_________

Published on Sep. 18, 2026

When a container ship arrives later than planned, a swarm of AI agents gets really busy. Trucks due to collect the cargo must be rescheduled. Customs clearance, the unloading of the cargo, storage allocation and berth allocations of all following ships must be synchronized from scratch.

COSCO Shipping’s automated logistics center lets AI agents handle all of these tasks simultaneously. The company that makes this possible is called IROOTECH, formerly known as Rootcloud. The Guangzhou-based enterprise is affiliated with the heavy machinery maker SANY Group. It started as an IoT platform and now bets on industrial AI by translating between different data sets and standards.

Before the AI agents took over, coordination in logistics was a real challenge. “In port logistics automation, individual technology modules are not particularly scarce. The real difficulty lies in systems being connected but not truly integrated. Vessel arrival schedules, customs clearance systems, truck dispatch networks and terminal operating systems are controlled by different organizations, use different standards and run on different platforms,” writes the investment portal Touzi Jie (in Chinese).

IROOTECH serves as an interpreter between different data silos, enabling its agents to access all data, rules and schedules and then send commands to industrial robots, which it also develops.

There is a solid basis for this new layer of intelligence that AI agents are currently introducing. For years, welding robots in car factories and automated cranes at container ports have performed repetitive tasks.

Each machine, however, followed its own program, and the systems above barely exchanged data. The systems controlling these machines were connected, but usually not integrated. Now, industrial AI agents are beginning to synchronize these machines, closing the loop from perception to action.

These agents observe events using cameras and sensors. They compare this information with production rules, scheduling data and with knowledge learned from seasoned engineers. Then they decide upon the next step. While doing so, they not only consider any single process, but the optimization of productivity overall.

The agents instruct robots, cranes, or software systems and verify if outcomes align with plans. As agents begin to make more of these decisions on their own, factories and ports can run with less human intervention. In this model, human engineers set objectives and intervene only when necessary.

IROOTECH was recently singled out as a benchmark case for a Chinese industrial AI company in a whitepaper by Frost & Sullivan. It was portrayed as an up-and-coming full-stack provider of agentic AI in China, alongside Siemens and Huawei.

This may be an optimistic projection of IROOTECH’s future, as the company is relatively young and most of its business seems to come from other enterprises in the SANY universe. For example, IROOTECH is currently reorganizing the production processes at SANY Heavy Industry’s piling machinery factory in Beijing, writes Touzi Jie.

The relevance of these fast deployments of agentic AI goes beyond one company’s balance sheet or prospects. They demonstrate how “the focus of China’s artificial intelligence industry shifts from competition in large models and computing power toward large-scale deployment of AI agents and application monetization,” as the Chinese state broadcaster CCTV recently put it.

CCTV quoted the “Research Report on the Development of AI Infrastructure in the Agent Era 2026” published by the China Telecom Research Institute. “According to the report, artificial intelligence is now entering a new stage of development in which agents are becoming the dominant form,” the Chinese TV station reported.

AI agents are projected to drive almost tenfold annual growth in China’s computing demand over the next two to three years, the report says. In other words, demand for tokens in China is exploding.

“China’s annual token consumption is expected to reach 100 quadrillion in 2026 and exceed 35 quintillion by 2030,” the institute writes. Slowing down AI development, anybody? Not in China.

Artificial-intelligence capital expenditure by China’s leading technology companies is forecast to reach 600 billion yuan (around US$89 billion) in 2026.

Very different approaches to AI development are emerging in the U.S. and in China. While Western AI developers such as Anthropic or OpenAI chase ever more powerful large models and even artificial general intelligence (AGI), the industry’s holy grail of “super intelligence”, Chinese manufacturers are quietly deploying practical agent networks to extract value from factory floor data.

“As leading models become more comparable in areas such as general question answering, coding and multimodal understanding, enterprise buyers are increasingly prioritizing inference cost, response latency, tool-calling capabilities, and inference cost, response latency, tool-calling capabilities, and industry-specific adaptability over raw benchmark performance," writes the news portal Zhongguo Jingji Wang.

The result is a Chinese pivot in the direction of AI development from generative content creation to the execution of tasks.

The shift is also reflected in the way Chinese companies use large models. The research firm Gartner claims in a recent report that as recently as two years ago, globally developed closed-source models accounted for around 80 percent of enterprise use in China. Three months ago, it was 50 percent and recently the figure has dropped to 28 percent.

Partly, this is because Chinese large models such as those from DeepSeek, Alibaba or Zhipu AI are getting stronger. Partly, it is because Chinese companies are starting to use the large models as one tool among others to build broader systems, such as AI agents.

Enterprises in China no longer expect large models to solve every problem and “will not treat them like a magic mirror and keep making wishes,” the Chinese business daily 21 Shiji Jingji Baodao quoted Fang Qi, Senior Research Director at Gartner.

So thoroughly have AI agents replaced large model at the center of attention in China, that Gartner places AI agents at the “Peak of Inflated Expectations” in its Hype Cycle report.

Figures like the capital expenditure quoted above and the heavy bets that carmakers such as Xpeng and Xiaomi are placing on embodied AI and humanoid robots suggest that while expectations might be somewhat inflated at the moment, China’s embrace of agentic AI is anything but a short-lived hype.

First, China has recently spent two decades upgrading itself from the world’s sweatshop to a powerhouse of intelligent manufacturing exporting not only umbrellas or toys, but also CNC machines, smart vehicles and telecom equipment.

In China, AI agents are meeting the diverse industrial supply chains that give their integrative powers real meaning. Agents might be one field of AI development where China first moves from follower to rule-setter.

Second, driven by the need to find new sources of growth as its population is simultaneously aging and shrinking, the Chinese government has decided that combining the country’s existing industries with AI should receive the same attention as the race to develop powerful large models. The Ministry of Industry and Information Technology (MIIT) in Beijing has issued a comprehensive “Artificial Intelligence + Manufacturing Action Plan.”

“Industrial agents are clearly emerging as another major competitive advantage for Chinese manufacturing. They may lack the explosive visibility of the consumer internet or the immediate visual impact of new energy vehicles, but the transformation is taking place deep inside factories, between machines and within flows of data. It is quiet and gradual, but irreversible,” writes Touzi Jie.

Everybody is also stressing the existing bottlenecks to the succesful large-scale deployment of AI agents. They are as pronounced in China as anywhere else. System architectures are fragmented. High-quality industrial data is scarce. Agents still struggle with independent decision-making, and collaboration mechanisms are often clumsy.

Yet Chinese engineers are busy addressing those bottlenecks one by one. They experiment with AI agents one welding machinery shop floor, one smart vehicle production line and one harbor terminal at a time.

They are rolling up their sleeves and getting down to business, treating artificial intelligence like a sophisticated new tool, not like a magic wand.

In international competition, manufacturers in Japan, Germany or the U.S. can ignore this shift in Chinese factory automation or talk it down, but only at their own risk.