--- title: "The design of Claude Code" date: "2026-02" category: "Working with AI" tags: ["Working with AI"] description: "Kind of interesting. Claude Code probably uses an agent team rather than a single agent. The top layer sorts out the requirements first, then produces an intermediate prompt and a todo list..." source: "https://enriquemark.com/en/posts/claude%20code%20%E7%9A%84%E8%AE%BE%E8%AE%A1" --- Kind of interesting. Claude Code probably uses an agent team rather than a single agent. The top layer sorts out the requirements first, then produces an intermediate prompt and a todo list, and finally hands the tasks out and pulls the results together at the end. That process may well involve skills and other tools too. It really is a good design approach, and if you want to build an agent yourself, Claude Code is absolutely the best thing to learn from. --- **Translation note.** I wrote this in Chinese. This English version is an LLM translation, so the wording is not mine even though the thinking is. Original: [claude code 的设计]().