Deepseek hires Beijing 'Harness' team to build 'Deepseek Code' AI coding agent

Deepseek has posted job openings for a new Beijing team called "Harness" to develop an AI coding agent under the working name Deepseek Code, signaling a move from model research toward agent — focused product work. The listings frame the effort as building a production — ready coding assistant that combines a base model with a supporting "harness" layer — tool use, planning and memory — to perform developer — oriented tasks, positioning Deepseek as a competitor to Claude Code, OpenAI's Codex and Cursor.

The roles advertised are explicitly product — and engineering — facing rather than pure model — research positions. Openings include a product manager who will own the roadmap, run feedback analysis and build the community, and a developer expected to implement agent — side systems and integrations while working closely with model researchers. The listings describe the Harness group as responsible for the components that sit beyond the base model: tool orchestration, planning flows and persistent memory.

The job ads, shared publicly by Deli Chen on X, call for candidates who are heavy users of existing agent tools such as Claude Code, Cursor, Codex or GitHub Copilot, suggesting early design choices will target parity with current developer workflows and integrations. Technical requirements listed include familiarity with agent loops, MCP, multi — agent systems and context engineering, as well as experience in what the postings term "vibe coding." That mix of skills points to an applied — research approach closely tied to deliverable product features rather than exploratory model science alone.

Deepseek explicitly frames its approach around the formula "model plus harness equals AI agent," placing the harness components at the center of its strategy. The combination of product roles, engineering hires and close collaboration with model researchers implies a roadmap that prioritizes integration work and agent orchestration capabilities. In practice, the openings signal that Deepseek is aiming to ship an agent that fits into developers' existing toolchains and workflows rather than focusing solely on base-model advances. If Deepseek follows through on these hires, developers may soon see another integrated coding agent option that emphasizes practical integrations and orchestration as much as model performance.