DeepSeek Launches Super-Agent to Rival Claude Code

DeepSeek, the Chinese AI lab, has released an open prototype of an autonomous coding agent called Super-Agent. The goal is to challenge Claude Code, Anthropic's coding agent. The DeepSeek Harness team announced this on WeChat, the Chinese messaging app, on Saturday. They said the agent reads developer chats on its own. It notices complaints about bugs. Then it proposes a fix, even without being asked.

An agent that jumps into a chat uninvited and proposes a fix is exactly where agents tend to go wrong. The task was never clearly specified, so a confident-looking suggestion can still turn out to be incorrect. The announcement gives no numbers on code quality or speed. It only describes the feature. This fits a familiar pattern with Chinese AI labs. Bold benchmark claims often do not hold up in real agentic coding work. A fair comparison with Claude Code will have to wait until developers test Super-Agent on real projects.