recently, Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of the Cloud and Intelligent Industry Business Group, published a signed article in the internal publication of Tencent titled "Running with AI: Some Thoughts Over These Years," systematically responding to various concerns from the outside world regarding the pace of Tencent's AI, core product directions, and implementation strategies. In this marathon-like long run, Tencent is trying to navigate through the cycle with its unique rhythm and solid foundation.
The competition in large models has only just started
Facing external comments that "Tencent is slow in AI," Tang Daosheng admitted he was not without anxiety, but he prefers to view this technological competition from a marathon perspective. He pointed out that the second half of the large model era has just begun, and the market structure and business models are far from settled. Just like the development process of the mobile internet era, being early does not necessarily mean winning in the end. The fall of early pioneers such as Nokia and BlackBerry serves as a cautionary tale. What is most important now is not competing for the speed of the first step, but maintaining long-term strategic composure, identifying real needs during technical iterations, and preserving strength and flexibility.
Scenarios are Tencent's thickest advantage
In terms of basic technology investment in large models, Tencent's attitude is very firm, and it will continue to follow the cutting-edge research. However, Tang Daosheng believes that simply focusing on the model layer is not enough; AI must truly be applied to practical scenarios to create value. Tencent's biggest advantage lies in its massive ToC and ToB products, each of which is deeply connected to real users, specific scenario needs, or business processes. By combining algorithmic and engineering capabilities, Tencent can transform data, context, and historical interactions within scenarios into personal memory or reusable skills for large models. Enterprise applications, by accessing intelligent agents through skills and MCP, will also build deeper moats.
The Journey of the WorkBuddy Intelligent Agent
The article gives a detailed review of the evolution of the star product WorkBuddy. Due to the unclear early business model, developer tools once faced losses and contraction pressures, but the team did not cut them off. In 2021, the release of GitHub Copilot brought AI programming into focus, and Tencent subsequently launched CodeBuddy, gaining exploration space through private deployment. By the end of 2025, with the significant improvement in the ability of large models to perform complex programming tasks, and in early 2026, when the team integrated the editing capabilities of Tencent Docs, WorkBuddy, more suitable for non-technical users, was born. Behind the high-speed iteration of more than forty versions in just three months is a new way of working—letting humans focus more on judgment, debugging, and supervision, while AI generates code autonomously in long tasks.
The Continuous Evolution of Yuanbao and Chatbot
Regarding the market's judgment that "the chatbot battle is over," Tang Daosheng holds a different view. He believes that information search and Q&A are long-term user needs, and as long as the product experience is good, opportunities still exist. During the development of Yuanbao, Tencent has accumulated key foundational capabilities such as experimental platforms, evaluation systems, and feedback loops. These capabilities not only support other products but also continuously improve Yuanbao's search experience, increase retention rates, and reduce operating costs. Tencent will refine Yuanbao's fundamental capabilities in a more rhythmic manner, focusing on the accuracy, authority, and timeliness of answers.
Bilateral Layout of Personal Agents and Service Agents
In terms of the product roadmap for intelligent agents (Agents), Tencent believes they will develop in both personal and service directions. On one hand, there are personal agents like WorkBuddy, which efficiently understand user intentions and complete tasks from the user side; on the other hand, there are service agents aimed at enterprises, which need to connect internal systems, meet high availability, 7x24-hour response requirements, and strict demands for data accuracy, security, and permissions. Tencent Cloud enables enterprises to build service agents through an intelligent agent development platform, offering ready-to-use solutions. Both coexist and complement each other, forming a complete architecture for enterprise agent applications.
