DeepSeek is forming a new team to build an agentic coding platform internally described as “Code Harness,” marking the Chinese AI lab’s clearest move yet into direct competition with coding products like Claude Code from Anthropic and Codex from OpenAI.
The hiring push was revealed by DeepSeek engineer Deli Chen, who posted openings for a product manager and R&D engineer based in Beijing.
🚀 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
According to the job descriptions, the company views the product through a simple framework: “Model + Harness = Agent.”
The listings specifically ask candidates to have experience with tools including Claude Code, Codex, Cursor, GitHub Copilot, Manus, and OpenClaw, highlighting the exact competitive landscape DeepSeek is targeting.
DeepSeek Expands Beyond Foundation Models
The move signals a broader shift in strategy for DeepSeek as competition increasingly moves from standalone models toward full AI agent ecosystems.
DeepSeek already has a foothold inside that ecosystem. Its recently released DeepSeek V4 models can run natively within Claude Code, allowing developers to use DeepSeek models inside Anthropic’s coding environment.
By building its own harness layer, DeepSeek would no longer just provide the underlying model, but also the interface and orchestration system developers interact with directly.
Pricing Could Become a Major Advantage
One of DeepSeek’s biggest advantages remains cost.
The company says its V4 Flash model costs $0.14 per million input tokens, while the more capable V4 Pro tier currently runs at $0.435 per million tokens under a temporary promotion.
That pricing dramatically undercuts premium frontier coding models like Claude Opus 4.7, which costs roughly $15 per million input tokens.
The gap matters because agentic coding systems continuously generate tool calls, terminal loops, planning steps, and long-running workflows that consume far larger token volumes than traditional chatbot interactions.
A lower-cost model stack could make DeepSeek-native coding agents significantly cheaper to operate at scale for enterprises and developers running persistent AI automation workflows.
Agentic Coding Becomes the New AI Battleground
The hiring push comes as nearly every major frontier AI company races to build developer-focused autonomous coding systems.
Anthropic recently expanded Claude Code with multi-agent orchestration and mobile controls through the ChatGPTecosystem, while OpenAI continues scaling Codex and enterprise coding workflows.
Google has also accelerated its own push with Google Antigravity and the new Gemini 3.5 Flash model focused on agentic software engineering tasks.
At the same time, companies including xAI, Cursor, and Replit are increasingly treating AI coding agents as the next core computing interface for developers.
Beijing-Based Expansion Reflects Strategic Importance
The new team is being built specifically in Beijing rather than other Chinese technology hubs like Shenzhen or Hangzhou.
That detail stands out because Beijing remains the political and regulatory center of China’s AI industry, where relationships between frontier AI companies and the government are most closely managed.
DeepSeek itself emerged from quantitative hedge fund High-Flyer and has rapidly become one of the most closely watched AI companies globally following the release of its R1 reasoning model earlier this year.
No launch timeline for the Code Harness platform has been announced.
