Skip to content

SouthwardTech

Technology, industry and investment across China and the Global South

5 projects tracked1 Contracted4 AnnouncedReviewed

Robotics

What Agibot’s robot work figures leave out

China's Agibot has proposed a new measure of robot work. Understanding its value means looking at completed tasks, reliable operation and the people and costs behind them.

Chinese robotics company Agibot has proposed a way to count the work its machines perform. Called Worken, the framework, unveiled in Shanghai on October 29, 2026, combines active robots, autonomous working time and output measured against a benchmark for a defined task.

The proposal raises a question that matters well beyond China: how can a customer tell whether a robot is doing useful work? A successful demonstration shows that a task is possible. A business also needs to know whether the machine can repeat it, how much help it needs and what each acceptable result costs.

Industrial automation already measures quality, speed and equipment availability. New humanoid robots face those practical tests too, whatever units are used to describe their work.

Define the job

At BMW’s Spartanburg plant in the United States, Figure’s humanoid robots loaded sheet-metal parts onto welding fixtures, according to the robot maker’s November 2025 account. Conventional industrial robots then performed the welding. The humanoids were handling a specific step in vehicle production.

Figure assessed the job through three measures: the time needed to complete a cycle, correct placement of the parts, and how often a person had to pause or reset the machine. Moving a part quickly is useful only if it reaches the required position in time for the next process. Defining that outcome gives a performance claim something concrete to measure. That deployment is now a past stage: BMW described a new logistics sequencing project in June 2026, and Figure said that month that Figure 03 had arrived at Spartanburg. In September 2026, Figure said it had decommissioned most of its Figure 02 fleet, retaining only a few units at headquarters.

Count the whole shift

Agility Robotics reported in September 2026 that Digit 4 maintained approximately 98% accuracy while performing tasks at logistics company GXO’s Flowery Branch facility in the US state of Georgia. The partners have since announced plans to expand the deployment: in October 2026 GXO and Agility said Digit would expand beyond Flowery Branch to GXO sites in Statesville, North Carolina, and Lebanon, Tennessee, with deployments expected in early 2027. The qualification matters: accuracy during a task does not tell a customer how much of the day a robot was available to work.

To understand a whole shift, a customer would also need to count waiting, charging, repairs and recovery from failed attempts. A brief interruption that the machine resolves itself has different consequences from one that requires a technician. A single success rate does not reveal that difference. The question is how much acceptable work reaches the next stage over the period that matters to the business.

Put people in the picture

Agibot’s own experience illustrates why human support belongs in the assessment. In an interview reported by Chinese business outlet Yicai, company executive Wang Chuang described changing staffing arrangements as robots were deployed at a Chimelong theme park. Staff had to manage issues including depleted batteries, damage and updates to performance routines. These were the supplier’s account of operations, rather than an independent measure of customer savings.

A person replenishing supplies, supervising several machines or helping with an unusual situation may be part of an effective division of work. The important distinction is between useful collaboration and repeated rescue. Both consume time, but their contribution to the result can differ. On-site and remote support should therefore be counted alongside the machine’s output.

What a work total leaves out

Worken combines the number of active robots, their autonomous working hours and an output efficiency ratio. Agibot says the last factor compares output with a reference for a defined task, such as acceptable work completed by a person under established operating standards. The company describes the framework as evolving, with task benchmarks to be refined through deployment experience.

The formula also shows why total work and efficiency need separate attention. Holding its other factors constant, adding robots raises the total without making each machine more productive. Combining different jobs also requires deciding what counts as an equivalent result.

Yicai reported that Worken calculations were being used internally at Agibot, while charging customers through the unit remained an idea under exploration. Yicai's 30 October report gave the internal calculation: 33,480 Worken for roughly 300 robots at the Chimelong deployment. The report did not specify the measurement period or task benchmark behind that total. Wang Chuang said staffing there had gone from one staff member per robot to one person for three robots, with a goal of a 10:1 robot-to-human ratio by year-end. A second Agibot explanation published on 30 October presents Worken as a proposed framework, not an industry standard. A work total cannot, by itself, establish what a customer should pay or whether the deployment earns its cost back.

Recalculate for the next customer

A performance figure needs context. Research by Adam Norton, Elena Messina and Holly Yanco explains why robot performance should be evaluated against user needs and described together with the environment in which it was achieved.

For a business considering a deployment in another market, that means examining its own workflow. Purchase or rental charges are part of the bill. Integration, maintenance, energy, human support and interruptions also affect the cost of each acceptable result. Compare that cost with the existing process or another feasible form of automation, using the same quality requirements and a realistic operating period.

Consider two hypothetical sites using the same robot model for the same task. If one waits longer for repairs or needs more staff support, the economics can differ even when the machine’s task accuracy is identical. A result from a US factory or warehouse provides a starting point for questions about another location; local operating records are needed to answer them.

Value may also come from more consistent quality or less physically demanding work. Those benefits need evidence of their own. The test returns to three questions: what acceptable work was completed, what support sustained it, and was the full effort worthwhile?

Published 4 November 2026 · Last updated 10 November 2026 · Corrections to corrections@southwardtech.com

Weekly newsletter

The Southward Brief

A weekly briefing on the technologies, companies and industrial shifts reshaping China and the Global South.

Required. Enter a valid email address.

By subscribing, you agree to receive The Southward Brief. You can unsubscribe at any time. Read our Privacy Policy.