AgiBot put six of its G2 humanoid robots to work inside a Chinese convenience store for a continuous 24-hour livestream running November 2 to November 3, branded "Robots Never Clock Out." The robots handled stocking shelves, preparing coffee and hot food, greeting customers and clearing trash across four operating windows: the lunch rush, the evening rush, overnight duty and breakfast preparation.
Unlike a controlled lab demo, the store stayed open to real customers and staff throughout, meaning the robots had to share aisles with people, respond to unscripted requests and recover from mistakes in front of a live audience rather than being reset between takes.
Why a 24-Hour Shift Is the Real Test
Most humanoid robot demonstrations run for a few minutes under favorable lighting and clear floor space. AgiBot's format inverts that: a full day-night cycle is long enough to expose battery endurance limits, grip failures on oddly shaped items, and how the robots behave when foot traffic is heaviest. The company has not claimed the robots ran without any human assistance or charging breaks, and outside observers have noted the livestream format documents the event rather than certifying continuous autonomous operation.
That distinction matters for how the industry should read the stunt. A single successful pour of coffee proves manipulation works once; a 24-hour window with four distinct traffic patterns tests whether the underlying software generalizes across tasks and conditions without engineers resetting the scene. It is a louder, more public version of the kind of endurance testing that logistics companies like Agility Robotics and Figure have run privately in warehouses.
AgiBot's Pattern of Public Deployment Tests
The convenience store stunt follows a string of AgiBot moves aimed at proving commercial readiness rather than lab performance, including shipping its 20,000th robot and introducing a productivity metric for comparing robot labor output across deployments. Where competitors like Unitree have leaned on spec sheets and benchmark scores, AgiBot has increasingly chosen public, unedited environments — traffic duty at a marathon, mountain hikes, now retail shifts — as its proof points.
That strategy carries risk. A visible stumble, a dropped item or a stalled task during a 24-hour livestream is harder to edit around than a curated demo reel. But it also generates a different kind of evidence for potential retail customers: not a benchmark score, but a record of how the robots behaved when a real shift got busy. For a company competing against both domestic rivals and Western humanoid makers for retail and hospitality contracts, that kind of footage, failures included, may carry more weight with buyers than another controlled showcase.
The next test for AgiBot is whether a retail chain signs a paid deployment contract off the back of this demonstration, rather than another one-off livestream.