DeepSeek shares a 'super-agent' built for minimal human attention
An AI agent is a program that acts for you: it reads, writes code and runs tasks. On October 31, DeepSeek's Harness team published an essay called 'Exploring Super-agent Design: Towards a More Autonomous Agent System'. With it came a free research prototype, dsh-superagent-prototype, under the MIT license. It got 10,768 GitHub stars in its first days.
It is not a new AI model. It is a way to organize many agents around one GitHub project. A 'Router' takes incoming requests and decides who does what, but does not do the work itself. Each accepted job gets its own 'Worker'. The goal, as community posts describe it, is minimal human attention. DeepSeek says it is only a prototype, not an official product.
Not everyone thinks agent swarms work yet. Tom's Hardware reports that teams of 50 agents in Minecraft failed to build a civilization and got stuck in 'paralyzing bureaucracy' (170 points on Hacker News). Still, MIT Technology Review says it plainly: 'the agentic web is here'.
This is where the industry is going: fewer humans in the loop, more agents talking to agents. After the agent incidents we saw in October, 'minimal human attention' should read as a warning, not a feature. If you try it, keep it in a sandbox and read what the workers actually do.
