Tencent Accelerates Agent Strategy with AI Organizational Restructuring

Tencent announced on July 23, 2026, the integration of its Hunyuan multimodal model division and large language model (LLM) division to establish a foundational model unit. Yao Shunyu, Tencent’s Chief AI Scientist, was appointed as the head of the new unit, tasked with overseeing comprehensive research and development across LLMs, multimodal models, reinforcement learning, and AI agents. Previously, Linus, who had held leadership roles at JD Digits and Alibaba Tongyi Lab, managed the multimodal model division. However, this restructuring significantly expands Yao Shunyu’s authority. As a result, Hunyuan is now positioned to build a complete model matrix centered on LLMs, covering images, videos, audio, and 3D generation. This marks a major acceleration in Tencent’s AI implementation-focused strategy shift.

Centralizing Authority Under Yao Shunyu

Through Hunyuan Integration

The crux of this organizational restructuring lies in eliminating the dual-track management system. Previously, the multimodal model division and LLM division operated under separate leadership, leading to challenges like dispersed R&D resources and inconsistencies between models. According to reports from Huxiu, Yao Shunyu’s role has been elevated from LLM division head to “Head of Foundational Models,” giving him overarching control over language models, multimodal models, reinforcement learning, and agents.

This integration enables Hunyuan to develop an ecosystem that integrates language, images, videos, audio, and 3D generation. Improving collaboration efficiency in model R&D and exploring the intelligence limits of multimodal models are set as core objectives. Tencent’s move to consolidate its AI talent and resources under a single umbrella demonstrates a significant strategic shift in its approach to AI.

Expansion of Agent-Based Product Portfolio

In conjunction with the restructuring, Tencent is accelerating the rollout of agent-based products. On July 22, Tencent officially unveiled “Miora,” China’s first multi-agent collaborative multimodal creative studio. Miora allows users to submit a single requirement, after which the system deploys different creative agents and specialized skills to complete tasks such as image, video, 3D, and UI creation. The embedded memory system continuously learns user preferences, brand guidelines, and creative habits. According to Huxiu, over half of Miora’s internal test users were not professional designers but staff from marketing and R&D departments who used it for promotional images and demo material creation.

Furthermore, Tencent’s enterprise-focused agent, “Tencent WorkBuddy,” reached 20 million monthly accesses in June, maintaining its lead in the domestic PC-based AI native office agent market, surpassing the combined total of the second- and third-ranked competitors. During WAIC, Tencent released standalone apps for iOS, Android, and HarmonyOS, marking the first general-purpose agent app on HarmonyOS. The company also collaborated with Li Weike to introduce X-AI memory glasses, expanding AI’s entry point into wearable devices. Additionally, Tencent launched the “AI Navigator+” tool for overseas deployment, which incorporates corpora from seven countries to support vertical applications in politics, law, and taxation.

Hunyuan Hy3, Tencent’s foundational model, topped OpenRouter’s global large-scale model call volume rankings within a week of its release. Amid industry competition in parameters and benchmark scores, Hy3 prioritizes high cost-performance for commercial use. Developers evaluated its stability and API experience based on actual call volumes. Currently, Hunyuan supports Tencent’s internal products and offers APIs via Tencent Cloud for external access. Tencent’s platform strategy involves establishing a robust model foundation while scaling agent applications.

Overcoming Enterprise Implementation Barriers

Wu Yunsheng, Tencent Cloud’s Vice President, highlighted that the primary bottleneck in agent implementation is not technology but the outdated legacy systems accumulated by companies over the years. Transitioning these “human-oriented” systems to “agent-oriented” designs requires significant adaptation and modification costs due to the lack of standardized interfaces.

Tencent is addressing this challenge by integrating the entire lifecycle of agents into its ADP development platform. This enables bidirectional calls between agents and workflows, balancing cost management and security audits. Product roles are clearly defined, with WorkBuddy serving as a ready-to-use office efficiency tool and ADP positioned as a cloud-based enterprise development platform tailored for core business scenarios.

Regarding token cost reductions, Wu Yunsheng identified three mature pathways within the industry:

1. Layered model calls, allocating models with varying capabilities based on task complexity.
2. Divided task processing, delegating ambiguous inference tasks to agents and structured processes to workflows.
3. Economies of scale, where increased user numbers drive down unit costs.

These strategies have contributed to a steady decline in token prices. Addressing the technological gap between Chinese and U.S. large-scale models, Wu noted that the two regions have pursued different market strategies. U.S. manufacturers focus on enhancing intelligence limits, whereas Chinese manufacturers prioritize cost-performance while maintaining intelligence levels. Wu added, “A year ago, the gap between Chinese and U.S. models was evident, but now, Chinese open-source models surpass the capabilities of last year’s closed-source models in the U.S.”

Platform Strategy and Ecosystem Development

Tencent’s enterprise agent business is entering a rapid growth phase. The number of enterprise agent customers on Tencent Cloud and agents on the ADP platform has seen over 100% year-on-year growth. Implementation spans diverse fields, from marketing and customer service to inspections and media. ADP has secured the top spot in the agent development platform market for the media sector.

Tencent’s strategy resolves the contradiction between customization and scalability through front-end customization integration and back-end solution standardization. SkillHub, Tencent’s skill marketplace, has amassed 78,000 AI skills, connecting the entire chain of skill distribution and payment, mimicking the ecosystem logic of an App Store.

Tencent’s AI product hierarchy spans four layers: foundational infrastructure, model layer, platform layer, and application layer. By consistently deploying Hunyuan as the model layer, ADP as the core platform layer, and WorkBuddy and Miora as the application layer, Tencent aims to establish its authority in the agent-driven era.

Editorial Opinion

Short-term Impact: The organizational restructuring centralizes Tencent’s AI R&D leadership under Yao Shunyu, enabling quicker decision-making. Over the next 3–6 months, strengthened collaboration between Hunyuan’s modalities and the introduction of additional agent-based products are expected. Notably, WorkBuddy’s cross-platform expansion and Miora’s ecosystem growth could enhance Tencent’s visibility in its AI strategy.

Long-term Perspective: Backed by robust platforms like WeChat, Tencent Meeting, and WeCom, Tencent has a favorable position in the enterprise agent market. However, intensifying competition with ByteDance (Doubao) and Baidu (Ernie Bot) poses challenges in differentiation. While Tencent’s cost-performance strategy aligns well with the Chinese market, maintaining its edge in international intelligence comparisons could be crucial for sustained competitiveness.

References

• ” 腾讯开始激进了 ”, by 黄青春 — 虎嗅网, 2026-07-26T22:50:00.000Z (ARR)
• Source URL: https://www.huxiu.com/article/4877285.html?f=rss