DeepSeek Unveils Super-Agent Prototype to Rival Claude Code
DeepSeek has revealed Super-Agent, a prototype AI agent that watches developer team chats, spots bugs and proposes fixes unprompted, aimed at Anthropic's Claude Code.
Chinese AI startup DeepSeek has unveiled a prototype called Super-Agent, an AI agent for software development that monitors engineering teams, detects problems and proposes fixes proactively. The move targets Anthropic's Claude Code, one of the most popular AI coding tools.
An agent that does not wait for commands
Unlike conventional chatbots that answer questions and return code on request, Super-Agent can follow group conversations among programmers, flag errors and technical issues in real time, and suggest solutions without waiting for a direct instruction. According to the usage guide published on GitHub, the project aims to cut the time and effort developers spend on everyday tasks, delivering as much help as possible with minimal human intervention. In effect, the AI becomes an active member of the team rather than a question-answering tool.
Competition and funding
The release comes amid intensifying competition between American and Chinese companies to build agents capable of handling complex professional work on their own. DeepSeek is positioning Super-Agent as an open-source, more efficient alternative to tools such as Claude Code, which has gained wide popularity for writing code, analyzing projects and carrying out development tasks. The company has also been developing models that improve coding, reasoning and long-context handling while lowering memory and compute requirements.
DeepSeek's ambitions are backed by substantial capital. Reports carried by Bloomberg say it has raised about 100 billion yuan, roughly $15 billion, in one of the largest funding rounds in China's tech sector. Investors reportedly include Tencent, owner of WeChat, and battery maker CATL. The money could speed up research as the company weighs a possible initial public offering.
Still a prototype
Super-Agent remains at the prototype stage, and according to the project's documentation it is not yet a finished commercial product ready for broad use. Still, it reflects a shift in the industry from helping users write code to running whole workflows and catching problems before a human steps in. If the approach proves reliable, it could change how development teams operate in the coming years and sharpen the race for the AI agent market.
Apps worth a look
SUMO: the number puzzleTap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.
Dialo: learn languages by speakingTalk to an AI conversation partner; every sentence comes back corrected and explained in your language. 9 languages, A1 to C2.
ediviewer: read EDI files like documentsOpen EDIFACT and X12 messages, see every segment explained, convert to JSON/XML, export PDF and Excel. Runs locally, no upload.
More from this day
See all →
Votour V5 crossover enters production at former Hyundai plant
ChatGPT Desktop App Tests Merged Chat and Work Tabs
Australian officer boards Spain's Álvaro de Bazán, finds a twin of HMAS Sydney
