Yicai) May 21 -- DeepSeek has begun hiring for agent harness product managers and research and development engineers, in a sign the Chinese artificial intelligence startup is pushing deeper into AI coding products to rival offerings such as Anthropic's Claude Code in a fast-growing corner of the AI market.

Successful candidates will join DeepSeek’s Agent Harness team in Beijing and take part in the full development cycle of a desktop agent, working to “transform DeepSeek's frontier model capabilities into leading agents,” according to the job descriptions posted on the Hangzhou-based company’s website.

The product manager roles call for deep hands-on experience with tools including Claude Code, Cowork, Codex, Cursor, OpenCode, GitHub Copilot, Manus, OpenClaw and Hermes, a lineup that spans nearly all of the industry’s leading coding and agentic AI tools. But DeepSeek did not disclose what form its planned desktop agent will take.

"Join DeepSeek to build Code Harness from scratch," Deli Chen, a senior researcher at DeepSeek, posted on social media, noting that the project aims to "benchmark against Claude Code." 

Chen, who earned a master’s degree from Peking University, joined DeepSeek in 2023 and has played a central role in several of its key research efforts. He has also become one of the startup’s most visible public-facing figures.

AI coding has emerged as one of the fastest-growing commercial applications of large language models. The global market for AI coding tools reached USD29.6 billion last year and is projected to expand to USD64.7 billion by 2030, implying a compound annual growth rate of about 17 percent, according to market researcher Research and Markets.

Competition in the AI-assisted coding market is already well-established. In China, ByteDance has launched Trae, Alibaba Group Holding offers Tongyi Lingma, Tencent Holdings has CodeBuddy and Baidu provides Wenxin Kuaima. Overseas, Anthropic offers Claude Code, while OpenAI has rolled out Codex.

Beyond commercial potential, coding capability is a key benchmark for evaluating the strength of foundation models, with top-tier coding agents also serving as an important gateway to attracting developer ecosystems.

DeepSeek's move into this space may be linked to its recent fundraising activity, which could give it more resources to pursue product development beyond pure research, industry observers noted.

Editor: Martin Kadiev