Every few months, a new set of projections circulates about how many jobs robots will eliminate by some future date. The numbers vary by orders of magnitude depending on the source — anywhere from ten million to eight hundred million, depending on methodology, definitions, and apparent willingness to speculate. The humanoid robot specifically has become the focal point of these anxieties in a way that earlier industrial robots never quite were, partly because the human form factor makes the substitution feel more direct and personal.
The anxiety is understandable. The specific projections are almost entirely unreliable. And the actual research on automation and employment — which is more careful, more limited in scope, and considerably less dramatic than the headlines — points to a more complicated picture that is worth understanding on its own terms.
This is not an argument that humanoid robots won't change work. They almost certainly will, in specific industries and specific roles, over a timeframe that is genuinely uncertain. It is an argument for reading the evidence carefully rather than borrowing the most alarming available number and treating it as settled.
What the Industrial Robot Research Shows
The most rigorous evidence on automation and employment comes not from speculation about humanoids but from decades of data on industrial robots — the fixed-arm machines that have been operating in automotive and manufacturing plants since the 1960s and 1970s. This research gives us a baseline for understanding what automation actually does to labour markets when it arrives at scale.
A widely cited 2017 study by economists Daron Acemoglu and Pascual Restrepo examined US labour markets between 1990 and 2007 and found that each additional robot per thousand workers was associated with a reduction of between 3.3 and 6.2 workers in that local labour market. That's a real displacement effect. It also found wage suppression in affected areas. The impacts were concentrated in manufacturing-heavy regions, which is consistent with where industrial robots were actually deployed.
What the same research does not show is that total employment fell as a result. Nationally, employment continued to grow through the same period. The displaced workers found other work — often at lower wages, often in different sectors, with real hardship in the transition. But the macro-level employment apocalypse predicted in the more dramatic projections did not materialise from industrial robotics, even over a multi-decade horizon.
This is the pattern that most labour economists point to when asked about automation: displacement in specific occupations and regions, followed by adjustment that is often slow and painful at the individual level, without the aggregate employment collapse that the most alarming projections suggest. That pattern has held across multiple waves of automation over two centuries. There is no obvious reason to assume humanoid robots will break it — though there is also no certainty that they won't.
Why Humanoids Are a Different Question
Industrial robots have a structural limitation that matters for the employment analysis: they are fixed in place and designed for a single, highly specific task. An automotive welding robot does one thing in one location. Redeploying it to a different task is expensive and often impractical. This limits the range of jobs industrial robots can substitute for, and it means their employment effects are concentrated in a relatively narrow set of occupations.
Humanoid robots, if they reach the capability levels their developers claim, would be different in a specific way: they are designed to operate in environments built for humans, doing tasks that require mobility, dexterity, and situational judgment. The theoretical substitution range is much broader. A robot that can navigate a warehouse, handle varied objects, and respond to an unstructured environment could potentially substitute for a wider range of physical labour than a fixed industrial arm.
This is why humanoids attract more attention from labour economists than their current deployment status would otherwise justify. The potential scope of substitution, if the technology matures as claimed, is genuinely different from prior automation waves. The key phrase is "if the technology matures as claimed" — which is doing a great deal of work in that sentence, and which the current state of the industry does not yet support treating as settled.
The Capability Gap and What It Means for Timelines
The gap between what humanoid robots can currently do reliably and what would be required for the broader labour substitution scenario is significant, and understanding it is necessary for any honest discussion of employment effects.
Current commercial humanoid deployments — Agility Robotics' Digit at Amazon, and the various pilot programmes announced by Figure, 1X, and others — are operating on narrowly defined, highly repetitive tasks in structured environments. Moving standardised containers between fixed points. Operating in sections of facilities that have been modified to accommodate the robot's current capabilities. Performing under human supervision, with human intervention available for anything the robot handles poorly.
These are real deployments doing real work. They are not evidence that humanoid robots can currently substitute for the broad range of physical labour that the more alarming projections describe. Unloading a delivery truck with varied package sizes, shapes, and weights — the kind of task a human warehouse worker handles as routine — is a problem that current humanoid systems approach inconsistently at best. Working safely alongside humans in a genuinely unstructured environment, making the kind of real-time situational judgments that workers make continuously, is not a solved problem.
The pace at which these capability gaps close matters enormously for the employment question. If humanoids progress quickly toward general-purpose physical capability over the next five to seven years, the labour market implications are more pressing than if that progression takes fifteen to twenty years. The honest answer is that no one knows the timeline with the precision that the specific projections require, and the companies with the strongest financial incentive to predict rapid progress are the ones making the most specific claims.
Which Workers and Which Industries Face the Most Exposure
Rather than aggregate projections, the more useful framing is: which specific occupations and industries are most likely to see humanoid deployment first, and at what scale?
The early deployment pattern points clearly toward structured logistics — warehouses, fulfilment centres, distribution facilities — and towards manufacturing environments where tasks are repetitive and the physical environment can be partly standardised around the robot's capabilities. These are the sectors where the cost-benefit case for humanoid deployment is clearest, where the technology is closest to sufficient, and where labour shortages and physical demand are already creating pressure on employers to find alternatives.
Construction is often cited as a sector with high potential for humanoid deployment, given the physical demands and the labour shortages that have characterised the industry. The challenge is that construction environments are among the least structured of any industry — sites change constantly, tasks vary widely, and safety requirements are stringent. The technical bar for reliable autonomous operation in construction is substantially higher than in a warehouse.
Healthcare is another frequently cited sector, primarily for physical care tasks — patient transfer, mobility assistance, supply logistics within facilities. Here the regulatory environment, the proximity to vulnerable people, and the requirements for reliable, safe operation create barriers to deployment that are distinct from the technical ones.
The common thread across all of these: the sectors most exposed to humanoid deployment in the near term are those where the tasks are most repetitive and the environments most controlled. The workers most exposed in the near term are those performing physically demanding, low-variation work in environments that can be partly adapted around the robot. That is a real group of people, and their exposure is worth taking seriously — not as a basis for projecting aggregate employment collapse, but as a basis for thinking carefully about how the transition is managed.
What the Honest Uncertainty Looks Like
The most credible position on humanoid robots and employment is one that holds several things simultaneously: the technology is advancing, real deployments are happening, the theoretical scope of labour substitution is broader than prior automation waves, the timeline is genuinely uncertain, and the historical pattern of automation suggests adjustment without aggregate collapse — while acknowledging that historical patterns are not guarantees.
What that position does not support is treating any specific projection as reliable, in either direction. The researchers who have done the most rigorous work on automation and employment are notably cautious about extending their findings to humanoid robots at scale, precisely because the deployment is not yet at a scale where the data exists to draw the same kinds of conclusions they have drawn about industrial robots.
The more useful question for policymakers, workers, and industries is not "how many jobs will humanoid robots eliminate?" — a question that cannot currently be answered with any reliability — but "which workers and industries should be preparing now, and what does that preparation look like?" That question has more tractable answers, and it points toward the specific, grounded analysis that is actually useful rather than the dramatic projections that generate attention but shed very little light.