manufacturing employment in Brazil while pushing the economy toward raw-material extraction. Foreign robots, in other words, were deindustrializing Brazil without ever crossing the border. The authors concluded that global automation may contribute to premature deindustrialization in emerging economies.

This is the cruelest asymmetry of the humanoid robotics revolution. The countries that build the robots will capture the productivity gains. The countries that currently supply the labor those robots replace will lose their primary engine of development. And unlike previous waves of automation, which at least required cheap human hands somewhere in the supply chain, humanoid robots sever that dependency entirely. A factory staffed by humanoid robots in Ohio or Bavaria does not need workers in Dhaka or Ho Chi Minh City at any price.

The development ladder that lifted hundreds of millions of people out of poverty over the past half-century is not merely getting harder to climb. It may be getting pulled up behind the countries that already reached the top.

---

There is a version of this story that ends well. In that version, the shoulder you wrecked in year six never gets wrecked at all, because the robot does the overhead lifting before you ever have to. You move into a technician role at the same facility, maintaining the fleet that took over your old station. The pay is better. You are home for dinner. Your kids see you on weekends. The company funded the retraining because it was cheaper than turnover, and the union negotiated the transition because it still had the leverage to do so.

There is another version. In that version, you get the packet and the practiced gentleness and then nothing. The retraining program has a six-month waitlist. The technician jobs require certifications you cannot afford. The warehouse three towns over automated last year. You drive for a delivery app for a while, but the delivery app is testing autonomous vehicles, and you can see where that road ends.

Both versions are plausible. Both are already happening, in fragments, in different buildings and different countries. The distance between them is not technological. The machines will arrive regardless. The distance is political: who decides how the gains are shared, who bears the cost of transition, whether the people who built the economy with their hands get to participate in the economy that no longer needs them.

This book has tried to lay out what is coming with enough specificity to be useful and enough honesty to be uncomfortable. Humanoid robots will change work, migration, care, war, and the basic contract between effort and reward that most societies are built on. The robots are getting ready. Whether the rest of us are is a different matter entirely.