China’s AI startup race has gained a new company with an unusually ambitious brief. Junyang Lin, a former technical leader on Alibaba’s Qwen team, has announced Pragmatik Labs, a Shanghai-based venture that says it will build agents for both digital and physical worlds. The company’s own website describes its mission as developing systems that reason, use tools, learn from feedback, coordinate action, and eventually move from knowledge work into real environments. The Next Web’s report says the company has attracted backing from Gaorong Ventures, HSG, Tencent, and the government-financed Shanghai Future Industries Fund.
The announcement is important less for a released product than for what it reveals about the next China AI narrative. The previous cycle was dominated by foundation models that could write, code, summarize, and answer questions. The next cycle is increasingly organized around agents: systems expected to complete multi-step tasks, operate tools, persist through feedback, and, in some cases, connect reasoning to robotics. Pragmatik has made that ambition explicit from the start. Its proposed scope extends from software agents for work and operations to embodied intelligence that can act in the physical world.
Lin left Alibaba on March 4 after contributing to Qwen’s development, and announced Pragmatik about five months later. That personnel movement reflects the continuing fragmentation of China’s AI talent pool, where senior researchers and builders are using the current investment cycle to create companies around narrower technical theses. EastFrontier recently covered ByteDance’s effort to build a dedicated AI data and safety organization, another indication that firms now see specialized teams, data pipelines, and agent systems as strategically decisive.
Pragmatik is betting on agents, not another generic chatbot
Pragmatik’s public description distinguishes between digital agents and physical agents. Digital agents are intended for knowledge work, operations, and industry-scale workflows. Physical agents are intended to bring intelligence into real environments. The company also emphasizes research-to-product cycles, real-world feedback, and long-horizon tasks.
That framing matters because it moves beyond the familiar assistant model. A conventional chatbot takes a prompt and returns text. An agent is expected to interpret a goal, break it into steps, use software or tools, evaluate the result, and continue until it reaches an outcome. The distinction is not purely semantic. A useful agent must manage permissions, memory, retrieval, tool reliability, user preferences, cost, latency, and failure recovery. It also needs a business case stronger than a visually impressive demo.
For physical agents, the challenges are even harder. The system must combine perception, planning, control, safety, and hardware constraints. Pragmatik’s website does not identify a robot partner, product schedule, model architecture, or demonstration. It would therefore be premature to describe it as a robotics company in the commercial sense. What it has announced is a research and product direction: digital agents that can eventually link to embodied systems.
The investor roster links private capital and Shanghai policy
The funding reported for Pragmatik brings together venture firms, a technology giant, and government-backed capital. Gaorong Ventures and HSG reportedly co-led the round, while Tencent and Shanghai’s Future Industries Fund participated. This blend is increasingly characteristic of China’s AI financing environment. Private investors can provide speed, networks, and commercial pressure, while large platforms offer distribution and infrastructure relationships. Government funds can connect startups to local industrial priorities.
Shanghai’s Future Industries Fund is a 10 billion yuan vehicle aimed at six designated future industries. Its involvement in Pragmatik gives the startup a place within the city’s broader effort to become a center for advanced AI, chips, biotech, future energy, and robotics. For the city, backing an agent startup led by a former Qwen figure supports a narrative that Shanghai can attract and retain frontier AI talent rather than only host downstream applications.
Still, there is a critical uncertainty around Pragmatik’s financial scale. The Information previously reported that Lin was targeting a $2 billion fundraise in May. Neither Lin nor the investors confirmed a valuation in the subsequent reporting. The number has drawn attention because it would be unusually large for a new startup, but it should not be treated as an established valuation or closed financing result.
That distinction is more than a technicality. China’s AI market is full of headline valuations attached to companies before they have shipped products. A high profile and strong backers may help Pragmatik recruit, buy compute, and negotiate partnerships. They do not eliminate the need to show a product that works reliably and a business that can support an expensive model and agent infrastructure.
From Qwen talent to a new startup formation cycle
Lin’s departure from Alibaba illustrates how the Qwen ecosystem is becoming a source of new company formation. Major Chinese AI labs have trained engineers and researchers in the practical work of scaling models, serving users, building toolchains, and managing infrastructure. When senior figures leave, they take that experience into startups that can pursue a more focused thesis than a large platform may allow.
Pragmatik is not the only company aiming to create agents. Alibaba, Tencent, ByteDance, DeepSeek, Zhipu, Manus, MiniMax, and many smaller teams are all working on systems that connect models to work. Global competitors are pursuing the same goal. The differentiator will likely be execution: which teams can make agents reliable enough for real tasks while controlling inference costs and earning trust from users and enterprises.
The physical-agent ambition adds another layer. China’s strength in manufacturing, supply chains, autonomous vehicles, drones, and service robotics gives startups a reason to believe that agent research can connect to tangible deployment. Yet that advantage can become a distraction if teams promise embodied intelligence before solving basic software-agent reliability. Pragmatik’s public materials acknowledge the long horizon by describing a progression from digital work to physical action.
For now, Pragmatik is a company defined by its people, backers, and thesis rather than by a released product. That is normal for a frontier AI startup, but it raises the standard for what comes next. The company must show how its agents differ from the growing number of systems that can call tools or control workflows. It must demonstrate whether its digital and physical ambitions reinforce each other rather than dilute focus. And it must do so in a market where the cost of AI capability is falling, but the cost of proving durable value remains high.
Pragmatik’s launch is therefore a meaningful talent and strategy story. It signals that China’s AI entrepreneurs are moving beyond the question of who can train a capable language model. The new question is who can turn models into agents that complete real work in software and, eventually, in the physical world.