The workforce conversation around humanoid robots has been almost entirely one-directional. How many jobs might robots displace? What happens to warehouse workers, assembly line operators, delivery drivers? These are legitimate questions, and the honest answer — that the effects are uncertain, gradual, and highly dependent on how companies and policymakers respond — rarely satisfies people who want a cleaner answer in either direction.
What gets almost no coverage is the other side of the workforce equation: the people humanoid robots will create demand for. Not just software engineers at the robot companies, but the technicians, operators, and maintenance specialists who will need to exist in large numbers at the facilities where these machines actually run. Right now, those people are in short supply — and the shortage is a real constraint on how fast humanoid deployment can actually scale, regardless of what the hardware achieves.
What Running a Humanoid Robot Actually Requires
When a humanoid robot is deployed in a warehouse or factory, it does not arrive as a turnkey solution that runs itself. Someone has to commission it — configure it for the specific environment, map the facility, define its task parameters, and verify that it operates within the bounds it's supposed to. Someone has to monitor it during operation, at least initially, and intervene when it encounters situations its software can't handle autonomously. Someone has to perform routine maintenance: inspecting joints and actuators, replacing worn components, updating software, recalibrating sensors. And someone has to diagnose and resolve failures when they occur — which, in early-stage deployments, is often.
This is not a single role. It's a cluster of roles that blend skills from several existing technical fields. Conventional industrial robotics — the fixed-arm robots that have populated automotive and electronics factories for decades — has a mature workforce to support it. There are established training programmes, industry certifications, and a large pool of technicians who know how to maintain those systems. The companies that make conventional industrial robots have invested heavily in training infrastructure, partly because keeping their installed base running is a significant part of their business model.
Humanoid robots are different enough from conventional industrial robots that the existing workforce doesn't transfer cleanly. A technician skilled in maintaining a fixed-arm welding robot has relevant background — experience with actuators, control systems, and sensor calibration — but a humanoid robot adds layers of complexity that go well beyond that experience. The locomotion system alone — the combination of sensors, control algorithms, and mechanical components that keeps a bipedal robot upright and moving — requires understanding of dynamics and balance control that has no direct equivalent in conventional robotics maintenance. The perception systems, which typically combine cameras, depth sensors, and AI-driven software to interpret the robot's environment, involve software debugging and machine learning concepts that traditional robotics technicians have not needed to develop.
The Talent Pool Problem
The people who currently have all the relevant skills — mechatronics, perception systems, machine learning, real-time control software, and physical hardware maintenance — are mostly employed at the companies building humanoid robots, or at research institutions. They are not available in the numbers that large-scale industrial deployment would require.
Consider the arithmetic. Agility Robotics has described ambitions to manufacture thousands of Digit units per year at its facility in Salem, Oregon. If a fleet of, say, twenty robots at a single facility requires even one dedicated technical specialist to keep it running reliably, scaling to thousands of deployed units means training thousands of specialists. That training pipeline does not currently exist at anything like that scale.
The robot companies are aware of this. Several have described service and support models that would centralise maintenance expertise rather than distributing it to every deployment site. The logic is appealing: instead of requiring each facility to develop in-house expertise, the robot company sends its own technicians or relies on a network of certified service partners. This is similar to how enterprise IT vendors have historically handled support for complex systems — a managed service model where the expertise stays with the vendor.
The managed service model has real advantages, particularly in early deployment phases when the robots themselves are still evolving and field fixes require current knowledge of the latest software versions. But it also has constraints. Response time when a robot fails depends on technician availability and proximity. For facilities running humanoid robots on shifts where robot uptime is operationally important, waiting hours for a service technician is a problem. The managed service model works better when robots can be taken offline without critical impact than when they are carrying real operational load.
What the Transition From Conventional Robotics Looks Like
The industrial robotics industry — the established players like FANUC, KUKA, ABB, and Yaskawa — has navigated a related challenge before, multiple times. Each major shift in automation technology created a gap between the existing workforce's skills and what the new technology required. Each time, the gap closed over years through a combination of vendor training programmes, community college and vocational curricula, on-the-job learning, and the gradual accumulation of practitioners who grew up with the new technology.
The current generation of conventional industrial robotics technicians trained through exactly this kind of layered system. Community colleges in manufacturing-intensive regions offer two-year programmes in industrial robotics and automation. Vendors like FANUC run their own training academies. Employers supplement that baseline with in-house training specific to their equipment and processes. The whole system has had several decades to mature.
For humanoid robots, that system is at day one. A handful of institutions have begun developing humanoid-specific curricula — Georgia Tech's robotics programme, Carnegie Mellon's Robotics Institute, and several European technical universities have all expanded their coverage of legged and manipulation robotics in the past few years. But translating research-level academic coverage into the kind of practical, hands-on technical training that produces field technicians takes time and requires the cooperation of the robot companies themselves, who need to provide curriculum input, access to hardware for training, and clear certification frameworks.
Some of that cooperation is beginning. Agility Robotics has described plans for a training and certification programme for Digit operators. Figure AI has talked about the importance of service infrastructure. But the specifics are thin, and the programmes are early-stage. The workforce pipeline that would need to exist to support, say, 50,000 deployed humanoid units across multiple industries is nowhere near mature enough to support that scale today.
The Operator Layer
Beyond the maintenance and technical support question, there is a distinct workforce challenge at the operator level — the people responsible for supervising humanoid robots during operation, not repairing them when they break.
Current humanoid deployments are not fully autonomous. They operate with human oversight, particularly for edge cases, error recovery, and situations the robot's software hasn't been trained to handle. This supervisory role — sometimes called fleet supervision or robot operations — is different from both conventional factory supervision and conventional robotics engineering. It requires understanding what the robot is supposed to be doing, recognising when it is deviating from expected behaviour, knowing how and when to intervene, and communicating problems to the technical support team in useful terms.
Some companies have framed this role as an evolution of existing operator jobs. A warehouse worker who currently performs the task the robot is being deployed to do could, in principle, transition into a supervisory role — their knowledge of the task and environment is directly relevant, and their stake in the operation working correctly is clear. The challenge is that this transition requires real training and a genuine change in how the job is understood. Treating it as a simple job reclassification, without investment in that transition, produces supervisors who are neither comfortable with the technology nor adequately compensated for the increased responsibility.
The facilities that have done this best with earlier automation waves — Amazon's deployment of autonomous mobile robots (AMRs) is the most-studied example — invested deliberately in workforce transition programmes that combined technical training with honest communication about how roles were changing. The facilities that did it poorly created workforce friction that slowed adoption and contributed to high turnover in the redefined roles. Humanoid robots, which are more capable and more visible than AMRs, will likely amplify both outcomes depending on how the transition is managed.
A Constraint That Scales Slowly
The hardware and software challenges facing humanoid robotics get most of the attention, and they are real. But the workforce constraint is different in character from a technical problem. You cannot solve a workforce gap with the next product release or a better training dataset. Building the training programmes, the certification frameworks, and the practitioner community that large-scale humanoid deployment requires takes years of consistent investment across multiple institutions and industries, coordinated in ways that no single company controls.
This suggests that even if the hardware and software advance faster than current trajectories suggest — even if a humanoid robot capable of performing a wide range of tasks reliably exists within the next few years — the workforce constraint would remain a real limitation on how quickly deployment could scale. You can't deploy 100,000 robots if there are only enough trained technicians to support 10,000.
The conversation about humanoid robots and work has been almost entirely about what happens to the workers whose tasks robots might perform. That question matters and deserves the attention it's getting. The question of who will be employed to keep those robots running, and whether the training infrastructure to produce those workers exists or is being built, has barely entered the public discussion. It should, because the answer will shape the deployment curve as much as anything happening in the engineering labs.