Humanoid robots: less sci-fi, more spreadsheet

State of play 2026: Figure, Tesla (Optimus), Unitree, Boston Dynamics, Agility — all scaling manufacturing, all running paid-ish pilots, all still doing narrow, supervised, industrial work. 'General-purpose robot butler' is not 2026. 'Robot that does one warehouse task reliably and is learning the second faster' is.

In the software. The hardware's been good enough for a while; the hard part was making a robot cope with an unscripted world. Neural networks are replacing hand-tuned control code, and world models (Decart's Oasis 3, etc.) let makers train robots in endless simulations instead of expensive real-world runs. The interesting humanoid story is just the chatbot story with legs — and the same caveats about hype apply.

So if you're evaluating the space (or the stocks), mute the highlight clips and read the boring metrics: units shipped, tasks generalised, time-to-second-deployment, real revenue per robot. The companies that win won't have the flashiest demo — they'll have the steepest learning-and-deployment curve. Everything else is marketing with servos.