China's AI Giants Pursue Divergent Agent Strategies
An analysis of China's AI sector reveals that its largest tech firms are taking different paths in the agent race. Tencent is reportedly shifting to a product-centric, vertically-integrated strategy, while Alibaba is focusing on platformization and open ecosystems. Baidu, in contrast, is betting on proprietary core technology like models and chips to create differentiation.
- Alibaba's open-source framework, AgentScope, is designed for building multi-agent applications and features a `MsgHub` for routing messages between agents, enabling complex conversational flows. The company also released an open-source "deep research" agent in September 2025, which it claims matches the performance of OpenAI's equivalent tool with greater efficiency due to a smaller parameter count. - Baidu's agent development platform, AgentBuilder, provides no-code and low-code modes for creating agents and is used by over 160,000 developers and 50,000 enterprises. The company's AI strategy centers on its Qianfan platform, a one-stop shop for large model services and agent development that supports its own Ernie foundation models alongside open-source alternatives. - Tencent's Agent Development Platform 3.0 (ADP) enables enterprises to build and integrate autonomous agents for scenarios like customer service and marketing. The platform is complemented by an "Agent Runtime" infrastructure designed to execute, sandbox, and observe hundreds of thousands of concurrent agents. - Recent research in multi-agent systems highlights a shift towards dynamic role discovery, where agents autonomously determine and assign roles based on evolving task demands rather than relying on static pre-assignments. Frameworks like Microsoft's AutoGen and CrewAI are gaining traction; AutoGen focuses on chat-centric orchestration for complex conversations, while CrewAI uses a higher-level abstraction based on agent "roles," "goals," and "backstories" to simplify development. - For consumer-facing agents, product design is shifting from shaping screens to shaping agent behaviors, personality, and the rules governing user control and oversight. A key challenge is designing for "legible autonomy," which involves providing clear indicators of what an agent is doing in the background to maintain user trust and a sense of control. - China's AI regulations, such as the 2023 Interim Measures for Generative AI, require service registration, content governance, and security reviews, creating a compliance roadmap for companies. These rules mandate transparency, requiring AI services to display model names and filing numbers with the Cyberspace Administration of China (CAC). - As engineering teams scale beyond 15-20 people, the CTO's role often shifts from direct technical oversight to building leadership layers and formal processes. A common scaling pitfall is the CTO becoming a bottleneck; successful scaling requires implementing decision-making frameworks that delegate autonomy while ensuring coordination and quality control. - A 2025 survey of AI agents in production found that successful teams often trade maximum capability for greater controllability, with 68% of agents executing ten or fewer steps before requiring human intervention. The study also revealed that 85% of teams build custom agent implementations rather than relying on standard frameworks, indicating the field's early stage of maturity.
