Humanoid robot demonstrations have improved to the point where they are difficult to distinguish from staged footage. Machines walk on uneven ground, recover from shoves, manipulate objects with something close to dexterity, and follow spoken instructions.
Almost none of that translates directly into deployment, and the reasons are instructive.
What a demo omits
Duration, mostly. A two-minute video shows the successful attempt. It does not show battery life measured in a couple of hours, the setup time before each run, the environment prepared for the task, or the failure rate across repeated attempts.
It also omits cost. A machine that performs a task less reliably than a person and costs more to buy and maintain has no business case, however impressive the motion looks.
Why the humanoid form at all
The argument is that the world is built for human bodies – stairs, door handles, tools, shelf heights – so a human-shaped machine can operate in it without modification.
That argument is real but weaker than it sounds. Modifying an environment is usually far cheaper than building a machine general enough to handle an unmodified one. Warehouses did not deploy humanoids; they redesigned around wheeled robots and did it a decade ago.
Where deployment is real
Narrow, repetitive, physically unpleasant tasks in controlled settings. Moving containers between fixed points. Inspection rounds in hazardous environments. Machine tending on a production line.
These are jobs where the task is fixed, the environment is known, and the value is in not requiring a person to be there.
The bottleneck
Manipulation, not locomotion. Walking is largely solved. Picking up an unfamiliar object of unknown weight and grip characteristics, without crushing it or dropping it, remains genuinely difficult, and it is what most useful physical work consists of.
A reasonable expectation
Steady expansion into structured industrial settings over the next several years. General-purpose machines doing household work remain considerably further out than the demonstrations suggest, and the gap is in the boring parts.