Chinese AI startup DeepSeek has released an open-source coding-agent prototype designed to handle software development tasks with less human supervision, stepping up its competition with Anthropic’s Claude Code and other AI coding tools.
The project, called Super-Agent or dsh-superagent, is being developed by DeepSeek’s Harness team. It is designed to monitor activity around a software project, identify authorised coding requests and work on them independently before sharing results with developers.
The move adds another layer to DeepSeek’s AI strategy under founder Liang Wenfeng
, whose company has become a major competitor to US AI developers through its open-source models and focus on cost efficient AI development.
How DeepSeek’s Super-Agent works
Unlike conventional coding assistants that largely wait for a developer to submit a prompt, Super-Agent is designed to respond to activity within a development team.
According to its GitHub documentation, the prototype works with a single GitHub repository and can also connect to Feishu, a workplace collaboration platform. When an authorised user requests work, it creates a task, runs it in a separate agent session and can return the result in the original conversation, including a pull request.
DeepSeek’s stated objective is to reduce the time developers spend coordinating and supervising AI agents.
However, the company has clearly labelled the public code a research prototype, not an official commercial product. It also warns that the software is not production-ready and that some features described in its design paper are not yet implemented.
Why the coding agent market matters
The release comes as AI companies move beyond chatbots towards agents capable of completing multi-step tasks.
Anthropic’s Claude Code has helped build interest in AI-assisted software development, while companies across the US and China are working on systems that can write code, identify bugs, run tests and manage development workflows.
The competitive question is increasingly about how much work an agent can complete reliably, rather than simply how well a model responds to a prompt.
For businesses, more autonomous coding tools could reduce routine development work and help engineering teams manage multiple tasks. But reliability, security and the need for human review will remain important considerations before companies allow agents to operate across critical codebases.
DeepSeek’s reported $15 Bn fundraising
The product release follows reports of a major fundraising round for DeepSeek. Bloomberg reported that the company raised around 100 billion yuan, or approximately $15 billion, with investors including Chinese battery maker CATL and technology group Tencent.
Reuters also reported the fundraising figure, citing Bloomberg’s account. The reported round is among China's largest AI fundraising deals and comes as DeepSeek prepares for a potential public listing.
The funding would give the company additional resources to compete with well-funded AI developers, invest in computing infrastructure and expand its research capabilities. The reported amount and investor participation should be understood as media-reported financing details, rather than figures independently confirmed through a public company filing.
For Liang Wenfeng, the challenge is now to translate DeepSeek’s model-building reputation into a broader ecosystem of tools that developers and businesses can use. Super-Agent is an early step in that direction, but its performance in real world development environments will determine how seriously it can compete with established coding agents.
