Former Qwen Lead Launches Pragmatik Labs For China’s Agent Race

One of the researchers behind Alibaba’s Qwen models has launched a new AI company in Shanghai, and the move says a lot about where China’s AI race is heading next.

Junyang Lin, the former technical lead for Qwen, has announced Pragmatik Labs, a company focused on next-generation agents across digital and physical worlds. The company’s own website says it is working on digital agents, physical agents, frontier research and products shaped by real-world feedback.

Pandaily reports that Pragmatik Labs has reached a $2 billion valuation at angel stage, with Gaorong and HSG co-leading the round, Tencent investing $20 million and the Shanghai Future Industry Fund also participating. The Next Web also reported the Shanghai government-linked backing and noted that no product has been publicly released yet.

That last point matters. This is not a company being valued on revenue or users. It is being valued on the talent behind it and the belief that agents are the next important AI platform. Lin worked on Alibaba’s Qwen family, one of China’s strongest open-model efforts, before leaving to build around long-horizon agents that can interact with tools, environments and eventually physical systems.

The shift from foundation models to agents is important. The first phase of the generative AI race was about who could train the strongest base model. The next phase is about who can turn models into systems that plan, use tools, remember context, operate software and act in the real world. That is where digital agents, robotics and embodied AI start to overlap.

China has a strong reason to move in this direction. The country already has hardware supply chains, robotics manufacturing depth, large internet platforms and a fast-moving open-model ecosystem. If agents become the layer that connects AI to work and machines, Chinese labs may have advantages that do not depend only on having the most advanced U.S. chips.

This connects naturally to the broader China AI story. We recently wrote that China could still win the AI race if the U.S. underestimates cheaper models, open ecosystems and deployment speed. Pragmatik Labs fits that thesis because it is not trying to be another model leaderboard company. It is trying to build agents that work in context.

The valuation is still risky. A $2 billion company without a released product has to justify a lot of expectation. AI talent is valuable, but agents are difficult. They need reasoning, memory, safety, tool use, reliability and, in the physical world, perception and robotics integration. A demo can look impressive. A product that works repeatedly is much harder.

Still, Pragmatik Labs is worth watching. It shows that China’s AI talent market is no longer only staying inside Alibaba, Tencent, Baidu or ByteDance. Senior researchers are spinning out into new companies backed by major investors and local government capital. That is how an AI ecosystem deepens. The next major Chinese AI story may not be a model. It may be an agent company.