DeepSeek推出了一款新型自主编码代理的原型，此举是中国人工智能实验室加大努力，以与Anthropic的Claude Code、OpenAI的Codex等其他流行编码工具竞争的一部分。
DeepSeek has unveiled a prototype of a new autonomous coding agent, as the Chinese AI lab steps up its efforts to compete with Anthropic’s Claude Code, OpenAI’s Codex and other popular coding tools.
DeepSeek Harness团队在周末开源了其描述为“超级代理”的项目。该团队表示，该研究原型“不是DeepSeek的官方产品”，是其持续努力的一部分，旨在构建一种需要最少人工干预的更自主代理。Harness团队基于DeepSeek的基础模型构建代理应用程序。
The DeepSeek Harness team open-sourced what it described as a “super agent” over the weekend. The team noted that the research prototype, which is “not an official DeepSeek product,” is part of its ongoing effort to build a more autonomous agent that requires minimal human intervention. The Harness team builds agentic applications based on DeepSeek’s foundational models.
当超级代理处理任务时，它可以在公开群聊中观察反馈，主动识别问题的根本原因并提出修复方案，Harness团队在一篇微信帖子中表示。用户可以通过多个渠道与该代理交互，包括即时通讯平台、GitHub以及字节跳动的工作协作应用飞书。例如，如果用户最初通过即时通讯应用指示代理使用特定方法，但随后在GitHub上要求其切换到不同方法，代理可以理解这一变化并相应执行任务。
When the super agent tackles a task, it can observe feedback in public group chats, proactively identify the root cause of a problem and propose a fix, the Harness team said in a post on WeChat. Users can interact with the agent across multiple channels, including instant messaging platforms, GitHub and ByteDance’s work collaboration app, Feishu. For example, if a user initially instructs the agent to use a specific method via a messaging app, but later asks it on GitHub to switch to a different method, the agent can understand the change and execute the task accordingly.
超级代理还旨在同时协调多个复杂任务，让人类花费更少时间进行会话管理，而专注于最终审查和批准，Harness团队表示。
The super agent is also designed to coordinate between multiple complex tasks simultaneously, allowing humans to spend less time on session management and focus on final review and approval, the Harness team said.