On August 12, Lin Junyang, former head of Alibaba's Qwen large model technology, officially announced his new entrepreneurial venture. He has founded a new company in Shanghai called Pragmatik Labs (Yuyong Technology), abbreviated as p7k. The company's research focus is on next-generation agents that bridge the digital and physical worlds.
Lin Junyang departed from Alibaba in March this year. Regarding his new company, he disclosed that Gaorong Ventures and Sequoia Capital China jointly led the current funding round, with TENCENT and the Shanghai Future Industry Fund providing support. Tianyancha data shows that Yuyong (Shanghai) Technology Co., Ltd. was established in May 2026, with Lin Junyang as the legal representative. It has a registered capital of 1.25 million yuan. Shareholders include Gaorong Ventures, an affiliate of Sequoia Capital China, TENCENT-controlled Shanghai Qishan Investment Co., Ltd., and state-controlled Shanghai Future Emergence Enterprise Consulting Partnership (Limited Partnership). Previous media reports indicated that the funding round for Pragmatik Labs could reach several hundred million US dollars.
Explaining the naming of the company, Lin Junyang said he initially studied linguistics on a friend's recommendation, later shifting to computational linguistics and natural language processing. The name "Yuyong" (pragmatics) means returning to the origin of language. He emphasized that it also signifies pragmatism, which he believes AGI (Artificial General Intelligence) should aim for.
According to the introduction, Yuyong Technology's technological development will focus on four main areas: digital agents, physical agents, research-to-product development, and advanced exploration. Digital agents refer to general-purpose agents for knowledge work, business operations, and industrial workflows. Physical agents refer to embodied intelligent systems capable of adapting to environments, taking actions, and completing long-term tasks in the real world. Additionally, the company aims to translate cutting-edge research into real products, incorporating real-world feedback to shape future development directions, and continuously explore systems that break existing paradigms and accelerate scientific progress.
Notably, in March this year, Lin Junyang posted a lengthy article after his departure, discussing his understanding of model development trends and predictions for the next phase of AI. He argued that the core industry question has shifted from "Can models think long enough?" to "Can models think in a way that supports effective action?" He believes the core object of model training needs to transform into the entire system of the model plus its environment. As AI moves into the physical world, Lin Junyang is betting that agent-based thinking will become the mainstream approach.
Comments