China’s internet users already live inside super apps. Now those apps want to think and act on their behalf.
Instead of tapping through menus for coffee, flights or groceries, millions are handing off entire tasks to AI systems. The agents plan. They compare. They buy. And they do it without the user switching apps.
The Agent Takes Over
Tencent connected an open-source AI agent framework called OpenClaw directly into WeChat on March 22, 2026. Overnight a contact named ClawBot appeared in the chat list for more than a billion users. Send it a request. It executes. No download. No new interface. (Beam AI, March 30, 2026)
But the move didn’t come from nowhere. Chinese developers had spent months building on OpenClaw after its release. Adoption exploded. Shenzhen even offered subsidies to companies using the framework despite some security concerns in Beijing. The result? A wave of agents that don’t just answer questions. They finish jobs.
Alibaba responded with its own push. In March 2026 the company launched Wukong, an enterprise platform that coordinates multiple AI agents at once. Document editing, spreadsheet updates, meeting transcription, research. All inside one interface. The launch followed a company reorganization that created a new AI-focused business group. (Reuters, March 17, 2026)
ByteDance took a different route. Its Doubao agent reached 159 million monthly active users by October 2025. The system opens four shopping platforms at the same time, scans for products, applies coupons and completes the cheapest purchase in under 30 seconds. Users don’t browse. The agent does. (Ivinco, March 5, 2026)
These aren’t experiments. More than 600 million people have used some form of agentic app, according to estimates shared by The Economist. Consumer spending in China remains weak. Tech giants see autonomous agents as a fresh growth engine. They turn passive users into customers who consume more because the friction disappears. (The Economist, May 17, 2026)
The original Digital Trends report captured the shift early. Chinese companies moved past the super app model. They now build agents that handle the entire flow from decision to delivery. The article highlighted how this changes daily routines. Users issue high-level goals. The system breaks them down, calls tools, confirms details and completes transactions. (Digital Trends)
Yet the real advantage lies in the architecture. WeChat already contains payments, messaging, mini-programs and logistics data. An agent inside that environment doesn’t need to ask for permission to act. It simply does. American developers wrestle with fragmented logins and permission walls. Chinese agents operate inside closed gardens built over a decade. The gap shows.
Khosla Ventures partner Ethan Choi pointed out the difference in a video discussion. China’s super apps give agents instant access to real-world actions. Booking a table, sending money, ordering food. All without leaving the conversation. The US ecosystem remains split across dozens of apps. (YouTube via The Information)
Alibaba went further in early June 2026. It opened its Qwen model to third-party AI agents. Brands including KFC, Mixue, Luckin Coffee and China Eastern Airlines now test features that handle customized orders through natural conversation. Qwen’s app already logs more than 100 million daily lifestyle service engagements. The company positions the assistant as the single gateway for digital tasks. Individual apps may become unnecessary. (South China Morning Post, reported across X on June 4, 2026)
Tencent keeps pushing too. The company nears the launch of a native AI agent for WeChat’s 1.4 billion users. Internal testing of a prototype has begun. Compliance checks are scheduled for this month. President Martin Lau told investors in 2025 that agents would sit at the core of the WeChat ecosystem. Integration across payments, logistics and social features matters most. (Financial Times, June 2026; CNBC, January 21, 2026)
Baidu chose breadth. Rather than one flagship agent it released a family of OpenClaw-based tools for desktop, cloud, mobile and smart home devices. ERNIE Bot maintains 200 million users. During last year’s Double 11 shopping event 83 percent of livestreamers used Baidu’s digital human technology. Livestream numbers jumped 119 percent. Gross merchandise value rose 91 percent. The agents don’t replace people. They multiply what one person can manage.
Startups add pressure. Butterfly Effect’s Manus agent drew global notice in March 2025 with claims it outperformed OpenAI tools on certain benchmarks. Zhipu AI released AutoGLM Rumination the same month. Both companies open-sourced key components. Chinese labs continue to publish model weights and code. The strategy accelerates experimentation. It also spreads capability across thousands of developers.
Enterprise adoption follows. Alibaba’s Accio agent for small businesses now sees 10 million monthly active users. It manages customer service, taxes, marketing, logistics and listings. One-person companies suddenly handle work that once required teams. President of Alibaba.com Kuo Zhang credits the OpenClaw wave with educating the market. The tools lowered the barrier. Businesses experiment faster. (Business Insider, March 28, 2026)
Challenges remain. Reliability varies. Agents still make mistakes on complex or ambiguous tasks. Privacy questions grow louder as systems gain spending power. Regulators watch closely. Beijing promotes large-scale AI application in its latest policy signals. Yet it also sets standards for agent development through bodies like the China Academy of Information and Communications Technology.
The pace feels relentless. From chatbots that answered questions to agents that finish errands. The distance closed in roughly 18 months. Chinese consumers already accept the trade-off. Convenience beats control in many daily situations. They let the agent pick the coffee blend, confirm the order and track the delivery. Then they move on.
Western observers call it agentic commerce. Chinese users simply call it normal. The super app didn’t die. It learned to act. And the next version of the internet may not need as many clicks. Or as many decisions.
Whether this model exports remains uncertain. Cultural comfort with centralized platforms helps. Data density inside each app helps more. American attempts at super apps have faltered. Their AI agents will likely face the same fragmented reality.
China’s giants don’t wait for perfect models. They ship agents that improve through use. Billions of interactions generate training signal. The loop tightens. Capabilities compound. The agents get better at doing it all. For you.
WebProNews is an iEntry Publication