For two years, robotics has competed on vanity numbers. Units shipped. Robots off the line. Fleet size.

But a robot that's powered on isn't a robot that's working, and one that needs a human babysitter every five minutes isn't autonomous.

AGIBOT's Peng Zhihui just proposed a metric that fixes this, one that pushes robots to earn their keep:

Worken (Work + Token) = active robots x autonomous working hours x output efficiency

Only robots actually doing tasks count. Only autonomous time counts. Output is measured against a human benchmark.

It's borrowed from industries that already matured, like cloud computing, where nobody brags about servers owned, they measure usage.

There's a lesson from the AI race here too: benchmarks often push models to optimize for the wrong things, while real usefulness comes down to how much work a user can actually get done.

Early days still, but shifting the scoreboard from robots built to work done is exactly the reframing this industry needs.