Dowson Tong, Senior Executive Vice President of Tencent Group and CEO of the Cloud and Smart Industries Group, recently shared his insights in an article published in Tencent's internal publication "Zhidian" annual edition, as disclosed by the official Tencent Cloud account on the afternoon of August 26.

Tong emphasized that for AI-native applications to truly succeed, the key lies in enabling free exploration, identifying opportunities based on a deep understanding of scenario pain points, acting swiftly when changes emerge, quickly building usable products, valuing user feedback, validating ideas, and continuously iterating.

Reflecting on the past two years of developing Yuanbao, Tong noted that the journey has been instrumental in accumulating experience and capabilities for building large-model products. While many might assume the chatbot battle is already over, he firmly believes that information search and Q&A remain long-term, essential user needs. As long as the answers provided are superior to those of competitors, opportunities will always exist. The process of building Yuanbao has significantly enhanced his understanding of AI work, recognizing that AI products are not merely about combining a model with a page and input box, but rather establishing a comprehensive, evolvable system capability. This includes experiment platforms, evaluation systems, and feedback loops to continuously validate the effectiveness of different models, retrieval strategies, and answer organization methods. Many foundational capabilities developed during Yuanbao's evolution are now being reused by products like WorkBuddy.

Regarding Yuanbao itself, Tong acknowledged that while major competitors have performed admirably in anthropomorphism and voice experiences, offering valuable lessons, the essence of a Q&A product lies in what users truly care about most: the accuracy of answers, the authority and credibility of information sources, and the timeliness of content. If Yuanbao can excel in these core capabilities, there remains substantial opportunity to catch up. Over the past year, amid immense pressure for user growth, significant efforts were directed toward promotion and traffic acquisition, though the models and products were not yet ready, yielding unsatisfactory results. However, with the company's AI strategy becoming increasingly clear and multiple products advancing in parallel, the Yuanbao team can now focus more rhythmically on refining the product experience and strengthening foundational capabilities. Recently, with the Co-Design integration of Yuanbao and Hunyuan Hy3, search experiences have been continuously improving, retention rates are rising steadily, and unit operating costs are declining consistently. As mentioned earlier, as long as the direction is correct, the product is refined with patience, and time is treated as a friend, he remains highly confident in Yuanbao's long-term development.