DeepSeek proposes Super-agent dual-layer architecture for autonomous multi-task AI
Briefing by News4U · First published
DeepSeek Harness released a design paper on 31 October describing a Super-agent architecture with Router and Worker layers. The Router maintains global context and dispatches tasks while multiple Workers execute them in isolated sessions. The system supports continuous cross-platform interaction, parallel task handling, proactive event response and long-term persistence. Experiments showed median task completion of 12.4 minutes versus 29.5 minutes for single-session baselines. An internal instance is already running for code review and bug fixes, but the prototype is not a formal product. [1] [2]
Sources
- DeepSeek 提出 Super-agent 架构设计：Router 与 Worker 双层架构，探索更自主的 Agent 系统
ITHome · · Excerpt
- DeepSeek 重磅发布全新 Agent 架构：AI 开始指挥 AI 干活
iFanr · · Excerpt