DeepSeek Unveils AI Prototype That Finds Coding Problems Without Being Asked

DeepSeek has showcased a prototype of an open-source coding tool called “Super-Agent” that monitors programmers’ group chats. The tool detects programming-related problems in conversations and suggests solutions. It does not require a programmer to submit a separate query or ask for help. In other words, the AI does not merely respond to tasks it is given; it tries to identify problems on its own.

The tool was developed by the “Harness” team, which the company formed earlier this year. The team is working to build AI agents that can do more while requiring as little intervention from programmers as possible. A software layer added to the large language model enables it to work with various tools and perform more complex tasks. “Super-Agent” uses the same approach.

Anthropic’s “Claude Code” tool already examines project code, finds errors, writes code, and runs tests. OpenAI’s “Codex” service performs tasks in parallel and reviews written code. DeepSeek’s tool stands out by continuously monitoring programmers’ conversations. It tries to determine on its own when help is needed.

“Super-Agent” is still at the prototype stage. The company has not disclosed how reliably it works on complex projects. There is also no information on whether the tool is ready to operate across multiple teams simultaneously. Therefore, “Super-Agent” is not currently considered a complete product capable of replacing “Claude Code” and “Codex.” Its main distinguishing feature is that it identifies problems and tries to intervene without a programmer asking for help.

DeepSeek raised approximately 100 billion yuan in its latest funding round ahead of a planned initial public offering. This is worth $15 billion (approximately 25.5 billion manats). The company plans to use the funds to expand its work in artificial intelligence. With the new prototype, DeepSeek is beginning to compete more directly with OpenAI and Anthropic in the coding agent market.