DeepSeek Harness Super-agent design: the bottleneck isn't smarter models, it's your attention
sujingshen · x · 2026-11-02
A detailed reading of DeepSeek Harness's article "Exploring Super-agent Design: Towards a More Autonomous Agent System". Key claim: single tasks are limited by model capability, but multiple tasks are limited by human attention — a person can realistically follow only 3-5 active sessions at once.
Key points
- The article is about architecture direction, not shipped features: models are already good enough; the problem is making agents that don't need to be fed window by window.
- The common shortcut — wiring CLI/MCP/connectors into a local single-session agent — only adds ears: it's still a synchronous session that stalls on long operations, mixes contexts across tasks, and dies when you close the laptop.
- The way out is decoupling the agent from any single manually-driven session into a persistent entity that keeps context across sessions and advances multiple complex tasks in parallel without interference — the "Super-agent".
- The author pairs each mechanism with a sketch and distills practical tips usable in ChatGPT/Claude/DeepSeek today: cap parallel windows at 3-5, prefer structure over persuasion, manage context like a cache, manage output (draft first and mark uncertainty, batch small judgments).
More from coding & agent
- rakyll's AX Studio prototype manages Gemini/OpenAI/Claude agent tasks — rakyll · 2026-11-02
- Claude built an app that auto-downloads Gradescope exams and runs IRT analysis — DataJunkie · 2026-11-02
- OpenAI DevDay London hackathon: Codex reset reportedly arriving within hours — paw_lean · 2026-11-02
- As AI Writes More Code Than Humans Can Review, What Understanding Still Matters? — Panel_pl · 2026-11-02
- Live by the Claude, Die by the Claude: Can Junior Devs Survive AI? — bashaZP · 2026-11-02
- Skip soul.md: just tell your agent never to use first person, argues jessfraz — repligate · 2026-11-02