AGIBOT's 24-Hour Store Trial Logged 1,730 Tasks, and 683 Were Customer Interactions
Hardware / analysis
AGIBOT's 24-Hour Store Trial Logged 1,730 Tasks, and 683 Were Customer Interactions
Four of six G2 humanoids passed the Shanghai assessment, and neither account published how often a human stepped in.

AGIBOT's 24-hour convenience store trial in Shanghai logged 1,730 tasks across six G2 humanoid robots, and the arithmetic of that total shows 683 of them were customer-service interactions rather than physical work. Four of the six robots passed the final assessment; two did not, and those two held the shelf and table-cleaning jobs.
The trial ran from 10 a.m. on Nov. 2 to 10 a.m. on Nov. 3 in a working store in Shanghai's Xuhui district, according to Humanoids Daily, whose page is labelled sponsored content with support from RoboStrategy. Gasgoo published a longer account on Nov. 6. Both rest on AGIBOT's own materials; neither publishes a task log.
Where the 1,730 comes from
The categories add up, and the sum is the useful check. Customer-service interactions were 683, pickups and deliveries 369, shelf inspections and restocks 277, coffee orders 147, table cleaning 128 and breakfast servings 126. Those six numbers total 1,730.
- Customer-service interactions683 tasks
- Pickups and deliveries369 tasks
- Shelf inspections and restocks277 tasks
- Coffee orders147 tasks
- Table cleaning128 tasks
- Breakfast servings126 tasks
Source: Humanoids Daily and Gasgoo, accessed 2026-11-16
Gasgoo describes the 683 as customers served, and AGIBOT's livestream drew more than 2 million viewers, a figure that counts people watching on screens, not shoppers in the aisle. So the headline count mixes a conversation with a customer and a cup of coffee into the same unit. Strip out the 683 and the robots completed 1,047 physical jobs in 24 hours, or about 175 per robot on average, one every 8 minutes each if the load was even, which neither account says it was.
Who passed and who did not
Gasgoo names each robot. The clerk Huobao handled 369 pickup-and-delivery orders and coped with order surges and sudden cancellations. The barista Kasha made 147 coffees. Dingjiang served 126 hot meals and passed, though a dropped corn cob, Gasgoo wrote, shows stability still needs work. Greeter Ahao earned the top customer rating.
The failures were physical. Juanwang, which patrolled and restocked 277 times, showed little initiative and could not keep the store's pace. Lengjie, which cleared tables 128 times, drifted off its route and wandered out of the store.
| Robot | Role | Result |
|---|---|---|
| Ahao | Greeter | Passed |
| Huobao | Clerk | Passed |
| Kasha | Barista | Passed |
| Dingjiang | Hot food | Passed |
| Juanwang | Shelf patrol and restock | Failed |
| Lengjie | Table cleaning | Failed |
Evaluators set their own tests: ordering regular and sugar-free Coca-Cola to see whether a robot can tell similar products apart, scrambling gum packs on a shelf and asking for them back in place, and blocking an aisle with a person and a box. Gasgoo adds that inspectors visited the store to make things difficult on purpose, which suggests the tests were adversarial in spirit, though the account does not say how many inspectors came or when. Humanoids Daily lists the criteria as time on duty, task completion, customer feedback, response times and handling of unexpected situations.
What the account leaves out
Neither report says how often a human took over. Gasgoo says AGIBOT tracked human takeover frequency, anomaly recovery time and system availability, and that the system hands a failed task to another robot or requests human help, but it prints none of those three numbers. Charging is not mentioned in either account, so the 24 hours describes the livestream window rather than proven unbroken runtime. Cost, payload and G2 specifications are also absent.
Ren Guanghui, vice president of AGIBOT's Embodied Intelligence Research Center, was more candid than the totals. Gasgoo quotes him: "Compared to a seasoned clerk, robots still lag noticeably, in how fast they adapt to on-site changes, the rhythm of fine manipulation, and anomaly handling." He named deformable objects, complex contacts and long multi-step operations as weak spots.
The orchestration layer matters as much as the hardware. Humanoids Daily says a system called Genie Master coordinates the six robots and feeds successful and failed attempts into later training, and Gasgoo says early on there were failed orders, missed status updates and positional conflicts between robots.
Gasgoo adds that AGIBOT introduced a unit it calls "Worken", for Effective Intelligent Work Hours, calculated as active robot count multiplied by autonomous work duration multiplied by an output efficiency ratio. The definition is useful because it puts autonomous duration inside the metric, which is the figure the trial summary lacks. Ren said the goal is for costs to fall steadily as the system matures over long operating cycles, and gave no dollar figure.
Set this beside the money. Agility raised $465 million against $174,000 of half-year revenue, and Fractile's $720 million Series C was raised before its first chip ships. Humanoid deployment is being financed ahead of evidence, and a store trial with four passes in six is a more honest data point than either.
What would change this read is the takeover rate. AGIBOT plans to keep testing across more roles and environments, and if it publishes human interventions per robot-hour, the 1,047 physical jobs become a productivity figure instead of a demonstration.
Sources
More in Hardware
- 01Starship's Ship 42 Is Caught After 14 Orbits, and Two Reports Give Different Reasons the Booster Was NotFlight 15 returned an upper stage to the tower for the first time and dropped 26 Starlink V3 satellites; SpaceX has no reflight date in these accounts.
- 02Genesis AI Raises $500 Million at $3.5 Billion, and Its CEO Says Eno's Hands Last WeeksThe Paris robot maker has built 30 to 40 Eno units and chose a US contract manufacturer, with no price and no paying customer count.
- 03Tesla Cybercab's Eight NHTSA Crashes Include a 53 mph Highway Hit With a Test Driver AboardThe four driverless crashes stayed under 5 mph; the Robotaxi median pre-crash speed is 1 mph, which says more about the geofence than the software.
- 04Anthropic's Three-Day AMD Port Covered Prefill for One Model on MI355XThe roadshow slide credits Claude Mythos Preview with a six-to-12-month job; it leaves out decode, the kernel denominator and any Nvidia comparison.