Eight days. That's how long Figure AI ran its humanoid robots continuously, sorting packages around the clock, with a live stream anyone could watch.
This isn't a polished 90-second video shot over a weekend under controlled conditions. Figure AI let the camera run nonstop, showing its robots working through real shifts in what appears to be an operational logistics environment. No cuts, no highlight reel, no ability to hide the failures.
The distinction matters more than it might seem. Robotics companies have a long history of releasing impressive demos that turn out to be cherry-picked runs - robots that perform for 20 minutes under ideal conditions and break down when something is slightly out of place. A continuous public live stream is a different kind of proof. If the robots are failing repeatedly, falling over, or requiring constant human intervention every hour, the world sees it.
Figure hasn't published detailed uptime statistics or error rates from the 8-day run. But the fact that the stream ran without an embarrassing public shutdown is itself informative. The hardware and software are reliable enough that Figure was comfortable with the exposure.
The Demo Problem in Robotics
Humanoid robots doing warehouse work is a crowded ambition right now. Tesla's Optimus, 1X's NEO, and Agility Robotics' Digit are all targeting logistics and manufacturing environments. The difference between companies at this stage tends to come down less to peak performance and more to consistency: can the robot do the same repetitive task 10,000 times without failing in a way that stops the line?
Eight days of package sorting doesn't answer that question completely. It doesn't tell us what the error rate was, how often a human had to intervene, or how the robots handled edge cases like damaged boxes or unusual package shapes. But it's a more honest data point than a demo reel, and it's a format competitors will now feel pressure to match.
For operations teams and logistics companies watching the humanoid robotics space, the move toward operational proof over marketing proof is the right signal to track. Staged demos can be engineered to look good once. Continuous public operation is harder to fake.