DeepSeek Doubles Headcount, Opens Internships Across All Roles, and Skips Sales and PR
A top-tier AI lab scaling up without sales or PR and with an explicit anti-credentialist hiring philosophy challenges the standard Silicon Valley playbook. The Agent Harness team signals that the next competitive battleground is not bigger models but the engineering layer that makes models useful, a shift that directly affects how developers will build and integrate AI.
DeepSeek's first major public recruitment drive targets a doubling of all departments, with 33 specific positions spanning algorithm development, engineering, data, testing, and operations, plus functional roles like HR, legal, and finance. Every single role is open to interns, a policy that flows from founder Liang Wenfeng's stated preference for raw ability and creativity over industry experience. The company still refuses to hire for sales, marketing, or public relations, channeling all resources into R&D and product.
The recruitment push also spotlights the new Agent Harness team, led by former Jane Street quant and Zhejiang University alum Cui Tianyi. Its core engineering formula — Model + Harness = Agent — treats the large model as a brain that needs a full engineering harness to act. The team builds the tooling, frameworks, and developer interfaces that turn raw model capability into usable agent products, moving DeepSeek beyond pure model research.
Internally, DeepSeek operates without KPIs or OKRs. Management philosophy rejects top-down task assignment in favor of natural, bottom-up division of labor, where individuals pursue ideas and pull in colleagues as needed. The company screens for long-termism, extreme focus on AGI fundamentals, and a genuine passion that, in Liang's view, no amount of experience can replace.
Refusing to hire sales and PR while doubling headcount is a bet that technical superiority alone creates enough pull to make traditional go-to-market functions unnecessary.
Opening every role to interns is not just a recruitment tactic; it operationalizes a belief that inexperience can be an asset when the problems are novel enough that received wisdom is a liability.
The Agent Harness team's formation marks a strategic pivot: DeepSeek is no longer content to be a model provider and is now building the integration layer that competitors like Cursor and Codex already occupy.
A no-KPI, bottom-up management model at this scale is rare and fragile — it works only if the hiring filter for intrinsic motivation is nearly perfect, which the company's values screening is designed to ensure.
DeepSeek's public silence and lack of PR machinery create an information vacuum that paradoxically amplifies its mystique and the impact of any announcement it does make.