Six AGIBOT G2 robots worked a Shanghai convenience store for 24 hours and logged 1,730 tasks

From 10:00 on November 2 to 10:00 on November 3, 2026, six AGIBOT G2 humanoids worked as staff in an operating convenience store in Shanghai’s Xuhui district, on a public livestream. AGIBOT said the robots completed 1,730 tasks: 683 customer receptions, 369 item fetches, 147 coffees, 126 breakfast services, 277 restocking and shelf checks, and 128 table cleanings. The stream drew more than 2 million viewers, the company said.

Why it matters: it is a 24-hour, multi-robot run in a public shop with real customers, the kind of setting where scheduling conflicts and small faults compound. It is also a company-run test with company-reported results.

How were the robots judged?

AGIBOT framed the run as a probation review. Criteria included time on duty, tasks completed, customer feedback, task success, response time and handling of exceptions, according to Jiqizhixin’s account. Four of the six robots “passed”; two did not, on stability and on-duty consistency. The robots ran on Genie Master, AGIBOT’s agent framework that schedules tasks and feeds execution results back into training, Jiemian reported.

What is missing from the results?

The published figures are counts. None of the reports from Jiemian, Securities Times or Time Weekly gives a success rate or the number of times humans took over, which are the measures that would show whether the robots could hold the jobs without staff. AGIBOT proposed a work-hour metric, Worken, for exactly this kind of reporting a week earlier; see our Innovation Day report.