DeepSeek Opens 17 Recruitment Positions, Core Direction Shifts Toward Agent Productization

Gate News, March 25 — DeepSeek has opened 17 new job positions, shifting its core research focus from foundational model research to Agent productization. Three dedicated Agent roles cover algorithm research, data evaluation, and infrastructure across the entire chain: Algorithm researchers focus on applying reinforcement learning to large model alignment (RLHF/RLAIF, reward shaping, preference learning, etc.); data evaluation specialists are responsible for building evaluation datasets and designing test cases for Agent capabilities such as planning, tool invocation, multi-turn interactions, and long-term memory; infrastructure engineers are tasked with building the Agent runtime platform, including integrating external tools into internal reinforcement learning infrastructure and setting up evaluation platforms. Several positions specify a preference for candidates who heavily use AI programming tools like Claude Code, Cursor, Copilot, etc., as a bonus. The full-stack developer role includes responsibilities such as “being a heavy user of Vibe Coding to continuously explore innovative applications of model capabilities in products,” with the core focus on building “the next-generation container scheduling and isolation platform to support the operation of massive AI Agents.” The Model Strategy Product Manager role has a dedicated Agent track, requiring familiarity with Agent products like Claude Code, OpenClaw, Manus, etc., and insight into high-value application scenarios (including personal assistants, Deep Research, automation workflows, multi-modal device control). The candidate will lead the design of Agent evaluation systems and training data schemes. Compared to January this year, when core positions focused on general directions like “Deep Learning Researcher - AGI,” this recruitment shift clearly emphasizes Agent productization.