DeepSeek 推出团队级 Super-agent，可自动监控群聊并修复 bug
Unlike OpenAI's Codex or Anthropic's Claude Code, which require constant 1-on-1 human oversight, DeepSeek's new Super-agent is proactive and team-native. It mon
DeepSeek 推出新的 Super-agent，可监控开发群聊和 PR，在无需 @ 提及的情况下自动复现并修复团队成员提到的 bug。该智能体采用事件驱动触发，通过解耦存储与计算在整个团队间同步上下文，并借助外部 Watchdog 系统实现持续后台运行；DeepSeek 内部还用它审查代码并迭代自身的 Harness 代码库。
推荐理由
AI 评分 74
原文对比了 DeepSeek 超级智能体与 Codex、Claude Code 在触发方式和协作规模上的差异，可据此观察团队级智能体的设计取向。
主题
正文
Unlike OpenAI's Codex or Anthropic's Claude Code, which require constant 1-on-1 human oversight, DeepSeek's new Super-agent is proactive and team-native. It monitors dev group chats, automatically reproducing and fixing bugs mentioned by team members without requiring explicit @ mentions.
How DeepSeek's Super-Agent Operates (in detail):
- Triggering (Reactive vs. Event-Driven): U.S. agents sit idle until prompted. DeepSeek's agent continuously monitors multi-channel streams (group chats, PRs) and self-triggers without explicit mentions.
- Scale (Team-Native vs. 1-on-1): Current tools are single-user assistants capped by individual attention spans. DeepSeek decouples storage and compute to sync context across an entire team.
- Resilience: An external Watchdog system enables continuous background operation instead of timing out in a browser tab.
- Self-Evolution: DeepSeek uses the agent internally to review code and autonomously iterate on its own Harness codebase.