Tencent is fully engaged in the Agent battlefield

Author | Huang Yu

Editor | Wang Xiaojun

With the catalysis of DeepSeek and Manus, the popularity of AI Agents continues to rise. 2025 is widely regarded as the inaugural year for AI Agents, with global tech giants and startups entering the field, and the commercialization process accelerating.

Recognizing this trend, Tencent has also decided to accelerate the application of AI Agents.

On May 21, at the 2025 Tencent Cloud AI Industry Application Summit, Tencent Cloud announced a comprehensive upgrade of its large model knowledge engine to the Tencent Cloud Agent Development Platform (TCADP), integrating Tencent Cloud's RAG (Retrieval-Augmented Generation) technology, comprehensive Agent capabilities, and features refined through practical experience to meet user needs.

By launching the Tencent Cloud Agent Development Platform, Tencent Cloud hopes to provide enterprise users with the ability to quickly build Agent applications.

Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of the Cloud and Smart Industry Group, stated that users can allow Agents to autonomously decompose tasks and plan paths, actively selecting and invoking tools. "We have achieved zero-code support for multi-Agent handover collaboration for the first time, further lowering the threshold for building Agents."

On the Tencent Cloud Agent Development Platform, Tencent Cloud has built a complete Agent tool system, supporting the MCP protocol and compatible with key definitions of the OpenAI Agents SDK. It also pre-installs a wealth of high-quality internal and external plugins, including Tencent Location Services and other ecological MCP Servers.

These capabilities can help AI Agents better invoke tools, query professional data, and expand service boundaries.

As a company with a vast array of application scenarios, many products within Tencent have already leveraged the Tencent Cloud Agent Development Platform to incorporate Agent capabilities, such as QQ Browser, Tencent Health, Tencent Cloud Code Assistant CodeBuddy, and Tencent Qidian Marketing Cloud.

Tang Daosheng introduced that QQ Browser recently launched the Agent QBot, which allows users to issue a task command, and QBot can directly perform a series of operations such as searching, browsing, finding, downloading, and analyzing.

Although Agent products are rapidly diversifying, there is currently no unified and clear product definition for Agents in the industry.

In the view of Wu Yunsheng, head of Tencent Cloud's AI division and head of Tencent Youtu Lab, at the user demand level, an Agent is a new application form that can autonomously plan and choose what tools to invoke, including multi-Agent collaboration to complete a complex task.

In other words, Agents differ from traditional AI Assistants, which require user prompts for each response generation, while Agents theoretically only need the user to issue a high-level task to autonomously plan the completion path The underlying large model capability is the key to making Agents truly useful, serving as a "brain" like existence.

Tencent has clearly committed to executing a multi-model strategy of "steadfastly investing in self-developed models + openly embracing advanced open-source models." Since the beginning of this year, Tencent has been actively integrating the DeepSeek large model, while the iteration speed of its self-developed Hunyuan model has also significantly accelerated.

The Tencent self-developed inference model Thinker (T1), which excels at complex tasks and deep reasoning, has been rapidly iterating since its launch on the Yuanbao App at the beginning of the year. Additionally, Tencent has released a new generation of fast-thinking model Hunyuan Turbo S, which focuses on faster task processing capabilities.

Based on the TurboS foundation, Tencent has also launched the visual deep reasoning model T1-Vision and the end-to-end voice call model Hunyuan Voice. Furthermore, a series of multimodal models such as Hunyuan Image 2.0, Hunyuan 3D v2.5, and Hunyuan Game Visual Generation have also been simultaneously "launched."

To achieve rapid product innovation and deep model research and development, Tencent has integrated its AI products and applications, including Tencent Yuanbao, QQ Browser, Sogou Input Method, and ima, into CSIG (Cloud and Smart Industry Group) this year, and has made organizational adjustments to TEG (Technical Engineering Group), which is responsible for Tencent's Hunyuan large model development tasks.

Last month, Wall Street News learned that Tencent has comprehensively restructured its Hunyuan large model R&D system. After the adjustment, TEG established two new departments: the Large Language Model Department and the Multimodal Model Department, responsible for exploring cutting-edge technologies in large language models and multimodal large models, continuously iterating on foundational models, and enhancing model capabilities.

At the same time, there is a further strengthening of large model data capabilities and platform foundation construction, with the Data Platform Department focusing on the full-process management and construction of large model data, while the Machine Learning Platform Department focuses on the construction of machine learning and big data integration platforms, providing a comprehensive and efficient PaaS platform foundation for AI model training and inference, as well as big data business, jointly supporting Tencent's Hunyuan large model technology R&D.

Tang Daosheng has pointed out that with the open-source of Deepseek and breakthroughs in deep thinking, AI large models are crossing the threshold of industrialization and standing at a new node of widespread application. The industry has evolved from being dominated by model training to more application and Agent-driven development today.

The future market space for Agents is vast, which is undoubtedly an important reason for Tencent Cloud to accelerate the implementation of AI Agent applications.

Minsheng Securities released a research report stating that it firmly believes that 2025 will be the inaugural year for AI Agents and the starting point for a software revolution: Agents may become an important catalyst for the revaluation of software, with the target market for software vendors expected to expand to the trillions of dollars labor market. AI Agents are also expected to enhance the consumption attributes of software, further opening up the valuation ceiling for software vendors.

According to the latest predictions from Gartner, the proportion of enterprise software integrating autonomous AI will leap from less than 1% in 2024 to 33% in 2028; at the same time, over 15% of daily work decisions will be autonomously completed by AI agents In this global AI arms race, AI Agents have become a necessary path, and it has become an industry consensus that internet giants are simultaneously focusing on both C-end and B-end.

Ying Ying, Chief Analyst of Computers at CITIC Securities, pointed out that compared to the current layout of Agents at home and abroad: North American cloud vendors mainly focus on helping customers efficiently deploy models and Agents, while B-end vendors are more focused on creating and managing Agent platforms; domestic internet giants continue to follow the user traffic logic of the internet era, seizing users through general Agent products similar to "Manus," while B-end companies are similar to those in North America.

On the C-end product side, Tencent has not yet launched a native Agent product similar to "Manus."

At Tencent's recent first-quarter earnings meeting, the management revealed that from Tencent's perspective, they will categorize Agent products into two types: one is a general Agent that anyone can create—it acts on behalf of users to complete tasks in the external world; the other is an AI agent embedded in the WeChat ecosystem, operating based on WeChat's unique ecosystem.

It is reported that in terms of general AI Agents, Tencent is building this capability through AI-native products like Yuanbao and IMA.

According to Tencent's plan, initially, they will be able to quickly answer questions; then they will incorporate "chain thinking" long reasoning models to handle more complex inquiries; subsequently, they will be able to execute more complex tasks, gradually evolving into "embodied intelligence" capabilities, interacting with other applications, programs, and even external APIs to assist users.

Tencent's management stated that this process will continue to evolve, and its capabilities are not fundamentally different from those of peers' general AI Agents.

The AI Agent that Tencent plans to build within the WeChat ecosystem is a differentiated product that is difficult for other vendors to replicate.

It is reported that this Agent will deeply connect with the core components of the WeChat ecosystem, including social relationship chains, communication and community capabilities, content systems like public accounts and video accounts, as well as millions of mini-programs. These components encompass information, transaction, and operational capabilities, spanning multiple verticals.

Similar to the previously launched native AI applications, the significant importance of internet giants creating AI Agents lies in competing for the super traffic entrance of the AI era, and no one dares to be absent.

By 2025, the hot topic in the AI field has shifted from large language models to AI Agents. The development of AI Agents is unstoppable, but current product capabilities are still in the early stages. In this context, whoever can create the "Deepseek of the AI Agent field" is likely to become the winner in the next phase