Summary of Key Points
Tencent has recently accelerated its efforts in the AI field: on one hand, it has integrated the Hunyuan Model department (combining the large language model and multi-modal teams) to unify the technical foundation; on the other hand, it has vigorously promoted the implementation of intelligent agent products (such as the creative agent Miora and the office agent WorkBuddy). It has also provided solutions to the core bottleneck in deploying enterprise-level intelligent agents (adapting existing systems) and is driving scalability through a platform-based ecosystem (SkillHub). Overall, Tencent has established a complete AI framework that spans from underlying computing power and models to mid-layer platforms and upper-layer applications, betting on intelligent agents as the core strategic direction for the future.
1. Integration of Hunyuan Models: Ending Dual Systems to Lay a Solid Foundation for Intelligent Agents
Previously, Tencent's large language model and multi-modal models (for processing images, videos, etc.) were two separate departments, which have now been merged into the "Basic Model Department" under the leadership of Yao Shunyu. This is like combining two independent workshops into one, allowing for the sharing of resources and technologies, thereby improving efficiency. The unified Hunyuan Model encompasses capabilities in language, images, videos, 3D, and other modalities, providing intelligent agents with a more comprehensive set of tools. Intelligent agents need support from underlying models to perform tasks such as creativity and office work, and this centralized management enables them to respond more quickly to higher-level requirements.
Hunyuan has avoided the traditional approach of simply increasing the number of parameters in the model. Instead, it focuses on cost-effectiveness: companies are not willing to spend significantly more for only a 1% improvement in performance, and individuals do not want to pay high fees for daily AI usage. Hunyuan currently supports Tencent's products like WeChat and QQ internally and offers APIs through Tencent Cloud externally. The more developers use it, the lower the cost, which has helped it become one of the most frequently used large models globally within a week of its release.
2. Implementation of Intelligent Agent Products: From Creativity to Office Work – Making AI Truly Useful
Tencent's intelligent agents are not just for show; they directly solve practical problems:
- Miora Creative Agent: You submit a request (e.g., "Create a new smartphone promotional image with a 30-second video"), and the system will coordinate multiple AI assistants to work together – some for drawing, some for video editing, and some for creating 3D effects. What’s more convenient is that it learns your preferences (such as preferred color schemes and brand styles), making it increasingly user-friendly over time. You can also save frequently used tasks as "skills" for quick access next time, eliminating the need for repeated operations. During internal testing, over half of the users were not designers; marketers and developers could use Miora to create materials, indicating that ordinary people can also become creative masters.
- WorkBuddy Office Agent: In June, the number of PC visits exceeded 20 million (more than the combined total of the second and third-ranked agents), and it has a dedicated app for mobile systems. It has even collaborated with eyewear manufacturers to develop "memory glasses" that automatically record meeting content, addressing the issue of AI's inability to retain context. The agent also provides tools for international operations, taking into account policies and tax information from seven countries to help companies avoid common pitfalls when expanding overseas.
The core idea behind these products is to integrate AI into daily work processes, freeing people from repetitive tasks (such as image editing and meeting note organization) so they can focus on creativity or decision-making.
3. The Biggest Barrier for Enterprise-Level Intelligent Agents: Outdated Systems
Many companies want to use intelligent agents but are hindered by outdated systems that lack standard interfaces, preventing them from directly accessing data or operating processes. For example, with old ERP systems, intelligent agents cannot access inventory information without additional customization and modification, which is more troublesome than developing new technologies.
Tencent’s solution is the ADP Enterprise-Level Intelligent Agent Platform:
- It enables "two-way interaction" between intelligent agents and outdated systems: vague tasks (like analyzing customer complaints) can be handled by intelligent agents, while specific processes (such as generating expense reports) can use the existing system workflows, saving costs without wasting the investments made in those systems.
- It emphasizes "trustworthiness and controllability": industries like finance and government require high data security, so ADP allows for strict control over who can access data and tracks every step of operations (for auditing purposes).
- There is a clear division of roles: WorkBuddy is a personal productivity tool, while ADP is an enterprise-level platform that can be integrated with CRM, OA, and other systems, covering core business functions such as finance and legal affairs.
4. Ecosystem Development: Enabling Mass Adoption and Scalability
Tencent does not aim to develop all intelligent agents on its own; instead, it builds a platform for everyone to participate:
- SkillHub Skill Ecosystem: Similar to the AI version of the App Store, developers can upload their AI skills (e.g., automatically generating contracts or processing invoices) for users to purchase. There are already 78,000 skills available, creating a cycle where developers create skills, users use them, and developers earn revenue.
- It addresses the contradiction between customization and scalability: Companies have diverse needs, so Tencent uses a combination of "customized front-end solutions and standardized back-end processes" – on-site engineers meet with clients to develop templates that can be reused by other similar customers, balancing both customization and scalability.
- Collaboration with Traditional SaaS: Instead of replacing traditional SaaS, intelligent agents act as a "scheduling center" that connects different SaaS systems (e.g., using intelligent agents to access CRM data and send notifications via OA) to complete complex tasks together.
5. Tencent’s Complete AI Framework: From Foundation to Application – A Closed Loop
Tencent has established a clear AI framework:
- Foundation: Computing power, storage, and other infrastructure that support all models and applications.
- Model Layer: Hunyuan large language model and multi-modal models provide core AI capabilities.
- Platform Layer: ADP intelligent agent development platform and Tairos embodied intelligence platform that connect models and applications.
- Application Layer: WorkBuddy (for office work), CodeBuddy (for coding), Miora (for creativity), etc., covering both individual and enterprise use cases.
The advantage of this system is its "layered interaction": improvements in the underlying models enhance the functionality of upper-layer applications, and more users of the applications provide richer data for the models, which in turn improve their performance. By integrating resources, implementing products, and building an ecosystem, Tencent is gaining a significant role in the era of intelligent agents.
In summary, Tencent’s current AI strategy is clear: it uses a unified model foundation to support intelligent agents, solves practical problems with these agents, and achieves scalability through platforms and ecosystems. This not only represents Tencent’s own transformation but may also influence the overall pace of AI adoption in the industry.