DeepSeek Doubles Headcount, Opens Internships Across All Roles, and Skips Sales and PR
It's fair to say DeepSeek has really broken new ground this time.
Since its official founding in 2023, in three years, when have you ever seen this company specifically make a big fanfare announcement to recruit people?
But just a few days ago, DeepSeek dropped a bombshell, officially announcing a large-scale recruitment drive. I believe many of you have seen it.
This time, DeepSeek's public recruitment copy, although not long in text, reveals several very informative points.
1. The Scale of Recruitment
To be fair, since its establishment in 2023, DeepSeek has only been around for three years, and its team size has never been particularly large.
But this time, DeepSeek seems to be getting serious and is preparing to massively expand its existing team.
The text of the recruitment announcement clearly mentions that they will strive to at least double the size of all departments.
Note the wording here: 'all departments', 'at least double'.
In terms of specific recruitment positions, this announcement directly released 33 specific positions across 7 major categories. Compared to previously released job postings, this scale is unprecedented.
For technical R&D positions, roles in algorithms, development, data, testing, and operations are all available. For non-technical R&D positions, DeepSeek has also recently provided many functional and support roles including HR, legal, finance, procurement, and administrative sales.
2. All Positions Accept Interns
Below the specific position list, it was also specially noted this time: All positions accept interns.
It seems this is rarely seen in other companies' recruitment announcements, right?
Regarding similar issues, Liang Wenfeng actually talked about this before when interviewed by the media.
Liang Wenfeng once stated that one of the key principles for DeepSeek's team recruitment is:
Look at ability, not experience.
Because when doing something, an experienced person will tell you without thinking that it should be done this way, but an inexperienced person will repeatedly explore, think very seriously about how to do it, and then find a solution that fits the current actual situation.
If pursuing short-term goals, finding experienced people is correct. But looking long-term, experience is not that important; foundational ability, creativity, and passion are more important.
So this explains why DeepSeek's core technical team is mainly composed of fresh graduates and young people who have just graduated for one or two years.
Correspondingly, in this recruitment announcement, it's not surprising that all positions accept interns.
3. Specifically Not Recruiting Sales, Marketing, and PR Positions
Look at the many specific positions DeepSeek released above. This time, they are recruiting quite a few functional positions, but if you look closely, you'll find that there are no positions for sales, marketing, or public relations.
You have to know, in the current market, for any large model company that can be named, its marketing team and PR team are basically mandatory. Every time these companies release a new model, they hold press conferences and flood the market with product placements and marketing, afraid that users won't know or won't buy in.
But when it comes to DeepSeek, it's completely different.
In the three years since its establishment, every time DeepSeek has produced something impressive, at least in the public eye, there have never been press conferences, and market promotion has never been done.
Even Liang Wenfeng himself rarely makes public appearances. The entire team remains as mysterious and low-key as ever.
Because at DeepSeek, they simply don't maintain a PR department or marketing department, and they don't recruit for sales, marketing, and PR positions during recruitment. The focus is entirely on R&D and products.
See, a truly powerful company is just this unpretentious.
Talking about DeepSeek's recruitment standards, I actually wrote a special article to share this two months ago, summarized in three points.
The first point is value screening.
'Devote oneself to the cause of exploring the essence of AGI, do not do mediocre things, bring curiosity, and use the most long-term perspective to answer the biggest questions.' Such a corporate culture determined the value cornerstone of DeepSeek's recruitment from the very beginning.
First is long-termism, rejecting the profit temptation brought by short-term business drivers, and instead investing a lot of energy into basic research and exploration of AGI.
Second is extreme focus, focusing on basic research related to AGI, to solve the most exploratory and difficult problems in the field.
Furthermore, there is the spirit of open source, embracing open source, enjoying the spiritual satisfaction brought by open source, and in this process gradually improving and enriching one's own technical system.
The second point is boundless passion.
Liang Wenfeng clearly mentioned in an interview that a key condition for team recruitment is passion.
Because only those who truly love this field will explore the unknown from the heart. On this basis, employees' internal drive, creativity, and work efficiency can be further stimulated.
From the experiences of DeepSeek's R&D team members, people who are full of passion and curiosity, have some unique experiences, and can stand out from non-traditional evaluation metrics are often noticed and favored by this team.
The third point is the philosophy of employing people.
DeepSeek's team assessment standards are somewhat different from typical internet companies. There are neither KPIs nor so-called tasks here.
Liang Wenfeng once said that innovation requires as little intervention and management as possible, allowing everyone free space to play and opportunities for trial and error, because innovation often arises on its own, not deliberately arranged, and certainly cannot be taught.
Due to the uncertainty of AI R&D, if traditional KPI metrics were used to constrain the team, it would instead strangle the team's innovation nerves and imprison the team members' creativity.
DeepSeek needs creators and adventurers, not skilled workers on an assembly line. The mechanisms commonly used by internet companies, such as OKRs and KPIs, have been completely abandoned at DeepSeek.
And this management model has gradually spawned its unique innovation ecosystem: first, an innovation culture that allows trial and error; second, an anti-empiricism working method.
At the same time, within the DeepSeek team, it is also bottom-up, with natural division of labor.
Because everyone has their own unique growth experience and comes with their own ideas, they don't need to be pushed. When encountering problems during exploration, they will pull people together to discuss. However, when an idea shows potential, the team will also allocate resources from the top down.
I wonder if you noticed, in this DeepSeek recruitment announcement, one technical team is listed as a key recruitment position.
That's right, it is precisely its: Agent Harness Team.
This is a new department established by DeepSeek this year, and its head is Liang Wenfeng's junior from Zhejiang University, the post-90s quantitative expert Cui Tianyi.
As the former co-founder of TSY Capital, Cui Tianyi has a computer science background from Zhejiang University and 9 years of work experience at the legendary quantitative finance firm Jane Street.
And this Agent Harness team he is responsible for is a new team formed by DeepSeek to transform cutting-edge large model capabilities into actual productivity tools.
The team's core engineering philosophy can be summarized by one formula:
'Model + Harness = Agent'.
This formula reveals a key insight:
No matter how powerful a large model is, it is essentially still a brain. But if it lacks a supporting engineering system, it is like a giant who can only think but cannot act. What Harness needs to do is precisely all the work outside the model.
There is a very vivid metaphor online: If the large model is compared to a steed that can travel a thousand miles a day, then what the Harness team does is to build the corresponding harness and carriage. Only by putting these engineering supports in place can this thousand-mile horse truly exert its strength.
For a long time, DeepSeek has seemed in everyone's impression to be a model company focused solely on underlying basic model research.
However, as competition in the AI industry intensifies, simply possessing a powerful model is not enough to form an absolute moat. Corresponding products, tools, frameworks, and developer interactions are equally important.
This also means that the focus of this team's work is no longer just training smarter basic large models, but is committed to building a model harnessing layer that connects large models and Agent products, responsible for handling all engineering work outside the model.
Currently, two representatives doing Harness relatively well in the industry are CC and Codex, but now that DeepSeek is personally entering this field, it is still very exciting and promising to think about.
These days, I see Cui Tianyi himself frequently active on various social media platforms, personally recruiting and personally answering questions.
It can be seen that their team's thirst for talent is currently very urgent. Interested students can decisively go for it.
DeepSeek's recruitment standards have always been like a breath of fresh air in the industry. The circle says this is the dream company for AI majors.
At the same time, this also serves as a wake-up call to those large internet companies accustomed to using experience, education, and resume background as yardsticks for recruitment.
I believe this also, to a certain extent, explains from the side: why DeepSeek can always stand out and amaze the world.
'Not tempted by praise, not afraid of slander, following the path, uprightly correcting oneself,' perhaps this is the inspiration this young startup brings to the industry.
Note: This article has been included in the GitHub open-source repository 'Road to Coding' https://github.com/rd2coding/Road2Coding, which contains self-study routes + knowledge point overviews for 6 major programming directions (positions) I organized, interview points, my resume, several hardcore pdf notes, as well as programmer life and reflections. Welcome to star.