I'm a Chinese marketing mom with no robotics background. One evening I decided the two robot arms on my table should dance. A few hours later, they did — a synchronized little routine, every motion computed live. I didn't write a single line of the code. Here's how that's possible in 2026.
Two years ago this would have been a weekend of reading robotics docs and watching tutorial videos. Instead I described what I wanted — "make both arms sway and wave to a beat, safely" — and Claude wrote the program, ran it, and fixed the problems as they came up. My contribution was taste and direction, not engineering.
Three ingredients made it cheap and fast:
Each arm has 6 joints. To make it move, you send new joint angles ~30 times a second. To make it dance, you drive those angles with the most rhythmic function in math — a sine wave. Different speeds and phases per joint give you swaying, waving, and rippling "waves" that travel up the arm. Claude choreographed four looping "phrases" so it never looks repetitive.
Our first attempt scraped the tabletop — the arm's default pose sits low. The fix was to lift the arms into the air first, then dance up there, with a side camera watching the height. Add hard limits on every joint and make both arms sway the same direction (so they can't collide), and it's safe to run unattended.
A dancing robot arm is a toy. But it's a receipt. It shows that the gap between "I have an idea for something physical" and "it's running on my table" has shrunk to an afternoon and $280 — if you sit inside the Shenzhen hardware ecosystem and let an AI do the engineering. For anyone non-technical, that's the real headline: physical AI is now something you can just… make.