The obvious story about robots taking jobs has the class politics backwards, and Forbes just published the numbers to prove it.
The Math Nobody Ran
A construction software company called Planera cross-referenced federal wage data against real automation vendor pricing for 30 of America’s most common jobs. The output flips a decade of automation panic on its head. Swapping a nursing assistant for a humanoid robot costs roughly $375,100 a year — nearly nine times the $42,200 those workers actually make. Construction laborers, home health aides, maintenance techs, teaching assistants: the entire “unskilled” category everyone assumed would go first turns out to be the most expensive category to replace.
The reason isn’t sentimental. It’s mechanical.
Hands Are Expensive
A Planera spokesperson told journalist John Koetsier the pattern plainly: the lowest-paid workers tend to do the most physically demanding, human-facing work, and that’s precisely what machines handle worst. Hands-on patient care, judgment calls on a messy job site, diagnosing a fault on equipment nobody documented properly — none of that maps onto a robot’s strengths yet. Dexterity remains the bottleneck.
TDK Ventures’ Ankur Saxena put a number on the floor: building a useful humanoid robot starts above $200,000. Most manufacturers still sell below cost to win pilot customers. It’s the same subsidized-economics story we found when comparing Digit’s manufacturing cost with its selling price. The gap between building a humanoid and selling one profitably remains wide across the industry—not just at one company.
That’s a subsidy, not a business model. It can’t hold at scale.
The Real Target Was Never the Floor
Koetsier’s reporting quietly reframes the whole automation narrative: white-collar work is the exposed flank now, not blue-collar. Software developers — the group everyone assumed automation-proof a few years back — are already losing headcount to AI coding tools that need no hands at all. No actuator costs, no supply chain, no $200K hardware floor. Just inference costs that keep dropping.
The automation story was never really about replacing humans wherever it was technically possible. It was always about replacing humans wherever it was cheapest to do it. For twenty years that pointed at factory floors and call centers. Right now, with humanoid hardware still absurdly expensive and LLM inference absurdly cheap by comparison, that same cost logic points at knowledge work instead.
The Expiration Date Problem
None of this is permanent, and Koetsier is careful not to pretend it is. Boston Dynamics’ newest Atlas runs almost an order of magnitude simpler than its predecessor. 1X has verticalized its own supply chain for its home robot, Neo. Chinese manufacturers are shipping thousands of units at prices Western competitors can’t match yet — part of a broader pattern of China’s humanoid rental market moving faster than its autonomy claims can keep up with. Barclays projects the humanoid robotics market will grow from roughly $2–3 billion today to $200 billion by 2035, and that kind of money doesn’t sit on the sidelines waiting for hardware costs to stay put.
Every investor writing that check is betting the $375,100 number shrinks, not that the $42,200 one grows.
That’s the actual stakes of this report. It isn’t a permanent reprieve for care workers, laborers, and teaching assistants. It’s a snapshot of a cost curve mid-collapse, taken at the one moment the math still favors the human. The jobs on this list aren’t safe because they’re valued. They’re safe because nobody’s built a hand good enough to replace them yet, and hands are hard.
Watch what happens to that list the next time hardware costs drop by half.
Related: How Humanoid Robots Walk: A Complete Guide to Bipedal Locomotion