After
the Job

The Numbers

What the Data Actually Says

The headline figures have circulated widely enough to lose their weight. Goldman Sachs estimated in 2023 that AI and automation could expose 300 million full-time jobs globally to displacement or significant alteration. The IMF published analysis in early 2024 suggesting that 40 percent of jobs in advanced economies are exposed to AI in ways that could reduce employment, with 60 percent of jobs in high-income countries facing that risk. McKinsey projected that by 2030, up to 30 percent of work hours performed in the United States could be automated using technologies that already exist. The World Economic Forum's 2025 Future of Jobs report struck a different tone, projecting a net creation of 12 million jobs as automation opens new roles — but also the elimination of 85 million roles, a displacement that will not fall evenly.

These numbers are difficult to compare because they measure different things. Some count jobs eliminated; others count tasks automated within jobs that persist. Some focus on the next two years; others model to 2035. The variation reflects genuine uncertainty — about the pace of deployment, about regulatory responses, about which industries will move fastest, and about the degree to which new jobs materialize for the people whose old ones disappear. What is not in dispute is the direction.

On the Floor

Who Is Deploying, and Where

The humanoid robot sector, which barely existed as a commercial category in 2021, has attracted billions in capital and produced a dozen credible platforms. The leaders differ in focus, architecture, and deployment strategy — but all are moving from demonstration to production.

The logistics sector is moving fastest, and for straightforward economic reasons. A warehouse tote-pulling job — repetitive, rules-based, physically demanding — is exactly the kind of task that robotics handles well. GXO Logistics, one of the world's largest third-party logistics providers, has been explicit that humanoid robots are part of its long-term labour strategy. Amazon operates 750,000 robots across its network already, the vast majority of which are not humanoid but mobile shelving units — but the company has been piloting humanoid systems and has invested directly in Agility Robotics.

Goldman Sachs research published in 2024 projected the humanoid robot addressable market at $38 billion by 2035, driven primarily by manufacturing and warehousing. That estimate has been revised upward since, as production costs have fallen faster than expected. The cost of a capable humanoid platform — around $150,000 to $250,000 in 2023 — is expected to fall below $50,000 within a decade, at which point the economics of substitution become unambiguous for an enormous range of tasks.

Dangerous Work

The Jobs Humans Should Not Do Anyway

There is a category of displacement that tends to get lost in the economic anxiety conversation — work that is not being taken from people but offered to machines because the work is genuinely hazardous, and because machines can do it without dying.

The oil and gas industry kills roughly 1,000 workers per year in the United States alone and injures tens of thousands more. Offshore platform inspection, pipeline survey, wellhead intervention — tasks requiring workers to operate in environments of fall risk, pressure hazard, and toxic exposure — are being automated with underwater drones and surface-level robotic inspection systems. BP, Shell, and Equinor have all piloted autonomous inspection platforms that can survey North Sea infrastructure without a human crew. The argument for automation here is not productivity: it is that the status quo is a body count.

Mining presents a similar case. Underground hard rock mining — the extraction of gold, copper, lithium, and the rare earth elements that power electric vehicles and electronics — is among the most lethal industries on earth. Rio Tinto operates fully autonomous truck fleets at its Pilbara iron ore operations in Australia: 130-tonne vehicles navigating 24 hours a day with no driver aboard, controlled from an operations centre in Perth, 1,500 kilometres away. Fatalities in those operations have dropped to near zero. BHP and Vale have announced similar transitions. The workers displaced are real people with real livelihoods — but the work they have been displaced from was killing them.

What Comes Next

The Displacement Problem Has No Easy Resolution

The optimistic version of the automation story runs like this: every previous wave of labour-displacing technology — the loom, the steam engine, the tractor, the computer — ultimately created more jobs than it destroyed, by lowering costs, expanding output, and creating demand for goods and services that did not previously exist. Automation raises productivity; productivity raises living standards; rising living standards create new industries and new employment. The World Economic Forum's net positive figure of 12 million jobs is in this tradition: a claim that the coming wave will follow the historical pattern.

The pessimistic version is that this time is different in a specific way: the speed of displacement may outrun the economy's capacity to retrain workers and create new absorptive demand. Previous automation waves unfolded over decades, giving labour markets time to adjust. The current wave is unfolding in years, and unlike previous automation that replaced physical labour while expanding demand for cognitive labour, AI is now moving into cognitive tasks simultaneously. The fallback that absorbed displaced factory workers — clerical work, service work, professional services — is itself under pressure.

The geography of displacement matters too. Warehouse and logistics work is not distributed evenly across the economy. It is concentrated in specific communities — often communities that were already absorbing the aftershocks of earlier deindustrialisation. The workers sorting packages in Memphis and Stockton and Allentown are not abstractly represented in a macro labour statistic. They are people whose skill set was developed for a specific task that is being automated away, in communities where alternative employment is limited, in a policy environment that has so far produced more discussion than action.

What the data cannot tell us is what comes after. The Industrial Revolution created conditions of profound misery for a generation before it created conditions of unprecedented prosperity for many. The transition period was not a footnote: it was decades of child labour, urban poverty, broken families, and political upheaval. Whether the current transition will follow a similar pattern — misery then prosperity, unevenly distributed — or whether policy intervention can smooth the curve, is the central political question of the next twenty years. Economists can model scenarios. Governments will determine which scenario materialises.

What is not in doubt is that the machines are already on the floor. The question of what we owe the people who built what those machines are now replacing is not a technical question. It is a moral one — and so far, it has mostly gone unanswered.