DeepSeek Details Super-agent Architecture for Autonomous, Multi-Task AI
DeepSeek Harness detailed a Super-agent architecture designed around Router and Worker layers, event-driven execution, persistent state, asynchronous tool calls, and a Watchdog for recovery and oversight. The team said an internal instance is already used for issue reproduction, code-review responses, and DeepSeek Harness development, while the dsh-superagent-prototype implementation was released only to validate the technical approach and is not an official product.
The proposal shifts agent design from isolated sessions toward continuous, concurrent systems that can observe events and coordinate work, but DeepSeek's performance figures come from synthetic workloads and do not measure real task execution quality.
11月1日消息，DeepSeek Harness 官方团队发布《Super-agent 设计探索：迈向更自主的 Agent 系统》一文，介绍了其面向工程实践的 Super-agent 架构设计思路。
随着大模型执行单个任务时越来越不需要人的干预，限制生产力进一步提升的因素正逐渐转向人的注意力，Agent 系统需要从被动响应指令的单任务工具，向持续在线、能够并行处理多项任务的自主协作者演进。
DeepSeek 表示，现有主流 Agent 设计仍以单一 Session（会话）为基本单位，不同 Session 之间相互独立。用户并行推进多项工作时，需要手动切换并维护各个 Session 的上下文。
根据 DeepSeek 团队观察，一个人通常只能同时跟进 3 至 5 个活跃 Session。即使通过 CLI、MCP 或外部连接器扩展本地 Agent，也仍可能面临单会话同步阻塞、多任务上下文混乱，以及本地设备离线后任务停止等问题。
他们将这种新型 Agent 形态及其系统架构称为 Super-agent，并归纳出四项基础特征：连续、高效、主动和持久。
#DeepSeek#DeepSeek Harness#dsh-superagent-prototype#Router#Worker#Watchdog