Summary
- DeepSeek is assembling a "Code Harness" team in Beijing to create a native coding tool that will rival Anthropic's Claude Code and OpenAI's Codex.
- The existing DeepSeek V4 already works seamlessly with Claude Code, marking a strategic shift to control the entire agentic stack rather than merely providing the underlying model.
- V4 Flash operates at $0.14 per million input tokens, significantly cheaper than Claude Opus 4.7, allowing any DeepSeek-native tool to have a clear pricing advantage.
What began as a side initiative is now poised to challenge one of Anthropic's flagship products.
DeepSeek, an AI lab originating from the quantitative investment firm High-Flyer, is in the process of forming a "Code Harness" team aimed at developing an autonomous coding tool that would directly compete with Anthropic's Claude Code and OpenAI's Codex. This news was shared by Deli Chen, a DeepSeek engineer, who recently posted two job openings on X for a product manager and an R&D engineer.
Both positions are located in Beijing.
Unfortunately for developers in the West advocating for remote work, the company mandates that candidates must be based specifically in China's capital. Living in other cities like Hangzhou, where High-Flyer was established, or Shenzhen, known as China's Silicon Valley, is not sufficient.
Beijing is the desired location—the political center of China's technological aspirations, closely monitored by the government and the U.S.
"You could call it DeepSeek Code or something along those lines," Chen humorously noted in his post, though the implications are serious.
🚀 We’re hiring! DeepSeek is forming a new Harness team to build Code Harness from the ground up—may be you can call it DeepSeek Code or something like this hhh🤣🤣🤣
📍 Based in Beijing. Two roles open:
🧠 Harness Product Manager → https://t.co/vb3aWbYV9L
👨💻 Harness R&D…— Deli Chen (@victor207755822) May 20, 2026
The job postings, listed the same day on High-Flyer's recruitment site, clearly outline what DeepSeek is aiming to build. The internal motto for the team is "Model + Harness = Agent." The product manager position specifically requires applicants to have practical experience with Claude Code, Codex, Cursor, GitHub Copilot, Manus, and OpenClaw—essentially a list of the leading agentic tools DeepSeek plans to compete against. The compensation for both roles has not been disclosed.
The Importance of Tools in AI Development
During historical gold rushes, some individuals became wealthy by mining gold, while others profited by selling essential tools. DeepSeek aims to excel in both areas.
Agentic coding tools, which autonomously plan, write, test, and debug entire software projects rather than merely suggesting code snippets, are currently a highly competitive sector in AI. Anthropic's Claude Code allows developers to delegate complex engineering tasks to an AI that operates independently. OpenAI introduced Codex, and many technology firms are striving to create the AI that will be integrated into every developer's terminal.
DeepSeek is eager to enter this arena, boasting a unique advantage: its latest model, DeepSeek V4, already operates within Claude Code.
The V4 series, launched on April 24, includes two variants. V4 Flash, designed for efficiency, executes agentic functions at $0.14 per million input tokens. (Tokens serve as the fundamental units of data processed by large language models.) The V4 Pro variant, which offers enhanced capabilities, is currently priced at $0.435 per million tokens as part of a promotional offer lasting until May 31. In contrast, Claude Opus 4.7, Anthropic's flagship model, costs $15 per million input tokens, representing a significant cost difference for teams managing ongoing, resource-intensive agent workflows.
DeepSeek's strategy of developing a Code Harness is about attaining control over the entire stack. By creating a harness, DeepSeek would manage the user interface that developers interact with, including checkpoints, terminal commands, rollback features, and integrations. This is where customer loyalty is built, and likely where revenue will follow.
DeepSeek has a track record of underestimating its potential while consistently delivering results. For instance, when it unveiled R1 in January 2025, the reasoning model wiped nearly $600 billion off Nvidia's market capitalization in a single day by matching OpenAI's offering at a fraction of the price. Initially dismissed as a fluke, DeepSeek continued to release innovative products.
While no launch date has been confirmed, the job postings are available for those interested, who can speak Chinese and are willing to relocate to Beijing.