DeepSeek, the Chinese AI lab that emerged from quantitative investment firm High-Flyer, is forming a new team to build an agentic coding tool called "Code Harness." The tool will directly compete with Anthropic's Claude Code and OpenAI's Codex. The announcement was made by Deli Chen, a DeepSeek engineer who posted job listings on X this week for a product manager and an R&D engineer, both based in Beijing.
Unlike many global tech companies that embrace remote work, DeepSeek requires these roles to be in the Chinese capital. The company explicitly states that candidates cannot be based in Hangzhou—where High-Flyer was founded—or Shenzhen, China's tech hub. Beijing represents the political and symbolic center of Chinese tech ambition, where government oversight of AI is most direct and closely watched by Washington.
"You can call it DeepSeek Code or something," Chen wrote in his post, adding laughing emojis. But the implications are serious. The job listings, posted on High-Flyer's recruitment platform, reveal an internal formula: "Model + Harness = Agent." The product manager role requires hands-on experience with Claude Code, Codex, Cursor, GitHub Copilot, Manus, and OpenClaw—essentially every major agentic tool on the market. Salaries for both positions are undisclosed.
The Rise of Agentic Coding Tools
Agentic coding tools represent a paradigm shift in software development. Unlike traditional code assistants that suggest lines or snippets, these programs autonomously plan, write, test, and debug entire projects. They operate in the developer's terminal, handling complex engineering tasks without constant supervision. Anthropic's Claude Code pioneered this space, followed by OpenAI's Codex and other tools like Cursor. The market is now a battleground where every major AI lab wants to own the interface that developers use daily.
DeepSeek's entry into this arena is strategic. The company already has a peculiar advantage: its latest model, DeepSeek V4, runs natively inside Claude Code. This means users can switch the underlying model in Claude Code to DeepSeek V4, benefiting from its cost efficiency. But DeepSeek wants more than just being the engine—it wants to control the entire stack, from model to user interface.
Pricing Advantage
DeepSeek V4 comes in two versions. V4 Flash, the workhorse, is priced at $0.14 per million input tokens—ideal for speed and agentic workflows. V4 Pro, the more capable tier, costs $0.435 per million tokens during an introductory promotion through May 31. In comparison, Anthropic's flagship Claude Opus 4.7 runs at $15 per million input tokens. That's a difference of up to 100 times, which is significant for continuous, loop-heavy agent pipelines where costs can escalate quickly.
This pricing advantage is not new for DeepSeek. In January 2025, the release of its R1 reasoning model wiped nearly $600 billion from Nvidia's market cap in a single day because it matched OpenAI's o1 at a fraction of the cost. Analysts initially dismissed it as a fluke, but DeepSeek continued to release innovative models at disruptive price points.
The Full Stack Strategy
Building a harness means DeepSeek would control the interface developers actually see and interact with—the checkpoints, terminal commands, rollback features, and integrations. That's where user loyalty is built and likely where revenue follows. Currently, many developers use Claude Code with third-party model providers, but the experience is fragmented. A native DeepSeek Code tool would unify the model and the interface, offering seamless integration and potentially lower latency.
The move also signals China's ambitions in the AI space. Beijing has been actively supporting domestic AI development, and DeepSeek's push into agentic tools aligns with the government's goal of building a self-sufficient AI ecosystem. By owning the entire stack—from foundational models to application-level tools—China can reduce reliance on foreign technology and set its own standards.
However, the path is not without challenges. DeepSeek must compete with established players like Anthropic and OpenAI, which have strong brand loyalty among developers. Additionally, the requirement to be based in Beijing may limit the talent pool, as many top AI engineers prefer remote or flexible arrangements.
No launch date has been announced for DeepSeek Code. But the job listings are clear: the company is serious about building a competitive product. For developers interested in working on cutting-edge AI tools, the opportunity is available—but only if they speak Chinese and are willing to relocate to Beijing.
DeepSeek's history shows it tends to underpromise and overdeliver. From R1 to V4, the company has repeatedly demonstrated that it can match or exceed Western models at a fraction of the cost. If Code Harness follows this pattern, the agentic coding tool market could see a major shakeup.
Source:Decrypt News