Liang Wenfeng recruited a junior schoolmate.

A top expert has joined DeepSeek.

News spread that DeepSeek is forming a Harness team internally, focusing on code intelligent agent products, targeting Claude Code under Anthropic. Subsequently, Chen Deli, a senior researcher at DeepSeek, confirmed the news. Simultaneously, DeepSeek opened recruitment for two positions: Harness product manager and R & D engineer.

A new team and new faces have emerged.

According to LinkedIn, Cui Tianyi, the former co - founder of quantitative trading firm TSY, joined DeepSeek's Harness team in March this year. Different from before, DeepSeek used to have few social recruitments. Cui Tianyi is not a fresh graduate. After graduating from the School of Computer Science at Zhejiang University, he worked at the top quantitative trading firm Jane Street for 9 years and then co - founded a business.

The much - talked - about financing news is yet to be finalized. At this moment, DeepSeek is targeting the product end and has taken the first step in commercial exploration.

A genius teenager admitted to Zhejiang University joins Liang Wenfeng

While the outside world focuses on DeepSeek's talent turnover rate, it ignores the company's attractiveness to top - notch talents.

At DeepSeek, the research team of about 200 people has a very high density of researchers, and many have backgrounds from top universities. Perhaps because a group of equally excellent talents are gathered here, there is less emphasis on departmental barriers or seniority. During his internship at DeepSeek, intern Xin Huajian led the development of the DeepSeek - Prover series of models focusing on mathematical proofs and was the first author of the corresponding paper.

"Even the HR at DeepSeek will forward news about the latest progress of large models and industry trends," said a former employee of DeepSeek.

Cui Tianyi, who has now joined DeepSeek, is also a genius teenager. Strangely enough, he and Liang Wenfeng both graduated from Zhejiang University and have both delved deeply into the quantitative field.

As a 2008 graduate of Anyang No.1 High School in Henan Province, Cui Tianyi was admitted to Zhejiang University through the national youth informatics olympiad bronze medal. Judging from the enrollment year, Cui Tianyi is probably a post - 90s generation.

When he was an undergraduate in the School of Computer Science, Liang Wenfeng was a postgraduate in the major of Information and Communication Engineering at Zhejiang University. During his four - year college life, Cui Tianyi represented Zhejiang University and won the gold medal in the ACM International Collegiate Programming Contest Asia Regional Contest six times.

Upon graduation in 2013, he was hired as an Assistant Quantitative Researcher by the Hong Kong branch of the quantitative giant Jane Street Capital with an annual salary of over one million yuan. For a time, he became an idol among many students at Zhejiang University. After that, Cui Tianyi worked at Jane Street Capital for 9 years, during which his future boss, Liang Wenfeng, founded Magic Square Quantitative.

In 2022, Cui Tianyi embarked on a similar entrepreneurial path - co - founding the quantitative trading institution TSY Capital, which is committed to developing systematic quantitative trading strategies for the global stock market. The team members also come from various prestigious universities. In February this year, news spread that Cui Tianyi had left TSY Capital.

Just one month later, Cui Tianyi announced that he had joined the DeepSeek Harness Team.

The second half of DeepSeek

Now DeepSeek wants to be more than just a model company.

In the official recruitment information, DeepSeek wrote: "All work other than the model itself falls within the scope of Harness." At the same time, it straightforwardly presented a formula: Model + Harness = Agent.

Looking back, Harness became extremely popular in the AI circle at the beginning of 2026. If the large model is like a "brain", Harness can determine what this brain can see, what it can do, what it can't do, how to verify after it's done, how to correct when it makes a mistake, and when to hand it back to humans.

Just as its name implies, Harness puts a harness on AI, allowing humans to tame AI's code - writing ability.

In this regard, Claude Code under Anthropic has done an excellent job. Thanks to its solid Agent SDK (i.e., Harness) design, it has become the most widely used AI programming tool among developers in a very short time. Although the code - writing ability of DeepSeek V4 is already very strong, there is still a certain gap between the bare model and Claude Code.

Therefore, DeepSeek has also started to improve its ability to turn the model into a product, trying to transform its model capabilities into actual desktop - based Agent products.

This may be another step for DeepSeek in commercialization.

Since its establishment in 2023, Liang Wenfeng has always declined investment. In early May, it was reported that DeepSeek's valuation soared five - fold in 21 days to 350 billion yuan. The target of the entire round of financing could reach up to 50 billion yuan. It is reported that Liang Wenfeng personally invested 20 billion yuan, accounting for 40% of the total financing amount, making him the largest investor in this round of financing.

The outside world sees this as a "compromise" by DeepSeek in the current fierce competition of large models.

On the one hand, the competition for top AI talents is intensifying, and DeepSeek needs to give its core internal members a definite valuation. On the other hand, the competition of large models has shifted from the parameter arms race to actual combat, and product R & D and iteration require funds.

The second - half competition for value has begun, and it's time to test the commercialization ability of large - model companies.

This article is from the WeChat official account "Investment AI", written by Feng Yuchen and published by 36Kr with permission.