跳到正文
Turing Post (Ksenia Se)· @TheTuringPost · X·· 2 小时前精选AI 评分78
AI 导读

Turing Post 的 Ksenia Se 用一周时间实测 DeepSeek 新发布的开源 Super-Agent,观察它完成了多少有用工作、占用了多少人工注意力。

推荐理由

作者以一周实测观察 DeepSeek Super-Agent 的 router 协调机制,可了解多智能体协作如何减少人工介入。

正文 · 原文

Another week, another big DeepSeek release.

This time its new Super-Agent wants to end AI babysitting.

I really get tired of managing multiple agents. It just turns into jumping between conversations, remembering who knows what, and copying information from one to another. Maybe you can relate.

That’s what @deepseek_ai is trying to address and also solve the constant avalanche of requests for approval. And yes, they did it again - it’s open source.

The idea is to give one agent, the router, the job of coordinating the others. You talk to it, it assigns tasks to workers, each with its own context, and keeps track of who’s doing what.

What I like most is that Super-Agent keeps routine updates quiet and only brings you in when there’s a result, a decision for you to make, or an error – something that indeed needs your attention.

So I wanted to explore how much useful work Super-Agent got done in a week, and how much human attention it took.

In this video I discuss how it works without a prompt, what we can use today, whether the architecture helps, and what happens when you close your laptop. Lots of interesting stuff!

来源:Turing Post (Ksenia Se) · x.com