<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Jason Ye · Founder OS</title><description>一线 AI 资料的中文译本与注解、Founder OS 方法论、手记。</description><link>https://jasonyeyuhe.github.io/</link><language>zh-Hans</language><item><title>我把 AI agent 做成了自己的个人操作系统</title><link>https://jasonyeyuhe.github.io/writing/agent-as-personal-os/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/agent-as-personal-os/</guid><description>agent 真正有价值的地方，不是会回答，而是可以被调度。我现在理解的个人指挥中心不是一个很酷的 dashboard，而是一套慢慢长出来的操作系统：手机调度、桌面执行、记得过去、明天接着今天往前走。</description><pubDate>Sat, 23 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>personal-os</category><category>agents</category><category>codex</category><category>claude-code</category><category>agentic-engineering</category><category>workflow</category></item><item><title>Agent 越能干，你越容易被挤出判断回路</title><link>https://jasonyeyuhe.github.io/writing/output-format-keeps-you-in-the-loop/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/output-format-keeps-you-in-the-loop/</guid><description>agent 越能干，人越容易被一大坨「看似完整」的输出淹没。不是 AI 写不出来，而是你没法继续判断、比较、编辑、交接。好的 agent output 不只是让你看懂，而是把你带回工作流。</description><pubDate>Sat, 23 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>agentic-engineering</category><category>html</category><category>markdown</category><category>claude-code</category><category>product-thinking</category><category>claude-md</category></item><item><title>Prompt 不是 Skill</title><link>https://jasonyeyuhe.github.io/writing/prompt-is-not-a-skill/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/prompt-is-not-a-skill/</guid><description>有朋友问：Skill 是不是就是收藏一段 prompt？不是。一个最小可用的 Skill 只需要一个文件夹 + 一个 SKILL.md。判断标准很简单：如果你会说「下次我还想让 AI 这样做」，它就值得被做成 Skill。</description><pubDate>Sat, 23 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>skill</category><category>agentic-engineering</category><category>automation</category><category>workflow</category><category>codex</category></item><item><title>把判断变成系统</title><link>https://jasonyeyuhe.github.io/writing/judgment-as-a-system/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/judgment-as-a-system/</guid><description>AI 把创业瓶颈从「能不能做」推到了「该不该做」。但判断不是玄学——它可以被写下来、可复用、可对抗。这就是我说的 Founder OS。</description><pubDate>Fri, 22 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>founder-os</category><category>judgment</category><category>methodology</category></item><item><title>[Library] Claude Code 官方文档</title><link>https://jasonyeyuhe.github.io/library/claude-code-docs/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/library/claude-code-docs/</guid><description>官方文档的入口。安装、CLAUDE.md、Plan Mode、hooks、settings、IDE 集成、MCP server——日常用 Claude Code 时绕不开的查阅地点。</description><pubDate>Thu, 21 May 2026 00:00:00 GMT</pubDate><category>MVP</category><category>Launch</category><category>Scale</category><category>claude-code</category><category>docs</category><category>reference</category></item><item><title>这个站是什么 · 一封发刊词</title><link>https://jasonyeyuhe.github.io/writing/hello-this-site/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/hello-this-site/</guid><description>这个站不是另一个 AI 教程站，是我自己的 founder operating system 的对外入口。资料库收的是一线官方资料的注解，OS 是我的方法论，Build 是我的真实项目。</description><pubDate>Thu, 21 May 2026 00:00:00 GMT</pubDate><category>note</category><category>meta</category><category>launch</category><category>founder-os</category></item><item><title>原型不是验证，对话才是</title><link>https://jasonyeyuhe.github.io/writing/prototype-is-not-validation/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/prototype-is-not-validation/</guid><description>agentic coding 把「我有想法」到「我有原型」的距离压成了几个下午。危险也在这里：一个能跑的原型，太容易被误当成「问题真实存在」的证据。它不是。</description><pubDate>Thu, 21 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>idea-stage</category><category>validation</category><category>ai-native</category></item><item><title>[Library] Claude Code 最佳实践</title><link>https://jasonyeyuhe.github.io/library/claude-code-best-practices/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/library/claude-code-best-practices/</guid><description>Anthropic 工程团队公开的 Claude Code 内部用法合集。讲清楚 CLAUDE.md、计划模式、并行会话、验证回路这些是怎么在他们自己团队里运转的。</description><pubDate>Wed, 20 May 2026 00:00:00 GMT</pubDate><category>MVP</category><category>Launch</category><category>claude-code</category><category>agentic-coding</category><category>context</category><category>workflow</category></item><item><title>[Library] 创始人手册：打造 AI 原生创业公司</title><link>https://jasonyeyuhe.github.io/library/founders-playbook/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/library/founders-playbook/</guid><description>Anthropic 官方手册的非官方中文全译版（36 页，按原版结构 1:1 重排）。这里只放摘要与我的注解，全译 PDF 走 GitHub 下载。</description><pubDate>Wed, 20 May 2026 00:00:00 GMT</pubDate><category>Idea</category><category>MVP</category><category>Launch</category><category>Scale</category><category>playbook</category><category>AI-native</category><category>founders</category><category>Anthropic</category></item><item><title>AI 时代的技术债，是复利的</title><link>https://jasonyeyuhe.github.io/writing/ai-technical-debt-compounds/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/ai-technical-debt-compounds/</guid><description>传统技术债线性累积，可以攒一个 sprint 还掉。AI 技术债不一样——它复利。原因不在代码质量，在于每次会话都重新推断结构假设，而这些假设会漂移。</description><pubDate>Wed, 20 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>mvp-stage</category><category>technical-debt</category><category>claude-md</category></item><item><title>[Library] MCP · Model Context Protocol</title><link>https://jasonyeyuhe.github.io/library/mcp/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/library/mcp/</guid><description>Anthropic 主导、面向开放的「让 AI agent 连接工具与数据源」的协议。理解 MCP 是把 agent 从演示变成产品的关键一环。</description><pubDate>Tue, 19 May 2026 00:00:00 GMT</pubDate><category>MVP</category><category>Launch</category><category>Scale</category><category>mcp</category><category>protocol</category><category>agents</category><category>tools</category></item><item><title>翻完 Anthropic 创始人手册，我记下的三件事</title><link>https://jasonyeyuhe.github.io/writing/translating-the-playbook/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/translating-the-playbook/</guid><description>我把官方 36 页 playbook 1:1 译成了中文（保留全部插图与排版）。逐字过一遍的副作用，是有几句话真的钻进了脑子。这是其中最顶的三条。</description><pubDate>Tue, 19 May 2026 00:00:00 GMT</pubDate><category>note</category><category>reflection</category><category>founder-os</category><category>translation</category></item><item><title>[Library] 构建有效的 AI 智能体</title><link>https://jasonyeyuhe.github.io/library/building-effective-agents/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/library/building-effective-agents/</guid><description>Anthropic 工程团队给「workflow」与「agent」划了一条清晰的线，并列出几种实战中真正有用的组合模式。是判断「这个场景该不该上 agent」的最快入门。</description><pubDate>Mon, 18 May 2026 00:00:00 GMT</pubDate><category>MVP</category><category>Launch</category><category>Scale</category><category>agents</category><category>workflows</category><category>architecture</category><category>patterns</category></item><item><title>你从做工作的人，变成设计系统的人</title><link>https://jasonyeyuhe.github.io/writing/founder-becomes-the-bottleneck/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/founder-becomes-the-bottleneck/</guid><description>创业生命周期里最难的转变之一，是从「亲手做工作」切换到「设计做这件工作的系统」。难就难在它没有一个明确的发生时刻——你很容易整个错过它。</description><pubDate>Mon, 18 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>launch-stage</category><category>systems</category><category>delegation</category></item><item><title>护城河来自累积深度，不是又一个新功能</title><link>https://jasonyeyuhe.github.io/writing/moat-from-accumulated-depth/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/moat-from-accumulated-depth/</guid><description>到了 Scale 阶段，护城河的来源很反直觉：不是功能数量，而是累积的深度——内置的领域专长、集成深度、被锁定的用户工作流。这些都不是冲刺几个 sprint 能换来的。</description><pubDate>Sun, 17 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>scale-stage</category><category>moat</category><category>compounding</category></item><item><title>[Library] Anthropic Startups · 案例与计划</title><link>https://jasonyeyuhe.github.io/library/anthropic-startups/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/library/anthropic-startups/</guid><description>Anthropic 官方的创业项目页 + 案例库。和官方 playbook 配着读，能看到那些抽象阶段在真实公司里长什么样。</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate><category>Idea</category><category>MVP</category><category>Launch</category><category>Scale</category><category>startups</category><category>case-studies</category><category>program</category></item><item><title>Chat、Cowork、Code：我怎么分工</title><link>https://jasonyeyuhe.github.io/writing/chat-cowork-code/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/writing/chat-cowork-code/</guid><description>Chat、Claude Cowork、Claude Code 底层共用同一个 Claude；区别是围绕它的工作空间。我按「任务形状」来分工，而不是按「哪个更强」。</description><pubDate>Sat, 16 May 2026 00:00:00 GMT</pubDate><category>essay</category><category>claude</category><category>workflow</category><category>tools</category></item><item><title>[Library] Anthropic Cookbook · 代码示例库</title><link>https://jasonyeyuhe.github.io/library/anthropic-cookbook/</link><guid isPermaLink="true">https://jasonyeyuhe.github.io/library/anthropic-cookbook/</guid><description>Anthropic 官方维护的代码 recipes。讲 patterns 的文章读完后，回到这里能找到对应的可运行代码。</description><pubDate>Fri, 15 May 2026 00:00:00 GMT</pubDate><category>MVP</category><category>Launch</category><category>cookbook</category><category>examples</category><category>claude-api</category><category>patterns</category></item></channel></rss>