Summary of Key Points
In 2026, Chinese AI products have witnessed a surge in the development of "desktop agents" (the Chinese version of Codex), with major companies such as Alibaba, Tencent, ByteDance, and Baidu, as well as model manufacturers, launching these products. These agents are no longer merely simple "chat and answer" tools; they have evolved into practical assistants that can integrate into real work processes—handling files, operating web pages, generating PPTs, and automating tasks across applications, even working in a collaborative manner similar to small teams. They compete for users by providing different ways to execute tasks, reflecting the transition of AI from being content generators to task executors, and they are becoming a new intermediary layer connecting users with the digital world.
Detailed Analysis
1. What is the Chinese version of Codex? — From "answering questions" to "accomplishing tasks on your behalf"
Previous AI assistants (like early versions of ChatGPT) were more like intelligent encyclopedias: you asked a question, and they provided an answer, but they didn't take any action. The Chinese version of Codex, however, is much more proactive. For example, if you ask it to "organize last week's emails, extract customer needs, and create an Excel spreadsheet," it will automatically open your email, filter the messages, extract the necessary information, and generate the spreadsheet for you.
The key difference lies in the shift from mere information exchange to a complete task execution process. Instead of just informing you how to do something, these agents actually complete the tasks for you. Take Tencent's WorkBuddy as an example; you can ask it to "write a product weekly report using data from Tencent Documents," and it will retrieve the data, organize the content, generate the report, and even sync it with Enterprise WeChat.
2. Three different approaches: Major companies each demonstrating their unique capabilities
Although these products all aim to help users with tasks, they approach this in distinct ways, mainly falling into three categories:
Path One: Expanding from coding tools to office applications
Examples include Kimi Work and MiniMax Code. These tools were originally designed to assist programmers with writing code but have since extended their functionality to office scenarios. For instance, Kimi Work can break down a task like "writing an industry research report" into steps such as "researching information," "organizing data," "writing the text," and "creating charts," and can even allocate tasks to different AI components. Their strength lies in handling structured tasks, but they may struggle with unstructured tasks like responding to customer messages.
Path Two: Directly taking control of your computer
Examples include Tencent's Marvis and Alibaba's QoderWork. These tools interact closely with the operating system, allowing them to control files, applications, and the disk directly. For example, Marvis can help you manage files on your computer and remotely execute tasks via mobile devices, while QoderWork can even "sense" what's on the screen and assist with software operations (such as automatically filling out forms). The advantage here is their strong control over the system, but challenges include potential permission issues and compatibility with various software.
Path Three: Integrating into existing office ecosystems
Examples are ByteDance's TRAE Work and Tencent's WorkBuddy. These tools do not replace your operating system but integrate with commonly used office applications. For example, TRAE Work can generate documents and extract meeting minutes directly within the Lark collaboration platform, while WorkBuddy connects with Tencent Documents and Enterprise WeChat to synchronize work processes. The advantage is their quick adoption; however, users who don't rely on these platforms may find them less useful.
3. Three emerging trends in AI assistance
There are three notable trends in the evolution of these products:
Trend One: From assisting programmers to helping everyone
Initially, AI was mainly used by programmers, but now it's being applied to all knowledge workers—organizing files, writing reports, creating PPTs, cleaning data, and more. Even free users of some tools (like DouBao Professional Edition) can use AI to generate PPTs. AI is no longer exclusive to programmers; it has become a universal assistant for professionals in the workplace.
**Trend Two: From a single AI to an "AI team"
Single AI systems may make mistakes when handling complex tasks, so many products now include multiple AI components that work together. For example, MiniMax Code's system includes components responsible for data collection, report writing, and error checking, functioning like a small team. This approach improves efficiency and accuracy but also increases costs due to the higher computational requirements.
Trend Three: From app-specific tools to system-level assistants
Previously, AI was limited to specific apps (such as ChatGPT's web version), but now desktop agents reside directly on the computer screen and can access all applications. For instance, you can use a floating assistant like JieyueAI to perform tasks at any time. Some agents even learn your habits (for example, remembering that you prefer to read news in the morning) and proactively suggest or execute tasks for you. AI is becoming a new "operating system layer" on computers—you no longer need to open apps one by one; you just need to communicate your needs.
4. Practical applications: What can these AI tools do for you?
Here are some real-life examples that demonstrate their usefulness:
- Remote control: Use your phone to automatically download files or organize your desktop (e.g., Tencent's Qclaw).
- Office automation: Automatically extract key information from emails, generate weekly reports, and create PPTs (e.g., DouBao Professional Edition).
- Complex tasks: Assist with writing industry reports by researching, organizing data, and creating documents (e.g., Kimi Work).
- Personal assistance: Remind you to drink water, schedule shutdowns, or transfer files remotely (e.g., JieyueAI).
- Entertainment: Engage in games like Werewolf or play cards with AI companions (e.g., Alice).
5. The essence of the competition: Why are companies competing for this "entry point"?
The reason major companies are aggressively launching desktop agents is to control the "task execution interface." In the future, user interactions with the digital world may no longer involve opening apps and performing actions manually; instead, users will simply instruct AI, which will handle everything. Whoever dominates this interface will gain control over users' work processes and data, potentially becoming the new "traffic hub" of the internet era. Just as search engines (Baidu) and social media platforms (WeChat) were once dominant, AI agents could become the new standard, offering greater user engagement.
The surge in desktop agents signifies that AI has finally evolved from a novelty in chat boxes to a practical tool that can truly assist users with their work. In the future, you may no longer need to learn complex software operations; with just your voice or keyboard input, AI will handle most of your tasks—this is the era of "AI-powered work."