By 2025, AI has transformed from a cutting-edge concept into one of the core tools driving productivity changes, influencing the global business community in unprecedented ways. From corporate strategic planning to employees’ daily tasks, it has become integral.
In contrast, the AI industry in China presents a unique landscape: on one hand, while models like Deepseek have emerged as dark horses, Chinese companies still face challenges in technological catch-up, high costs, and a shortage of talent in the foundational AI technology represented by large language models (LLM), making it a difficult endeavor to “forge AI”.
However, when we shift our focus to AI Agents, we find that China, with its unique market environment and industrial genes, is entering a field full of opportunities. It can even be said that in China, developing Agents is a much easier path.
01 What is an AI Agent? The Evolution from “Digital Assistant” to “Digital Employee”
To understand AI Agents, we can start from a specific application scenario.
Imagine this: you are an operations manager at an e-commerce company, and tomorrow morning at 10 o’clock, you need to hold an emergency review meeting regarding the unexpected drop in sales of the hot-selling product “xx smart noise-canceling headphones.“ To prepare, you need to manually log into the company’s ERP and BI systems, export sales data, user profile data, and inventory turnover data; then, open the pages of competitors on Tmall, JD.com, etc., to check their promotional activities, price changes, and user reviews from last week; next, browse Weibo, Xiaohongshu, and Zhihu to search for the latest user feedback and negative sentiment; finally, organize and analyze all the scattered information, spending several hours to create a logically clear PPT report. After that, log into the OA system to book a meeting room and send out meeting invitations.
The entire process takes up at least half a workday.
Now, with the help of AI Agents, you only need to open the system and give a command: “Please prepare for tomorrow morning at 10 o’clock a review meeting on the sales drop of ‘smart noise-canceling headphones’, analyze the reasons, provide initial suggestions, and organize the meeting.“
Next, this “digital employee” will start working autonomously according to the following process:
1. Understanding and Planning: Quickly grasp the core objective, “hold a successful review meeting”. This task will be autonomously broken down into a series of clear sub-tasks: ① Analyze internal sales data; ② Research external competitor dynamics; ③ Collect user feedback from across the internet; ④ Integrate information to generate an analysis report; ⑤ Propose initial solutions; ⑥ Book the meeting and notify relevant personnel.
2. Invocation and Execution: Automatically invoke the company’s internal database API, retrieve all relevant data, and conduct multi-dimensional analysis; initiate web crawlers to gather all publicly available information from competitors within minutes; connect to sentiment monitoring systems to aggregate thousands of user comments and perform sentiment analysis; invoke pre-set PPT templates, automatically filling in all charts, data, and conclusions; finally, access the company’s OA system to find all relevant personnel’s available time slots, book the meeting room, and send out meeting invitations with the attached PPT report.
Half an hour later, a detailed analysis report and meeting arrangement notification will appear on your desk. More importantly, it will also record the process and results of this task, making it easier for you to conduct similar reviews next week.
In this scenario, the core value of AI Agents is highlighted. They are no longer just passive tools responding to commands, but intelligent agents capable of autonomously understanding objectives, planning complex steps, invoking diverse tools, and ultimately completing tasks.
A typical AI Agent architecture usually includes the following core modules:
First, the Perception Module: The “senses” of the Agent, which receive various information from the digital world or physical environment through APIs, sensors, database interfaces, etc.
Second, the Decision Module: The “brain” of the Agent, which analyzes and judges the perceived information based on its internal knowledge base, workflow models, and strong reasoning capabilities, and plans the best action path to achieve the objectives.
Third, the Action Module: The “limbs” of the Agent, responsible for executing the plans formulated by the decision module, interacting with the external world through invoking APIs, operating software, controlling hardware, etc., to produce actual effects.
Fourth, the Learning and Memory Module: The key to the Agent’s self-optimization, allowing it to learn from the successes and failures of each task execution, continuously optimizing its knowledge base and decision model, forming long-term memory, and becoming increasingly “capable” and “intelligent”.
02 The Business Logic of AI Agents: From “Chatting” to “Doing” Value Leap
AI Agents have a fundamentally different business logic compared to the familiar ChatGPT or traditional AI Chatbots (chatbots). If the latter’s business value mainly lies in “information services”, then the core value of AI Agents is in “process automation”.
The fundamental difference from ordinary AI Chatbots is that they have upgraded from “answering questions” to “solving problems”.
Traditional chatbots, no matter how intelligent, are limited in their capabilities to merely “speak”; they cannot truly “do” things, cannot operate the company’s CRM system, nor can they help you order coffee. The core capability of AI Agents is to bridge these “walls” in the digital world, connecting different software and systems to complete actual workflows.
Therefore, the commercialization logic of AI Agents is closer to an important upgraded version of SaaS (Software as a Service) — RaaS (Result-as-a-Service). Companies no longer purchase just a software tool that requires human learning and operation, but rather a “digital productivity” that can directly deliver business results.
This model is evidently attractive, as it directly ties to the core interests of enterprises ( cost reduction, efficiency increase, revenue generation).
03 In China, “Forging” AI Large Models is Difficult, but “Doing” AI Agents is Easy
In China, entrepreneurs engaged in the AI industry must deeply understand that at the foundational large model level, “forging” a world-class model is an extremely challenging path.
First, there are the high costs of R&D and training. Training a foundational model comparable to GPT-4 incurs astronomical costs in computing power, high-quality data acquisition, and top talent salaries, often reaching tens of billions or even hundreds of billions of RMB. Only a few tech giants can afford to participate.
Second, there is the gap in core technologies and talent. Although China’s large models are striving to catch up in performance, there remains a certain gap with the world’s top levels in terms of original model architecture, underlying algorithm innovation, and solving core technical challenges such as “hallucination”. At the same time, top AI scientists and architects are scarce resources globally, and the competition for talent is fierce, leading to a structural talent shortfall domestically.
Potential risks in the supply chain: High-performance AI chips are the “heart” of training and inference for large models, and China still faces risks of being “choked” in the design and manufacturing of high-end chips. Changes in international geopolitics bring uncertainty to the underlying hardware supply chain of the AI industry.
However, if we shift our perspective from “forging models” to “using models”, China’s unique market environment provides many inherent advantages for practitioners of AI Agents.
Massive and complex digital application scenarios: China has the largest, most active, and most “intensely competitive” digital economy ecosystem in the world. From e-commerce, food delivery, short videos, ride-hailing, to industrial manufacturing, smart cities, and digital governance, almost all industries have undergone deep digital transformation. This means that AI Agents have rich and real practical scenarios. An AI Agent capable of skillfully handling the entire process from marketing, customer acquisition, live streaming, customer service to fulfillment in China’s complex e-commerce environment will far exceed its counterparts trained in simpler environments.
First, it is the technology path of “application-driven innovation”.
Chinese tech companies are more pragmatic, believing in the principle of “application is king”. We may not be obsessed with inventing the most fundamental “hammer”, but we excel at finding various “nails” and efficiently hammering them in at the fastest speed. By utilizing excellent domestic and international open-source models or commercial model APIs as the technical foundation, we quickly combine them with localized scenarios for fine-tuning and application development, creating truly problem-solving AI Agent products, which is what Chinese companies are better at.
Second, there is a complete digital infrastructure.
Developed mobile payments, efficient logistics networks, widespread cloud services, and a rich API ecosystem provide a solid foundation for AI Agents to complete “end-to-end” task loops in the Chinese market. An AI Agent in China can easily achieve full process automation from planning, booking, payment to check-in, which is unimaginable in many countries where payment and credit systems are not yet well developed.
Third, there is government policy guidance and support.
The Chinese government places great importance on the application of AI, vigorously promoting the “AI +” initiative, encouraging deep integration of AI technology with the real economy. In the process of promoting industrial digital transformation and building a “digital government”, a large market demand and policy support for the application of AI Agents have been created.
Currently, the track for AI Agents in China is showing a flourishing array of application scenarios. Internet giants (such as Alibaba, Tencent, Baidu, ByteDance) are leveraging their vast business ecosystems to use Agents as a “neural network” connecting internal services. For example, Alibaba’s “Tongyi Qianwen” is gradually evolving into a super Agent capable of integrating its complex e-commerce, finance, logistics, and other businesses.
Moreover, a large number of innovative companies in vertical fields are emerging, embedding the value of AI Agents deeply into specific industry soil.
For example, Zhongke Shiyu integrates vision and language, allowing agents to delve into the transportation and other real economy sectors to solve complex problems. Zhuoshiketech focuses on healthcare, with its “AI Family Doctor” simulating professional workflows to provide personalized health management. Meanwhile, Bantuo Goose acts as an “enabler”, providing a low-code development platform that allows companies to quickly build their own Agents, significantly lowering the technical threshold.
These three represent the three development directions of deepening into the real economy, focusing on verticals, and platform empowerment.
04 Conclusion
In stark contrast to the challenges faced in foundational AI R&D, AI Agents are ushering in a “spring” of development in China. With the world’s unique vast digital application scenarios, a culture of “application-driven innovation” deeply rooted in entrepreneurial spirit, and a complete digital infrastructure, Chinese companies have unique advantages in “making good use of AI”. From the rise of empowering platforms to the deep cultivation of vertical industries, we are witnessing the emergence of a pragmatic and efficient development path for China’s AI Agent industry.
For many Chinese tech companies and investors, rather than pouring all resources into the competition for foundational models, it is more beneficial to focus on the application fields of AI Agents. Here, many business processes waiting to be optimized and unmet user needs provide ample space for the development of AI Agents. Opportunities are emerging.
Source | Chaos Academy (ID: hundun-university)
Author | Chaos Academy ; Editor | Lizhi
Content represents the independent views of the author and does not reflect the position of Zao Du Ke.
