Chinese AI company DeepSeek has unveiled a prototype of a highly autonomous programming agent, designed to compete with products such as Claude Code from the American firm Anthropic and Codex from OpenAI. This was announced by the DeepSeek “Harness” team in a post on WeChat.
“To require as little human attention as possible while delivering maximum value”—this is how the development objective is described in the user manual published by DeepSeek on GitHub.
The new open-source tool, dubbed “Super-Agent,” is capable of proactively monitoring developer group chats, gathering feedback on code issues, and suggesting solutions—all without needing to be prompted.
DeepSeek established its “Harness” team earlier this year to focus on AI agents capable of performing programming tasks—a field already being explored by leading American AI companies. The term “harness” refers to a software framework built around a large language model, enabling agents to tackle more complex tasks. Anthropic and other developers use such frameworks to optimize their products for programming and various other professional workflows.
According to Bloomberg, Chinese firms are locked in an intense rivalry with their American counterparts to develop more advanced yet affordable AI tools. While OpenAI’s ChatGPT ignited the global AI race, many companies have since shifted their focus toward the burgeoning market for AI agents—tools designed to automate tasks for software developers and business professionals.
To bolster its resources for competing with Anthropic and OpenAI, DeepSeek recently raised approximately 100 billion yuan ($15 billion) in a funding round—one of the largest in Chinese history—ahead of a potential initial public offering (IPO). It is scheduled for 2027.
Previously, the U.S. banned the use of DeepSeek in government systems, citing a threat to national security.
About the Author
