Huawei Developer Conference (HDC) 2026 is set to open on June 12 at Songshan Lake in Dongguan, with HarmonyOS and frontier AI technologies taking center stage. Beyond updates to the HarmonyOS ecosystem, the market is also watching AI agents as a key test of Huawei's broader AI roadmap.
As large language models mature, the technology industry is shifting attention to the next stage of human-machine interaction. AI agents, which can plan tasks, call tools, and operate across applications, are increasingly seen as a crucial step in moving AI from "chatting" to "doing."
Over the past year, AI agents have become a new battleground for China's technology giants. Huawei has continued to integrate Huawei PanGu LM with the HarmonyOS ecosystem. Tencent is reportedly preparing to bring agent capabilities into WeChat. Alibaba is linking its Qwen model with e-commerce and local services, while ByteDance is using Doubao to connect with TikTok's e-commerce resources.
As major platforms accelerate deployment, AI agents have moved from a technical concept to a mainstream direction for the industry.
Cooling hype, persistent demand
The market's enthusiasm has not been without turbulence. In March 2026, Tencent's open-source project OpenClaw, known in Chinese as "Lobster," triggered a wave of user experimentation that became informally known as the "everyone raising lobsters" craze. Within months, however, the trend quietly cooled, with early users reportedly uninstalling the tool and raising questions over whether AI agents were already losing momentum.
Industry observers generally argue that what has cooled is excessive market expectation, not the long-term outlook for AI agents. Unlike conventional chatbots that mainly generate text responses, OpenClaw can use AI to act on behalf of users. It can understand instructions, plan steps, call tools, operate interfaces, and complete tasks.
As real-world use cases have increased, however, OpenClaw's limitations have also become clearer. One issue is higher-than-expected operating costs, as agents must frequently call large models and various tool services. Even simple tasks can consume large amounts of tokens. Another is unstable execution. When tasks involve multiple systems or longer workflows, interruptions, misjudgments, and execution failures can occur.
High-level access permissions have also raised personal cybersecurity concerns, including risks related to data leaks, malware injection, and system control.
Yet while the OpenClaw craze has faded, the AI agent industry continues to heat up.
China's AI agent race splits into two tracks
China's major technology companies have already launched different forms of agent products. Beyond its enterprise collaboration tool WorkBuddy, Tencent is reportedly preparing to integrate agent capabilities into WeChat, connecting millions of mini-program services to form a task-oriented assistant. Alibaba is continuing to strengthen Qwen's agent capabilities by linking AMap, Taobao, Ele.me, Fliggy, and other local service functions. ByteDance's Doubao can directly call TikTok e-commerce resources to search for products, place orders, and process payments.
Startups are also moving quickly. Moonshot AI has launched Kimi Agent and Kimi Claw. Z.ai, formerly Zhipu AI, has released AutoGLM and AutoGLM Rumination. Yonyou Network has built enterprise-grade intelligent agent platforms, YonAgent and YonClaw, bringing agent functions deeper into ERP, supply chain, and corporate management systems.
The market is gradually splitting into two development paths. One is consumer-facing agents built around WeChat, TikTok, Taobao, and the HarmonyOS ecosystem, using super apps to connect life services and content platforms. The other is enterprise-grade agents, focused on business process automation, knowledge management, and supply chain management.
Based on current deployment, the enterprise market is becoming an important breakthrough point for agent commercialization. Some financial institutions have begun using agents for data governance and risk analysis. Manufacturers are applying them to supply chain coordination, production scheduling, and equipment maintenance. Government agencies are also testing agents to improve the efficiency of public services.
Industry participants widely believe, however, that the real challenge begins once AI agents enter core enterprise workflows. Longstanding problems such as data silos, difficult system integration, and inconsistent data quality remain major barriers to deployment. Once agents gain the ability to execute processes and modify data, companies must also manage permissions, build audit systems, and protect internal data security.
With HDC about to open, market attention is turning to whether Huawei will disclose further progress on Huawei PanGu LM and its HarmonyOS agent architecture. For the industry, the next stage of competition is no longer just about model parameter scale. It is about which company can integrate ecosystems, control data, build secure and trustworthy execution mechanisms, and prove those capabilities in real-world applications.
Article translated by Levi Li and edited by Jack Wu
