A robot that can carry a box is not automatically the right way to move boxes.

That sounds obvious. Yet much of the conversation about the next generation of automation gravitates toward machines shaped like us. We ask whether a humanoid can walk through a factory, use existing tools, and perform work designed for people.

Those are fair questions. But they are not the first questions I would ask.

The RSM article on the next wave of manufacturing automation prompted me to think about this distinction. I agree with its emphasis on readiness: manufacturers should know where automation creates value and what the work requires. But I keep coming back to a more basic question: Are we choosing a robot for the job, or choosing a job for the robot we are excited to buy?

Start with the route, not the robot

Imagine cartons moving from a receiving area to a packing line along a stable route. One vision might involve a fleet of humanoid robots, each picking up and carrying boxes. Another might use a palletizing cell at one end, a conveyor or tugger along the route, and a second cell to unload or restack at the other end.

That is a hypothetical, not a claim about a specific plant or a universal answer. The point is that the two designs solve the same material-flow problem in very different ways.

That choice also changes the work around the machines: who moves each box, who clears a jam, and where people spend time when volume shifts.

If volume is steady and the route rarely changes, a dedicated system may move more material with fewer handoffs. If destinations change often, aisles are shared, and the work varies from hour to hour, a mobile general-purpose robot may earn its place. A humanoid could be useful where the environment is built around people and changing the facility would cost more than buying flexibility.

But the shape of the machine does not settle the business case. We still need to compare throughput, uptime, safety, floor space, maintenance, integration, recovery from jams, and the cost of exceptions. The relevant number is not the price of one robot. It is the cost and performance of the whole system doing the work.

Human-shaped is a design choice, not a goal

There is a real reason humanoids attract attention: our buildings, tools, stairs, carts, and workstations were designed for people. A robot that can use those spaces might reduce the need to redesign them.

That advantage matters most when work is varied, the environment is difficult to change, and the robot can move between useful tasks. It matters less when the job is repetitive and predictable, or when a simple conveyor, lift, fixture, or automated cart can remove the bottleneck.

Human form is one way to buy flexibility. It is not flexibility for free. A humanoid design can require more joints, actuators, sensors, and control than a task-specific machine. If a conveyor, palletizer, or cart can do the same work with fewer, simpler components, a human form means paying for flexibility the process may not need. That cost can show up in purchase, maintenance, integration, and downtime. Human likeness may feel familiar or compelling. But aesthetics alone do not create operating value. The tradeoff is not automatically wrong. It should be explicit.

The International Federation of Robotics makes a similar distinction: humanoids may complement and expand existing robotics, rather than replace it. Its 2025 position paper also notes that the timing of mass adoption remains uncertain. That is a useful counterweight to treating a promising form factor as an inevitable destination.

Human likeness is not a proxy for intelligence. A machine does not need a human body to spot a misplaced carton. An AI system need not reason like a person to find a defect or route an exception. Judge both by the work: can they perform reliably and detect and handle failure?

Compare complete systems

Before selecting a form factor, I would map the work from end to end:

  • What is moving, how often, and where does the flow slow down?
  • Which parts are stable, and which depend on judgment or change?
  • What happens when a box is damaged, a route is blocked, or a label is unreadable?
  • How much supervision, maintenance, and recovery does each option require?
  • What would it take to change the process or facility instead?

Then I would compare several designs, not just several robot vendors. The candidates might include process changes, conveyors, tuggers, fixed robotic cells, collaborative robots, mobile robots, and humanoids. For each one, I would measure the complete workflow: units moved, time lost, labor redirected, safety exposure, downtime, and the work required to keep it running.

That last part matters. A machine may move boxes quickly when everything goes right and still create a new exception queue when it does not. A deployment is only as useful as its performance across ordinary shifts, awkward loads, blocked routes, maintenance windows, and system outages.

Use the right amount of generality

Specialized automation can excel at a narrow task, though it may need changes when the product, route, or volume changes. Adoption may hinge less on making every machine general and more on making specialized tools cheaper and easier to deploy. Imagine cutting a vision system’s price in half and simplifying configuration. That could bring inspection capability to more stations without waiting for a general-purpose robot. A humanoid’s adaptability matters only when it solves a real operating constraint.

So I would not frame this as humanoids versus traditional automation. I would ask how much generality the work actually needs.

For stable, high-volume movement, a purpose-built flow may be the stronger answer. For work that shifts between locations and tasks in spaces that are costly to redesign, a more flexible robot may make sense. And some processes may not need a robot at all; clearer routing, better containers, or a redesigned handoff may solve the problem with less complexity.

For an executive team, this is a capital-allocation choice: fund flexibility where it earns its cost, and reduce the price and setup friction of specialized tools where they can scale.

Humanoid robots may have an important role. But the goal should not be to make every robot look like a worker. It should be to make the work safer, more dependable, and more productive—with the simplest system that fits the job.

The question I would put in front of any automation proposal is: What is the least complicated system that reliably removes this bottleneck?

References