Summary of Key Points
In 2026, the domestic AI industry has shifted from competing based on model parameters and rankings to focusing on desktop office assistants. These AI systems are no longer just “teachers” within chat boxes; they aim to become an integral part of users’ computers, helping with tasks such as reading documents, running code, writing reports, and creating PowerPoint presentations. Currently, 17 such products have received over 60 million visits, but there are two main challenges: first, their ability to create professional-looking PPTs and images is lacking (often resembling the work of interns working late into the night); second, large companies refuse to allow these AI systems on their computers due to security concerns. Seven leading firms—Tencent, ByteDance, Alibaba, Baidu, and others—each have their own approaches. The key to success in the future will be addressing these shortcomings and developing web-based versions of their AI that can securely integrate with corporate systems.
From “Chat Box Teachers” to “Computer Managers”: AI is Taking on Real Tasks
In the past, AI product launches would highlight things like “the number of billions of model parameters” or “ranking first in certain lists.” Now, product managers refer to these tools as Work, Buddy, Mate, etc., which sound more like assistants with significant ambitions: they aim to take over real work tasks on users’ computers. For example, if you give them a folder containing PDFs and Excel files, they can organize the content and produce Word documents, spreadsheets, and PPTs.
According to Analysys, the total number of visits to these desktop assistants exceeded 60 million in June this year, with Tencent’s WorkBuddy leading the pack, followed by TRAE and QoderWork. What were once considered geeky toys have now been transformed into “new office creatures” by these large companies. The core change is that AI has evolved from providing answers to gaining control over users’ computers—now they can not only answer questions but also read and write local files, control browsers, run code, and even help with creating weekly reports.
Seven Competitors with Unique Strengths: Whose Approach is the Most Reliable?
Each of the seven companies offers different features, along with clear advantages and weaknesses:
- WorkBuddy (Tencent): Similar to an “AI marketplace,” it can use multiple models (such as Hunyuan and Kimi) and can assign tasks remotely via WeChat or Enterprise WeChat. However, its versatility leads to stability issues, such as frequent permission prompts and lost historical records.
- TRAE Work (ByteDance): Developed by programmers, it excels at converting product requirements into functional web pages. Its main weakness is the lack of professional-quality PPT design capabilities; recently, it added a “design mode” to address this issue.
- Kimi Work (Zhipu): Possesses the strongest models and is good at handling complex tasks like analyzing multiple reports. However, it can become laggy when handling multiple tasks simultaneously.
- QoderWork (Alibaba): Specializes in creating plugins that incorporate methods from legal, investment banking, and finance industries. While more professional than general chat tools, data privacy concerns arise as files may be transmitted to the AI service provider.
- DuMate (Baidu): Relys on Baidu’s search, cloud storage, and document libraries for information retrieval. Its output quality is average, and it might just seem like another version of Baidu’s AI services.
- MiniMax Mavis: A versatile tool for text, voice, and video processing, but it lacks a unified and stable interface for enterprise users.
- WPS AI: Focusing on the Office software ecosystem, it understands PPT templates and font standards well and can refine AI-generated content into professional documents suitable for meetings. In the future, it might collaborate with general AI to divide tasks: AI handles the actual work, while WPS focuses on refining the final output.
Why Are PPTs and Images Still Poor? Beauty Is AI’s Weak Point
Text and tables can be used as long as they are structured correctly, but PPTs and images require attention to visual design and brand guidelines. Current AI systems create content first and then use templates or generate images randomly, resulting in subpar quality (often resembling the work of inexperienced users).
The companies are working to improve this: Qoder has added a Slides Desk for PPT editing, TRAE has introduced a Design Mode, and WorkBuddy has added a PPT plugin. However, there is no standard for visual design, so developing a comprehensive system that includes layout engines, material libraries, and visual review models will be crucial to gaining a prominent position in the office environment.
How Can AI Systems Enter Large Companies? Web-Based Versions Are the Key
Desktop assistants work well with personal computers, but valuable data is stored in corporate ERP, CRM, and OA systems. These systems have strict security measures, making AI clients that can read files and execute commands appear risky from a security perspective.
The solution lies in web-based versions: employees can use them directly via browsers, with IT departments managing accounts, permissions, and data sources. All operations are logged, and high-risk actions require approval. Companies like TRAE offer both desktop, web, and mobile versions; Qoder provides cloud-based solutions. To truly gain traction in large companies, five additional features are needed: single sign-on (integrating with corporate accounts), fine-grained permissions (e.g., access to specific folders only), comprehensive auditing (tracking all user actions), private deployment (preventing data leakage), and integration with corporate data sources (such as ERP systems).
Who Will Win in the Future?
There is no clear winner yet, as all companies have shortcomings. The next stage of competition will focus on:
- For individual users: better-looking PPTs, fewer errors, and less need for manual adjustments.
- For enterprises: secure and reliable AI systems that can access corporate data, manage permissions effectively, and maintain audit trails.
Core Recommendation: Companies should develop both desktop and web-based versions. The desktop version should enable AI to handle personal files, while the web version should allow it to work with organizational data. What truly matters in China is an AI system that can safely complete tasks within corporate systems and processes.
In Summary: For AI to become a “full-time employee” in the office, it must overcome two major hurdles: visual design and security. The desktop version acts as a resume, while the web version serves as the “entry pass.” Those who master both will be the ones to gain the next step in this evolving field.