This morning, the 2026 World Intelligence Industry Expo opened at the Tianjin National Convention and Exhibition Center. The exhibition area reaches 130,000 square meters, with more than 700 companies participating and six themed exhibition halls.
One highlight I didn't expect: for the first time this year, there is a dedicated venue showcasing embodied intelligence. It isn't squeezed in with the robots, nor is it a corner of the AI venue—rather, it's an entire exhibition hall. JD.com has brought "Robot Town" here, and Unitree's robot dog is making coffee for visitors—picking up a cup, placing it on the conveyor belt, turning around to handle the next one. Its movements aren't especially fast, but they are very steady.
If this had happened last year, people might still have regarded it as a novelty. But today, few people doubt one thing: artificial intelligence is acquiring physical form. And the pace of this process is faster than anyone predicted in 2024.
Policies were introduced intensively in April and May; intelligent agents are no longer just a “trend”
In May, the Cyberspace Administration of China, the National Development and Reform Commission, and the Ministry of Industry and Information Technology jointly issued the "Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents." This is not a set of guiding opinions of the "encouraging exploration" type; rather, it provides a formal official definition for AI agents:
An intelligent agent is an intelligent system with autonomous perception, memory, decision-making, interaction, and execution capabilities.
Note the last two characters: execution. Not chatting, not Q&A, not text generation, but execution. It explicitly lists nineteen typical application scenarios, covering five directions: scientific research, industrial development, boosting consumption, people's wellbeing, and social governance. Scientific research ranks first; this order itself is a signal.
In the same month, the "AI Plus" initiative was upgraded from "deepening and expanding" at the late-April Politburo meeting to "full implementation." The two-character difference means that the pilot phase has ended and large-scale deployment has begun.
Connect these threads: the national-level expo opens today, with embodied intelligence getting its own independent pavilion for the first time; the document from three departments provides an official definition and a list of scenarios; “AI+” is moving from pilot programs to full-scale implementation. These three threads converge in the same month—this is by no means a coincidence.
OpenClaw ignited the "AI Action," and all major players have entered the fray.
The trigger for this change is an open-source AI agent framework called OpenClaw (Lobster), which can see the screen, click the mouse, and work automatically. That itself is not surprising—what is surprising is the speed at which it went viral.
Robin Li of Baidu proposed a concept called DAA (Daily Active Agents), intended to replace DAU from the mobile internet era. His exact words were: "The rise of agents shows that what users are really paying for is not 'whether you can do it,' but 'whether you can get it done for me.' This marks AI's transformation from a chat tool to a digital employee."
On Alibaba’s side, Tongyi Qianwen launched a “citation” feature in March this year—showing a bright green “trusted” label for reliable sources and a red warning for ambiguous sources. On May 11, Qwen became fully integrated with Taobao; behind “ordering takeout with one sentence” is the integrated support of hundreds of small projects. Taobao covers more than 300 prefecture-level cities and more than 3,000 districts and counties. At such a scale, not using AI is a waste, and using AI is inevitable.
Moonshot AI (Kimi) has already produced data in financial and educational scenarios: work that used to take securities researchers 2-3 days now takes only 2-3 hours; a literature review that used to take doctoral students 2-3 weeks to complete now yields a first draft in 1-2 days. Its trillion-parameter MoE architecture activates 32 billion parameters per inference and has 384 domain experts—not everyone can understand these technical parameters, but the figure of reducing 2-3 days to 2-3 hours is enough to make any enterprise CIO uneasy.
360 AI Security Research Institute released a report titled "When AI Has 'Hands and Feet,' Enterprise Security Boundaries Must Be Rebuilt". The core message is: in the past, AI security issues mainly involved preventing "talking nonsense," but now they involve preventing "acting recklessly." There are two paths: using AI to strengthen traditional security protection, and enabling uncertain tasks to be executed under security constraints.
Tencent, ByteDance, and Zhipu have also entered this track, but their public moves are not as clear as those of the aforementioned companies. However, this is not important—what matters is that when everyone is running in the same direction at the same time, that direction is likely the right one.
Capital markets are also repricing
According to Securities Times DataBao statistics, since May, the average daily turnover of AI agent concept stocks has been 7.84 billion yuan, up 30% month-on-month. The average daily turnover of 22 individual stocks doubled month-on-month. Nanwei Software's average daily turnover surged 688% month-on-month to 505 million yuan—the company operates a full chain of "computing power + large models + intelligent agents" and has passed certification by the China Academy of Information and Communications Technology.
Seventeen concept stocks have been forecast by more than five institutions to see net profit growth of over 20% in both this year and next year. Interestingly, 10 of them have posted drawdowns of more than 20%—Hehe Information is down 50% from its January high, and Dian Diagnostics is down 47%.
Intsig Information's AI product is called Docflow, focused on automated document processing—capture, parsing, classification, extraction, and review—positioned as the "document infrastructure for enterprise digital transformation." This kind of thing may not sound as sexy as general-purpose large models, but it can generate revenue.
Dean Diagnostics is more direct: tertiary hospitals and county-level medical consortia are already paying to use their AI pathology and intelligent diagnostic products.
The market's reaction this time illustrates a principle: capital no longer pays for “AI hype”; it is looking for companies with paying customers, concrete use cases, and delivery capabilities.
852 billion: China's AI agent market size is forecast to reach 852 billion in 2028 (554 billion in 2023, CAGR 72.7%)
What It Means for Enterprise CIOs
Connecting some things from today together, the thread I can see is this:
First, the infrastructure is already in place. The document from the three departments provides a compliance framework, Robin Li's "DAA" provides a measuring yardstick, and the products at the Smart China Expo are right before our eyes. 2026 is not a year of "whether to use AI," but a year of "how to use AI without being eliminated."
Second, embodied intelligence is not a gimmick. The fact that it has a dedicated exhibition hall itself shows that the industry chain is maturing—from chips to sensors, and then to motion control and AI brains, the supply chain is already sufficient to support an independent industry exhibition. Manufacturing companies should start paying attention: this is not a matter of buying robots, but of closing the loop of AI decision-making on the production line.
Third, security is no longer optional. When agents move from “dialogue” to “action,” errors are no longer merely embarrassing—they mean real financial loss. The “security constraint framework” in the 360 report and the “full-chain security requirements” in the documents of the three departments point to the same thing: before adopting AI, enterprises must first adopt governance.
Fourth, taking small, rapid steps is more practical than trying to get everything right in one go. Neither this exhibition nor the policy points to a “comprehensive transformation”—the policy mentions gradually advancing 19 scenarios, while Kimi and Tongyi Qianwen both started from a single scenario. Rather than formulating a grand AI strategy, it is better to find a specific scenario that can show results within a month and get it running first.
The expo runs until the 31st. Over the next few days, there will be signings, launches, and bilateral talks. But the first clear takeaway today is this: AI is shifting from "can it work?" to "can it actually get things done?" This is not something for next year, nor for next quarter—it is happening this month.
Data sources for this article: Xinhua News Agency (report dated May 26, 2026), Securities Times · DataBao (May 28, 2026), China News Network (May 28, 2026), and the "Implementation Opinions on the Standardized Application and Innovative Development of Intelligent Agents" (May 2026) issued by the Cyberspace Administration of China / National Development and Reform Commission / Ministry of Industry and Information Technology.
Disclaimer: This article does not constitute any investment advice. All views in this article represent only the author's personal opinions.
