Elon Musk Is Building the Wrong Robot
This week the internet watched Tesla's Optimus walk into a Los Angeles mansion, take grocery bags out of a car trunk, and set them on a kitchen island. Nine million views so far. I'd bet most of them were applauding the wrong thing.
Because if you think about it, the robot didn't do anything remarkable. Carrying a bag from a trunk to a counter is a solved problem; warehouse machines have done harder work for a decade without anyone filming them. What drew the applause was how Optimus looked while doing it. Two legs, two arms, a head where a head belongs, a walk that reads as human. People were cheering for the resemblance.
And the resemblance deserves a closer look than it's getting, because the actual engineering case for a humanoid robot covers a lot less of it than Tesla built. I want to lay that case out honestly before I start poking holes in it — arguing with a strawman is a waste of everyone's time, and Musk's position is stronger than his critics tend to admit.
The best version of Musk's argument
Every doorway you've ever walked through was sized for a human being. Same for every staircase, handle, light switch, washing machine, and shelf. Our civilization isn't a neutral environment. It's a user interface, and the target user has two arms, two legs, opposable thumbs, and eyes about a meter sixty off the ground.
A wheeled robot can't climb your stairs. A robot dog can't reach the top shelf or drive your car. A specialized kitchen arm will never take out the trash. So the humanoid form, the argument goes, is simply compatibility with legacy infrastructure: one machine that operates everything already built for people, without rebuilding any of it.
Two quieter arguments sit behind that one. First, training data. The internet holds an effectively unlimited archive of video showing humans doing things with human bodies, and a robot shaped like us can learn from those demonstrations almost directly. Some custom form factor would need custom data nobody has. Second, manufacturing: a single universal chassis produced by the million beats ten specialized machines produced by the thousand. It's Model T logic applied to labor itself.
All of this is true, and I say that as someone who builds products for a living. If you grant the premises, the humanoid bet is defensible. It is not Silicon Valley theater.
Read the fine print, though
Hardly anyone takes that argument and checks what it actually pays for. So let's.
Stairs require legs, or at least a body that can negotiate steps. Top shelves require reach and articulated hands. Driving requires limbs in roughly human positions. Learning from human video requires joints that map onto human joints. Notice that every single requirement is about kinematics — geometry of movement, reach, degrees of freedom.
None of them requires a face. Or a head shaped like ours, or a gait tuned to look relaxed instead of tuned to save power, or five fingers where three would grip better, or the little performance Optimus gives with every delivery — the nod, the palm-up presentation of the bag, the body language of a well-mannered house guest.
So the engineering argument buys you a machine that can move through a human world. What Tesla built is a machine that performs humanness at you. Those are different products. One is a tool. The other is aimed at its owner.
They paid extra for the theatrical parts
Walking on two legs is among the hardest problems in robotics. A biped burns continuous computation just staying upright, while a quadruped gets stability nearly for free. Two legs cost more energy, more actuators, more ways to fail. Engineers eat those costs where the payoff is real — stairs, narrow corridors, car seats. Fine.
Now watch the demo again with that in mind. The pace of the walk isn't optimized for energy or speed; it's optimized to be pleasant to watch. The hands aren't the cheapest grippers that hold a grocery bag; they're the most human-looking ones. Wherever the engineering answer and the theatrical answer part ways, Optimus takes the theatrical path, and the theatrical path is consistently the more expensive one.
When a company keeps paying a premium for something its stated use case doesn't need, at some point you have to conclude the stated use case isn't the whole product. The resemblance isn't overhead they forgot to trim. It's what they're selling.
Who is the face for?
Most critiques of humanoid robots stop at "it's inefficient," and I find that unsatisfying. Inefficiency is just a cost, and costs get amortized at scale. The more interesting question is what all that extra humanness is for.
Not for the robot — a robot needs no face to fold laundry. The face is for us. Human brains respond to human form with a speed and depth that skips deliberation entirely: we trust the human-shaped thing sooner, forgive it faster, grant it intentions and competence and moral weight it hasn't earned. That circuitry is ancient and involuntary, and every potential customer already has it installed.
Design aimed at the reflex rather than the task has quietly left engineering and entered persuasion. A humanoid body solves Tesla's problem beautifully: the machine must be loved before it's useful and trusted long before it has done anything to deserve trust. Whether it solves your problem is a separate question — one the resemblance is rather good at keeping you from asking.
I don't think anyone at Tesla drew up a manipulation machine on purpose. Nobody had to. The market applauds resemblance, so resemblance gets built. Which is exactly why it's worth talking about: nobody is steering this, and the incentives point somewhere strange.
An honest robot
Over the years I've settled on a design principle for this, and this essay series is my attempt to argue it out loud. I call it ontological honesty:
The principle that a machine must be recognizable as a machine at the level of reflex, not disclaimer: human kinematics where the world demands it, machine semiotics everywhere else. A robot may have hands like ours; it should not wear a face like ours.
Give the robot legs if stairs demand legs. Give it arms that reach the top shelf and hands that hold a grocery bag. That much the world dictates, and there the steel man stands. But the face, the gait, the little nods — none of that is compatibility with our world. That's a key cut for the oldest lock in our heads, and I'd like us to notice before there's one of these in every home.
The strangest part of the story isn't what robots are becoming anyway. It's what they reveal about how easily we can be read — by anyone willing to build a body shaped like ours and walk it, slowly and warmly, into our kitchen.
Next essay: why the anthropomorphism was never in the robot to begin with. Soldiers have held funerals for bomb-disposal machines that look like tank treads, and plenty of people name their vacuum cleaner. The reflex lives in us; design just drills into it. That changes the question entirely.
This is Essay 01 of Honest Machines, a series on anthropomorphism in robotics — what human-shaped machines cost in engineering, what they do to our psychology, and how to design them honestly. New essays every few days: RSS · Telegram · léelo en español.