Just now, Lin Junyang, the person in charge of Qianwen, returned: the new company is targeting Agent, and Tencent is following suit

Only five months have passed since Lin Junyang bid farewell to Qwen late at night, and this former head of Alibaba Qwen in 1993 has returned with his new company Pragmatik Labs!

Sequoia China and Gaorong Venture Capital jointly led the investment, and Tencent and Shanghai Future Industry Fund also participated in supporting this round of financing.

The name of the new company also carries some traces of Lin Junyang’s career trajectory.

Pragmatik， Abbreviated as p7k, it is called pragmatic technology in Chinese.

Lin Junyang explained that he started studying linguistics because his friend recommended pragmatics to him. Later on, he moved from linguistics to computational linguistics, and then into NLP (natural language processing) and big models, circling around, and now he has returned to this name.

Meanwhile, Pragmatik also represents Pragmatism.

According to Lin Junyang, they believe that AGI (General Artificial Intelligence) should ultimately move towards “pragmatic” solutions, truly solving problems and creating value.

It’s a bit like starting from linguistics and finally preparing to use AGI to transform the real world.

What exactly is Pragmatik going to do? The official website states: Agents for digital and physical worlds.

In other words, the new generation of agents they want to create is one that plunges into the digital world of computers and the other that extends into the real world where robots are located.

https://pragmatik.com/

The first thing mentioned on the official website is Digital Agents.

The goal is to enable AI to truly enter knowledge work, enterprise operations, and even large-scale industrial processes. It needs to reason on its own, call tools, adjust actions based on feedback, and coordinate a series of tasks.

You can understand it as a generic agent that everyone is currently crazily rolling, but Pragmatik wants to enter more complex and long-term work scenarios.

Next up are physical agents.

AI is no longer just helping you write documents and code through a screen, but entering a real environment with adaptability and action capabilities, ultimately capable of handling real-world tasks that require tens of minutes, hours, or even longer.

That is to say, what Mr. Lin wants now is a real Agent Holy Body.

Being able to be an advanced cow and horse in the cyber world, using digital agents to handle knowledge work and enterprise processes; It is also possible to enter the real world through physical agents, stuffing intelligence into robots and various physical devices to work on their own.

It is worth noting that according to the information disclosed on Pragmatik’s official website, the company’s goal is not just to create a robot or a certain Agent product.

The official website specifically lists Research to Product.

The meaning is that cutting-edge research cannot just stay in papers and laboratories, but must be truly deployed to the real world, and then based on feedback from real users and environments, decide what to study next.

There are also Long horizons.

Pragmatik hopes to enable agents to break through the current relatively short-term task paradigm, continuously advance the same goal for a longer period of time, and ultimately even apply it to more complex scenarios such as scientific research.

So strictly speaking, Pragmatik’s current ambitions are broader than those of a ’embodied intelligence startup’.

Digital agents, Physical agents, Research to product, Long horizons, pieced together, are actually a new generation of agent system that spans the digital world and the real world.

The official website even directly stated the recruitment target as’ super intelligence ‘. If you also want to create super intelligence that can create long-term value for humanity, please feel free to contact them.

Well, as soon as I started my business, I wrote the final outcome directly on the official website.

Very Lin Junyang.

Looking back at his extreme struggles over the past six months, this experience is even more daring than writing a cool article.

Lin Junyang was born in 1993 and had a cross disciplinary background in linguistics, computer science, and other fields during his university years. Later, he moved from computational linguistics to NLP (Natural Language Processing).

In 2019, he joined Alibaba Damo Academy. In the following 6 years, he rose from P6 to P10 and became one of the youngest P10 technical experts at Alibaba.

After the establishment of the Tongyi Laboratory at the end of 2022, Lin Junyang began to be responsible for the Tongyi Qianwen series of large models and led the development of Qwen’s 0-1 models and the construction of an open source ecosystem.

If Qwen can become one of the most prominent open-source big model families in the world today, Lin Junyang can hardly avoid it. As a result, in early March of this year, everything suddenly took a sharp turn for the worse.

On March 2nd, the Qwen team just opened sourced the Qwen 3.5 small-sized model series, and even Musk came to publicly evaluate its “impressive intelligence density”.

Lin Junyang is still replying in the comment section. Thank you.

Who knows the result, my family? The day before, we were still having a peaceful time with Musk, but on the afternoon of March 3rd, the art style suddenly changed! After an internal meeting, due to issues such as route and organizational adjustments, Lin Junyang left the meeting and submitted his resignation application on the same day.

On the same day, Yu Bowen, the training manager of Qwen, also confirmed his resignation. In the early morning of March 4th, Lin Junyang left only a brief farewell on X: “I’m stepping down. bye my beloved qwen

At almost the same time, several core members of Qwen also experienced personnel changes.

Core contributors Kaixin Li and others have spoken out one after another, and a response from a member of the Qwen team has left the onlookers confused: “I know leaving is not your choice

This personnel shock quickly alarmed Alibaba’s top management.

Soon, Tongyi Laboratory held a staff meeting, and Alibaba Group Chairman and CEO Wu Yongming sent an internal email confirming Lin Junyang’s resignation. He emphasized that he would continue to adhere to the open source model strategy, increase investment in AI research and development, and attract outstanding talents.

On March 7th, Lin Junyang officially put an end to his Qwen career on his social media.

He said, ‘It wasn’t these days that I didn’t know so many people in this world loved me.’ He also wrote that he felt that he had done ‘good for my brothers, good for Alibaba Cloud, and good for the group’ in his heart.

From a late night greeting ‘bye my beloved qwen’ to the final moments of saying goodbye on social media, it has a bit of the flavor of a cyberworld feud.

Now, Lin Junyang has come out to establish his own sect.

And from the direction announced by Pragmatik today, it seems that he has also brought some of the technologies he used during the Qwen era to the new company.

Emphasizing General Agents, Research to Product, and Long horizons, the battlefield extends from big language models to the digital and real worlds.

It can only be said that in today’s world of big models, talent mobility itself has become a strategic asset.

The model can be retrained, computing power can be bought, and data can continue to be accumulated. But there are still plenty of people in the entire industry who have truly guided a world-class basic model from start to finish, gone through pre training, post training, open-source ecosystem, and large-scale engineering.

As for whether Lin Junyang can create another Qwen level project after leaving Qwen, no one knows yet. The only thing that can be confirmed is that the second half of the sentence “bye my beloved qwen” from five months ago has finally arrived.

Hello, Pragmatik.