After 20,000 units rolled off the line, AgiBot proposes Worken and releases GO-3 Preview, aiming to achieve a robot-to-engineer ratio of 20:1

When will customers be willing to pay?

In September this year, AgiBot's 20,000th robot officially rolled off the production line.

At the 2026 AgiBot Innovation Day held on October 29, Peng Zhihui, co-founder, president and CTO of AgiBot, disclosed two figures: in the first half of this year, AgiBot shipped 9,700 robots and generated revenue of nearly 2 billion yuan.

Robots rolling off the line numbering 20,000 does not mean that 20,000 robots have already been put into actual work. "Rolling 10,000 units off the line is only a victory for manufacturing; 10,000 units being able to take up posts is the victory of physical AI," Peng Zhihui said in an interview with the media after the meeting.

During the interview, several AgiBot executives discussed the challenges robots still face in moving from mass production to large-scale deployment. For AgiBot, getting more robots to truly take up posts has become a core problem that urgently needs solving.

That day, AgiBot unveiled a number of achievements in a concentrated release, including the heavy-load embodied robot Genie G2 Max, the embodied foundation model GO-3 Preview, and the omni-modal interaction large model WITA-OMNI, and for the first time proposed a new unit called Worken for measuring a robot's effective intelligent working hours.

Robots "taking up posts": a yardstick for production efficiency

Is the length of time a robot is powered on equal to the length of time it creates labor value?

Peng Zhihui gave a negative answer: "A robot being powered on for 8 hours does not mean it has created 8 hours of value, and a high success rate on a single task does not mean it can continuously work as stably as a human for a month."

So how should a robot's production efficiency be measured? At this Innovation Day, AgiBot proposed Worken, the concept of "effective intelligent working hours," with the formula "Worken = number of active robots × autonomous working hours × output efficiency ratio."

Peng Zhihui introduces the concept of Worken (effective intelligent working hours). Photo/AgiBot

Peng Zhihui said that in the past, robot companies mostly showcased performance parameters and flashy demos, whereas what Worken aims to answer is how much sustainable and effective intelligent labor robots can actually contribute in real environments.

At the Innovation Day venue, AgiBot showcased Worken data for different application scenarios. Among them, in the tourist service scenario at Kuanzhai Alley in Chengdu, 4 robots worked 310 hours and generated a cumulative 744 Worken; in the loading and unloading scenario on the production line of the Longqi factory, 8 robots worked 528 hours and generated 2,956.8 Worken; and in the security check guidance scenario at Guangzhou Metro, 120 robots worked 240 hours and generated 14,400 Worken.

Worken attempts to provide a unified labor value indicator for different robots, but how the output efficiency ratio in its formula is determined—for example, whether losses caused by manual takeover, equipment maintenance and abnormal downtime need to be deducted, and whether the work value of different positions can be compared—AgiBot has not yet published a unified accounting standard.

From 1:1 to 20:1, "withdrawing people step by step"

In September this year, AgiBot deployed its first batch of more than 300 robots at Chimelong Spaceship Park in Hengqin, distributed across more than 100 locations and undertaking tasks such as guided tours and interactive entertainment. Data disclosed by AgiBot shows that as of today, nearly 500 robots have been deployed on a regular basis, and the number will increase to 1,000 by the end of the year.

Wang Chuang, AgiBot partner and president of the general business unit, said in the interview that these robots are not concentrated in the same venue but scattered across different areas—some play chess or finger-guessing games with tourists, some play table tennis, and others lead tour groups and give explanations—and the environment at each location, the tasks and the flow of people all differ.

With such complex scenario deployment, at the very beginning of the project almost every robot needed to be paired with an engineer, so several hundred robots required several hundred engineers on site.

After the robots were put into use, new problems emerged. "Some children like to hug the robots and push them, and when playing games they may not follow the set rules," Wang Chuang said.

"The most important thing is how to withdraw people step by step," Wang Chuang said. In the early stage of the project, the ratio of robots to engineers was about 1:1, later changed to 3:1, with the goal of reaching 10:1 by the end of this year and 20:1 next year.

In Wang Chuang's view, 20:1 means this position only needs an additional 5% of the human resources budget, which is an acceptable range for many customers.

Beyond operation and maintenance costs, the capabilities of models and actuators also limit the scope of robot deployment. At the 2026 Innovation Day, AgiBot released the new-generation embodied foundation model GO-3 Preview, along with the embodied intelligence native multi-modal interaction large model WITA Omni 1.0.

However, the results a model achieves in specific test tasks are not equivalent to its performance in long-term operation in a factory. Wang Chuang admitted that the model level still needs continuous iteration—for example, WITA-Omni has only just been developed, and its consistency and stability when deployed on a large number of robots still need optimization, while the cost of on-device deployment is also being continuously optimized. The ultimate goal is to make every robot affordable and usable.

Behind 20,000 robots rolling off the line, when will customers be willing to pay?

Regarding the possible reshuffle in the robotics industry, Yao Maoqing, co-founder of AgiBot and president of the embodied business system, believes that this is not a cycle or stage specific to the robotics industry: "Every industry, especially hot technology tracks, will generally, by objective laws, see a large influx of capital and talent and the emergence of homogeneous development. After a certain stage, based on the problems found in the real implementation process, corresponding technical and capital reserves are needed, and then the industry gradually converges."

In Yao Maoqing's view, for robots to truly achieve healthy long-term development, the entire chain needs the ability to convert technology and products into measurable productivity.

At the same time, he also gave a key milestone: if in the future the Chinese market can have more than 10,000 robots working long-term and stably in factories, that will be an important milestone for the industry's development.

However, whether robots can achieve large-scale deployment ultimately still depends on whether customers are willing to pay.

"When will customers be willing to pay? I think it is when customers begin to seriously think about it, shifting from 'I am willing to try your product and see whether you can build it' to 'whether I can actually afford to keep your product.'" Peng Zhihui said that such a shift also means the industry has begun to seriously face the complete lifecycle of robots.

Behind this is a real economic calculation. After robots are delivered, long-term equipment health management, software updates, spare parts supply, remote diagnostics, and on-site maintenance are still needed. Peng Zhihui said bluntly that only by quantifying actual output, enabling tracking and scheduling of abnormal issues, and continuously reducing operating costs as deployment scale expands, so that customers' investment and returns reach a balance, will the scale of robot procurement and deployment be likely to grow further.

"Otherwise, you might have to assign a nanny to every robot you sell, and that day will never come," Peng Zhihui said.

English text is being prepared.

2 A braised egg shop in Luoyang went viral, and the owner begged people to come less; township government staff: some people bought more than a hundred at a time 50,000

3 Stop-loss technology shifts to dividends! Jiao Wei completes a "resources-technology-dividends" round trip in three quarters 49,000

4 The CPC Central Committee holds a press conference on the spirit of the Fifth Plenary Session of the 20th CPC Central Committee 31,000

5 Freshmen posed for photos with a gross anatomy teacher's skull specimen and posted them on Moments; school: disciplinary action given 22,000

Latest

Ministry of Commerce: Firmly opposes the U.S. Federal Communications Commission's vote to pass restrictive measures on testing laboratories 18 minutes ago

JD 11.11: 150 million service orders, over 20 million AI products have entered thousands of households 18 minutes ago

KPMG's Chen Yi: China's economy is moving toward a new stage of intelligent development 18 minutes ago

TAL President Peng Zhuangzhuang: Overall growth is expected to slow, and the learning devices business faces challenges 19 minutes ago

Connecting the security chain for AI implementation, Xunzhong Communication-Xunjie Computing Network grandly releases the "Secure Token Channel" 24 minutes ago

Hot discussion

69 people killed; first report on Spain's high-speed rail accident released

Muxi: Net profit attributable to parent in the first three quarters was 639 million yuan, turning profitable year-on-year

Leaderdrive's third-quarter revenue rose 56% year-on-year, and net profit rose 29% year-on-year

How can deposit costs be reduced? How should personal loan risks be managed? A direct look at the first batch of third-quarter earnings briefings of A-share listed city commercial banks

Amazing! As his wife was about to give birth, small birds came to build a nest and hatch eggs in the quilt drying on the balcony