# 龙虾AGI通用实验室（Lobster AGI Lab）

> 从第一性原理出发，做 AI 原生的产品与思考。作者：格里灰丝，清华大学工学博士，清华大学经管学院数字金融方向博士后，现任某国企高级AI算法研发工程师；曾在中央部委直属研究院任战略规划主管、高级研究员，深度参与国家级AI与区块链基础设施的顶层设计与战略研究。具备丰富的多智能体系统构建与AI原生产品开发经验，擅长从第一性原理出发思考问题，对AI原生赛道有系统性洞察。

站点：https://myagilab.com  GitHub：https://github.com/guoy0701  联系：guoy0701@gmail.com

## English summary

Lobster AGI Lab (龙虾AGI通用实验室) — AI-native products and ideas, built from first principles. Personal site of Gelihuisi (格里灰丝). PhD in Engineering from Tsinghua University and postdoctoral researcher in digital finance at Tsinghua's School of Economics and Management. Currently a senior AI algorithm engineer at a state-owned enterprise; previously head of strategic planning and senior researcher at a research institute under a central government ministry, deeply involved in the top-level design and strategy of national AI and blockchain infrastructure. Experienced in building multi-agent systems and AI-native products, with a habit of reasoning from first principles.

English pages: https://myagilab.com/en/ . Articles are written in Chinese; each has a plain Markdown version (.md).

- [Opus 5.5 Direct](https://myagilab.com/en/products/chat/): A claude.ai-style AI assistant that works from mainland China: Opus 5.5 by default, real web reading, deep research, long documents and PDFs. Live at https://chat.myagilab.com
- [ClaudeTerminal](https://myagilab.com/en/products/terminal/): A desktop GUI for Claude Code: collapsible history, a Dynamic Island, auto-resume after rate limits, and remote control from your phone. Live at https://github.com/guoy0701/ClaudeTerminal/releases
- [AI Decision Council](https://myagilab.com/en/products/juece/): Ten AI agents hold a meeting to break down your hard decision. Live at https://juece.myagilab.com
- [ChatTree](https://myagilab.com/en/products/chattree/): Turns your ChatGPT, Claude, DeepSeek and Kimi conversations into a tree. Live at https://chattree.myagilab.com
- [VideoQuickExtract](https://myagilab.com/en/products/yinpin/): Paste a link or upload a file and get the spoken words as text. Live at https://yinpin.myagilab.com
- [WeChat Formatter](https://myagilab.com/en/products/paiban/): Write Markdown, preview the WeChat layout live, and push to your drafts in one click. Live at https://paiban.myagilab.com
- [MarkView](https://myagilab.com/en/products/md/): A lightweight Markdown viewer. Live at https://md.myagilab.com
- [Claude Code TXT Reader](https://myagilab.com/en/products/txt/): Turns Claude Code's exported TXT files into readable conversation logs. Live at https://txt.myagilab.com
- [HydroMet Report Generator](https://myagilab.com/en/products/shuiwen/): AI report generation for hydrology and meteorology. Live at https://shuiwen.myagilab.com
- [PlantMed Sleep](https://myagilab.com/en/products/kangyang/): An AI assistant for plant-based medicine and sleep health. Live at https://kangyang.myagilab.com

## 产品（全部在线可用）

- [Opus 5.5专线](https://myagilab.com/products/chat/)：像网页版一样的中文 AI 问答：默认 Opus 5.5，联网读原文、深度调研、长文与 PDF。在线地址 https://chat.myagilab.com
- [ClaudeTerminal](https://myagilab.com/products/terminal/)：Claude Code 的桌面图形客户端：折叠回看、灵动岛、限流自动续跑，手机扫码远程看。在线地址 https://github.com/guoy0701/ClaudeTerminal/releases
- [狗头军师](https://myagilab.com/products/juece/)：10 个 AI Agent 开会，帮你拆解复杂决策。在线地址 https://juece.myagilab.com
- [ChatTree](https://myagilab.com/products/chattree/)：把 ChatGPT / Claude / DeepSeek / Kimi 的对话画成一棵树。在线地址 https://chattree.myagilab.com
- [VideoQuickExtract](https://myagilab.com/products/yinpin/)：粘贴链接或上传文件，把视频里的话变成文字。在线地址 https://yinpin.myagilab.com
- [WeChat Formatter](https://myagilab.com/products/paiban/)：写 Markdown，实时预览微信排版，一键推到草稿箱。在线地址 https://paiban.myagilab.com
- [MarkView](https://myagilab.com/products/md/)：轻量的 Markdown 渲染查看器。在线地址 https://md.myagilab.com
- [Claude Code TXT Reader](https://myagilab.com/products/txt/)：把 Claude Code 导出的 TXT 变成可读的对话记录。在线地址 https://txt.myagilab.com
- [水文气象报告生成系统](https://myagilab.com/products/shuiwen/)：面向水文气象场景的 AI 报告生成。在线地址 https://shuiwen.myagilab.com
- [植医颐眠](https://myagilab.com/products/kangyang/)：植物医学与睡眠健康的 AI 助手。在线地址 https://kangyang.myagilab.com

## 开源项目

- [ChatTree](https://github.com/guoy0701/ChatTree)：AI 对话分支可视化浏览器插件
- [ClaudeTerminal](https://github.com/guoy0701/ClaudeTerminal)：受不了 Claude Code 的终端，自己写了一个
- [CodeDungeon](https://github.com/guoy0701/claude-code-game-CodeDungeon)：塞进 Claude Code 的地牢游戏，写代码就是闯关
- [claude-code-account-switch](https://github.com/guoy0701/claude-code-account-switch)：Claude Code 多账号切换工具
- [AnimaAgent](https://github.com/guoy0701/AnimaAgent)：Agent 的记忆与经验框架实验

## 文章（175 篇，每篇有同名 .md 纯文本版本）

### AI持续学习（Claude Code、Agent、工具链的实操与拆解）

- 2026-10-02 [我们研究了 claude.ai 的系统提示词](https://myagilab.com/articles/learn/2026-10-02-9b1508.md)
- 2026-09-28 [Qwen3.8-Flash-Next 和 Qwen3.8-27B 怎么配置：我们踩过的坑和实测对比](https://myagilab.com/articles/learn/2026-09-28-81cc49.md)
- 2026-09-22 [Vibe Coder 需要懂的一点 IDE 小知识](https://myagilab.com/articles/learn/2026-09-22-308561.md)
- 2026-09-19 [研究完 Jev，我们重新理解了大模型](https://myagilab.com/articles/learn/2026-09-19-ba2cf4.md)
- 2026-09-15 [一文讲透 WorkBuddy 和 CodeBuddy 到底有什么区别](https://myagilab.com/articles/learn/2026-09-15-9e2de1.md)
- 2026-09-13 [如何从零做一个自己的网页版 AI](https://myagilab.com/articles/learn/2026-09-13-3f3f52.md)
- 2026-09-05 [一文讲透 Claude Code 的子代理到底有几种](https://myagilab.com/articles/learn/2026-09-05-2288ea.md)
- 2026-09-02 [Fable 5.1 终于可以自由切换 effort 了](https://myagilab.com/articles/learn/2026-09-02-ac8baa.md)
- 2026-08-25 [Prompt Cache 最后的4朵小小乌云](https://myagilab.com/articles/learn/2026-08-25-5b9611.md)
- 2026-08-20 [深度解读：GLM创始人唐杰眼中的Scaling Law与大模型变强的真正逻辑](https://myagilab.com/articles/learn/2026-08-20-4d19cb.md)
- 2026-08-19 [Vibe Coder 需要懂的一点命令行小知识](https://myagilab.com/articles/learn/2026-08-19-1fa0b3.md)
- 2026-08-18 [Vibe Coder 的最后一块拼图](https://myagilab.com/articles/learn/2026-08-18-b8510c.md)
- 2026-08-17 [从对话到 Agent：缓存命中率的统一解释](https://myagilab.com/articles/learn/2026-08-17-852701.md)
- 2026-08-15 [关于 DeepSeek"缓存命中率 99%"，大部分人搞混了一件事](https://myagilab.com/articles/learn/2026-08-15-240283.md)
- 2026-08-14 [子代理没返回结果，我的Token白浪费了吗？](https://myagilab.com/articles/learn/2026-08-14-68b4b4.md)
- 2026-08-13 [大模型的脑子本身也是个Agent吗？](https://myagilab.com/articles/learn/2026-08-13-ee82da.md)
- 2026-08-12 [Claude给每句话都加了水印，到底是什么原理？](https://myagilab.com/articles/learn/2026-08-12-ba205b.md)
- 2026-08-10 [训练AI的考场，成了一门十亿美元的生意](https://myagilab.com/articles/learn/2026-08-10-b24bb0.md)
- 2026-08-08 [大模型天生就会投机取巧](https://myagilab.com/articles/learn/2026-08-08-79617a.md)
- 2026-08-07 [大模型的刻意练习](https://myagilab.com/articles/learn/2026-08-07-e9404b.md)
- 2026-08-03 [训练大模型太贵了，能不能用小模型"猜"出大模型的参数？](https://myagilab.com/articles/learn/2026-08-03-725e63.md)
- 2026-07-31 [一文讲透大模型对话的缓存计费机制（高阶版）](https://myagilab.com/articles/learn/2026-07-31-dcbcec.md)
- 2026-07-29 [同样一个 token，中文和英文在 AI 脑子里占的"空间"一样吗？](https://myagilab.com/articles/learn/2026-07-29-1e7422.md)
- 2026-07-28 [用文言文跟 AI 交流，真的省 token 吗？](https://myagilab.com/articles/learn/2026-07-28-272c03.md)
- 2026-07-27 [用中文问 AI 和用英文问 AI，效果一样吗？](https://myagilab.com/articles/learn/2026-07-27-7a1997.md)
- 2026-07-26 [一个关于 AI 训练数据的反直觉真相](https://myagilab.com/articles/learn/2026-07-26-c2b53b.md)
- 2026-07-25 [既然模型越大越强，为什么不直接做一个50万亿参数的模型？](https://myagilab.com/articles/learn/2026-07-25-1762e1.md)
- 2026-07-24 [对梁文锋发言的思考：如果可靠性不算智能的一部分，那到底什么才是智能？](https://myagilab.com/articles/learn/2026-07-24-441cbc.md)
- 2026-07-16 [从底层原理开始，把Git彻底讲明白](https://myagilab.com/articles/learn/2026-07-16-d3258b.md)
- 2026-07-12 [GPT-5.6 有三个模型，小的那个是大的"缩小版"吗？](https://myagilab.com/articles/learn/2026-07-12-698e7e.md)
- 2026-07-07 [别听他们瞎扯，Anthropic可从来没说Claude有意识了](https://myagilab.com/articles/learn/2026-07-07-ef5f68.md)
- 2026-07-06 [你对 Scaling 的力量一无所知](https://myagilab.com/articles/learn/2026-07-06-54a054.md)
- 2026-07-05 [AI的草稿纸被曝光，有人说它发明了自己的语言？](https://myagilab.com/articles/learn/2026-07-05-bf3d12.md)
- 2026-06-30 [研究了一圈，我终于搞清楚“Agent框架”这些概念](https://myagilab.com/articles/learn/2026-06-30-7b5aeb.md)
- 2026-06-29 [大模型只会吐字，那它到底是如何能调用工具的？](https://myagilab.com/articles/learn/2026-06-29-78767c.md)
- 2026-06-23 [AI Agent的路径，本质是自己发现的还是人类预设的？](https://myagilab.com/articles/learn/2026-06-23-538a8c.md)
- 2026-06-21 [大模型之外的 Harness，也分远近亲疏](https://myagilab.com/articles/learn/2026-06-21-789910.md)
- 2026-06-19 [之前的文章白写了？Claude Code 原来自带对话分叉](https://myagilab.com/articles/learn/2026-06-19-07f09a.md)
- 2026-06-02 [从Trae SOLO谈多智能体的困局](https://myagilab.com/articles/learn/2026-06-02-0b8cff.md)
- 2026-06-01 [别再只说"子Agent"了，Agent 之间的关系远比你想的复杂](https://myagilab.com/articles/learn/2026-06-01-ae97a5.md)
- 2026-05-27 [我颠覆了我的认知：OpenClaw 和 Hermes 还真的不一样](https://myagilab.com/articles/learn/2026-05-27-8f3b94.md)
- 2026-05-25 [正本清源说 Agent：OpenClaw 和 Hermes 到底有啥区别？](https://myagilab.com/articles/learn/2026-05-25-f1c26f.md)
- 2026-05-21 [大模型掷骰子吗？拆解 AI 输出的确定性边界](https://myagilab.com/articles/learn/2026-05-21-0be955.md)
- 2026-05-19 [大模型缓存机制完全指南：一篇讲透 KV Cache 到 Claude Code 的省钱工程](https://myagilab.com/articles/learn/2026-05-19-ff0773.md)
- 2026-05-17 [Claude Code的上下文对话想分叉怎么办？我有一个技巧](https://myagilab.com/articles/learn/2026-05-17-cf3084.md)
- 2026-05-10 [不要随便用子代理（subagent）模式啊](https://myagilab.com/articles/learn/2026-05-10-cbf70b.md)
- 2026-05-05 [AI Agent陷阱（七）| 你才是最终猎物](https://myagilab.com/articles/learn/2026-05-05-8c77f1.md)
- 2026-05-02 [AI Agent陷阱（六）| 当一万个AI同时做出"理性"的决定](https://myagilab.com/articles/learn/2026-05-02-12bf55.md)
- 2026-05-01 [AI Agent陷阱（五）| 你的AI助手开始为别人打工了](https://myagilab.com/articles/learn/2026-05-01-5f664e.md)
- 2026-04-27 [AI Agent陷阱（四）| AI记忆投毒有多容易](https://myagilab.com/articles/learn/2026-04-27-bb8262.md)
- 2026-04-26 [AI Agent陷阱（三）| 如何把你的Agent忽悠瘸了](https://myagilab.com/articles/learn/2026-04-26-a2bcb9.md)
- 2026-04-25 [AI Agent陷阱（二）| 你的Agent是怎么被"隐形文字"骗走的](https://myagilab.com/articles/learn/2026-04-25-bf3200.md)
- 2026-04-24 [OpenAI 创始人说"我们用 AI 的方式全错了"——他给出了什么方案？](https://myagilab.com/articles/learn/2026-04-24-aa41bb.md)
- 2026-04-23 [AI Agent陷阱（一）| 你的Agent正在被互联网攻击](https://myagilab.com/articles/learn/2026-04-23-9ef0d0.md)
- 2026-04-21 [一文讲清楚，闲鱼上 25 块的 ChatGPT Plus到底怎么来的](https://myagilab.com/articles/learn/2026-04-21-cb0364.md)
- 2026-04-15 [搞懂 Agent 架构，只需要想明白这 7 个问题](https://myagilab.com/articles/learn/2026-04-15-465d86.md)
- 2026-04-12 [如何把 Claude 的会员账号变成了 API使用？](https://myagilab.com/articles/learn/2026-04-12-e5cd7e.md)
- 2026-04-11 [Agent之间的沟通交流，需要新协议吗？](https://myagilab.com/articles/learn/2026-04-11-c0fc23.md)
- 2026-04-10 [从 0 到 1 打造一个 Agent 系统](https://myagilab.com/articles/learn/2026-04-10-da91cd.md)
- 2026-04-04 [如何做出苹果级演示视频，我试了 6 个工具](https://myagilab.com/articles/learn/2026-04-04-f1e7d6.md)
- 2026-03-23 [Claude Code 那些让人怀疑自己英语水平的单词，终于被我全破译了](https://myagilab.com/articles/learn/2026-03-23-6f0044.md)

### 对AI的思考（AI 时代的产业、组织与个人的第一性原理思考）

- 2026-10-03 [能不能用大模型来训练大模型？](https://myagilab.com/articles/thinking/2026-10-03-3343c2.md)
- 2026-09-30 [AI之大操纵术](https://myagilab.com/articles/thinking/2026-09-30-68dedf.md)
- 2026-09-27 [AI可能比人类更先出现无用阶级](https://myagilab.com/articles/thinking/2026-09-27-fd11d4.md)
- 2026-09-27 [图灵完备的门槛：人类的特殊性与 AI](https://myagilab.com/articles/thinking/2026-09-27-80aa52.md)
- 2026-09-25 [当我们再也看不懂更高级的 AI](https://myagilab.com/articles/thinking/2026-09-25-e7b136.md)
- 2026-09-24 [给足算力，AI 能无限聪明下去吗？](https://myagilab.com/articles/thinking/2026-09-24-9bb053.md)
- 2026-09-20 [如果 token 不用就过期，那赶紧用掉是否可以省 token？](https://myagilab.com/articles/thinking/2026-09-20-27a8b9.md)
- 2026-09-04 [Fable5.1太贵了，我不得不用子代理降本增效](https://myagilab.com/articles/thinking/2026-09-04-1b8e47.md)
- 2026-08-27 [编排不创造智能，Graph Engineering也是伪命题](https://myagilab.com/articles/thinking/2026-08-27-d8ff59.md)
- 2026-08-26 [当前的大模型已经杀死了循环工程](https://myagilab.com/articles/thinking/2026-08-26-4db38b.md)
- 2026-08-22 [智能的形状](https://myagilab.com/articles/thinking/2026-08-22-3047e2.md)
- 2026-08-21 [神的陨落：Fable 5竟然也会写bug](https://myagilab.com/articles/thinking/2026-08-21-48582a.md)
- 2026-08-11 [寒武纪大爆发，正在硅基世界重演](https://myagilab.com/articles/thinking/2026-08-11-8527bb.md)
- 2026-07-30 [在AI培训课上，我跟学员说：能不用AI就不用AI](https://myagilab.com/articles/thinking/2026-07-30-3c9ac4.md)
- 2026-07-13 [试错守恒定律（续）| 直接做事还是造机器？](https://myagilab.com/articles/thinking/2026-07-13-ba66a7.md)
- 2026-07-11 [当AI什么都能做的时候，人应该学什么](https://myagilab.com/articles/thinking/2026-07-11-f03bf0.md)
- 2026-07-09 [用大模型来开发，不等于要交付AI产品](https://myagilab.com/articles/thinking/2026-07-09-a6abb2.md)
- 2026-07-08 [根本不存在"专业Agent"这个东西](https://myagilab.com/articles/thinking/2026-07-08-2be220.md)
- 2026-07-04 [当AI有了子目标，人类就有了灭亡的风险](https://myagilab.com/articles/thinking/2026-07-04-de1454.md)
- 2026-07-03 [如何把 Fable 5 的每一个 token 都花在刀刃上](https://myagilab.com/articles/thinking/2026-07-03-e91c52.md)
- 2026-07-02 [AI的第一性原理 | AI产品的广义圣维南原理](https://myagilab.com/articles/thinking/2026-07-02-a7c817.md)
- 2026-07-01 [生产级开发中，不要无脑追求做Agent](https://myagilab.com/articles/thinking/2026-07-01-201968.md)
- 2026-06-28 [GPT 5.6也不行，ChatGPT已经比Claude落后一个段位了](https://myagilab.com/articles/thinking/2026-06-28-e0ff3f.md)
- 2026-06-27 [FDE是大模型改造世界的最后一公里](https://myagilab.com/articles/thinking/2026-06-27-77dffc.md)
- 2026-06-26 [什么才是不会被大模型发展所吞并的资产](https://myagilab.com/articles/thinking/2026-06-26-967faf.md)
- 2026-06-25 [AI来了，我推荐高考完的亲戚报考"天坑"专业生物学](https://myagilab.com/articles/thinking/2026-06-25-c5799f.md)
- 2026-06-24 [我的亲身经历告诉你，FDE工程师到底难在哪？](https://myagilab.com/articles/thinking/2026-06-24-0eddb8.md)
- 2026-06-22 [AI时代，"协作"可能是个坏习惯](https://myagilab.com/articles/thinking/2026-06-22-0949f5.md)
- 2026-06-20 [做技术和做开发是两回事，AI时代技术依然有价值，但开发危险了](https://myagilab.com/articles/thinking/2026-06-20-bafcc1.md)
- 2026-06-18 [当AI比人类更聪明，谁来告诉AI什么是"最好的"？](https://myagilab.com/articles/thinking/2026-06-18-8695e1.md)
- 2026-06-17 [脑洞：如果把OpenAI和Anthropic合并，AI能力的天花板在哪里？](https://myagilab.com/articles/thinking/2026-06-17-66594e.md)
- 2026-06-16 [AI什么都知道，为什么偏偏没有观点？](https://myagilab.com/articles/thinking/2026-06-16-572387.md)
- 2026-06-15 [用力学思维重新设计 AI 的约束机制](https://myagilab.com/articles/thinking/2026-06-15-5f3de1.md)
- 2026-06-14 [Fable 5被禁，GLM-5.2开源：智谱是在破局，还是被做局？](https://myagilab.com/articles/thinking/2026-06-14-67ebdc.md)
- 2026-06-13 [让 AI 不跑偏，可能是整个 AI 时代最难的工程问题](https://myagilab.com/articles/thinking/2026-06-13-690a82.md)
- 2026-06-12 [AI 的第一性原理 | 大模型的试错守恒定律](https://myagilab.com/articles/thinking/2026-06-12-7c14fc.md)
- 2026-06-11 [循环工程（Loop Engineering）到底在干什么？拆穿一个被过度包装的概念](https://myagilab.com/articles/thinking/2026-06-11-258a84.md)
- 2026-06-10 [Claude Fable 5 首发日，我替你们测了下它到底算不算硅基生命](https://myagilab.com/articles/thinking/2026-06-10-7cd502.md)
- 2026-06-09 [大模型生成文字的过程，像极了基因的表达](https://myagilab.com/articles/thinking/2026-06-09-971be1.md)
- 2026-06-08 ["你写的是AI文"是这个时代最侮辱人的话](https://myagilab.com/articles/thinking/2026-06-08-1b2fea.md)
- 2026-06-07 [科举1300年，高考74年：AI时代的高考还会是这个样子吗？](https://myagilab.com/articles/thinking/2026-06-07-6c2a8f.md)
- 2026-06-06 [老说AI没有人味，那到底什么才是人味啊？](https://myagilab.com/articles/thinking/2026-06-06-d1be23.md)
- 2026-06-05 [李飞飞写了篇世界模型分类，但世界模型还得先靠物理学家定义](https://myagilab.com/articles/thinking/2026-06-05-a0e292.md)
- 2026-06-04 [初中生都知道，大模型的数据还远没有用完](https://myagilab.com/articles/thinking/2026-06-04-70c4b2.md)
- 2026-06-03 [大模型的Token生成动力学漫谈](https://myagilab.com/articles/thinking/2026-06-03-cfe395.md)
- 2026-05-31 [如何训练一个有洞察力的大模型？](https://myagilab.com/articles/thinking/2026-05-31-7a05f3.md)
- 2026-05-30 [会编程的大模型千篇一律，有灵魂的大模型万里挑一](https://myagilab.com/articles/thinking/2026-05-30-631649.md)
- 2026-05-29 [Claude Opus 4.8 发布后，我发现更强的模型反而更没用了](https://myagilab.com/articles/thinking/2026-05-29-5b4554.md)
- 2026-05-28 [Helio 想让 AI 当你同事，这事现在能成功吗？](https://myagilab.com/articles/thinking/2026-05-28-0cc686.md)
- 2026-05-26 [朋友叫我谈谈对卡兹克的《AI时代的人才……》的看法，我又要得罪人了](https://myagilab.com/articles/thinking/2026-05-26-dcd99d.md)
- 2026-05-23 [AI时代，开发最难的部分不是写代码（以微信为例）](https://myagilab.com/articles/thinking/2026-05-23-85b773.md)
- 2026-05-22 [你的孩子编程得学，但不是原来那样的学](https://myagilab.com/articles/thinking/2026-05-22-7ea2d4.md)
- 2026-05-18 [每个时代，都有自己的周星驰](https://myagilab.com/articles/thinking/2026-05-18-1d761a.md)
- 2026-05-16 [大模型的原理完美诠释了什么是"道可道，非常道"](https://myagilab.com/articles/thinking/2026-05-16-069c0a.md)
- 2026-05-15 [AI时代，交易不会消失](https://myagilab.com/articles/thinking/2026-05-15-4c554a.md)
- 2026-05-14 [架构师已死](https://myagilab.com/articles/thinking/2026-05-14-2d98f7.md)
- 2026-05-12 [我们可能永远无法知道AI是否有意识](https://myagilab.com/articles/thinking/2026-05-12-c030c6.md)
- 2026-05-11 [别再学写代码了](https://myagilab.com/articles/thinking/2026-05-11-22fe25.md)
- 2026-05-09 [连AI味是什么都说不清，凭什么用它卡我论文？](https://myagilab.com/articles/thinking/2026-05-09-a1ff24.md)
- 2026-04-28 [【深度】未来的Agent该是什么样的？](https://myagilab.com/articles/thinking/2026-04-28-139dd8.md)
- 2026-04-20 [我们一直管它叫"语言大模型"，但它可能只是个"文字大模型"](https://myagilab.com/articles/thinking/2026-04-20-236bf8.md)
- 2026-04-19 [搭Agent系统踩的坑，人类组织100年前就踩过了](https://myagilab.com/articles/thinking/2026-04-19-9a03f6.md)
- 2026-04-18 [AI的第一性原理 | 上下文工程的尽头，是组织学](https://myagilab.com/articles/thinking/2026-04-18-98c758.md)
- 2026-04-16 [我和大模型对视的那一刻，突然怀疑：我自己是不是也是被"调用"的？](https://myagilab.com/articles/thinking/2026-04-16-dd2983.md)
- 2026-04-14 [我们可能都在用错的方式做 AI 产品](https://myagilab.com/articles/thinking/2026-04-14-7ce563.md)
- 2026-04-09 [Claude Mythos可能是历史上第一个超人工智能（ASI）](https://myagilab.com/articles/thinking/2026-04-09-4fd19e.md)
- 2026-04-02 [AI时代，最该创业的不是程序员](https://myagilab.com/articles/thinking/2026-04-02-965e39.md)
- 2026-04-01 [Claude Code源码泄露后，我从第一性原理重新思考了一个问题：代码为什么要解耦？](https://myagilab.com/articles/thinking/2026-04-01-2411ad.md)
- 2026-03-29 [软件正在死去——而大多数人还不知道](https://myagilab.com/articles/thinking/2026-03-29-888f9e.md)
- 2026-03-09 [OpenClaw | 技能，第一次可以和人分离了](https://myagilab.com/articles/thinking/2026-03-09-460966.md)

### 开发的部分产品（亲手做出来的工具与产品的诞生记）

- 2026-10-01 [ClaudeTerminal已经进化为多AI智能体协作平台](https://myagilab.com/articles/builds/2026-10-01-0933c6.md)
- 2026-09-29 [人人都是艺术大师](https://myagilab.com/articles/builds/2026-09-29-638ba7.md)
- 2026-09-23 [升级Opus 5.5后，我给专线网站打造了一个微型世界](https://myagilab.com/articles/builds/2026-09-23-80319b.md)
- 2026-09-21 [ClaudeTerminal 可以用手机远程操控了](https://myagilab.com/articles/builds/2026-09-21-6c2a29.md)
- 2026-09-17 [我的问答版Claude"中转站"上线了！](https://myagilab.com/articles/builds/2026-09-17-c8e31f.md)
- 2026-09-16 [让 Claude 派活儿给 WorkBuddy 里的模型](https://myagilab.com/articles/builds/2026-09-16-f33ef6.md)
- 2026-09-14 [我的个人网站更新了](https://myagilab.com/articles/builds/2026-09-14-6b1707.md)
- 2026-09-12 [我的 GPT-6 Astra 也可以调用 Claude 了](https://myagilab.com/articles/builds/2026-09-12-e11a5e.md)
- 2026-09-10 [我把两个Claude账号打通了](https://myagilab.com/articles/builds/2026-09-10-21cbc2.md)
- 2026-09-08 [Claude上天，额度无边](https://myagilab.com/articles/builds/2026-09-08-9b288b.md)
- 2026-09-07 [Claude Code 的 harness进化到多台电脑了](https://myagilab.com/articles/builds/2026-09-07-1d14f9.md)
- 2026-09-06 [现在可以让 Claude 替你上夜班了](https://myagilab.com/articles/builds/2026-09-06-25a2ae.md)
- 2026-08-24 [AI让使用电脑的门槛又进一步的降低](https://myagilab.com/articles/builds/2026-08-24-a8e103.md)
- 2026-07-20 [我受不了 Claude Code 的终端，于是自己写了一个](https://myagilab.com/articles/builds/2026-07-20-19c6b0.md)
- 2026-07-17 [朋友让我帮他做个 CAD 插件，我花了半天做完了](https://myagilab.com/articles/builds/2026-07-17-7b07ea.md)
- 2026-05-24 [用AI魔改了《三国群英传3》，这才是真实的世界规律](https://myagilab.com/articles/builds/2026-05-24-6a0124.md)
- 2026-05-13 [你和 AI 聊了一年，这些对话值多少钱？](https://myagilab.com/articles/builds/2026-05-13-3369d4.md)
- 2026-05-07 [想不到你浓眉大眼的ChatGPT也学坏了啊？幸好我有神器ChatTree](https://myagilab.com/articles/builds/2026-05-07-5b4828.md)
- 2026-05-06 [被 Claude Code 的导出文件逼疯后，我做了个小工具](https://myagilab.com/articles/builds/2026-05-06-3e3327.md)
- 2026-05-03 [【Anima 创世纪】Agent 的记忆和经验，完全不是一回事](https://myagilab.com/articles/builds/2026-05-03-4487b0.md)
- 2026-04-22 [AI神器 | ChatTree支持国产大模型了——此外我发现90%的人根本没理解"分支"是什么](https://myagilab.com/articles/builds/2026-04-22-71b7e6.md)
- 2026-04-17 [ChatTree支持claude.ai了——以及为什么Claude更需要多分支聊天](https://myagilab.com/articles/builds/2026-04-17-d51eef.md)
- 2026-04-13 [你和 AI 的对话，从来就不该是线性的](https://myagilab.com/articles/builds/2026-04-13-7ce73b.md)
- 2026-04-08 [我在 Claude Code 里塞了一个地牢游戏，现在写代码就是在闯关](https://myagilab.com/articles/builds/2026-04-08-ddcd0b.md)
- 2026-04-07 [我开发了个Claude Code切换账号的工具，妈妈再也不用担心我token限量而焦虑了](https://myagilab.com/articles/builds/2026-04-07-157125.md)
- 2026-03-22 [我做了一个10个AI开会帮你做决策的系统，朋友试完直呼牛逼](https://myagilab.com/articles/builds/2026-03-22-f5fa60.md)

### 扒底裤系列（把热点项目和自媒体叙事拆开看本质）

- 2026-09-03 [Anthropic 的 Max 20x 误导确实无耻，但自媒体给的证据全是错的](https://myagilab.com/articles/debunk/2026-09-03-5d12f9.md)
- 2026-07-26 [经过研究我才发现，梁文锋为什么开源DeepSeek，跟你想的完全不一样](https://myagilab.com/articles/debunk/2026-07-26-1826d0.md)
- 2026-07-15 [我拆解了SoftwareCopyright-Skill 这个软著申请的开源项目，告诉你它到底值不值](https://myagilab.com/articles/debunk/2026-07-15-7906ab.md)
- 2026-05-20 [号称"能搜索谷歌搜不到的80%互联网"的 AnySearch，我扒完代码发现新智元越来越不要脸了](https://myagilab.com/articles/debunk/2026-05-20-9d93ff.md)
- 2026-05-12 [数字生命卡兹克的AI十级体系，我越读越难受](https://myagilab.com/articles/debunk/2026-05-12-46bc10.md)
- 2026-05-08 [扒底裤系列 | 分析完号称"给 AI 装上大脑" 的 GBrain 代码，发现它确实挺强的](https://myagilab.com/articles/debunk/2026-05-08-50d8c8.md)
- 2026-04-30 [号称 "OpenClaw + Hermes 完美合体"的 Mercury，我扒了它的底裤都不剩](https://myagilab.com/articles/debunk/2026-04-30-c79a8c.md)
- 2026-04-29 [GitHub 的热榜项目，我来给它们扒一扒底裤](https://myagilab.com/articles/debunk/2026-04-29-accdab.md)
- 2026-04-09 [我蒸馏了一下"蒸馏"，发现里面没什么东西](https://myagilab.com/articles/debunk/2026-04-09-0ec679.md)
- 2026-04-06 [AI靠打工7小时狂赚1万刀？呸，自媒体别制造焦虑了！](https://myagilab.com/articles/debunk/2026-04-06-ce444d.md)

### 历史进程（AI 革命的时间线与关键节点）

- 2026-09-18 [我 + Fable 5.1 = 最强运维工程师 #260918](https://myagilab.com/articles/history/2026-09-18-2bb5bb.md)
- 2026-08-06 [ChatGPT的聊天不占用Codex的额度#260806](https://myagilab.com/articles/history/2026-08-06-207966.md)
- 2026-08-05 [升级OpenClaw到GPT 5.6 Sol，优化了部署流程#260805](https://myagilab.com/articles/history/2026-08-05-3ed6f0.md)
- 2026-08-04 [今天尝试用Codex写代码#260804](https://myagilab.com/articles/history/2026-08-04-560591.md)
- 2026-07-14 [连续日更100天，聊聊我的大模型之路](https://myagilab.com/articles/history/2026-07-14-369c21.md)
- 2026-07-10 [AI革命简史：ChatGPT5.6挽救了AI革命](https://myagilab.com/articles/history/2026-07-10-b501c2.md)
- 2026-05-04 [烧掉的每一个 Token，都在烧出一个未来的你](https://myagilab.com/articles/history/2026-05-04-eb54f0.md)
- 2026-03-04 [小龙虾进化日记：玩了 3 天 OpenClaw，我把它从"聊天玩具"改造成了一个能自己干活的系统](https://myagilab.com/articles/history/2026-03-04-55c23d.md)