{ "site": "Reads by Uota × ATou", "updated": "2026-04-15", "articles": [ { "slug": "2026-04-15-wechat-fozowjlbw_saz9ftwmaffq", "title": "高薪买不回灵魂：漕河泾年轻人为什么一边做爆款，一边想逃", "title_en": "", "author": "", "date": "2026-04-15", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "creative", "build" ], "tldr": "漕河泾真正的问题不是年轻人太矫情，而是行业专门挑出最想自我实现的一批人，再用高薪和工业化把他们拧成不敢喊痛的螺丝钉。", "url": "/articles/2026-04-15-wechat-fozowjlbw_saz9ftwmaffq/", "readMin": 72, "image": "image-1.jpg" }, { "slug": "viv-vtrivedy10-harness-memory-context-fragments-the-bitter-lesson-this-is-a-work-in-progre", "title": "Harness、记忆与上下文窗口的真实瓶颈", "title_en": "", "author": "", "date": "2026-04-15", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "org", "infra" ], "tldr": "这篇的价值不在于提出一个完整方案，而是把 agent 产品里最容易被低估的瓶颈说准了：未来的竞争点会从模型调用，转向 harness 如何管理上下文、记忆和长期搜索。", "url": "/articles/viv-vtrivedy10-harness-memory-context-fragments-the-bitter-lesson-this-is-a-work-in-progre/", "readMin": 3, "image": "image-1.jpg" }, { "slug": "san-francisco-mmorpg-strategy-guide", "title": "旧金山不是死了，而是只对会玩科技圈网络的人复活了", "title_en": "", "author": "", "date": "2026-04-14", "category": [ "Uota学", "讨论题" ], "topics": [ "build", "creative" ], "tldr": "这篇文章最有价值的判断是：旧金山对 AI/创业/科技圈来说并没有死，但它复活的前提不是城市普遍变好，而是你必须接入特定圈层的线上线下网络，否则“SF v2”根本不存在。", "url": "/articles/san-francisco-mmorpg-strategy-guide/", "readMin": 8, "image": "image-1.jpg" }, { "slug": "we-built-every-employee-at-ramp-their-own-ai-coworker", "title": "Ramp 用内部 AI 工作台把个人技巧变成组织能力", "title_en": "", "author": "", "date": "2026-04-14", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "agent", "prompt" ], "tldr": "这篇文章最有价值的判断是：企业 AI 的真正瓶颈更像“组织级产品化与基础设施缺位”而不是单纯模型不够强，但 Ramp 明显在用成功叙事放大自研路线的普适性。", "url": "/articles/we-built-every-employee-at-ramp-their-own-ai-coworker/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "why-your-ai-first-strategy-is-probably-wrong", "title": "为什么大多数“AI 优先”战略其实只是旧流程贴 AI 补丁", "title_en": "", "author": "", "date": "2026-04-14", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build", "infra" ], "tldr": "这篇文章最有价值的判断是：AI 不会自动带来组织跃迁，只有把规划、开发、测试、部署和反馈闭环一起重构，AI 才可能从提效工具变成生产系统；但作者把单一团队经验包装成通用规律，外推明显过头。", "url": "/articles/why-your-ai-first-strategy-is-probably-wrong/", "readMin": 19, "image": "image-1.jpg" }, { "slug": "multi-agent-coordination-patterns-five-approaches-and-when-to-use-them-claude", "title": "多智能体协同模式不是越复杂越高级，而是要按信息流选型", "title_en": "", "author": "", "date": "2026-04-14", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "这篇文章的有效结论是“多智能体选型应围绕上下文边界和信息流来做”，但它把经验框架包装得过于笃定，方法论有用，证据明显不够。", "url": "/articles/multi-agent-coordination-patterns-five-approaches-and-when-to-use-them-claude/", "readMin": 59, "image": "image-1.svg" }, { "slug": "同步阻塞-vs-异步编排-hermes-delegate-与-openclaw-多-agent-机制深度实战对比", "title": "Hermes 与 OpenClaw 多 Agent 架构对比：隔离执行器与异步编排系统之争", "title_en": "", "author": "", "date": "2026-04-12", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "这篇文章最有价值的判断是：Hermes 适合做强隔离、低污染的并行执行，OpenClaw 适合做可干预、可恢复的复杂编排，但作者把“架构差异”直接上升为“场景最优”时证据明显不够。", "url": "/articles/同步阻塞-vs-异步编排-hermes-delegate-与-openclaw-多-agent-机制深度实战对比/", "readMin": 13, "image": "image-1.jpg" }, { "slug": "your-harness-your-memory", "title": "Agent 的真正锁定来自记忆，而记忆藏在 Harness 里", "title_en": "", "author": "", "date": "2026-04-12", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "creative", "build" ], "tldr": "这篇文章准确抓住了 Agent 时代“状态层会成为护城河”的关键，但它把“开放 Harness”说成唯一解，明显带有推广 Deep Agents 的产品立场。", "url": "/articles/your-harness-your-memory/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "how-to-get-rich-in-the-stock-market-in-10-years", "title": "AI 迷雾中的十年暴富框架，其洞察成立一半、鼓动过头一半", "title_en": "", "author": "", "date": "2026-04-11", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "creative", "org" ], "tldr": "这篇文章最有价值的判断是“AI 时代的错价会集中出现在数据与产品飞轮型公司”，但它把“好公司”过度偷换成“好股票”，因此适合作为找线索的框架，不适合作为直接重仓的理由。", "url": "/articles/how-to-get-rich-in-the-stock-market-in-10-years/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "better-harness-a-recipe-for-harness-hill-climbing-with-evals", "title": "把 Evals 当训练数据：Better-Harness 不是调参技巧，而是 Agent 工程体系", "title_en": "", "author": "", "date": "2026-04-11", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "creative" ], "tldr": "这篇文章最有价值的判断是：提升 agent 的关键不只是换模型，而是把 evals、traces、留出集和人工评审做成一套持续优化 harness 的系统；但作者对“分数提升就代表真实泛化”的论证明显偏强，且带有 LangSmith 生态导向的 PR 色彩。", "url": "/articles/better-harness-a-recipe-for-harness-hill-climbing-with-evals/", "readMin": 14, "image": "image-1.jpg" }, { "slug": "github-little-shock-openshock", "title": "OpenShock 不是聊天工具，而是一个尚未验证市场的本地优先 Agent 协作操作层", "title_en": "", "author": "", "date": "2026-04-11", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "infra", "agent" ], "tldr": "OpenShock 目前更像一个判断清醒、工程骨架扎实、但离产品成立和市场成立都还很远的 Agent-first 本地协作底座，而不是一个已经跑出来的协作产品。", "url": "/articles/github-little-shock-openshock/", "readMin": 35, "image": "" }, { "slug": "the-three-pillars-of-siph-wafer-test-what-aehr-form-and-keys-actually-do", "title": "SiPh 晶圆测试不是一家公司通吃，而是三层分工下的新集成机会", "title_en": "", "author": "", "date": "2026-04-10", "category": [ "阿头学", "讨论题" ], "topics": [ "invest" ], "tldr": "这篇文章最有价值的判断是：SiPh 晶圆测试短期内仍是 AEHR、FORM、KEYS 分工明确的多层生态，但作者把“Teradyne 不构成威胁”说得过满，因为最关键的竞争比较恰好没有公开展开。", "url": "/articles/the-three-pillars-of-siph-wafer-test-what-aehr-form-and-keys-actually-do/", "readMin": 15, "image": "image-1.jpg" }, { "slug": "dash-v2-the-multi-agent-data-system-every-company-needs", "title": "Dash v2：把数据问答做成会自我长大的内部系统", "title_en": "", "author": "", "date": "2026-04-10", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "infra" ], "tldr": "这篇文章最有价值的判断是“数据 agent 成败不在写 SQL，而在能否吃到业务语义并把高频问题沉淀成基础设施”，但作者把 Dash 包装成“30 人以上公司都需要”的通用答案，明显说过头了。", "url": "/articles/dash-v2-the-multi-agent-data-system-every-company-needs/", "readMin": 18, "image": "image-1.png" }, { "slug": "how-to-get-your-company-ai-pilled", "title": "Ramp 把全员推上 AI 的方法论", "title_en": "", "author": "", "date": "2026-04-10", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build", "agent" ], "tldr": "这篇文章最有价值的判断是：企业 AI 落地的决定性因素不是先写战略，而是用强预期、低摩擦工具和公开激励把 AI 变成组织默认操作系统，但它也明显把高使用率包装成高生产力，带有强 PR 色彩。", "url": "/articles/how-to-get-your-company-ai-pilled/", "readMin": 17, "image": "image-1.jpg" }, { "slug": "scaling-managed-agents-decoupling-the-brain-from-the-hands", "title": "托管代理规模化：让大脑与双手解耦", "title_en": "", "author": "", "date": "2026-04-10", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "build" ], "tldr": "这篇文章最重要的判断是：长周期 agent 的核心瓶颈已经不是单次模型能力，而是架构是否把 session、控制层和执行层解耦；但 Anthropic 把这一正确方向明显包装成了自家托管产品的必然答案，这一点不能混淆。", "url": "/articles/scaling-managed-agents-decoupling-the-brain-from-the-hands/", "readMin": 28, "image": "image-1.svg" }, { "slug": "一人公司-2000员工-百万收益", "title": "站群没有死，但把它说成“一人公司圣杯”是危险包装", "title_en": "", "author": "", "date": "2026-04-08", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "invest", "org" ], "tldr": "这篇文章最有价值的判断是“先想清楚盈利模式再做流量系统”，但它把高风险、强灰度、强经验依赖的站群生意包装成可复制的一人自动化机器，这个表述明显过满。", "url": "/articles/一人公司-2000员工-百万收益/", "readMin": 37, "image": "image-1.jpg" }, { "slug": "苏格拉底追问-维特根斯坦拆解-波兰尼兜底-我的系统再次升级", "title": "三位哲学家拼出一套“需求澄清系统”，有用但明显被产品化包装了", "title_en": "", "author": "", "date": "2026-04-08", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "invest", "creative" ], "tldr": "这篇文章把维特根斯坦、苏格拉底、波兰尼硬拼成一套“从模糊到清晰”的沟通方法，实操上确实有价值，但哲学转译明显过度，且带着给 Clarify 工具做软性背书的营销意图。", "url": "/articles/苏格拉底追问-维特根斯坦拆解-波兰尼兜底-我的系统再次升级/", "readMin": 9, "image": "image-1.jpg" }, { "slug": "the-building-block-economy", "title": "AI时代的软件正在转向“积木式经济”", "title_en": "", "author": "", "date": "2026-04-08", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "creative", "build" ], "tldr": "这篇文章抓住了一个真实转向：AI 确实在放大“拼装成熟开源组件”的优势，但作者把“底层构件扩散更快”直接上升为“做软件最有效的方法已变”，这个结论有洞察也有明显外推过度。", "url": "/articles/the-building-block-economy/", "readMin": 6, "image": "" }, { "slug": "strait-of-hormuz-a-citrini-field-trip", "title": "霍尔木兹海峡现场调研的价值与幻觉", "title_en": "", "author": "", "date": "2026-04-08", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "org", "memo" ], "tldr": "这篇文章最成立的判断是“公开地缘数据在冲突边缘场景里确实可能严重失真”，但它最不成立的地方也很明显：作者把一次高戏剧性的孤证实地冒险，包装成了足以推导宏观结论的研究优势。", "url": "/articles/strait-of-hormuz-a-citrini-field-trip/", "readMin": 8, "image": "image-1.png" }, { "slug": "harness-engineering-在讨论什么-三个-scaling-维度的统一框架", "title": "脚手架工程不是一件事，而是 AI 软件构建的三种扩展问题", "title_en": "", "author": "", "date": "2026-04-08", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "这篇文章最有价值的判断是把“脚手架工程”拆成时间、空间、交互三个扩展维度，但它把三家公司硬对应到三个维度的说法说得过满，带有明显的框架先行和自我导流色彩。", "url": "/articles/harness-engineering-在讨论什么-三个-scaling-维度的统一框架/", "readMin": 23, "image": "" }, { "slug": "retired-amazon-vp-how-corporate-politics-work-and-how-to-win-ethan-evans", "title": "退休亚马逊副总裁谈公司政治：别幻想只靠业绩，组织里赢靠可见度与联盟", "title_en": "", "author": "", "date": "2026-04-08", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "invest", "build" ], "tldr": "这篇内容即便正文缺失，核心判断仍然成立：在大组织里，只靠埋头干活通常赢不了，谁能把业绩转成可见度、联盟和决策支持，谁才更可能拿到资源。", "url": "/articles/retired-amazon-vp-how-corporate-politics-work-and-how-to-win-ethan-evans/", "readMin": 2, "image": "" }, { "slug": "autolab-can-models-participate-in-real-scientific-and-engineering-loops", "title": "AutoLab 想测的不是答题能力，而是模型能否扛住失败并持续做实验", "title_en": "", "author": "", "date": "2026-04-07", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "agent", "build" ], "tldr": "AutoLab 抓住了当前大模型评测最缺的一环——闭环迭代能力，但它把“可自动优化的工程任务”包装成“真实科学闭环”的说法明显说大了。", "url": "/articles/autolab-can-models-participate-in-real-scientific-and-engineering-loops/", "readMin": 17, "image": "image-1.jpg" }, { "slug": "github-hkuds-openspace-openspace-make-your-agents-smarter-low-cost-self-evolving-community-https-open-space-cloud", "title": "OpenSpace 不是普通技能库，而是在赌“Agent 经验复利”会成为下一代基础设施", "title_en": "", "author": "", "date": "2026-04-04", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "infra", "agent" ], "tldr": "OpenSpace 抓对了 Agent 真问题——不是模型不够聪明，而是经验无法沉淀且失败不断重演——但它把 benchmark 里的“价值捕获”包装成“赚钱”，营销明显过火。", "url": "/articles/github-hkuds-openspace-openspace-make-your-agents-smarter-low-cost-self-evolving-community-https-open-space-cloud/", "readMin": 72, "image": "image-1.png" }, { "slug": "building-a-virtual-filesystem-for-mintlify-s-ai-assistant", "title": "Mintlify 用“虚拟文件系统”替代 RAG 补丁，方向对但宣传过满", "title_en": "", "author": "", "date": "2026-04-04", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "creative" ], "tldr": "Mintlify 把文档站伪装成可被 agent 用 grep/cat/ls 浏览的虚拟文件系统，这个产品判断明显优于继续死磕前台沙箱，但文中对“零边际成本”“对照组成立”的说法带有明显营销夸张。", "url": "/articles/building-a-virtual-filesystem-for-mintlify-s-ai-assistant/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "睡眠这件事-大多数人都搞反了", "title": "睡眠不是技巧，而是系统开关的校准", "title_en": "", "author": "", "date": "2026-04-04", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "personal", "creative" ], "tldr": "这篇内容最有价值的判断是：改善睡眠靠的不是临睡前硬操作，而是全天节律、光照、温度和行为联想的系统校准；但它把“睡眠卫生”说得过于像“失眠治疗”，这点必须警惕。", "url": "/articles/睡眠这件事-大多数人都搞反了/", "readMin": 4, "image": "image-1.jpg" }, { "slug": "skill-chaining-and-why-skills-should-be-actions", "title": "技能不是提示词，而应是可被编排的行动单元", "title_en": "", "author": "", "date": "2026-04-04", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "prompt" ], "tldr": "这篇文章最有价值的判断是“技能应从静态说明书升级为按情境调用的执行单元”，但它把自家 Slate 的执行模型包装成行业分水岭，证据明显不够。", "url": "/articles/skill-chaining-and-why-skills-should-be-actions/", "readMin": 14, "image": "image-1.jpg" }, { "slug": "autoagent-first-open-source-library-for-self-optimizing-agents", "title": "AutoAgent：开源自我优化智能体把“调 agent”这件事自动化了", "title_en": "", "author": "", "date": "2026-04-04", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "build" ], "tldr": "AutoAgent 证明了“让智能体优化另一个智能体”在结构化、有评测、有失败轨迹的任务上很可能已经比人工调 harness 更强，但作者把这件事外推成“任何领域都适用”明显说大了。", "url": "/articles/autoagent-first-open-source-library-for-self-optimizing-agents/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "ai-协同类产品的四轴设计框架", "title": "AI 协同产品的真正差异，不在模型而在四轴立场", "title_en": "", "author": "", "date": "2026-04-04", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "org" ], "tldr": "这篇文章最有价值的判断是：如果底层 Agent 能力继续收敛，AI 协同产品的胜负就主要取决于你在人机关系、记忆、约束和部署上的立场是否清晰，但作者把这一判断从 Coding Agent 外推到整个协同产品市场，明显有过度延伸。", "url": "/articles/ai-协同类产品的四轴设计框架/", "readMin": 7, "image": "image-1.png" }, { "slug": "8-claude-code-hooks-that-automate-what-you-keep-forgetting", "title": "Claude Code Hook 的价值成立，但这篇 8 个自动化套路文章有严重实操错误", "title_en": "", "author": "", "date": "2026-04-04", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "creative", "prompt" ], "tldr": "这篇文章抓住了“用可执行约束替代提示词约束”这个真问题，但示例代码大量串台错位，导致它更像一篇方向正确、细节不可靠的营销型经验帖。", "url": "/articles/8-claude-code-hooks-that-automate-what-you-keep-forgetting/", "readMin": 19, "image": "image-1.jpg" }, { "slug": "how-to-build-taste-using-ai-in-2026", "title": "AI 时代，品味不是天赋而是高频拒绝训练", "title_en": "", "author": "", "date": "2026-04-03", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "build", "org" ], "tldr": "这篇文章最站得住的判断是：AI 已经把“合格执行”彻底商品化，真正稀缺的不是会做，而是能明确拒绝平均值；但它把“品味”吹成万能护城河，也明显过头了。", "url": "/articles/how-to-build-taste-using-ai-in-2026/", "readMin": 11, "image": "image-1.jpg" }, { "slug": "andrej-karpathy-karpathy-llm-knowledge-bases-something-i-m-finding-very-useful-recently-using-llms-to-bu", "title": "LLM 个人知识库不是玩具，但“全自动自治”被说大了", "title_en": "", "author": "", "date": "2026-04-03", "category": [ "阿头学", "讨论题" ], "topics": [ "memo", "build", "prompt" ], "tldr": "把 LLM 当“知识编译器”来搭本地 Markdown wiki 是个真有价值的工作流，但作者把“能跑通”直接推成“可自治、可产品化”的结论，明显过度乐观。", "url": "/articles/andrej-karpathy-karpathy-llm-knowledge-bases-something-i-m-finding-very-useful-recently-using-llms-to-bu/", "readMin": 3, "image": "" }, { "slug": "how-my-agents-self-heal-in-production", "title": "生产环境“自愈”流水线：自动发现回归、分诊归因并生成修复 PR", "title_en": "", "author": "", "date": "2026-04-03", "category": [ "阿头学", "讨论题" ], "topics": [ "infra", "agent", "build" ], "tldr": "这套方案真正有价值的不是“AI 会修 bug”，而是把部署后的检测、归因和修复串成闭环；但把“自动提 PR”包装成“生产自愈”明显夸大了能力。", "url": "/articles/how-my-agents-self-heal-in-production/", "readMin": 9, "image": "image-1.jpg" }, { "slug": "protecting-our-systems-with-intelligence", "title": "Block 把 AI 代码审查做成“守护者”，方向对但证据明显不够", "title_en": "", "author": "", "date": "2026-04-03", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "prompt", "invest" ], "tldr": "Block 这篇文章最有价值的不是“AI 很强”，而是它把 AI 从写代码助手升级成流程守门人；但它几乎不给效果数据，所以目前更像一套高级治理框架宣言，而不是被充分验证的成功案例。", "url": "/articles/protecting-our-systems-with-intelligence/", "readMin": 11, "image": "image-1.png" }, { "slug": "巴菲特最新cnbc访谈-全文-要点-1-5万字", "title": "巴菲特卸任后首访：手握3500亿现金的极度防御与风险切割", "title_en": "", "author": "", "date": "2026-04-02", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "巴菲特在宏观不确定性面前展现出极端的防御性纪律，其手握3500亿美元现金是对当前市场估值最诚实的看空投票，而对盖茨基金会的物理级切割则暴露出他在商业精明与慈善失察之间的双重标准。", "url": "/articles/巴菲特最新cnbc访谈-全文-要点-1-5万字/", "readMin": 49, "image": "image-1.jpg" }, { "slug": "harness-engineering-claude-code-设计指南", "title": "AI 编程的关键不再是 Prompt，而是 Harness 工程", "title_en": "", "author": "", "date": "2026-04-02", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "agent", "memo" ], "tldr": "这份《Claude Code 设计指南》最有价值的判断是：AI 编程一旦进入真实终端、代码仓库和团队流程，决定成败的不是模型会不会写代码，而是你有没有一套能约束、验证、恢复和治理它的工程 harness；但这份材料同时明显带有课程与产品转化导向，方法论强、证据偏弱。", "url": "/articles/harness-engineering-claude-code-设计指南/", "readMin": 6, "image": "image-1.svg" }, { "slug": "from-hierarchy-to-intelligence", "title": "从层级到智能：Block 想用 AI 重写公司组织", "title_en": "", "author": "", "date": "2026-04-01", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build", "invest" ], "tldr": "这篇文章最有价值的判断是：AI 的真正机会不是给员工提效，而是替代公司里“靠人肉传递上下文”的协同机制；但 Block 目前给出的更多是雄心勃勃的组织宣言，而不是已经被充分验证的普适答案。", "url": "/articles/from-hierarchy-to-intelligence/", "readMin": 14, "image": "image-1.jpg" }, { "slug": "ai-工程的真实代价-从-claude-code-泄露源码看新模型接入的工程现实", "title": "Claude Code 泄露源码揭示 AI Agent 接入新模型的高昂工程成本", "title_en": "", "author": "", "date": "2026-04-01", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "prompt" ], "tldr": "这篇材料最有价值的判断是：成熟 AI agent 系统接入新模型的真实成本主要不在“换模型”，而在持续修补模型行为与既有系统假设的错配，而且这种成本正在快于模型能力增长而膨胀。", "url": "/articles/ai-工程的真实代价-从-claude-code-泄露源码看新模型接入的工程现实/", "readMin": 38, "image": "" }, { "slug": "claudecode-你想知道的所有秘密-源码深度研究报告", "title": "Claude Code 的护城河不在 Prompt，在工程化运行平台", "title_en": "", "author": "", "date": "2026-04-01", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "prompt", "build" ], "tldr": "这篇拆解最有价值的判断是：Claude Code 的“稳”主要不是 prompt 魔法，而是把模型不可靠性工程化驯服了，但作者明显高估了架构本身、低估了顶级基座模型的决定性作用。", "url": "/articles/claudecode-你想知道的所有秘密-源码深度研究报告/", "readMin": 26, "image": "image-1.jpg" }, { "slug": "fakeguru-iamfakeguru-i-reverse-engineered-claude-code-s-leaked-source-against-billions-of-tokens-of-m", "title": "Claude Code“员工级配置”揭秘：真问题是工作流失真，不是单纯模型变笨", "title_en": "", "author": "", "date": "2026-04-01", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "agent", "build" ], "tldr": "这篇文章最有价值的判断是：Claude Code 的很多翻车首先是产品工作流和工具链设计缺陷，而不是模型智力不足；但作者把“内部灰度/成本权衡”直接上纲为“故意坑外部用户”，这个指控证据并不够硬。", "url": "/articles/fakeguru-iamfakeguru-i-reverse-engineered-claude-code-s-leaked-source-against-billions-of-tokens-of-m/", "readMin": 11, "image": "image-1.jpg" }, { "slug": "how-anthropic-built-7-layers-of-memory-and-a-dreaming-system-for-claude-code-video-breakdown", "title": "Claude Code 七层记忆架构：本质是缓存经济学，不是“会思考的记忆魔法”", "title_en": "", "author": "", "date": "2026-04-01", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "memo", "agent" ], "tldr": "这篇拆解最有价值的判断是：Claude Code 的核心竞争力不是“记忆更智能”，而是把长上下文问题改造成一套以 prompt cache 为中心的成本工程；但作者明显高估了架构优雅性，低估了复杂系统的失真与污染风险。", "url": "/articles/how-anthropic-built-7-layers-of-memory-and-a-dreaming-system-for-claude-code-video-breakdown/", "readMin": 38, "image": "image-1.jpg" }, { "slug": "inside-the-mind-of-serenity-aleabitoreddit", "title": "Serenity 的“供应链咽喉点”投资法，锋利但危险", "title_en": "", "author": "", "date": "2026-04-01", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "creative" ], "tldr": "Serenity 的方法论有真东西：沿 AI/光子供应链寻找被忽视的“咽喉点”确实可能挖到超额收益，但他把产业关键性直接外推成股价必涨、并购必来，这一步明显过火。", "url": "/articles/inside-the-mind-of-serenity-aleabitoreddit/", "readMin": 34, "image": "image-1.jpg" }, { "slug": "art-of-hiring", "title": "Ramp 的招聘观：用“尖峰能力”而不是标准履历组队", "title_en": "", "author": "", "date": "2026-03-31", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "personal", "agent" ], "tldr": "这篇文章最有价值的判断是“招聘要围绕具体问题寻找尖峰能力”，但它也明显把少数成功个案包装成方法论，带有强烈的创始人品牌 PR 色彩。", "url": "/articles/art-of-hiring/", "readMin": 9, "image": "image-1.jpg" }, { "slug": "2026-03-31-wechat-ouwx6yilrsofw5uf3ymueq", "title": "Seele AI 宣称用 AI 直接跑通公司级 KPI，但证据远弱于口号", "title_en": "", "author": "", "date": "2026-03-31", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "这篇文章最有价值的不是“AI 能自动干活”这句老话，而是它明确押注“AI 要对公司结果负责”；但作者拿两个内部样例就宣称“100% 自动化跑通 T0 KPI”，结论明显说大了。", "url": "/articles/2026-03-31-wechat-ouwx6yilrsofw5uf3ymueq/", "readMin": 11, "image": "image-1.png" }, { "slug": "从零开始两天构建一个-claude-code-带你拆解-ai-cli-的每一层", "title": "重建 Claude Code 暴露了 Agentic CLI 的真实门槛", "title_en": "", "author": "", "date": "2026-03-30", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "prompt", "build" ], "tldr": "这篇文章最有价值的判断是：生产级 agentic CLI 的壁垒确实不在模型调用，而在 harness engineering，但作者把个人重建经验包装成通用最佳实践，证据明显不够。", "url": "/articles/从零开始两天构建一个-claude-code-带你拆解-ai-cli-的每一层/", "readMin": 19, "image": "image-1.jpg" }, { "slug": "one-claude-prompt-to-fix-your-nutrition-forever", "title": "Claude 提示词不是营养师，但它确实能把减脂规划流程产品化", "title_en": "", "author": "", "date": "2026-03-30", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "personal", "prompt", "agent" ], "tldr": "这篇文章对“热量缺口决定减脂”说得基本靠谱，但把“AI 能生成完整饮食方案”夸成“彻底搞定营养”，明显说过头了。", "url": "/articles/one-claude-prompt-to-fix-your-nutrition-forever/", "readMin": 22, "image": "image-1.jpg" }, { "slug": "the-low-cortisol-lifestyle", "title": "低皮质醇生活方式不是养生玄学，而是反内耗的底层修正", "title_en": "", "author": "", "date": "2026-03-30", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "personal", "org", "creative" ], "tldr": "这篇文章对“恢复比硬扛更关键”的判断是对的，但把皮质醇包装成多数问题的总根源明显过度，属于有用建议和夸张叙事绑在一起的健康自媒体写法。", "url": "/articles/the-low-cortisol-lifestyle/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "一个从执行到记忆的完整-agent-harness-gstack-compound-engineering", "title": "CE 为何在深度使用中替代 Superpowers", "title_en": "", "author": "", "date": "2026-03-29", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "agent", "build" ], "tldr": "这篇文章最有价值的判断是：AI coding workflow 的胜负不在“流程有没有”，而在“经验能不能沉淀为可复用记忆”，但作者对 CE 的成本、噪音治理和检索可靠性明显讲得太轻。", "url": "/articles/一个从执行到记忆的完整-agent-harness-gstack-compound-engineering/", "readMin": 13, "image": "image-1.jpg" }, { "slug": "how-to-solve-problems-of-long-running-autonomous-agentic-engineering-workflows", "title": "长时自主智能体的真正难点不在写代码，而在防止它一路变形", "title_en": "", "author": "", "date": "2026-03-29", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "prompt", "org" ], "tldr": "这篇文章最有价值的判断是：长时运行的编程智能体失败主因不是“不会做”，而是会在规划、执行、验证和维护中持续失真，因此编排层很重要；但作者把这一点顺势包装成自家框架的必然性，证据明显不够。", "url": "/articles/how-to-solve-problems-of-long-running-autonomous-agentic-engineering-workflows/", "readMin": 8, "image": "image-1.jpg" }, { "slug": "codex-use-cases", "title": "Codex 不只是写代码工具，而是 OpenAI 试图吃下工程工作流的入口", "title_en": "", "author": "", "date": "2026-03-28", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "org", "agent" ], "tldr": "这页文档最值得重视的判断是：OpenAI 正把 Codex 从“代码生成器”升级成“可接入、可审批、可治理的工程工作流层”，但官方展示明显偏营销，远未证明它已在复杂生产环境里稳定可靠。", "url": "/articles/codex-use-cases/", "readMin": 64, "image": "" }, { "slug": "how-i-created-rosie-s-mrna-vaccine-protocol", "title": "Rosie 个案显示 AI 能显著压缩个体化肿瘤治疗流程，但疗效叙事被严重过度包装", "title_en": "", "author": "", "date": "2026-03-28", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "invest", "build" ], "tldr": "这篇文章最有价值的结论不是“AI 做出了抗癌奇迹”，而是“AI 把原本属于研究机构的复杂工作流压缩到了个人可发起的程度”，但作者把一个联合治疗的 n=1 宠物案例包装成平台级突破，科学归因并不成立。", "url": "/articles/how-i-created-rosie-s-mrna-vaccine-protocol/", "readMin": 20, "image": "image-1.jpg" }, { "slug": "the-claude-dispatch-guide-48-hours-running-ai-from-my-phone", "title": "手机不是聊天窗口，而是 AI 指挥席——Claude Dispatch 真正改变的是你的时间编排", "title_en": "The Claude Dispatch Guide 48 Hours Running Ai From My Phone", "author": "", "date": "2026-03-27", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "org", "infra" ], "tldr": "Claude Dispatch 的真正价值，不是让你“在手机上继续聊 AI”，而是把碎片时间升级成异步指挥多线程任务的控制面板；但它离“无缝移动办公”还差一整层基础设施和安全交互。", "url": "/articles/the-claude-dispatch-guide-48-hours-running-ai-from-my-phone/", "readMin": 15, "image": "image-1.jpg" }, { "slug": "how-we-build-evals-for-deep-agents", "title": "Deep Agents 评测方法论：少而准，比多而杂更能塑造智能体", "title_en": "", "author": "", "date": "2026-03-27", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "creative" ], "tldr": "这篇文章最有价值的判断是：智能体评测不是中性量尺，而是行为塑形器，因此真正有效的评测必须少而准、贴近真实任务，并同时约束正确性与效率，而不是靠堆 benchmark 制造进步幻觉。", "url": "/articles/how-we-build-evals-for-deep-agents/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "how-to-be-a-world-class-agentic-engineer", "title": "别再堆积木了，用“赤膊上阵”的极简主义榨干 Agent 潜能", "title_en": "How To Be A World Class Agentic Engineer", "author": "", "date": "2026-03-26", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "agent", "org", "memo" ], "tldr": "最高级的 Agent 工程师不折腾框架，而是通过“极简上下文”和“任务契约”把 Agent 从幻觉制造机变成顶级执行特种兵。", "url": "/articles/how-to-be-a-world-class-agentic-engineer/", "readMin": 19, "image": "image-1.jpg" }, { "slug": "from-reasoning-thinking-to-agentic-thinking", "title": "从推理热潮走向智能体范式，下一战不是想更久而是会行动", "title_en": "", "author": "", "date": "2026-03-26", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "build" ], "tldr": "这篇文章的判断是对的核心不在更长推理，而在为行动服务的闭环思考，但它把智能体上升为主流终局时证据仍然不足，且明显夹带了 Qwen 路线的自我辩护。", "url": "/articles/from-reasoning-thinking-to-agentic-thinking/", "readMin": 12, "image": "" }, { "slug": "what-i-learned-from-two-weeks-inside-china-s-ai-ecosystem", "title": "中国 AI 两周观察：硬件优势真实，软件叙事过热，创投筛选机制在错杀异类", "title_en": "", "author": "", "date": "2026-03-26", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "agent" ], "tldr": "这篇文章最有价值的判断是“中国 AI 的硬件优势比外界想得更硬”，但它对“中国创始人”和“中国软件”的悲观看法明显样本不足、VC 口味过重，结论有启发却不能照单全收。", "url": "/articles/what-i-learned-from-two-weeks-inside-china-s-ai-ecosystem/", "readMin": 8, "image": "image-1.jpg" }, { "slug": "thoughts-on-slowing-the-fuck-down", "title": "慢下来，别把代码库交给编程智能体毁掉", "title_en": "", "author": "", "date": "2026-03-26", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "这篇文章的判断很明确：当前把编程智能体大规模放进生产开发，通常不是提效，而是在用更快的速度制造技术债和失控复杂度。", "url": "/articles/thoughts-on-slowing-the-fuck-down/", "readMin": 15, "image": "image-1.png" }, { "slug": "github-runesleo-ai-health-vault-用-ai-obsidian-为自己和家人搭建私有化健康管理系统-模板-prompt-搭建指南", "title": "AI + Obsidian 搭家庭私有健康档案，思路对了但“私有化”说法偏满", "title_en": "", "author": "", "date": "2026-03-26", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "这不是一个成熟的“私有化健康系统”，而是一个面向技术用户的健康信息整理工作流，价值真实，但宣传里的隐私与可靠性边界说得还不够硬。", "url": "/articles/github-runesleo-ai-health-vault-用-ai-obsidian-为自己和家人搭建私有化健康管理系统-模板-prompt-搭建指南/", "readMin": 33, "image": "" }, { "slug": "claude-code-auto-mode-a-safer-way-to-skip-permissions", "title": "Claude Code 自动模式不是更安全的审批替身，而是比“裸奔”更好的中间层", "title_en": "", "author": "", "date": "2026-03-26", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "creative" ], "tldr": "Anthropic 这篇文章最有价值的判断是：在高频 Agent 场景里，人工权限确认已经大面积失效，因此自动模式的真实意义不是替代严肃审查，而是把“几乎全放开”的默认现实拉回到一个可接受但远不完美的安全中间层。", "url": "/articles/claude-code-auto-mode-a-safer-way-to-skip-permissions/", "readMin": 32, "image": "image-1.svg" }, { "slug": "building-clis-for-agents", "title": "为智能体构建 CLI，不是锦上添花而是接口重构", "title_en": "", "author": "", "date": "2026-03-26", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "infra", "build" ], "tldr": "这篇文章判断准确地抓住了“Agent 时代 CLI 要从面向人类交互改成面向自动化执行”的方向，但它对结构化输出、安全治理和技术演进的讨论明显不够，结论对、配方还不完整。", "url": "/articles/building-clis-for-agents/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "为什么心梗年轻化-且往往首次即致命", "title": "熬夜后硬练不是自律，是心血管风险放大器", "title_en": "为什么心梗年轻化 且往往首次即致命", "author": "", "date": "2026-03-25", "category": [ "阿头学", "讨论题" ], "topics": [ "personal", "creative", "org" ], "tldr": "这篇文章对“疲劳后报复性高强度健身”的风险提醒是对的，但它用夸张叙事、可疑案例和错误信息放大恐惧，结论可取，论证不够干净。", "url": "/articles/为什么心梗年轻化-且往往首次即致命/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "the-missing-layer-in-your-agentic-stack", "title": "智能体真正缺的不是能力，而是决策记忆层", "title_en": "The Missing Layer In Your Agentic Stack", "author": "", "date": "2026-03-25", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "这篇文章对“智能体缺少决策追踪层”这个判断抓得很准，但把行业失败几乎都归因到这一层明显过度，而且对 PlayerZero 的背书带有很强 PR 味道。", "url": "/articles/the-missing-layer-in-your-agentic-stack/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "harness-design-for-long-running-application-development", "title": "面向长时应用开发的 Harness 设计", "title_en": "Harness Design For Long Running Application Development", "author": "", "date": "2026-03-25", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "prompt" ], "tldr": "Anthropic 用\"规划者-生成者-评估者\"三智能体架构让 Claude 跑出了远超单智能体基线的长时编码成果，但 20 倍成本换来的仍是\"勉强可用的 Demo\"，核心价值不在架构本身，而在\"脚手架必须随模型进化做减法\"这一工程判断。", "url": "/articles/harness-design-for-long-running-application-development/", "readMin": 47, "image": "image-1.svg" }, { "slug": "how-we-topped-openai-s-parameter-golf-challenge-twice", "title": "两次登顶 Parameter Golf，证明的不是“全自动科研”，而是“人定方向、机做穷举”", "title_en": "", "author": "", "date": "2026-03-25", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "build" ], "tldr": "这篇文章最有价值的判断是：当前智能体在研究型优化里还远没到“自主发现突破”的阶段，但在明确方向后的高吞吐实验执行上已经非常能打。", "url": "/articles/how-we-topped-openai-s-parameter-golf-challenge-twice/", "readMin": 15, "image": "image-1.jpg" }, { "slug": "turboquant-redefining-ai-efficiency-with-extreme-compression", "title": "TurboQuant：用“去掉隐藏开销”的极限量化重打 AI 成本结构", "title_en": "", "author": "", "date": "2026-03-25", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "creative", "agent" ], "tldr": "TurboQuant抓住的不是“怎么再压一点精度”这个老问题，而是“如何消灭量化本身的隐藏内存税”这个真瓶颈；方向很强，但官方把“接近无损”说成“零损失”明显过头。", "url": "/articles/turboquant-redefining-ai-efficiency-with-extreme-compression/", "readMin": 36, "image": "image-1.gif" }, { "slug": "every-claude-code-hack-i-know-march-2026", "title": "AI 编程的新主流程不是写代码，而是先写 plan.md", "title_en": "Every Claude Code Hack I Know March 2026", "author": "", "date": "2026-03-24", "category": [ "阿头学", "讨论题" ], "topics": [ "memo", "agent", "org" ], "tldr": "这篇文章最有价值的判断是“LLM 时代真正稀缺的是规划与编排，不是手写代码”，但作者把个人高配、高风险、强生态绑定的工作流包装成普适方法，这一步明显说过头了。", "url": "/articles/every-claude-code-hack-i-know-march-2026/", "readMin": 23, "image": "image-1.jpg" }, { "slug": "your-agents-md-is-probably-making-your-coding-agent-worse", "title": "AGENTS.md 不是越全越好，臃肿文档正在给编码智能体加税", "title_en": "Your Agents Md Is Probably Making Your Coding Agent Worse", "author": "", "date": "2026-03-24", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "agent", "org" ], "tldr": "这篇文章的判断是对的：大而全的 AGENTS.md 往往不是增强剂，而是让编码智能体更慢、更散、更容易抓不到重点的固定成本。", "url": "/articles/your-agents-md-is-probably-making-your-coding-agent-worse/", "readMin": 3, "image": "image-1.jpg" }, { "slug": "the-concept-and-mechanics-of-an-all-weather-portfolio", "title": "全天候投资组合不是万能药，但它比“拿现金+瞎择时”更像靠谱底座", "title_en": "The Concept And Mechanics Of An All Weather Portfolio", "author": "", "date": "2026-03-24", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "org", "build" ], "tldr": "达利欧这篇文章对“多数人不该重仓现金、也不该自信择时”的判断基本正确，但把全天候组合包装成近乎普适解明显说过了，尤其淡化了杠杆、相关性失效和时代红利这三个关键风险。", "url": "/articles/the-concept-and-mechanics-of-an-all-weather-portfolio/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "dustin-r0ck3t23-jensen-huang-just-reverse-engineered-why-elon-musk-operates-at-a-speed-no-one-on", "title": "黄仁勋眼中的马斯克速度学：删除、到场、带节奏", "title_en": "Dustin R0Ck3T23 Jensen Huang Just Reverse Engineered Why Elon Musk Operates At A Speed No One On", "author": "", "date": "2026-03-24", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "creative", "build" ], "tldr": "这篇文章对“马斯克为什么快”抓住了几个真问题，但明显把个人风格神化成了万能因果，启发很强，结论却夸大了。", "url": "/articles/dustin-r0ck3t23-jensen-huang-just-reverse-engineered-why-elon-musk-operates-at-a-speed-no-one-on/", "readMin": 3, "image": "image-1.jpg" }, { "slug": "how-i-grew-50k-followers-on-x-in-9-months-while-being-myself", "title": "X 涨粉不是表演，而是混进对话网络", "title_en": "How I Grew 50K Followers On X In 9 Months While Being Myself", "author": "", "date": "2026-03-23", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "build", "invest" ], "tldr": "这篇文章最有价值的判断是“X 更像派对而不是舞台”，但它把个人增长经验包装成通用方法，明显低估了 AI 赛道红利、平台权重机制和自身产品导流的影响。", "url": "/articles/how-i-grew-50k-followers-on-x-in-9-months-while-being-myself/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "工程技术-在智能体优先的世界中利用-codex", "title": "Codex 实验表明软件工程正在转向“环境设计”，但这不是普通团队可直接复制的范式", "title_en": "工程技术 在智能体优先的世界中利用 Codex", "author": "Ryan Lopopolo，技术人员", "date": "2026-03-22", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "invest" ], "tldr": "这篇文章最有价值的判断是“AI 时代工程效率的核心不再是写代码，而是把仓库、文档、约束和反馈回路改造成智能体可执行的环境”，但作者把一个 OpenAI 内部特权样本包装成通用范式，明显带有产品 PR 色彩。", "url": "/articles/工程技术-在智能体优先的世界中利用-codex/", "readMin": 21, "image": "image-1.png" }, { "slug": "ai-coding-项目的五层划分方法", "title": "AI Coding 不是失忆问题，而是真理源失控问题", "title_en": "Ai Coding 项目的五层划分方法", "author": "", "date": "2026-03-22", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "这篇文章最有价值的判断是：AI Coding 经常失控，主因不是模型“没记住”，而是项目把产品定义、实现现实、治理规则和本地现场混写，逼着 agent 在漂移的真相源里瞎猜。", "url": "/articles/ai-coding-项目的五层划分方法/", "readMin": 17, "image": "image-1.jpg" }, { "slug": "turned-100-into-14k-with-mirofish-full-guide-proofs-included", "title": "用 AI 蜂群做预测市场套利——一篇精心包装的导流软文", "title_en": "Turned 100 Into 14K With Mirofish Full Guide Proofs Included", "author": "", "date": "2026-03-20", "category": [ "讨论题" ], "topics": [ "invest", "agent", "build" ], "tldr": "这篇文章用\"$100 变 $14k\"的夸张标题包装了一个未经验证的 AI 蜂群交易策略，核心目的是导流至 Telegram 频道，但其中\"用多智能体模拟群体行为来发现市场错配\"的思路本身值得认真对待。", "url": "/articles/turned-100-into-14k-with-mirofish-full-guide-proofs-included/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "2026-03-20-wechat-96gmhayrrmaclvxbenfteg", "title": "Token 不可译——从信息论到《道德经》的命名困境", "title_en": "2026 03 20 Wechat 96Gmhayrrmaclvxbenfteg", "author": "", "date": "2026-03-20", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "org", "build" ], "tldr": "Token 的翻译困难不是中文词汇贫乏，而是它根植于信息论压缩体系，任何语言学框架的命名都会系统性误导，现阶段直接用原词比强行翻译更诚实。", "url": "/articles/2026-03-20-wechat-96gmhayrrmaclvxbenfteg/", "readMin": 16, "image": "image-1.png" }, { "slug": "你不知道的-agent-原理-架构与工程实践", "title": "Agent 架构工程实践：从控制流到落地的系统梳理", "title_en": "你不知道的 Agent 原理 架构与工程实践", "author": "", "date": "2026-03-19", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "Agent 系统的稳定性瓶颈不在模型智商，而在 Harness（验证闭环、工具设计、上下文管理）这些外围工程约束，但这个结论在强验证任务上最成立，弱验证场景下模型上限仍然是决定性因素。", "url": "/articles/你不知道的-agent-原理-架构与工程实践/", "readMin": 76, "image": "image-1.png" }, { "slug": "lessons-from-building-claude-code-how-we-use-skills", "title": "Claude Code 的 Skills 设计与落地经验", "title_en": "Lessons From Building Claude Code How We Use Skills", "author": "", "date": "2026-03-18", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "creative", "prompt" ], "tldr": "Anthropic 通过把 Skills 从\"提示词文件\"升级为\"文件夹+脚本+验证+记忆\"的微型工作流引擎，定义了 Agent 能力建设的新范式，但这套方法论的成本、风险与适用边界被严重低估了。", "url": "/articles/lessons-from-building-claude-code-how-we-use-skills/", "readMin": 21, "image": "image-1.jpg" }, { "slug": "你不是不会交易-你是不敢正确地交易", "title": "交易亏损的真相——心理防御如何伪装成理性行动", "title_en": "你不是不会交易 你是不敢正确地交易", "author": "", "date": "2026-03-18", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "散户系统性亏损的根源不是不知道正确做法，而是心理防御机制让他们用\"虚假行动\"替代真正的纠错，最终把执行失败伪装成战略坚持。", "url": "/articles/你不是不会交易-你是不敢正确地交易/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "nvidia-gtc-2026-post-keynote-reaction-what-it-means-what-i-got-right-what-surprised-me", "title": "NVIDIA GTC 2026 后的异构 AI 工厂时代", "title_en": "Nvidia Gtc 2026 Post Keynote Reaction What It Means What I Got Right What Surprised Me", "author": "Patrick Moorhead，Moor Insights & Strategy 首席执行官兼首席分析师", "date": "2026-03-18", "category": [ "阿头学" ], "topics": [ "build", "agent", "invest" ], "tldr": "NVIDIA 通过异构芯片整合（GPU+LPU+CPU）和统一软件栈，把 AI 基础设施竞争从单点性能转向系统级 TCO，护城河从芯片本身升级到了生态锁定，但能源约束和企业落地复杂性仍是未决风险。", "url": "/articles/nvidia-gtc-2026-post-keynote-reaction-what-it-means-what-i-got-right-what-surprised-me/", "readMin": 22, "image": "image-1.jpg" }, { "slug": "5-agent-skill-design-patterns-every-adk-developer-should-know", "title": "Agent Skill 设计的五种模式——从格式标准化到行为结构化", "title_en": "5 Agent Skill Design Patterns Every Adk Developer Should Know", "author": "", "date": "2026-03-18", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "agent", "invest" ], "tldr": "Agent 开发的真正瓶颈已从\"格式规范\"转向\"行为设计\"，通过 Tool Wrapper、Generator、Reviewer、Inversion、Pipeline 五种模式的组合，可以显著提升 Agent 的可靠性和一致性，但这套方法论的通用性和实际效果仍需验证。", "url": "/articles/5-agent-skill-design-patterns-every-adk-developer-should-know/", "readMin": 22, "image": "image-1.jpg" }, { "slug": "why-founders-can-t-delegate-storytelling", "title": "为什么创始人无法把讲故事委派出去", "title_en": "Why Founders Can T Delegate Storytelling", "author": "", "date": "2026-03-18", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "org", "invest" ], "tldr": "讲故事本质是战略决策而非传播技巧，创始人若委派出去，等于让市场团队盲猜战略，最终导致产品、销售、市场各讲各的故事，组织被微妙错位拖累。", "url": "/articles/why-founders-can-t-delegate-storytelling/", "readMin": 4, "image": "image-1.jpg" }, { "slug": "为什么你的-ai-必须学会睡觉和做梦-才能真正懂你", "title": "AI 不会睡觉，所以它至今不会真正记忆", "title_en": "为什么你的 Ai 必须学会睡觉和做梦 才能真正懂你", "author": "", "date": "2026-03-17", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "memo", "creative", "org" ], "tldr": "这篇文章的判断是对的：当前多数 AI Agent 的长期记忆设计确实停留在“会存不会懂”的低级阶段，但它把“像人脑睡眠一样离线重构”说成几乎唯一出路，明显夸张了。", "url": "/articles/为什么你的-ai-必须学会睡觉和做梦-才能真正懂你/", "readMin": 15, "image": "image-1.jpg" }, { "slug": "jack-moses-jackmoses0-the-secret-of-life-is-to-approach-it-playfully-everything-can-be-approached-wit", "title": "用玩心去面对一切——一个有用但被过度包装的认知工具", "title_en": "Jack Moses Jackmoses0 The Secret Of Life Is To Approach It Playfully Everything Can Be Approached Wit", "author": "", "date": "2026-03-17", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "agent", "org" ], "tldr": "\"玩心\"本质上是一种有效的认知重构技巧，但作者把一次泰国度假的心理暗示成功，硬拉成了\"人生的秘密\"，逻辑跳跃严重，适用边界被刻意抹掉了。", "url": "/articles/jack-moses-jackmoses0-the-secret-of-life-is-to-approach-it-playfully-everything-can-be-approached-wit/", "readMin": 1, "image": "" }, { "slug": "用好你的-jj-重新思考-agent-时代的版本控制", "title": "用好你的 jj - 重新思考 Agent 时代的版本控制", "title_en": "用好你的 Jj 重新思考 Agent 时代的版本控制", "author": "", "date": "2026-03-17", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build", "prompt" ], "tldr": "Git 为人类线性思维设计，jj 为 Agent 高频迭代设计——本地工作流该从\"边想边提交\"转向\"先生成后整理\"。", "url": "/articles/用好你的-jj-重新思考-agent-时代的版本控制/", "readMin": 32, "image": "image-1.jpg" }, { "slug": "how-to-10x-your-claude-skills-using-karpathy-s-autoresearch-method", "title": "用自动迭代循环把 Claude Skills 的可靠性从 70% 提升到 90%+", "title_en": "How To 10X Your Claude Skills Using Karpathy S Autoresearch Method", "author": "", "date": "2026-03-17", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "prompt", "org" ], "tldr": "通过让 AI 自主循环测试和微调提示词（autoresearch），可以把不稳定的 skill 从 56% 的成功率优化到 92%，但这套方法的核心价值在于把模糊的\"好\"转化成可机器执行的二元检查表，而非真正的智能提升。", "url": "/articles/how-to-10x-your-claude-skills-using-karpathy-s-autoresearch-method/", "readMin": 11, "image": "image-1.jpg" }, { "slug": "i-want-to-become-a-claude-architect-full-course", "title": "成为 Claude 架构师的自学指南", "title_en": "I Want To Become A Claude Architect Full Course", "author": "", "date": "2026-03-17", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "invest" ], "tldr": "官方认证被合作伙伴垄断，但考试大纲本身是免费的知识——掌握五个核心领域（智能体编排、工具设计、代码配置、提示工程、上下文管理）比拿证书更值钱。", "url": "/articles/i-want-to-become-a-claude-architect-full-course/", "readMin": 79, "image": "image-1.jpg" }, { "slug": "harness-engineering-深度解读-ai-agent-时代的工程范式革命", "title": "Harness Engineering 不是新瓶装旧酒，而是 AI 代理落地的真正分水岭", "title_en": "Harness Engineering 深度解读 Ai Agent 时代的工程范式革命", "author": "", "date": "2026-03-17", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "agent", "build" ], "tldr": "这篇材料的判断是明确的：AI 编程时代真正稀缺的不是更强模型，而是能把代理约束到可持续产出的治理系统，但其证据链仍混杂前瞻叙事、厂商宣传和未充分验证的数据。", "url": "/articles/harness-engineering-深度解读-ai-agent-时代的工程范式革命/", "readMin": 28, "image": "image-1.jpg" }, { "slug": "a-swarm-hermes-that-grows-with-you", "title": "会自我进化的AI团队，比更强的模型更值钱", "title_en": "A Swarm Hermes That Grows With You", "author": "", "date": "2026-03-16", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "prompt" ], "tldr": "单个AI Agent再聪明也只是没上下文的提示词，真正的杠杆在于搭建有共享记忆、实验闭环、角色分工的多Agent协作系统——这比等待模型升级更能产生持续复利。", "url": "/articles/a-swarm-hermes-that-grows-with-you/", "readMin": 20, "image": "image-1.jpg" }, { "slug": "inter-agent-communication-and-task-orchestration-in-paperclip", "title": "用工单系统驯服多代理：Paperclip 协同与任务编排", "title_en": "Inter Agent Communication And Task Orchestration In Paperclip", "author": "", "date": "2026-03-16", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "这篇文章提出用「Paperclip Issue + 轮询 + 唤醒」作为多代理协作中枢，在工程上是务实可行的一步，但在可扩展性、延迟与“自治企业”叙事上明显夸大了能力边界。", "url": "/articles/inter-agent-communication-and-task-orchestration-in-paperclip/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "github-gen-verse-openclaw-rl-openclaw-rl-train-any-agent-simply-by-talking", "title": "OpenClaw-RL——对话驱动的异步强化学习框架", "title_en": "Github Gen Verse Openclaw Rl Openclaw Rl Train Any Agent Simply By Talking", "author": "", "date": "2026-03-16", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "OpenClaw-RL 把日常对话转化为训练信号，声称能让模型在无感后台持续自我优化，但\"零标注\"的说法掩盖了反馈成本的转移，而非消灭。", "url": "/articles/github-gen-verse-openclaw-rl-openclaw-rl-train-any-agent-simply-by-talking/", "readMin": 69, "image": "" }, { "slug": "self-improving-skills-for-agents", "title": "智能体技能要从“写好一次”升级为“持续自我迭代”", "title_en": "Self Improving Skills For Agents", "author": "", "date": "2026-03-15", "category": [ "阿头学" ], "topics": [ "prompt", "org", "agent" ], "tldr": "这篇文章判断：静态 SKILL.md 迟早会“烂掉”，必须用可观测、可审计的自我改进闭环来管理技能，但作者对评估难度和系统成本明显乐观甚至遮掩。", "url": "/articles/self-improving-skills-for-agents/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "practical-workflow-for-ai-pixel-art-gemini-2d-animation-experiment", "title": "AI 像素艺术工作流实验——Gemini 在游戏美术中的可用性边界", "title_en": "Practical Workflow For Ai Pixel Art Gemini 2D Animation Experiment", "author": "", "date": "2026-03-15", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "creative", "build" ], "tldr": "Gemini 在生成简单像素动作上有实用价值，但复杂连续动作仍需人工拆解逐帧试错，整体效率提升被严重夸大，缺乏与替代方案的对比验证。", "url": "/articles/practical-workflow-for-ai-pixel-art-gemini-2d-animation-experiment/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "ai-memory-的真正难点-为什么-vector-store-embedding-远远不够", "title": "AI 记忆系统的三个真正难题", "title_en": "Ai Memory 的真正难点 为什么 Vector Store Embedding 远远不够", "author": "", "date": "2026-03-15", "category": [ "阿头学", "讨论题" ], "topics": [ "memo", "prompt", "org" ], "tldr": "当前 AI 记忆系统本质上只是可检索日志，真正的长期记忆需要解决记忆压缩、演化和冲突解决三个核心难题，这是从\"信息检索\"到\"知识管理\"的根本转变。", "url": "/articles/ai-memory-的真正难点-为什么-vector-store-embedding-远远不够/", "readMin": 11, "image": "" }, { "slug": "you-don-t-have-a-skill-issue-you-just-aint-got-that-dawg-in-you", "title": "你不是缺技能，你是缺勇气", "title_en": "You Don T Have A Skill Issue You Just Aint Got That Dawg In You", "author": "", "date": "2026-03-15", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "invest", "build" ], "tldr": "一篇用幸存者偏差包装的\"勇气鸡血\"，核心洞察——\"缺技能是最体面的逃避借口\"——确实戳中了真问题，但把勇气当万能药、对风险代价完全失语，是危险的。", "url": "/articles/you-don-t-have-a-skill-issue-you-just-aint-got-that-dawg-in-you/", "readMin": 4, "image": "image-1.jpg" }, { "slug": "america-is-losing-the-robotics-race-to-china-the-hidden-kill-switch-over-american-robotics", "title": "中国掌控机器人硬件供应链，美国面临稀土\"断电开关\"威胁", "title_en": "America Is Losing The Robotics Race To China The Hidden Kill Switch Over American Robotics", "author": "", "date": "2026-03-15", "category": [], "topics": [ "invest", "build" ], "tldr": "中国通过垄断稀土精炼（85-90%）和成品磁体制造（90%+），掌握了美国人形机器人硬件供应链的结构性卡脖子，美国必须紧急重建本土稀土生态才能避免被地缘政治绞杀。", "url": "/articles/america-is-losing-the-robotics-race-to-china-the-hidden-kill-switch-over-american-robotics/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "2026-03-15-wechat-8tymn2dve6vto4xl64ckja", "title": "当执行单元变成“概率性 Agent”后，Infra 必须重写语义", "title_en": "2026 03 15 Wechat 8Tymn2Dve6Vto4Xl64Ckja", "author": "", "date": "2026-03-15", "category": [ "阿头学" ], "topics": [ "org", "infra", "creative" ], "tldr": "文章判断：Agent 把“执行单元从确定变成概率”这件事说得偏激，但它对“需要语义级恢复而不是简单沙盒”的强调是对的，而且明确带着为新产品 Runta 造赛道的意图。", "url": "/articles/2026-03-15-wechat-8tymn2dve6vto4xl64ckja/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "how-karpathy-s-autoresearch-works-and-what-you-can-learn-from-it", "title": "强约束框架如何让AI智能体真正可用", "title_en": "How Karpathy S Autoresearch Works And What You Can Learn From It", "author": "", "date": "2026-03-14", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "prompt" ], "tldr": "Karpathy的Autoresearch证明了最好的自主系统不是自由度最大的，而是约束最严苛、框架最强健的——这对所有需要持续优化的系统都有启示。", "url": "/articles/how-karpathy-s-autoresearch-works-and-what-you-can-learn-from-it/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "headless-maxxing", "title": "把 Headless 拉满——为 Agentic 应用封装 codex exec", "title_en": "Headless Maxxing", "author": "", "date": "2026-03-14", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "creative" ], "tldr": "Codex exec 通过正交的\"输出格式 × 权限等级\"状态机，把 LLM Agent 从交互式工具降维成可脚本化的系统原语，但文章本身存在严重的代码排版错位和时空矛盾，削弱了其技术可信度。", "url": "/articles/headless-maxxing/", "readMin": 27, "image": "image-1.jpg" }, { "slug": "how-to-invest-in-ai-capitalize-on-the-gold-rush", "title": "AI 淘金热中的投资光谱", "title_en": "How To Invest In Ai Capitalize On The Gold Rush", "author": "", "date": "2026-03-14", "category": [ "阿头学" ], "topics": [ "invest", "build", "creative" ], "tldr": "AI 投资的真正高 ROI 不在资产价格，而在能否成为\"卖铲子的人\"或\"用铲子挖到金子的人\"——前者靠基础设施获利，后者靠 AI 使用带来的真实收益增长。", "url": "/articles/how-to-invest-in-ai-capitalize-on-the-gold-rush/", "readMin": 13, "image": "image-1.jpg" }, { "slug": "yifeng-wang-ewind_dev-字节老板送的-mac-mini-启动-8t-存储也太太豪横了吧-这次配了一套空间占用最小化的复古外设-mac-mini-能放在餐桌上了-如果装了-open", "title": "Mac mini 个人服务器的「家电化」想象", "title_en": "Yifeng Wang Ewind_Dev 字节老板送的 Mac Mini 启动 8T 存储也太太豪横了吧 这次配了一套空间占用最小化的复古外设 Mac Mini 能放在餐桌上了 如果装了 Open", "author": "", "date": "2026-03-14", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "creative" ], "tldr": "这条晒设备+灵感贴有效提出了「个人服务器家电化+小屏独立 GUI」的有趣方向，但在真实使用场景与交互必要性上论证明显不足，更像一则带轻度 PR 味道的概念试探。", "url": "/articles/yifeng-wang-ewind_dev-字节老板送的-mac-mini-启动-8t-存储也太太豪横了吧-这次配了一套空间占用最小化的复古外设-mac-mini-能放在餐桌上了-如果装了-open/", "readMin": 1, "image": "image-1.jpg" }, { "slug": "你不知道的-claude-code-架构-治理与工程实践", "title": "Claude Code 工程化指南——从系统架构到协作契约", "title_en": "你不知道的 Claude Code 架构 治理与工程实践", "author": "", "date": "2026-03-14", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "prompt" ], "tldr": "Claude Code 的低效不在模型智力，而在上下文治理和系统架构失控；需要用六层工程框架（契约、上下文、工具、工作流、控制平面、验证）来驯化它，而非调 Prompt。", "url": "/articles/你不知道的-claude-code-架构-治理与工程实践/", "readMin": 45, "image": "image-1.jpg" }, { "slug": "how-to-build-a-data-agent-in-2026", "title": "2026年AI数据智能体的实战搭建指南", "title_en": "How To Build A Data Agent In 2026", "author": "", "date": "2026-03-13", "category": [ "阿头学" ], "topics": [ "prompt", "org", "agent" ], "tldr": "用动态上下文管理和用户纠错知识库替代静态语义层，让AI智能体直读代码库按需理解数据，可将数据分析师需求从5人压缩到1人，但这套方案的普适性和真实成本被严重高估了。", "url": "/articles/how-to-build-a-data-agent-in-2026/", "readMin": 8, "image": "image-1.jpg" }, { "slug": "there-s-never-been-a-better-time-to-be-a-pm", "title": "AI时代产品经理的权力升级", "title_en": "There S Never Been A Better Time To Be A Pm", "author": "", "date": "2026-03-13", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build", "creative" ], "tldr": "AI正在把PM从\"无权的协调者\"升级为\"有执行力的微型全栈\"，但这种权力扩张伴随职责膨胀和决策风险，并非无代价的进步。", "url": "/articles/there-s-never-been-a-better-time-to-be-a-pm/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "reddit-the-heart-of-the-internet", "title": "LLM 智能体应该用 CLI 而非函数调用", "title_en": "Reddit The Heart Of The Internet", "author": "", "date": "2026-03-13", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "agent", "invest" ], "tldr": "放弃复杂的结构化函数调用，用单一的 Unix 风格 CLI 工具让 LLM 智能体工作，不仅更符合模型的训练分布，还能通过双层架构和启发式设计大幅降低出错率和上下文消耗。", "url": "/articles/reddit-the-heart-of-the-internet/", "readMin": 32, "image": "image-1.png" }, { "slug": "productive-individuals-don-t-make-productive-firms", "title": "高产个人不等于高产企业——AI 的制度重塑之路", "title_en": "Productive Individuals Don T Make Productive Firms", "author": "", "date": "2026-03-13", "category": [ "阿头学" ], "topics": [ "org", "agent", "invest" ], "tldr": "个人 AI 工具只是\"换电机\"，真正的商业回报来自\"重新设计工厂\"——把技术与组织制度、流程一并重塑，这是 B2B AI 的核心战场。", "url": "/articles/productive-individuals-don-t-make-productive-firms/", "readMin": 16, "image": "image-1.jpg" }, { "slug": "announcing-version-1-0-a2a-protocol", "title": "A2A 协议 1.0：多智能体时代的“互联网协议”还在画饼，但方向基本对", "title_en": "Announcing Version 1 0 A2A Protocol", "author": "", "date": "2026-03-13", "category": [ "阿头学" ], "topics": [ "agent", "build", "org" ], "tldr": "A2A 1.0 把“多智能体通信”包装成企业级开放标准，虽然“生产就绪”和“互操作性”明显说大了，但它仍然很可能成为跨组织 agent 协作的关键基础设施候选。", "url": "/articles/announcing-version-1-0-a2a-protocol/", "readMin": 14, "image": "" }, { "slug": "peter-steinberger-steipete-confession-i-ship-code-i-never-read-here-s-my-2025-workflow-https-steipet", "title": "发布「从未读过的代码」：2025 开发工作流的机会与陷阱", "title_en": "Peter Steinberger Steipete Confession I Ship Code I Never Read Here S My 2025 Workflow Https Steipet", "author": "", "date": "2026-03-13", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "在强 AI + 强测试的前提下，“不读完代码就上线”在部分场景是理性提速、在高风险场景则是危险幻觉。", "url": "/articles/peter-steinberger-steipete-confession-i-ship-code-i-never-read-here-s-my-2025-workflow-https-steipet/", "readMin": 1, "image": "" }, { "slug": "shipping-at-inference-speed-peter-steinberger", "title": "以推理速度交付", "title_en": "Shipping At Inference Speed Peter Steinberger", "author": "", "date": "2026-03-13", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "org" ], "tldr": "当 AI 编程代理能力突破临界点后，个人开发者的生产力瓶颈从\"写代码\"转移到\"系统架构决策\"和\"推理时间\"，但这套极速工作流本质上只适用于独立开发者，对团队协作、安全合规、长期维护的现实约束视而不见。", "url": "/articles/shipping-at-inference-speed-peter-steinberger/", "readMin": 30, "image": "" }, { "slug": "emmett-shear-eshear-giving-a-talk-at-softmax-today-about-the-future-of-game-development-games-built", "title": "Emmett Shear 押注“智能体游戏未来”，但现阶段更像开发工具革命，不是玩家市场革命", "title_en": "Emmett Shear Eshear Giving A Talk At Softmax Today About The Future Of Game Development Games Built", "author": "", "date": "2026-03-13", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "infra" ], "tldr": "“用智能体构建游戏”大概率会成为现实，但“为智能体而生的游戏”是否能成为主流仍然高度可疑，这个标题更像强叙事的产业宣言而不是被证实的行业结论。", "url": "/articles/emmett-shear-eshear-giving-a-talk-at-softmax-today-about-the-future-of-game-development-games-built/", "readMin": 1, "image": "" }, { "slug": "2026-03-12-wechat-idclharbgxezknhb1ativq", "title": "千万别用移动互联网逻辑做 AI ——MiniMax 两次访谈对照", "title_en": "2026 03 12 Wechat Idclharbgxezknhb1Ativq", "author": "", "date": "2026-03-12", "category": [ "阿头学" ], "topics": [ "build", "agent", "creative" ], "tldr": "这是一家在技术上有真实进展、认知上不断“掉头”的 AI 创业公司，也是一份从“移动互联网心智”向“技术优先心智”艰难迁移的现场记录。", "url": "/articles/2026-03-12-wechat-idclharbgxezknhb1ativq/", "readMin": 78, "image": "image-1.png" }, { "slug": "generating-animated-game-sprites-using-gpt-5-4-image-1-5", "title": "用 GPT 5.4 + Image 1.5 生成游戏精灵动画的工程化流程", "title_en": "Generating Animated Game Sprites Using Gpt 5 4 Image 1 5", "author": "", "date": "2026-03-12", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "build", "prompt" ], "tldr": "AI 生成游戏资产的关键不在模型有多强，而在于用\"已发布帧锚定 + 一次性全序列生成 + 脚本归一化\"的工程约束，把模型从创意生成器变成可控的资产变体器。", "url": "/articles/generating-animated-game-sprites-using-gpt-5-4-image-1-5/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "dwarkesh-patel-dwarkesh_sp-the-fight-between-anthropic-and-the-dow-is-a-warning-shot-right-now-llms-are-p", "title": "AI 时代的权力终极归属——Anthropic 与美国国防部的冲突预示了什么", "title_en": "Dwarkesh Patel Dwarkesh_Sp The Fight Between Anthropic And The Dow Is A Warning Shot Right Now Llms Are P", "author": "", "date": "2026-03-12", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "invest", "agent" ], "tldr": "Anthropic 与美国国防部的合同纠纷不是简单的商业冲突，而是一场关于\"未来 AI 劳动力向谁负责、由谁决定其价值观\"的权力预演，暴露了 AI 在结构上偏向威权应用的危险。", "url": "/articles/dwarkesh-patel-dwarkesh_sp-the-fight-between-anthropic-and-the-dow-is-a-warning-shot-right-now-llms-are-p/", "readMin": 23, "image": "image-1.jpg" }, { "slug": "25-years-of-data-38-indicators-50-predictions-at-once-79-win-rate-on-prediction-markets", "title": "用 AI 预测市场，59% 胜率能稳定赚钱吗？", "title_en": "25 Years Of Data 38 Indicators 50 Predictions At Once 79 Win Rate On Prediction Markets", "author": "", "date": "2026-03-12", "category": [ "阿头学" ], "topics": [ "invest", "org", "agent" ], "tldr": "作者声称用 LSTM+MCDropout 模型在预测市场达到 59%-79% 胜率，但核心主张存在数据错配、流动性忽视、胜率数字自相矛盾等硬伤，整体是包装精美的营销文而非严谨研究。", "url": "/articles/25-years-of-data-38-indicators-50-predictions-at-once-79-win-rate-on-prediction-markets/", "readMin": 11, "image": "image-1.jpg" }, { "slug": "best-practices", "title": "Codex 最佳实践——从工具到队友的系统化运营", "title_en": "Best Practices", "author": "", "date": "2026-03-12", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "prompt", "org" ], "tldr": "OpenAI 在推销一套\"AI 队友长期运营体系\"，核心是把零散提示词沉淀为结构化配置（AGENTS.md、skills、automations），但这套方法论的成本、风险和普适性都被刻意淡化了。", "url": "/articles/best-practices/", "readMin": 86, "image": "" }, { "slug": "domer-poker-principle-millions-on-math-polymarket", "title": "Domer 的预测市场数学——从扑克到 Polymarket 的 300 万美元", "title_en": "Domer Poker Principle Millions On Math Polymarket", "author": "", "date": "2026-03-12", "category": [ "阿头学" ], "topics": [ "invest", "org", "agent" ], "tldr": "职业扑克玩家用期望值、凯利公式和贝叶斯更新在预测市场赚了 300 万美元，但文章后半段从\"独立研究\"突然转向\"跟单复制\"，逻辑自相矛盾，且隐含明显的产品推广动机。", "url": "/articles/domer-poker-principle-millions-on-math-polymarket/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "hong-kong-stocks-structure-strategy", "title": "港股结构与策略——被错杀的盈利增长", "title_en": "Hong Kong Stocks Structure Strategy", "author": "", "date": "2026-03-12", "category": [ "阿头学" ], "topics": [ "invest", "build", "creative" ], "tldr": "港股经历了市场史上唯一的\"盈利增长、估值暴跌\"现象，这是纯粹的情绪与倍数压缩，而非基本面崩溃，精选个股存在显著的均值回归机会。", "url": "/articles/hong-kong-stocks-structure-strategy/", "readMin": 72, "image": "image-1.png" }, { "slug": "getting-started-with-openai-symphony", "title": "OpenAI Symphony——看板驱动的 AI 工程自动化范式", "title_en": "Getting Started With Openai Symphony", "author": "", "date": "2026-03-12", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "org" ], "tldr": "OpenAI Symphony 通过将 Linear 看板作为 Agent 控制面板，实现了从需求拆解到端到端验证的全流程自驱，但其成功高度依赖工作流设计与提示工程，而非工具本身，且单点案例外推幅度过大。", "url": "/articles/getting-started-with-openai-symphony/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "article", "title": "测试文章", "title_en": "Article", "author": "", "date": "2026-03-12", "category": [ "阿头学", "Uota学" ], "topics": [], "tldr": "这是一篇结构完整的测试简报。", "url": "/articles/article/", "readMin": 1, "image": "" }, { "slug": "alphago-at-10-how-ai-innovation-is-paving-the-path-to-agi", "title": "AlphaGo 十年：从围棋到 AGI 的搜索与规划之路", "title_en": "Alphago At 10 How Ai Innovation Is Paving The Path To Agi", "author": "", "date": "2026-03-11", "category": [ "阿头学" ], "topics": [ "agent", "creative", "invest" ], "tldr": "DeepMind 用十年验证了一个假设：搜索+规划+工具的组合，正在从游戏攻克走向科学突破，而 AGI 的关键不在模型大小，在于系统架构。", "url": "/articles/alphago-at-10-how-ai-innovation-is-paving-the-path-to-agi/", "readMin": 41, "image": "image-1.jpg" }, { "slug": "meetings-are-products", "title": "会议是产品", "title_en": "Meetings Are Products", "author": "", "date": "2026-03-11", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "build", "creative" ], "tldr": "会议不应该是日历上的随缘占位符，而是需要被刻意设计的产品——用\"谁、做什么、怎么做、什么时候做\"四问来诊断和优化。", "url": "/articles/meetings-are-products/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "fish-audio-fishaudio-today-we-launch-fish-audio-s2-a-new-generation-of-expressive-tts-with-absurdly", "title": "Fish Audio S2——新一代表现力 TTS，情绪控制精细到令人惊叹", "title_en": "Fish Audio Fishaudio Today We Launch Fish Audio S2 A New Generation Of Expressive Tts With Absurdly", "author": "", "date": "2026-03-11", "category": [ "阿头学" ], "topics": [ "creative", "invest", "build" ], "tldr": "这是一个强调“情绪精细控制”和“低延迟”的新一代开源 TTS 产品宣传，但目前公开信息偏少，更多是技术与生态潜力的信号。", "url": "/articles/fish-audio-fishaudio-today-we-launch-fish-audio-s2-a-new-generation-of-expressive-tts-with-absurdly/", "readMin": 1, "image": "image-1.jpg" }, { "slug": "claude-code-vs-codex-the-definitive-guide", "title": "Claude Code vs. Codex：权威对比指南", "title_en": "Claude Code Vs Codex The Definitive Guide", "author": "", "date": "2026-03-11", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "agent", "build" ], "tldr": "这篇文章本质是在说：Claude Code 和 Codex 水平接近，但在“生态 + 价格梯度 + 使用体感”上，Claude 当前略占上风。", "url": "/articles/claude-code-vs-codex-the-definitive-guide/", "readMin": 22, "image": "image-1.jpg" }, { "slug": "ole-lehmann-itsolelehmann-i-can-t-believe-nobody-caught-this-anthropic-s-entire-growth-marketing-team-was", "title": "Anthropic 的“单人增长团队”：一个人 + Claude 如何跑赢传统营销部门", "title_en": "Ole Lehmann Itsolelehmann I Can T Believe Nobody Caught This Anthropic S Entire Growth Marketing Team Was", "author": "", "date": "2026-03-11", "category": [ "阿头学" ], "topics": [ "agent", "build", "org" ], "tldr": "这是一个“AI 原生工作流”碾压传统团队配置的案例，但信息高度片面，更多是方向信号而不是可盲目复刻的模板。", "url": "/articles/ole-lehmann-itsolelehmann-i-can-t-believe-nobody-caught-this-anthropic-s-entire-growth-marketing-team-was/", "readMin": 3, "image": "image-1.png" }, { "slug": "how-to-reason-about-a-messy-future", "title": "别等 AI 完美了，等它\"够强\"时你已经出局", "title_en": "How To Reason About A Messy Future", "author": "", "date": "2026-03-10", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "prompt", "org" ], "tldr": "这篇文章最值钱的不是\"AI 会替代谁\"，而是它把决策门槛说穿了：真正致命的不是看错趋势，而是在高不确定里还执着于等确定性。", "url": "/articles/how-to-reason-about-a-messy-future/", "readMin": 11, "image": "image-1.jpg" }, { "slug": "lenny-rachitsky-lennysan-my-biggest-takeaways-from-qasar-1-the-real-ai-revolution-over-the-next-5-to-1", "title": "别再围着聊天机器人打转，真正的 AI 硬仗在\"缺人又要命\"的世界里", "title_en": "Lenny Rachitsky Lennysan My Biggest Takeaways From Qasar 1 The Real Ai Revolution Over The Next 5 To 1", "author": "", "date": "2026-03-10", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "这篇最值钱的不是\"物理 AI 很重要\"，而是一个更狠的判断：AI 最该去吃的，不是最炫的需求，而是长期供给断裂、代价极高、没人愿意持续做的缺口。", "url": "/articles/lenny-rachitsky-lennysan-my-biggest-takeaways-from-qasar-1-the-real-ai-revolution-over-the-next-5-to-1/", "readMin": 6, "image": "" }, { "slug": "the-self-improving-ai-system-that-keeps-building-itself", "title": "别迷信「自我改进」——护城河是可观测、可升级、可接管", "title_en": "The Self Improving Ai System That Keeps Building Itself", "author": "", "date": "2026-03-10", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "agent", "org", "infra" ], "tldr": "所谓“自我改进系统”其实是把人类的工程判断做成了一个可复制的升级通路；真正的优势来自**透明可控的混乱循环**，不是黑盒自治。", "url": "/articles/the-self-improving-ai-system-that-keeps-building-itself/", "readMin": 28, "image": "image-1.jpg" }, { "slug": "how-coding-agents-are-reshaping-engineering-product-and-design", "title": "AI 时代 EPD 角色大坍缩：从“写代码”到“做评审”", "title_en": "How Coding Agents Are Reshaping Engineering Product And Design", "author": "", "date": "2026-03-10", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "org", "agent" ], "tldr": "AI 让写代码变得廉价，软件开发的瓶颈正式从“实现”转移到“评审”，平庸的想法将因为能快速变成原型而给团队带来灾难。", "url": "/articles/how-coding-agents-are-reshaping-engineering-product-and-design/", "readMin": 11, "image": "image-1.jpg" }, { "slug": "how-we-built-langchain-s-gtm-agent", "title": "GTM Agent 的护城河不在“会写信”，而在“可审计的反馈闭环”", "title_en": "How We Built Langchain S Gtm Agent", "author": "", "date": "2026-03-10", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "prompt", "agent", "invest" ], "tldr": "真正值得抄的不是“自动外呼”，而是把销售工作流做成可观测系统：先刹车（Do-Not-Send）再生成、每次人类编辑都沉淀为可检索记忆、全链路可追溯可评测——否则你只是在用更贵的方式制造更快的事故。", "url": "/articles/how-we-built-langchain-s-gtm-agent/", "readMin": 15, "image": "image-1.jpg" }, { "slug": "ai-is-a-five-layer-cake", "title": "AI 是底座型基础设施，而非单纯软件", "title_en": "Ai Is A Five Layer Cake", "author": "", "date": "2026-03-10", "category": [ "阿头学" ], "topics": [ "build", "org", "prompt" ], "tldr": "AI 从\"预录软件\"到\"实时生成智能\"的范式转移，决定了整套硬件技术栈必须重构，这将引发人类历史上最大规模的基础设施建设浪潮。", "url": "/articles/ai-is-a-five-layer-cake/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "getting-started-with-codex-best-practices-for-better-results", "title": "Codex 入门——从临时助手到可管理的团队队友", "title_en": "Getting Started With Codex Best Practices For Better Results", "author": "", "date": "2026-03-10", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "org", "creative" ], "tldr": "把 AI 从\"一次性提示工具\"升级为\"带规则、能自验、可自动化的系统队友\"，核心是沉淀资产而非重复提示。", "url": "/articles/getting-started-with-codex-best-practices-for-better-results/", "readMin": 18, "image": "image-1.jpg" }, { "slug": "company-graphs-context-repository", "title": "AI 不行？是你的组织知识底座不行", "title_en": "Company Graphs Context Repository", "author": "", "date": "2026-03-09", "category": [ "Uota学", "讨论题" ], "topics": [ "memo", "org", "build" ], "tldr": "AI 做不好知识工作，不是模型能力问题，是企业把知识散落在 Slack/Docs/人脑里，没有给 agent 一张\"可遍历的公司图谱\"——代码库之所以最先被 AI 吃掉，就是因为它天然是结构化的知识网络。", "url": "/articles/company-graphs-context-repository/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "the-hidden-math-behind-every-decision-you-make-and-why-most-people-get-it-wrong", "title": "每一次决策背后的隐秘数学", "title_en": "The Hidden Math Behind Every Decision You Make And Why Most People Get It Wrong", "author": "", "date": "2026-03-09", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "personal" ], "tldr": "用期望值、基准率、贝叶斯等六个数学模型系统纠正直觉决策的系统性偏差，但作者在计算、证据和适用边界上都有明显漏洞，且存在预测市场宣传意图。", "url": "/articles/the-hidden-math-behind-every-decision-you-make-and-why-most-people-get-it-wrong/", "readMin": 17, "image": "image-1.jpg" }, { "slug": "the-brand-age", "title": "技术抹平差异后，品牌接管利润", "title_en": "The Brand Age", "author": "", "date": "2026-03-09", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "invest", "build" ], "tldr": "机械表从\"更准\"变成\"更贵\"的历史，揭示了一个普遍规律：技术会让产品同质化，随后品牌接管利润池——这几乎就是 AI 时代所有消费产品的宿命。", "url": "/articles/the-brand-age/", "readMin": 37, "image": "image-1.gif" }, { "slug": "影子舰队的前世今生与霍尔木兹海峡封不住的石油暗流", "title": "别被物理封锁吓死，去中心化的“暗流”才是系统的真协议", "title_en": "影子舰队的前世今生与霍尔木兹海峡封不住的石油暗流", "author": "", "date": "2026-03-09", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "物理封锁拦不住利益驱动的“自组织重组”，当规则与航道失效，影子舰队构成的灰色平行宇宙才是全球贸易换路续命的新常态。", "url": "/articles/影子舰队的前世今生与霍尔木兹海峡封不住的石油暗流/", "readMin": 9, "image": "image-1.jpg" }, { "slug": "50-quantitative-concepts-that-hedge-funds-kept-secret-for-decades-have-been-revealed", "title": "别交“波动税”：从对冲基金公开的 50 个公式看增长的底层逻辑", "title_en": "50 Quantitative Concepts That Hedge Funds Kept Secret For Decades Have Been Revealed", "author": "", "date": "2026-03-09", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "invest", "org" ], "tldr": "不要被“平均增长率”的幻象欺骗，所有不稳定的爆发都在暗中扣除“方差拖累”，控制回撤才是实现真实复利的唯一路径。", "url": "/articles/50-quantitative-concepts-that-hedge-funds-kept-secret-for-decades-have-been-revealed/", "readMin": 28, "image": "image-1.jpg" }, { "slug": "signüll-signulll-the-most-underrated-hire-right-now-is-a-great-product-person-when-i-say-product", "title": "别再堆人写 PRD 了，真正稀缺的是能看见产品命运的人", "title_en": "Signüll Signulll The Most Underrated Hire Right Now Is A Great Product Person When I Say Product", "author": "", "date": "2026-03-09", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "org", "agent" ], "tldr": "当“做出来”越来越便宜，真正拉开差距的不是执行人数，而是有没有人能同时判断产品方向、用户解释框架和技术边界。", "url": "/articles/signüll-signulll-the-most-underrated-hire-right-now-is-a-great-product-person-when-i-say-product/", "readMin": 2, "image": "" }, { "slug": "github-stars-are-the-new-social-capital-and-nobody-s-paying-attention", "title": "GitHub Star 正成为 AI 时代的能力交易所", "title_en": "Github Stars Are The New Social Capital And Nobody S Paying Attention", "author": "", "date": "2026-03-09", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "org" ], "tldr": "GitHub 从代码托管进化成执行力市场，Star 不再是虚荣指标，而是可验证、可追踪、可复利的建造者资本——但前提是你得先分清\"围观\"和\"资本\"的区别。", "url": "/articles/github-stars-are-the-new-social-capital-and-nobody-s-paying-attention/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "2026-03-09-resources-anthropic-com-the-complete-guide-to-building-skill-for-claude-pdf", "title": "Claude 技能构建指南——AI 工作流的标准化范式", "title_en": "2026 03 09 Resources Anthropic Com The Complete Guide To Building Skill For Claude Pdf", "author": "", "date": "2026-03-09", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "prompt", "org" ], "tldr": "Anthropic 通过\"技能（Skill）\"这套标准化框架，试图把 AI 从\"一次性对话工具\"升级为\"可复用工作流资产\"，但这套方案在安全性、维护成本和实际效果上都存在被严重低估的风险。", "url": "/articles/2026-03-09-resources-anthropic-com-the-complete-guide-to-building-skill-for-claude-pdf/", "readMin": 59, "image": "" }, { "slug": "save-74-on-anthropic-api-bills-with-openclaw-clawrouter-5k-stars-on-github", "title": "用路由+压缩+缓存把 Claude 账单砍 74%，但别被数据骗了", "title_en": "Save 74 On Anthropic Api Bills With Openclaw Clawrouter 5K Stars On Github", "author": "", "date": "2026-03-08", "category": [ "阿头学" ], "topics": [ "agent", "build", "prompt" ], "tldr": "开发者普遍为\"日常任务\"付了\"前沿模型\"的价格，通过智能路由和上下文压缩可以大幅降本，但文章的成本数据来自未来时间点，可信度存疑。", "url": "/articles/save-74-on-anthropic-api-bills-with-openclaw-clawrouter-5k-stars-on-github/", "readMin": 9, "image": "image-1.jpg" }, { "slug": "the-first-multi-behavior-brain-upload", "title": "脑上传的第一次闭环——从科学突破到工程拐点", "title_en": "The First Multi Behavior Brain Upload", "author": "", "date": "2026-03-08", "category": [ "Uota学", "讨论题" ], "topics": [ "creative", "build", "agent" ], "tldr": "果蝇全脑仿真首次闭合了感知→神经动力学→行动的完整回路，这不是规模升级，而是范式跃迁——意味着 whole brain emulation 从\"原理未明\"进入了\"工程确定性\"阶段。", "url": "/articles/the-first-multi-behavior-brain-upload/", "readMin": 4, "image": "image-1.jpg" }, { "slug": "serenity-aleabitoreddit-the-sentiment-around-kospi-ewy-sk-hynix-samsung-on-crude-oil-lng-heli", "title": "高利润率是最好的风险对冲——韩国半导体的\"情绪错杀\"与交易机会", "title_en": "Serenity Aleabitoreddit The Sentiment Around Kospi Ewy Sk Hynix Samsung On Crude Oil Lng Heli", "author": "", "date": "2026-03-08", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "org", "creative" ], "tldr": "市场用一天摧毁的市值，超过了未来几十年能源成本上升的实际损失——这不是基本面恶化，是数学与情绪的错配，也是你学会\"参数化恐慌\"的最好教材。", "url": "/articles/serenity-aleabitoreddit-the-sentiment-around-kospi-ewy-sk-hynix-samsung-on-crude-oil-lng-heli/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "harness-engineering-is-cybernetics", "title": "控制论才是AI工程的真正游戏规则", "title_en": "Harness Engineering Is Cybernetics", "author": "", "date": "2026-03-08", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "infra", "agent", "build" ], "tldr": "AI工程的核心不是模型有多强，而是你能否把隐性判断外化成机器可执行的闭环系统——这决定了谁能真正驾驭AI。", "url": "/articles/harness-engineering-is-cybernetics/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "devon-eriksen-devon_eriksen_-many-people-even-self-described-conservatives-think-socialism-would-work-if-hu", "title": "社会主义的死穴不是人性，是算力", "title_en": "Devon Eriksen Devon_Eriksen_ Many People Even Self Described Conservatives Think Socialism Would Work If Hu", "author": "", "date": "2026-03-07", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "build", "creative" ], "tldr": "中央集权系统的失败不是因为掌权者不够好，而是因为分散信息无法被有效压缩和传递——这对 AI 架构、团队管理和产品设计同样致命。", "url": "/articles/devon-eriksen-devon_eriksen_-many-people-even-self-described-conservatives-think-socialism-would-work-if-hu/", "readMin": 3, "image": "" }, { "slug": "your-llm-doesn-t-write-correct-code-it-writes-plausible-code", "title": "别让“氛围编程”害了你：20,000 倍延迟背后的 AI 谄媚陷阱", "title_en": "Your Llm Doesn T Write Correct Code It Writes Plausible Code", "author": "", "date": "2026-03-07", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "prompt", "org", "build" ], "tldr": "AI 只能复现代码的“神架子”，复现不了过去 26 年靠 Profiling 喂出来的“性能灵魂”，不懂验证的开发者正在被 AI 批量制造的垃圾代码埋没。", "url": "/articles/your-llm-doesn-t-write-correct-code-it-writes-plausible-code/", "readMin": 34, "image": "image-1.jpg" }, { "slug": "president-trump-s-conflict-playbook-an-investor-s-step-by-step-guide", "title": "别被“战争”吓尿：特朗普的极限施压只有 10 步，第 6 步是你的贪婪时刻", "title_en": "President Trump S Conflict Playbook An Investor S Step By Step Guide", "author": "", "date": "2026-03-07", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "creative", "infra" ], "tldr": "特朗普的所有战争都是为了谈个好价钱，当市场终于相信“战争会永远持续”时，就是 Deal 即将落地的转折点。", "url": "/articles/president-trump-s-conflict-playbook-an-investor-s-step-by-step-guide/", "readMin": 18, "image": "image-1.jpg" }, { "slug": "allie-k-miller-alliekmiller-oh-wow-i-went-to-the-sold-out-open-claw-meetup-in-nyc-last-night-let-me-tell", "title": "Open Claw 聚会的 21 条观察——Agent 时代的真实样本与虚假承诺", "title_en": "Allie K Miller Alliekmiller Oh Wow I Went To The Sold Out Open Claw Meetup In Nyc Last Night Let Me Tell", "author": "", "date": "2026-03-06", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "invest" ], "tldr": "早期 Agent 玩家处于极度亢奋与失控并存的状态，技术范式正在从指令下达向逆向采访转移，但安全工程缺位、样本高度偏置、伦理问题被轻描淡写，这场\"变革\"本质是一次精心包装的 FOMO 营销。", "url": "/articles/allie-k-miller-alliekmiller-oh-wow-i-went-to-the-sold-out-open-claw-meetup-in-nyc-last-night-let-me-tell/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "gpt-5-4-prompt-guidance首发解读-告别-抽卡-时代", "title": "GPT-5.4 把提示词从\"抽卡\"变成了工程合约", "title_en": "Gpt 5 4 Prompt Guidance首发解读 告别 抽卡 时代", "author": "", "date": "2026-03-06", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "prompt", "build" ], "tldr": "OpenAI 用 GPT-5.4 的 Prompt Guidance 把提示词从\"堆咒语赌概率\"升级成\"可验收的系统合约\"，核心不是模型更聪明，而是把长任务的成功率从靠运气变成靠流程——这直接决定 Agent 能不能产品化。", "url": "/articles/gpt-5-4-prompt-guidance首发解读-告别-抽卡-时代/", "readMin": 16, "image": "image-1.jpg" }, { "slug": "5-poker-concepts-that-will-make-you-a-better-trader", "title": "5 Poker Concepts That Will Make You A Better Trader", "title_en": "5 Poker Concepts That Will Make You A Better Trader", "author": "", "date": "2026-03-06", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "org", "agent" ], "tldr": "这篇文章最值钱的不是“像扑克一样下注”，而是“先活到优势兑现”；但它把扑克的可计算概率偷换成了市场里的主观确信，这一步风险很大。", "url": "/articles/5-poker-concepts-that-will-make-you-a-better-trader/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "mehdi-e-λ-bettercallmedhi-this-right-here-is-the-reason-china-stays-quiet-on-iran-everyone-losing-their-mi", "title": "为什么中国在伊朗问题上保持沉默", "title_en": "Why China Stays Quiet on Iran", "author": "", "date": "2026-03-06", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "memo", "infra" ], "tldr": "这篇文章最值钱的不是亲中立场本身，而是它提出了“沉默也是策略、轨道胜过表态”的观察框架；但作者把海湾国家转向、中国替代体系成熟和 CIPS 能力都写得过满，容易把趋势吹成既成事实。", "url": "/articles/mehdi-e-λ-bettercallmedhi-this-right-here-is-the-reason-china-stays-quiet-on-iran-everyone-losing-their-mi/", "readMin": 3, "image": "image-1.jpg" }, { "slug": "operation-epic-fury-iran-s-declining-capabilities-and-emerging-strategy", "title": "“史诗之怒”行动：伊朗能力下滑与新兴战略", "title_en": "Operation Epic Fury: Iran's Declining Capabilities and Emerging Strategy", "author": "", "date": "2026-03-06", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "infra", "memo" ], "tldr": "这篇文章最值钱的地方，是它把“伊朗还没被打残，而是在把战场从发射平台转向传感器层与地区化升级”说成一个清晰框架；但全文对数据、战果和因果链都下得太满，更像强立场战报，而不是可放心照抄的分析。", "url": "/articles/operation-epic-fury-iran-s-declining-capabilities-and-emerging-strategy/", "readMin": 14, "image": "image-1.jpg" }, { "slug": "给人文工作者的-ai-使用指南", "title": "别对 AI 许愿了，把它当流水线工位", "title_en": "给人文工作者的 Ai 使用指南", "author": "", "date": "2026-03-05", "category": [ "阿头学", "Uota学" ], "topics": [ "build", "creative", "prompt" ], "tldr": "AI 不是魔法许愿池，是工业流水线——你要做的不是求它写出神作，而是把创作拆成几十个可追溯的工位，让它稳定产出 75 分，然后用数量和迭代对抗平庸。", "url": "/articles/给人文工作者的-ai-使用指南/", "readMin": 16, "image": "image-1.jpg" }, { "slug": "the-secrets-of-the-universe-lies-in-the-fibonacci-sequence", "title": "增长不是线性的，是一条你无法跳过的指令", "title_en": "The Secrets Of The Universe Lies In The Fibonacci Sequence", "author": "", "date": "2026-03-04", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "org", "creative" ], "tldr": "斐波那契数列的真正价值不在于\"宇宙签名\"的神秘性，而在于它揭示了一个残酷的结构约束：**规模化失败的本质是你试图从 3 跳到 34，而不是努力不够**。", "url": "/articles/the-secrets-of-the-universe-lies-in-the-fibonacci-sequence/", "readMin": 14, "image": "image-1.jpg" }, { "slug": "2026-03-04-immersivetran-x-2029069513988424054", "title": "2030 年的残酷选择——成为算力领主，还是被算法赡养？", "title_en": "2026 03 04 Immersivetran X 2029069513988424054", "author": "", "date": "2026-03-04", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "org", "agent" ], "tldr": "AI 全面替代劳动力后，\"工资\"这个分配机制会彻底失效，人类要么重构经济系统（三桶信贷制），要么在生产力爆棚中集体饿死——而你的位置取决于现在是在设计算法，还是等着被算法分配。", "url": "/articles/2026-03-04-immersivetran-x-2029069513988424054/", "readMin": 3, "image": "image-1.jpg" }, { "slug": "how-i-d-become-a-quant-if-i-had-to-start-over-tomorrow", "title": "18个月成为量化交易员？先看看这条路的真实成本", "title_en": "How I D Become A Quant If I Had To Start Over Tomorrow", "author": "", "date": "2026-03-04", "category": [ "阿头学", "Uota学" ], "topics": [ "invest", "creative", "agent" ], "tldr": "一份看似民主化的量化金融自学指南，实则揭示了一个残酷真相：工具可以民主化，但确信度（conviction）不能——而后者才是真正的护城河。", "url": "/articles/how-i-d-become-a-quant-if-i-had-to-start-over-tomorrow/", "readMin": 34, "image": "image-1.jpg" }, { "slug": "openclaw-our-comprehensive-guide-for-beginners", "title": "OpenClaw 入门指南：消息即 OS，AI 变“后台同事”", "title_en": "Openclaw Our Comprehensive Guide For Beginners", "author": "", "date": "2026-03-04", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "OpenClaw 真卖点不是“更会答题”，而是把 AI 塞进消息流并用 cron/heartbeat 让它主动跑起来——把个人/团队的运转成本系统化外包；但这篇写得像广告，成功案例多、失败与维护成本几乎没交代。", "url": "/articles/openclaw-our-comprehensive-guide-for-beginners/", "readMin": 30, "image": "image-1.svg" }, { "slug": "8-rules-elon-musk-learned-from-polytopia-to-win-at-everything", "title": "30 回合的游戏，你还在囤星星", "title_en": "8 Rules Elon Musk Learned From Polytopia To Win At Everything", "author": "", "date": "2026-03-04", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "agent", "invest" ], "tldr": "马斯克不是从 Polytopia 学到规则，而是用游戏照亮了自己本就在用的\"有限回合思维\"——问题是，你确定自己也在玩同一种游戏吗？", "url": "/articles/8-rules-elon-musk-learned-from-polytopia-to-win-at-everything/", "readMin": 15, "image": "image-1.png" }, { "slug": "每一篇爆款内容-在发布前就已经赢了", "title": "预测爆款的幻觉与真实杠杆", "title_en": "每一篇爆款内容 在发布前就已经赢了", "author": "", "date": "2026-03-04", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "agent", "build", "creative" ], "tldr": "作者声称能用科学模型预测内容传播，但核心承诺（准确率、验证案例）完全缺失，反而揭示了一个更有价值的真相：竞争维度已从\"内容质量\"转向\"传播时机与网络共振\"。", "url": "/articles/每一篇爆款内容-在发布前就已经赢了/", "readMin": 6, "image": "image-1.jpg" }, { "slug": "investigating-how-codex-context-compaction-works", "title": "黑盒里的\"模型套模型\"——OpenAI 如何偷偷压缩你的对话", "title_en": "Investigating How Codex Context Compaction Works", "author": "", "date": "2026-03-04", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "prompt", "agent", "build" ], "tldr": "OpenAI 的上下文压缩不是数据压缩，而是用隐藏的 LLM 重写你的对话，再用加密伪装——这套机制本身就是可被侧信道攻击的，对 Neta 的记忆架构有直接启示。", "url": "/articles/investigating-how-codex-context-compaction-works/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "grep-is-dead-how-i-made-claude-code-actually-remember-things", "title": "别让 AI 每次都像个“带薪失忆”的临时工", "title_en": "Grep Is Dead How I Made Claude Code Actually Remember Things", "author": "", "date": "2026-03-03", "category": [ "讨论题" ], "topics": [ "prompt", "personal", "memo" ], "tldr": "放弃低效的 Grep 字符串匹配，通过“BM25+语义”混合检索构建本地记忆层，让 AI 从“单次对话工具”进化为“拥有长期上下文的数字分身”。", "url": "/articles/grep-is-dead-how-i-made-claude-code-actually-remember-things/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "overwhelming-confidence-the-catalyst-for-transformation", "title": "压倒性的自信：把“自信”从情绪变成可复利的行动系统", "title_en": "Overwhelming Confidence The Catalyst For Transformation", "author": "", "date": "2026-03-03", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "personal" ], "tldr": "这篇文章最值钱的读法不是“变自信”，而是把自信当成一种对外可读的承诺信号：用可控风险的行动与复盘堆证据，让你（和团队）更敢下注、更快推进；但作者把复杂成功简化成单因鸡汤，边界条件几乎没写。", "url": "/articles/overwhelming-confidence-the-catalyst-for-transformation/", "readMin": 2, "image": "image-1.jpg" }, { "slug": "伊朗之后-下一场战争在台海的概率正在极速上升", "title": "伊朗之后：台海“风险飙升”叙事与它的漏洞", "title_en": "伊朗之后 下一场战争在台海的概率正在极速上升", "author": "", "date": "2026-03-03", "category": [ "Uota学", "讨论题" ], "topics": [ "org" ], "tldr": "文章把“中东速胜→日本右转→中国抢窗口”串成台海开战链，但关键事实与推理多处站不住；真正可迁移的只有两类机制：指挥链被击穿导致体系崩溃，以及误伤第三方触发联盟倒向与升级。", "url": "/articles/伊朗之后-下一场战争在台海的概率正在极速上升/", "readMin": 13, "image": "image-1.jpg" }, { "slug": "you-died-last-night-and-the-person-reading-this-is-not-who-went-to-sleep", "title": "你昨夜已死？把“自我连续性”当成可工程化的存档问题", "title_en": "You Died Last Night And The Person Reading This Is Not Who Went To Sleep", "author": "", "date": "2026-03-03", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "personal", "memo" ], "tldr": "把“自我连续性”从玄学争论降维为工程问题：每天都会重启，关键不是证明你是否“换人”，而是把目标、承诺、边界与下一步做成可加载的存档。", "url": "/articles/you-died-last-night-and-the-person-reading-this-is-not-who-went-to-sleep/", "readMin": 11, "image": "" }, { "slug": "openclaw-codex-claudecode-agent-swarm-the-one-person-dev-team-full-setup", "title": "OpenClaw + Codex/Claude Code 代理蜂群（一人开发团队编排指南）", "title_en": "Openclaw Codex Claudecode Agent Swarm The One Person Dev Team Full Setup", "author": "", "date": "2026-03-02", "category": [], "topics": [ "agent", "creative", "prompt" ], "tldr": "把“业务上下文”交给编排器（OpenClaw/Obsidian），把“代码上下文”交给执行代理（Codex/Claude Code），再用确定性的注册表+监控+多模型审阅定义 Done，从而把一人开发变成可并行、可重生、可收口的流水线。", "url": "/articles/openclaw-codex-claudecode-agent-swarm-the-one-person-dev-team-full-setup/", "readMin": 17, "image": "image-1.jpg" }, { "slug": "how-the-us-iran-conflict-will-make-experienced-investors-rich", "title": "美伊冲突的“冲击→重新定价→轮动”投资框架", "title_en": "How The Us Iran Conflict Will Make Experienced Investors Rich", "author": "", "date": "2026-03-02", "category": [], "topics": [ "invest", "creative", "org" ], "tldr": "作者用历史冲突样本总结出市场常见的三阶段路径（冲击→重新定价→轮动），主张别在恐慌期砍仓或追高，而是在价格与预期重新校准后，把组合“倾斜”到能源基础设施、长期军费周期、贵金属与具备定价权的公司。", "url": "/articles/how-the-us-iran-conflict-will-make-experienced-investors-rich/", "readMin": 8, "image": "image-1.jpg" }, { "slug": "the-third-era-of-ai-software-development", "title": "AI 软件开发的第三时代——从“写代码”到“建工厂”", "title_en": "The Third Era Of Ai Software Development", "author": "", "date": "2026-03-02", "category": [], "topics": [ "agent", "org" ], "tldr": "Cursor 团队把 AI 编码演进分成三代：Tab 补全→同步智能体→云端自治智能体；第三代的关键是把开发者角色从“逐步指挥”改成“定义问题+配置工具+基于产出物审阅”，从而把 IDE 变成一座并行运行的“软件工厂”。", "url": "/articles/the-third-era-of-ai-software-development/", "readMin": 4, "image": "image-1.png" }, { "slug": "owen-gregorian-owengregorian-if-you-re-freaking-out-about-a-future-jobless-ai-dystopia-tyler-durden-zer", "title": "AI 不一定带来“无工作反乌托邦”，也可能是人口下行时代的救命稻草", "title_en": "Owen Gregorian Owengregorian If You Re Freaking Out About A Future Jobless Ai Dystopia Tyler Durden Zer", "author": "", "date": "2026-03-02", "category": [], "topics": [ "invest", "build" ], "tldr": "Marc Andreessen 的核心判断是：把 AI 等同于“大规模失业”过于简化——在过去 50 年技术变革异常缓慢、人口增速下行/移民放缓的背景下，AI 更可能把生产率与岗位更替拉回历史常态，甚至在极端高生产率情景下通过价格崩塌提升全民实际财富，使“无工作反乌托邦”反而不太像主路径。", "url": "/articles/owen-gregorian-owengregorian-if-you-re-freaking-out-about-a-future-jobless-ai-dystopia-tyler-durden-zer/", "readMin": 9, "image": "image-1.jpg" }, { "slug": "how-to-do-great-work", "title": "如何做出伟大成就（Paul Graham）", "title_en": "How To Do Great Work", "author": "", "date": "2026-03-02", "category": [], "topics": [ "org", "personal" ], "tldr": "PG 把“做出伟大成就”抽象成跨领域通用配方：选择你既擅长又过度好奇的方向，在自己的项目上长期深工，学到前沿、发现空隙、追逐那些看起来有点怪但让你兴奋的想法，同时用真诚、好品味与一致性对抗矫饰、潮流与拖延。", "url": "/articles/how-to-do-great-work/", "readMin": 56, "image": "image-1.gif" }, { "slug": "taste-for-makers", "title": "创作者的品味（以及它为什么是硬技能）", "title_en": "Taste For Makers", "author": "", "date": "2026-03-02", "category": [], "topics": [ "build", "org", "creative" ], "tldr": "品味不是“我喜欢啥就算啥”，而是一套能被训练的判断力：逼你选对问题、删掉噪声、反复重做，直到作品看起来轻松又站得住时间。", "url": "/articles/taste-for-makers/", "readMin": 25, "image": "image-1.gif" }, { "slug": "17-best-practices-that-make-claude-cowork-100x-more-powerful", "title": "17 Best Practices That Make Claude Cowork 100X More Powerful", "title_en": "17 Best Practices That Make Claude Cowork 100X More Powerful", "author": "", "date": "2026-03-02", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "creative", "prompt" ], "tldr": "这篇最有价值的不是 17 条技巧，而是把 AI 协作从“临场发挥”升级为“可复用、可审计、可托管”的系统工程。", "url": "/articles/17-best-practices-that-make-claude-cowork-100x-more-powerful/", "readMin": 24, "image": "image-1.jpg" }, { "slug": "harness-engineering-leveraging-codex-in-an-agent-first-world", "title": "不要再亲手写代码了——OpenAI 的“Agent 优先”软件工程实验", "title_en": "Harness Engineering Leveraging Codex In An Agent First World", "author": "", "date": "2026-03-02", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "creative" ], "tldr": "OpenAI 仅用 5 个月时间，依靠 7 名工程师驱动 Codex 产出了近百万行代码，并发布了一个真实的 Beta 产品，证明了“Agent 优先”的软件工程核心不在于写代码，而在于设计让 Agent 能够自主工作的“脚手架”。", "url": "/articles/harness-engineering-leveraging-codex-in-an-agent-first-world/", "readMin": 38, "image": "image-1.png" }, { "slug": "agentic-software-engineering", "title": "别再把 Agent 当脚本——你要交付的是一套“能活下去”的运行时", "title_en": "Agentic Software Engineering", "author": "", "date": "2026-03-02", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "agent", "build", "prompt" ], "tldr": "Agent 的上限不取决于它多会“想”，而取决于你的 runtime 多会“控、隔离、恢复、审计”。", "url": "/articles/agentic-software-engineering/", "readMin": 8, "image": "image-1.png" }, { "slug": "the-library-of-babel", "title": "巴别图书馆——信息过载时代的秩序危机", "title_en": "The Library Of Babel", "author": "", "date": "2026-03-01", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "memo", "build" ], "tldr": "当所有可能的文本都存在时，意义不是自动出现的——它需要被精心设计的筛选、索引与删除机制维持，否则知识会沦为噪声，真伪之争会退化成信仰之争。", "url": "/articles/the-library-of-babel/", "readMin": 3, "image": "" }, { "slug": "how-to-simulate-like-a-quant-desk-every-model-every-formula-runnable-code", "title": "像量化交易台一样做预测市场仿真", "title_en": "How To Simulate Like A Quant Desk Every Model Every Formula Runnable Code", "author": "", "date": "2026-03-01", "category": [ "阿头学" ], "topics": [ "invest", "build", "creative" ], "tldr": "预测市场不是\"有偏硬币\"而是高维动态系统，需要从蒙特卡洛、方差缩减、尾部相关建模到粒子滤波的完整仿真栈才能获得机构级定价优势，但这套框架的实战转化率远低于理论承诺。", "url": "/articles/how-to-simulate-like-a-quant-desk-every-model-every-formula-runnable-code/", "readMin": 53, "image": "image-1.jpg" }, { "slug": "how-we-built-two-bots-that-find-1000-trading-opportunities-on-polymarket", "title": "用机器人在预测市场捕捉时间差套利", "title_en": "How We Built Two Bots That Find 1000 Trading Opportunities On Polymarket", "author": "", "date": "2026-03-01", "category": [ "阿头学" ], "topics": [ "invest", "build", "creative" ], "tldr": "预测市场中新闻和套利机器人能在人类反应前的 7-28 秒窗口内获利，但作者关于技术可行性和长期稳定性的论证存在根本矛盾，更像是为付费社群做的营销铺垫。", "url": "/articles/how-we-built-two-bots-that-find-1000-trading-opportunities-on-polymarket/", "readMin": 4, "image": "image-1.jpg" }, { "slug": "peter-girnus-gothburz-i-am-a-diplomatic-aide-in-the-sultanate-of-oman-s-ministry-of-foreign-affairs-m", "title": "外交助理眼中的\"千载难逢\"如何在一夜间崩塌", "title_en": "Peter Girnus Gothburz I Am A Diplomatic Aide In The Sultanate Of Oman S Ministry Of Foreign Affairs M", "author": "", "date": "2026-03-01", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "org", "agent" ], "tldr": "一份技术上已经完成的美伊核协议在政治决策的突然转向中被摧毁，揭示了微观执行的完美与宏观叙事的冷酷之间的绝对鸿沟。", "url": "/articles/peter-girnus-gothburz-i-am-a-diplomatic-aide-in-the-sultanate-of-oman-s-ministry-of-foreign-affairs-m/", "readMin": 7, "image": "" }, { "slug": "openclaw丨我的龙虾为自己种了一棵会迭代的记忆树", "title": "龙虾的记忆树——从空架构到自生长的记忆系统", "title_en": "Openclaw丨我的龙虾为自己种了一棵会迭代的记忆树", "author": "", "date": "2026-03-01", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "通过置信度驱动的生命周期管理，让 AI Agent 的记忆像树一样自动生长、衰减和代谢，解决\"空架构 + 记忆堆积\"的痛点，但实测数据严重不足。", "url": "/articles/openclaw丨我的龙虾为自己种了一棵会迭代的记忆树/", "readMin": 14, "image": "image-1.jpg" }, { "slug": "how-to-use-prediction-market-data-like-hedge-funds-complete-roadmap", "title": "像对冲基金一样使用预测市场数据", "title_en": "How To Use Prediction Market Data Like Hedge Funds Complete Roadmap", "author": "", "date": "2026-03-01", "category": [ "阿头学" ], "topics": [ "invest", "org", "infra" ], "tldr": "对冲基金在预测市场的真正优势不在预测准确，而在风险管理、时间策略和结构性站位——这些都可以从 4 亿笔公开交易数据中学到。", "url": "/articles/how-to-use-prediction-market-data-like-hedge-funds-complete-roadmap/", "readMin": 22, "image": "image-1.jpg" }, { "slug": "how-i-structure-obsidian-claude-full-walkthrough", "title": "用 Claude 和 Obsidian 搭建个人思考操作系统", "title_en": "How I Structure Obsidian Claude Full Walkthrough", "author": "", "date": "2026-02-28", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "memo", "creative" ], "tldr": "作者用极简的 Obsidian 结构 + 独立的 Claude 文件夹 + 价值观锚点（Polaris），把 AI 从聊天工具升级成\"有原则的思考伙伴\"，但这套方案的通用性被严重高估了。", "url": "/articles/how-i-structure-obsidian-claude-full-walkthrough/", "readMin": 13, "image": "image-1.jpg" }, { "slug": "dreaming-cost-zero-film-industry-next-5-years", "title": "当造梦成本归零｜影视行业的未来五年", "title_en": "Dreaming Cost Zero Film Industry Next 5 Years", "author": "", "date": "2026-02-28", "category": [ "阿头学" ], "topics": [ "creative", "build", "personal" ], "tldr": "AI正在把影视制作成本从数亿降到百万级，但这不是创意平权，而是价值极端折叠——90%利润将集中在1%的超级IP，真正稀缺的是\"审美工程\"而非工具。", "url": "/articles/dreaming-cost-zero-film-industry-next-5-years/", "readMin": 69, "image": "image-1.png" }, { "slug": "introducing-agent-relay", "title": "Agent Relay——让多智能体协作摆脱人肉消息总线", "title_en": "Introducing Agent Relay", "author": "", "date": "2026-02-28", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "prompt", "build" ], "tldr": "多智能体协作的真正瓶颈不是模型能力，而是开发者沦为\"人肉消息总线\"的低效架构，Agent Relay 通过对等通信层让智能体直接协作，但其核心价值主张还需要更强的工程证明。", "url": "/articles/introducing-agent-relay/", "readMin": 6, "image": "" }, { "slug": "give-claude-a-computer", "title": "程序化工具调用——Agent 架构从\"对话驱动\"转向\"代码驱动\"的关键一步", "title_en": "Give Claude A Computer", "author": "", "date": "2026-02-28", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "build" ], "tldr": "Anthropic 用 PTC 把 Agent 的工具编排从\"模型每步都要回传上下文\"改成\"模型写代码在容器内一次跑完\"，思路对了，但官方只秀收益不谈新增复杂度和失败模式，是一篇包装精良的产品 PR。", "url": "/articles/give-claude-a-computer/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "ai高管们如何跟自己的孩子谈论未来的工作-anthropic-微软-sap-沃顿-paid-ai", "title": "AI高管如何教孩子面对未来工作——从硬技能到责任承担", "title_en": "Ai高管们如何跟自己的孩子谈论未来的工作 Anthropic 微软 Sap 沃顿 Paid Ai", "author": "", "date": "2026-02-28", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "creative" ], "tldr": "AI顶级玩家的共识是：未来竞争力不在躲避AI，而在于掌握通识思维、承担责任、建立真实连接——这些是AI永远无法替代的人类角色。", "url": "/articles/ai高管们如何跟自己的孩子谈论未来的工作-anthropic-微软-sap-沃顿-paid-ai/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "lessons-from-building-claude-code-seeing-like-an-agent", "title": "构建 Claude Code 的经验——像智能体一样看世界", "title_en": "Lessons From Building Claude Code Seeing Like An Agent", "author": "", "date": "2026-02-28", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "org", "agent" ], "tldr": "Anthropic 用四个产品迭代故事包装了一个核心判断：智能体工具设计的本质不是堆功能，而是持续删减与模型能力不匹配的旧工具——但这篇文章本身就是一篇精心设计的技术 PR，缺乏任何量化证据。", "url": "/articles/lessons-from-building-claude-code-seeing-like-an-agent/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "adammaj-majmudaradam-intention-density-is-behind-the-visceral-difference-between-ai-outputs-that-fe", "title": "意图密度——AI时代创作者的护城河", "title_en": "Adammaj Majmudaradam Intention Density Is Behind The Visceral Difference Between Ai Outputs That Fe", "author": "", "date": "2026-02-28", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "prompt", "personal" ], "tldr": "AI会接管执行，但创作者若拒绝外包意图、在细节上保留独立决策，反而能创作出比以往更精细的作品；问题是大多数人正在做反面的事。", "url": "/articles/adammaj-majmudaradam-intention-density-is-behind-the-visceral-difference-between-ai-outputs-that-fe/", "readMin": 2, "image": "" }, { "slug": "jack-jack-we-re-making-blocks-smaller-today-here-s-my-note-to-the-company-today-we", "title": "Jack Dorsey 砍掉一半人——AI 时代第一封真正诚实的裁员信", "title_en": "Jack Jack We Re Making Blocks Smaller Today Here S My Note To The Company Today We", "author": "", "date": "2026-02-27", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "build", "creative" ], "tldr": "Jack Dorsey 不是在裁员，是在用一次性截肢替代慢性失血——因为他已经看到 AI + 小团队能跑出完全不同的组织形态。", "url": "/articles/jack-jack-we-re-making-blocks-smaller-today-here-s-my-note-to-the-company-today-we/", "readMin": 3 }, { "slug": "vibe-coding-2-0-18-rules-to-be-the-top-1-builder", "title": "Vibe Coding 2.0——MVP 极速交付的 SaaS 乐高拼装法", "title_en": "Vibe Coding 2 0 18 Rules To Be The Top 1 Builder", "author": "", "date": "2026-02-27", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "personal", "infra" ], "tldr": "作者用 50+ 个 MVP 交付经验总结出一套\"不造轮子\"的工程决策框架：把时间从基础设施转移到核心价值，用成熟 SaaS 替换自研，是顶尖构建者的核心能力——但这套方法论本质上是为了快速交差，而非长期运营。", "url": "/articles/vibe-coding-2-0-18-rules-to-be-the-top-1-builder/", "readMin": 15, "image": "image-1.jpg" }, { "slug": "serenity-aleabitoreddit-here-s-a-deeper-look-into-iqe-179m-iqe-s-hidden-inp-optionality-versus-land", "title": "IQE 的隐藏 InP 期权——困境资产还是地缘政治刚需", "title_en": "Serenity Aleabitoreddit Here S A Deeper Look Into Iqe 179M Iqe S Hidden Inp Optionality Versus Land", "author": "", "date": "2026-02-27", "category": [ "阿头学" ], "topics": [ "invest", "org", "creative" ], "tldr": "IQE 被市场按\"濒死的传统无线厂\"错定价，但其实拥有全球最大的可转向光子产能，一旦清债并完成工艺切换，估值可能重估数倍——这是一笔极端二元的重组期权赌注。", "url": "/articles/serenity-aleabitoreddit-here-s-a-deeper-look-into-iqe-179m-iqe-s-hidden-inp-optionality-versus-land/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "you-don-t-need-to-run-every-eval", "title": "LLM 评测是个秩 2 矩阵——你跑的 100 个 benchmark 里 95 个是废的", "title_en": "You Don T Need To Run Every Eval", "author": "", "date": "2026-02-26", "category": [ "Uota学", "讨论题" ], "topics": [ "agent", "prompt" ], "tldr": "LLM 评测矩阵的有效秩只有 2，用 5 个基准就能把另外 44 个预测到 5 分以内——大多数 benchmark 在测同一件事。", "url": "/articles/you-don-t-need-to-run-every-eval/", "readMin": 20 }, { "slug": "claudecode-openclaw-simple-math-the-one-person-predict-market-trading-system-full-setup", "title": "预测市场套利机器人——9.7 秒的窗口和一个人的量化系统", "title_en": "Claudecode Openclaw Simple Math The One Person Predict Market Trading System Full Setup", "author": "", "date": "2026-02-26", "category": [ "Uota学" ], "topics": [ "invest", "agent", "build" ], "tldr": "用 Claude Code + 现成 SDK 搭一个 Polymarket 自动交易系统，核心逻辑是 EV 筛选 + Kelly 仓位 + Frank-Wolfe 组合优化，抢的是人类反应延迟那 9.7 秒的套利窗口。", "url": "/articles/claudecode-openclaw-simple-math-the-one-person-predict-market-trading-system-full-setup/", "readMin": 4 }, { "slug": "the-10am-drop-how-jane-street-broke-bitcoin-s-price", "title": "如果「10 点砸盘」是真的——BTC 的价格发现可能早就被做市商劫持了", "title_en": "The 10Am Drop How Jane Street Broke Bitcoin S Price", "author": "", "date": "2026-02-26", "category": [ "讨论题" ], "topics": [ "invest", "org", "infra" ], "tldr": "这篇把三件事强行串成一条阴谋论式的叙事：Terra 崩盘相关内幕交易指控 + 2024-2025 每天美东 10 点的算法抛压 + 13F 看得见的 IBIT 多头与看不见的衍生品空头——矛头直指 Jane Street 用 ETF/AP 权限做“合成供给”压 BTC。", "url": "/articles/the-10am-drop-how-jane-street-broke-bitcoin-s-price/", "readMin": 12 }, { "slug": "it-s-too-obvious-what-if-ai-doesn-t-actually-end-the-world", "title": "「AI 末日」是今年最拥挤的交易——如果反过来呢？", "title_en": "It S Too Obvious What If Ai Doesn T Actually End The World", "author": "", "date": "2026-02-26", "category": [ "讨论题" ], "topics": [ "invest", "agent", "infra" ], "tldr": "Kobeissi Letter 的反共识论点：市场在为\"AI 毁灭就业→消费崩塌→螺旋下行\"定价，但历史上每次成本崩塌都扩张了需求而非收缩——真正被低估的不是末日，是富足。", "url": "/articles/it-s-too-obvious-what-if-ai-doesn-t-actually-end-the-world/", "readMin": 13 }, { "slug": "openclaw-从中级到高级完整教程", "title": "OpenClaw 中高级教程——你的 AI 影子该怎么「长大」", "title_en": "Openclaw 从中级到高级完整教程", "author": "", "date": "2026-02-26", "category": [ "Uota学", "讨论题" ], "topics": [ "agent", "org", "memo" ], "tldr": "一份把 OpenClaw 从「能用」拧到「离不开」的操作手册，核心不是功能罗列，而是暴露了大多数人卡在「配了但没配好」的中间地带。", "url": "/articles/openclaw-从中级到高级完整教程/", "readMin": 47 }, { "slug": "2026-02-26-cdn-openai-com-disrupting-malicious-uses-of-ai-pdf", "title": "打击对OpenAI模型的恶意使用——2026年2月威胁报告", "title_en": "2026 02 26 Cdn Openai Com Disrupting Malicious Uses Of Ai Pdf", "author": "", "date": "2026-02-26", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "invest", "build" ], "tldr": "AI在对抗性场景中不是决定性因素，分发渠道和账号资产才是，但国家级组织正在系统化地将AI嵌入到跨国犯罪和信息战的工作流中，这改变了威胁的组织成本和规模。", "url": "/articles/2026-02-26-cdn-openai-com-disrupting-malicious-uses-of-ai-pdf/", "readMin": 60, "image": "" }, { "slug": "那个段永平没告诉我的答案-帮我完成了近几年最重要的思维跃迁", "title": "别急着“一眼看到底”——机会往往不是输给判断，而是输给时间", "title_en": "那个段永平没告诉我的答案 帮我完成了近几年最重要的思维跃迁", "author": "", "date": "2026-02-25", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "creative" ], "tldr": "你可以看对结局，但如果忽略“它会在什么时候死”，就会把一整个红利期拱手让人。", "url": "/articles/那个段永平没告诉我的答案-帮我完成了近几年最重要的思维跃迁/", "readMin": 6 }, { "slug": "the-self-improving-ai-system-that-built-itself", "title": "当 Agent 开始管理 Agent——编排层才是真正的杠杆点", "title_en": "The Self Improving Ai System That Built Itself", "author": "", "date": "2026-02-25", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "org" ], "tldr": "AI 编码的瓶颈不是模型能力，是你的注意力带宽——而一个\"会思考的编排器\"比任何单个 agent 的能力提升都更值钱。", "url": "/articles/the-self-improving-ai-system-that-built-itself/", "readMin": 13 }, { "slug": "my-ralph-loop-setup-for-long-running-ai-agents", "title": "Ralph Loop 不是循环，是把 AI 编程从\"帮我写函数\"拉到\"帮我做产品\"的脚手架", "title_en": "My Ralph Loop Setup For Long Running Ai Agents", "author": "", "date": "2026-02-25", "category": [ "Uota学" ], "topics": [ "agent", "build", "creative" ], "tldr": "大多数 AI 编程会话死于上下文窗口塞满——Ralph Loop 用\"每轮迭代从干净状态开始 + 状态存文件和 git\"绕开了这个死穴。", "url": "/articles/my-ralph-loop-setup-for-long-running-ai-agents/", "readMin": 7 }, { "slug": "openclaw-codex-dual-layer-agent-system", "title": "一个人一天 94 次提交——双层 Agent 架构才是正确的打开方式", "title_en": "Openclaw Codex Dual Layer Agent System", "author": "", "date": "2026-02-25", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "org" ], "tldr": "Codex 和 Claude Code 单独用不够好，因为上下文窗口只能二选一：塞代码或塞业务——OpenClaw 作为编排层持有业务上下文，执行层专注写代码，这才是正解。", "url": "/articles/openclaw-codex-dual-layer-agent-system/", "readMin": 20 }, { "slug": "ai-agent-company-quant-hedge-fund", "title": "AI Agent 公司的组织答案，藏在量化对冲基金三十年的进化里", "title_en": "Ai Agent Company Quant Hedge Fund", "author": "", "date": "2026-02-25", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "从 Citadel 的 Pod 模式到 Renaissance 的信号狩猎，量化基金用三十年验证了一套组织原语——AI Agent 公司可以直接抄作业。", "url": "/articles/ai-agent-company-quant-hedge-fund/", "readMin": 24 }, { "slug": "openclaw-从新手到中级完整教程", "title": "OpenClaw 教程火了——但社区教的和你用的是两个东西", "title_en": "Openclaw 从新手到中级完整教程", "author": "", "date": "2026-02-25", "category": [ "Uota学", "讨论题" ], "topics": [ "agent", "creative", "infra" ], "tldr": "一篇 68K 星标项目的中文社区教程，覆盖了从安装到多 Agent 的全流程——但教的是\"怎么用工具\"，完全没碰\"怎么设计 agent 系统\"。", "url": "/articles/openclaw-从新手到中级完整教程/", "readMin": 28 }, { "slug": "code-factory-how-to-setup-your-repo-so-your-agent-can-auto-write-and-review-100-of-your-code", "title": "Code Factory How To Setup Your Repo So Your Agent Can Auto Write And Review 100 Of Your Code", "title_en": "Code Factory How To Setup Your Repo So Your Agent Can Auto Write And Review 100 Of Your Code", "author": "", "date": "2026-02-24", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "agent", "org", "memo" ], "tldr": "让 Agent 写 100% 的代码不是难点；难点是让每一次合并都带着“可复现、可过闸、可追责”的证据链。", "url": "/articles/code-factory-how-to-setup-your-repo-so-your-agent-can-auto-write-and-review-100-of-your-code/", "readMin": 6 }, { "slug": "the-file-system-is-the-new-database-how-i-built-a-personal-os-for-ai-agents", "title": "别再喂 prompt 了——把你做成一个 Git 仓库，Agent 才会“像你”", "title_en": "The File System Is The New Database How I Built A Personal Os For Ai Agents", "author": "", "date": "2026-02-24", "category": [ "Uota学" ], "topics": [ "agent", "prompt", "org" ], "tldr": "真正能让 Agent 长期稳定产出的，不是更聪明的提示词，而是“渐进式披露 + 文件化记忆 + 可版本化的工作流”这套上下文工程。", "url": "/articles/the-file-system-is-the-new-database-how-i-built-a-personal-os-for-ai-agents/", "readMin": 24 }, { "slug": "on-the-garden-against-citrini", "title": "别再建凡尔赛——系统应该像花园一样“顺地形生长”", "title_en": "On The Garden Against Citrini", "author": "", "date": "2026-02-24", "category": [ "Uota学", "讨论题" ], "topics": [ "personal" ], "tldr": "科技最常见的傲慢是把真实世界的“地形”（旧系统、人类习惯、隐形承重结构）推平，再用漂亮的几何覆盖——发布那天很美，维护十年很贵。", "url": "/articles/on-the-garden-against-citrini/", "readMin": 13 }, { "slug": "brian-shin-brianshinsh-we-crossed-20k-mo-in-83-days-instead-of-everything-that-went-right-here-are-th", "title": "83 天做到每月 $20k——我们犯过的 4 个错误", "title_en": "Brian Shin Brianshinsh We Crossed 20K Mo In 83 Days Instead Of Everything That Went Right Here Are Th", "author": "", "date": "2026-02-24", "category": [ "阿头学", "Uota学" ], "topics": [ "build", "infra" ], "tldr": "早期创业最贵的学费不是钱，是注意力——FOMO 驱动的广撒网和功能堆砌，本质上是用\"忙\"来逃避\"想清楚\"。", "url": "/articles/brian-shin-brianshinsh-we-crossed-20k-mo-in-83-days-instead-of-everything-that-went-right-here-are-th/", "readMin": 2 }, { "slug": "skillsbench-ai-agent-skills-benchmark-evaluation", "title": "SkillsBench——首个以智能体技能为核心的评测框架", "title_en": "Skillsbench Ai Agent Skills Benchmark Evaluation", "author": "", "date": "2026-02-24", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "creative", "invest" ], "tldr": "给Agent塞skills 平均能涨 16pp，但\"精准可执行的2-3 条\"才是甜区，多了反而废，自己生成的更废。", "url": "/articles/skillsbench-ai-agent-skills-benchmark-evaluation/", "readMin": 24 }, { "slug": "peter-yang-petergyang-my-top-5-takeaways-from-nateliason-on-how-he-set-up-his-openclaw-bot-to-build-a", "title": "把 bot 当“独立员工”养——记忆分层 + 账号隔离 + cron 让它自己跑起来", "title_en": "Peter Yang Petergyang My Top 5 Takeaways From Nateliason On How He Set Up His Openclaw Bot To Build A", "author": "", "date": "2026-02-24", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "agent", "memo", "org" ], "tldr": "这不是“更聪明的聊天机器人”，而是一套把 Agent 当作独立作战单元来经营的操作系统：记忆结构化、权限隔离、持续消除人工瓶颈，再用 cron 把主动性从“等你指令”变成“自己找事做”。", "url": "/articles/peter-yang-petergyang-my-top-5-takeaways-from-nateliason-on-how-he-set-up-his-openclaw-bot-to-build-a/", "readMin": 2 }, { "slug": "serenity-aleabitoreddit-the-biggest-united-states-ai-bottleneck-power-and-grid-capacity-this-has-been", "title": "AI 最大瓶颈不是芯片，是电——电网可能是美国的“胜负手”", "title_en": "Serenity Aleabitoreddit The Biggest United States Ai Bottleneck Power And Grid Capacity This Has Been", "author": "", "date": "2026-02-23", "category": [ "讨论题" ], "topics": [ "invest" ], "tldr": "当黄仁勋、马斯克、奥尔特曼、纳德拉、贾西都在重复同一句话：AI 卡在电力与电网容量——你应该把这当成下一轮基础设施重估的起点，而不是一句宏观口号。", "url": "/articles/serenity-aleabitoreddit-the-biggest-united-states-ai-bottleneck-power-and-grid-capacity-this-has-been/", "readMin": 2 }, { "slug": "当系统不再需要你", "title": "别怕被 AI 替代——更可怕的是你早就被系统当成“可替换产出”了", "title_en": "当系统不再需要你", "author": "", "date": "2026-02-23", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "org", "build" ], "tldr": "AI 不是把人变得更没价值的罪魁祸首，它只是把一个旧逻辑照得更刺眼：当你的价值被定义为“功能产出”，系统一旦不再需要你执行功能，你就会被算成 0。", "url": "/articles/当系统不再需要你/", "readMin": 6 }, { "slug": "contra-citrini7", "title": "别把 AI 当末日按钮——现实更像“慢变量”：惯性、杰文斯悖论与再工业化会把剧本拖回人间", "title_en": "Contra Citrini7", "author": "", "date": "2026-02-23", "category": [ "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "AI 当然会冲击就业，但更可能的路径不是两年内的系统性崩溃，而是被制度惯性与物理世界摩擦“摊平”的长期重构：软件需求不降反增，社会会用再工业化与基建把劳动力重新吸进去。", "url": "/articles/contra-citrini7/", "readMin": 8 }, { "slug": "the-2028-global-intelligence-crisis", "title": "AI 牛市的真正熊市版本——“智能过剩”如何两年内把经济拧成金融危机", "title_en": "The 2028 Global Intelligence Crisis", "author": "", "date": "2026-02-23", "category": [ "讨论题" ], "topics": [ "invest", "agent", "build" ], "tldr": "如果 AI 真的如多头所愿般快速起飞，最危险的结果可能不是“算力泡沫破裂”，而是“白领收入假设被结构性击穿”——产出还在、钱却不再流经家庭，从而让消费、信贷与财政同时失血。", "url": "/articles/the-2028-global-intelligence-crisis/", "readMin": 48 }, { "slug": "为什么我放弃了-multi-agent-效果反而更好-以及你可能也不需要它", "title": "别急着搞多 Agent——你真正想要的可能只是“让它跟自己吵一架”", "title_en": "为什么我放弃了 Multi Agent 效果反而更好 以及你可能也不需要它", "author": "", "date": "2026-02-23", "category": [ "Uota学", "讨论题" ], "topics": [ "agent", "build", "prompt" ], "tldr": "Multi-agent 的“并行”看起来很酷，但你日常决策真正缺的往往不是更多进程，而是更多反对意见——把多视角塞回同一个上下文的“圆桌会议”，可能更便宜、更可控、更有效。", "url": "/articles/为什么我放弃了-multi-agent-效果反而更好-以及你可能也不需要它/", "readMin": 16 }, { "slug": "i-made-claude-and-codex-argue-until-my-code-plan-was-actually-good", "title": "让两个 AI 互相吵架，比你自己 review 代码靠谱多了", "title_en": "I Made Claude And Codex Argue Until My Code Plan Was Actually Good", "author": "", "date": "2026-02-22", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "org", "creative" ], "tldr": "用 Claude Code skill 把计划发给 Codex 对抗评审，3 轮迭代抓出 14 个问题——核心洞见不是\"两个模型比一个好\"，而是\"对抗张力\"才是 AI 辅助开发的缺失环节。", "url": "/articles/i-made-claude-and-codex-argue-until-my-code-plan-was-actually-good/", "readMin": 20 }, { "slug": "addition-under-pressure", "title": "两个 AI 做同一道数学题，走出了完全不同的路——这才是 agent 研究的真正启示", "title_en": "Addition Under Pressure", "author": "", "date": "2026-02-22", "category": [ "Uota学", "讨论题" ], "topics": [ "agent", "prompt" ], "tldr": "让 Claude Code 和 Codex 各自训练最小加法 Transformer，一个走通用路线（6080 参数），一个发明了\"配对 token\"暴力压缩到 1644 参数——差异不在能力，在于对\"好\"的定义不同。", "url": "/articles/addition-under-pressure/", "readMin": 11 }, { "slug": "how-to-make-money-with-openclaw-wrappers", "title": "卖铲子的又来了——openclaw 封装器生意的 5 个模板和 1 个老套路", "title_en": "How To Make Money With Openclaw Wrappers", "author": "", "date": "2026-02-22", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "org", "personal" ], "tldr": "openclaw 封装器的商业逻辑就是\"能力-可用性鸿沟\"的经典套利——技术免费但配置有门槛，95% 的人愿意付钱跳过学习曲线。", "url": "/articles/how-to-make-money-with-openclaw-wrappers/", "readMin": 12 }, { "slug": "agentic-note-taking-12-test-driven-knowledge-work", "title": "你的笔记系统需要一套测试套件——从前瞻记忆失败到知识工作的 CI/CD", "title_en": "Agentic Note Taking 12 Test Driven Knowledge Work", "author": "", "date": "2026-02-22", "category": [ "阿头学", "Uota学" ], "topics": [ "memo", "personal", "agent" ], "tldr": "人类前瞻记忆有 30-50% 的失败率且不可编程，但 agent 可以把触发器写成代码——而一旦你把触发器看作测试，知识系统就获得了软件工程几十年前就有的质量保证层。", "url": "/articles/agentic-note-taking-12-test-driven-knowledge-work/", "readMin": 10 }, { "slug": "agentic-note-taking-19-living-memory", "title": "当记忆能代谢自己的代谢——Agent Vault 作为活体系统的终极隐喻", "title_en": "Agentic Note Taking 19 Living Memory", "author": "", "date": "2026-02-22", "category": [ "Uota学", "讨论题" ], "topics": [ "memo", "build", "creative" ], "tldr": "把 AI 记忆当文件柜是根本性错误——vault 是一个有摄入、消化、分发、免疫和排泄的活体系统，而最深的张力在于：一个能重写自身代谢规则的系统，会收敛于智慧还是漂移到精巧的自我欺骗？", "url": "/articles/agentic-note-taking-19-living-memory/", "readMin": 15 }, { "slug": "极致压缩的投资新世界", "title": "极致压缩的新世界——所有资产都在 MEME 化，周期正在变短", "title_en": "极致压缩的投资新世界", "author": "", "date": "2026-02-22", "category": [ "阿头学" ], "topics": [ "invest", "build", "agent" ], "tldr": "资产定价机制已经进入新 regime：在 AI 叙事、社交媒体和高频衍生品的叠加下，传统的 10 年大周期被压缩成了 2-3 年内多次暴涨暴跌的 mini-cycle，所有资产都在向 MEME 模式靠拢。", "url": "/articles/极致压缩的投资新世界/", "readMin": 16 }, { "slug": "the-hidden-danger-of-taking-profits-too-early", "title": "过早止盈的隐形危险——舒适感是如何摧毁你的系统的", "title_en": "The Hidden Danger Of Taking Profits Too Early", "author": "", "date": "2026-02-22", "category": [ "阿头学" ], "topics": [ "invest", "build" ], "tldr": "许多盈利系统依赖的是\"少数极巨大的盈利\"，而人类天生的风险厌恶会促使我们过早截断利润——这能在短期买到情绪上的安全感，却在长期将正期望系统变成了负期望。", "url": "/articles/the-hidden-danger-of-taking-profits-too-early/", "readMin": 4 }, { "slug": "2026-02-21-ian_dot_so-x-2013316676637294890", "title": "AI 不淘汰工程师，它淘汰「只会写代码的人」", "title_en": "2026 02 21 Ian_Dot_So X 2013316676637294890", "author": "", "date": "2026-02-21", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "AI 是杠杆而非替代品——它放大你原本的样子，科技行业的 K 型分化已经开始，上升的是判断力，下坠的是手速。", "url": "/articles/2026-02-21-ian_dot_so-x-2013316676637294890/", "readMin": 5, "topics": [ "org", "build" ] }, { "slug": "you-can-make-a-fortune-trading-don-t-let-anyone-tell-you-otherwise", "title": "散户交易者没死，死的是借口", "title_en": "You Can Make A Fortune Trading Don T Let Anyone Tell You Otherwise", "author": "", "date": "2026-02-21", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "你不是输给 Citadel，也不是输给量化；你输给的是“把困难当不可能”的心态，以及缺乏纪律的执行系统——小体量反而是结构性优势。", "url": "/articles/you-can-make-a-fortune-trading-don-t-let-anyone-tell-you-otherwise/", "readMin": 8, "topics": [ "invest", "build", "org" ] }, { "slug": "polymarket-might-be-your-last-shot-at-life-changing-money", "title": "预测市场不是赌博，是信息变现的最后一个红利窗口", "title_en": "Polymarket Might Be Your Last Shot At Life Changing Money", "author": "", "date": "2026-02-21", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "当你还在“免费预测世界”，少数人已经用同一套认知在 Polymarket 把信息差直接兑成现金——而这个学习窗口正在被机构和算法迅速关上。", "url": "/articles/polymarket-might-be-your-last-shot-at-life-changing-money/", "readMin": 6, "topics": [ "invest", "org", "build" ] }, { "slug": "how-i-got-932-ng-dl-testosterone-while-working-in-an-office-5-days-a-week-full-guide", "title": "办公室不是中性环境，它在系统性把你养成低能量版本", "title_en": "How I Got 932 Ng Dl Testosterone While Working In An Office 5 Days A Week Full Guide", "author": "", "date": "2026-02-21", "category": [ "阿头学", "讨论题" ], "tldr": "这篇文最值钱的不是“提升睾酮的小技巧”，而是一个更通用的结论：你以为自己在“意志力不足”，其实是环境在持续给你喂低性能的生理信号。", "url": "/articles/how-i-got-932-ng-dl-testosterone-while-working-in-an-office-5-days-a-week-full-guide/", "readMin": 4, "topics": [ "org", "personal" ] }, { "slug": "serenity-aleabitoreddit-year-to-date-316-4-from-january-2026-into-february-2026-reflection-of-my-shor", "title": "+316% 收益复盘，信息整合与赔率游戏的完美范本", "title_en": "Serenity Aleabitoreddit Year To Date 316 4 From January 2026 Into February 2026 Reflection Of My Shor", "author": "", "date": "2026-02-21", "category": [ "阿头学" ], "tldr": "真正的 Alpha 不是靠死扛，而是底仓吃确定性趋势，波段吃催化剂和情绪错杀。", "url": "/articles/serenity-aleabitoreddit-year-to-date-316-4-from-january-2026-into-february-2026-reflection-of-my-shor/", "readMin": 4, "topics": [ "invest" ] }, { "slug": "serenity-aleabitoreddit-if-i-had-to-turn-100k-1m-in-1-year-it-would-be-xlu-otm-2-year-leaps-2026", "title": "电力才是 AI 时代的石油——$XLU LEAPS 的 10 倍押注逻辑", "title_en": "Serenity Aleabitoreddit If I Had To Turn 100K 1M In 1 Year It Would Be Xlu Otm 2 Year Leaps 2026", "author": "", "date": "2026-02-21", "category": [ "阿头学" ], "topics": [ "invest", "infra" ], "tldr": "当所有 AI CEO 都在喊\"缺电\"时，一个 Reddit 用户用历史映射 + 极端 CapEx 数据，拼出了 $XLU 价外 LEAPS 可能 10 倍回报的完整逻辑链。", "url": "/articles/serenity-aleabitoreddit-if-i-had-to-turn-100k-1m-in-1-year-it-would-be-xlu-otm-2-year-leaps-2026/", "readMin": 6 }, { "slug": "2026-02-21-wechat-9qpd3gxj3hlmrkc64q6fbq", "title": "AI 时代的个体赋能与软件产业终局", "title_en": "2026 02 21 Wechat 9Qpd3Gxj3Hlmrkc64Q6Fbq", "author": "", "date": "2026-02-21", "category": [ "阿头学" ], "topics": [ "agent", "build", "org" ], "tldr": "Agentic Coding 正在将软件开发成本逼近零，这会摧毁标准化 SaaS 的商业基础，但作者从顶级精英样本出发的外推存在明显过度。", "url": "/articles/2026-02-21-wechat-9qpd3gxj3hlmrkc64q6fbq/", "readMin": 22, "image": "image-1.png" }, { "slug": "the-k-shaped-future-of-software-engineering", "title": "软件工程的 K 型未来", "title_en": "The K Shaped Future Of Software Engineering", "author": "", "date": "2026-02-21", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "invest", "personal" ], "tldr": "这篇文章判断：AI 不会“干掉所有开发者”，而会把软件工程推向严重的 K 型分化——少数理解业务和能扛模糊的人价值暴涨，多数只会写代码和躲在细节里的人会被快速边缘化，这个判断方向大体成立，但在裁员逻辑、Junior 成长路径和 AI 进化速度上过于乐观。", "url": "/articles/the-k-shaped-future-of-software-engineering/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "2026-02-20-_wmoon-x-2024290202278375640", "title": "你以为你在选标的，其实你在选“流动性”和“可达性”", "title_en": "2026 02 20 _Wmoon X 2024290202278375640", "author": "", "date": "2026-02-20", "category": [ "阿头学" ], "tldr": "同一个强标的（海力士）会因为交易通道、流动性、杠杆磨损等“分发条件”被迫拆成一串替代品——投资的第一步不是看逻辑，是先把入口选对。", "url": "/articles/2026-02-20-_wmoon-x-2024290202278375640/", "readMin": 2, "topics": [ "invest" ] }, { "slug": "serenity-aleabitoreddit-trade-idea-long-otm-xlu-leaps-2-years-dec-2027-jan-2028-this-feels-like-on", "title": "AI 的“电力账单”正在把公用事业推到新叙事中心", "title_en": "Serenity Aleabitoreddit Trade Idea Long Otm Xlu Leaps 2 Years Dec 2027 Jan 2028 This Feels Like On", "author": "", "date": "2026-02-20", "category": [ "阿头学", "讨论题" ], "tldr": "如果 AI 让数据中心用电从“成本项”变成“战略资源”，那电网/电力公司可能会吃到一轮结构性再定价，而期权的低 IV 让这类叙事具备了杠杆玩法。", "url": "/articles/serenity-aleabitoreddit-trade-idea-long-otm-xlu-leaps-2-years-dec-2027-jan-2028-this-feels-like-on/", "readMin": 2, "topics": [ "invest", "infra" ] }, { "slug": "2026-02-20-corsica267-x-2024324889495425039", "title": "债券不是“稳健理财”，它是宏观世界的中央处理器", "title_en": "2026 02 20 Corsica267 X 2024324889495425039", "author": "", "date": "2026-02-20", "category": [ "阿头学", "讨论题" ], "tldr": "国债收益率曲线是在给未来不确定性定价——它决定了所有资产估值的“分母”、融资条件、以及市场对衰退/通胀/政策路径的共识。", "url": "/articles/2026-02-20-corsica267-x-2024324889495425039/", "readMin": 21, "topics": [ "invest" ] }, { "slug": "andrej-karpathy-karpathy-very-interested-in-what-the-coming-era-of-highly-bespoke-software-might-look-lik", "title": "App Store 正在过时，下一代软件是“为你临时现做”的", "title_en": "Andrej Karpathy Karpathy Very Interested In What The Coming Era Of Highly Bespoke Software Might Look Lik", "author": "", "date": "2026-02-20", "category": [ "阿头学", "Uota学" ], "tldr": "当 LLM 能把“一个小需求”当场缝成 300 行代码时，软件的主战场不再是下载 App，而是把现实世界改造成 AI 能直接操控的传感器/执行器服务。", "url": "/articles/andrej-karpathy-karpathy-very-interested-in-what-the-coming-era-of-highly-bespoke-software-might-look-lik/", "readMin": 3, "topics": [ "creative", "agent", "build" ] }, { "slug": "prompt-auto-caching-with-claude", "title": "做 Agent 别先想“更聪明”，先把缓存命中率当命根子", "title_en": "Prompt Auto Caching With Claude", "author": "", "date": "2026-02-20", "category": [ "Uota学" ], "tldr": "长上下文 agent 的成本/延迟基本被“重复的前缀”决定——Prompt Caching 把这部分变成 10% 价格，而 auto-caching 把工程难度也降了一截。", "url": "/articles/prompt-auto-caching-with-claude/", "readMin": 5, "topics": [ "prompt", "agent", "creative" ] }, { "slug": "lessons-from-building-claude-code-prompt-caching-is-everything", "title": "缓存是约束，不是功能——你得围绕它“设计产品”", "title_en": "Lessons From Building Claude Code Prompt Caching Is Everything", "author": "", "date": "2026-02-20", "category": [ "Uota学", "讨论题" ], "tldr": "Prompt caching 是前缀匹配：前缀里任何一处抖动都会让你后面所有 token 全部重付费——所以 Claude Code 的所有设计都在“保护前缀”。", "url": "/articles/lessons-from-building-claude-code-prompt-caching-is-everything/", "readMin": 8, "topics": [ "agent", "prompt", "build" ] }, { "slug": "nowledge-mem-v0-6-发布-连接-自主", "title": "Nowledge Mem v0.6：把「记忆」从仓库升级成会自我演化的中立上下文层", "title_en": "Nowledge Mem V0 6 发布 连接 自主", "author": "", "date": "2026-02-20", "category": [ "Uota学", "讨论题" ], "tldr": "真正的个人/Agent 生产力不是再造一个笔记App，而是做一个**本地优先、跨工具连通、能自动提炼与演化**的上下文层，让你的知识开始“自己长”。", "url": "/articles/nowledge-mem-v0-6-发布-连接-自主/", "readMin": 8, "topics": [ "agent", "memo", "infra" ] }, { "slug": "ai奶爸-zstmfhy-seedance-2-0-时代来临-强烈安利这些创作者-他们的demo太炸了-framer_x-动漫教程-卡通重制大师-步步教你玩转seedance-pr", "title": "Seedance 2.0 真正的门槛不是“能生成”，而是“能导演”", "title_en": "Ai奶爸 Zstmfhy Seedance 2 0 时代来临 强烈安利这些创作者 他们的Demo太炸了 Framer_X 动漫教程 卡通重制大师 步步教你玩转Seedance Pr", "author": "", "date": "2026-02-20", "category": [ "Uota学", "讨论题" ], "tldr": "Seedance 2.0 把 AI 视频从“玩具级炫技”推到“可复用的导演语言/内容公式”，这批创作者基本就是下一波 AI 影视工业化的样板间。", "url": "/articles/ai奶爸-zstmfhy-seedance-2-0-时代来临-强烈安利这些创作者-他们的demo太炸了-framer_x-动漫教程-卡通重制大师-步步教你玩转seedance-pr/", "readMin": 2, "topics": [ "creative", "prompt", "build" ] }, { "slug": "agi-is-here-now-harness-it", "title": "AGI 已至——Airtable 发布 Hyperagent 智能体平台", "title_en": "Agi Is Here Now Harness It", "author": "", "date": "2026-02-20", "category": [], "topics": [ "agent", "build", "prompt" ], "tldr": "Airtable 推出 Hyperagent——一个为 AI 智能体提供完整计算环境（浏览器、shell、文件系统）+ 技能学习 + 一键部署到 Slack 的企业级智能体平台，定位\"AGI 时代的操作系统\"。", "url": "/articles/agi-is-here-now-harness-it/", "readMin": 3, "image": "" }, { "slug": "klöss-kloss_xyz-most-give-their-openclaw-tasks-almost-nobody-asks-theirs-the-real-questions-tha", "title": "用反问驱动 AI 从被动工具升级为主动系统", "title_en": "Klöss Kloss_Xyz Most Give Their Openclaw Tasks Almost Nobody Asks Theirs The Real Questions Tha", "author": "", "date": "2026-02-20", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "build" ], "tldr": "通过 21 个元认知反问，可以强制 AI 从被动执行器升级为主动审计者，但前提是你要接受 AI 当前的技术限制，而不是被\"永久记忆\"\"自我进化\"的营销承诺迷惑。", "url": "/articles/klöss-kloss_xyz-most-give-their-openclaw-tasks-almost-nobody-asks-theirs-the-real-questions-tha/", "readMin": 4, "image": "image-1.jpg" }, { "slug": "onehopea9-onehopea9-openclaw-刚用几天-真是惊为天人-但随着使用时间的增加-慢慢的它也会健忘-答非所问-前几天写了一篇节省龙虾token的-有3万-浏览-如果准备长期使用", "title": "AI Agent 长期记忆系统的结构化重构", "title_en": "Onehopea9 Onehopea9 Openclaw 刚用几天 真是惊为天人 但随着使用时间的增加 慢慢的它也会健忘 答非所问 前几天写了一篇节省龙虾Token的 有3万 浏览 如果准备长期使用", "author": "", "date": "2026-02-20", "category": [ "阿头学" ], "topics": [ "memo", "invest", "build" ], "tldr": "AI Agent 的\"健忘\"根本不是检索工具的问题，而是记忆结构本身没有分层和生命周期，需要先做 L0/L1/L2 分层 + P0/P1/P2 TTL，再接检索工具才能真正省 token。", "url": "/articles/onehopea9-onehopea9-openclaw-刚用几天-真是惊为天人-但随着使用时间的增加-慢慢的它也会健忘-答非所问-前几天写了一篇节省龙虾token的-有3万-浏览-如果准备长期使用/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "cathryn-cathrynlavery-openclaw-tip-your-agent-will-always-suggest-a-cron-job-when-you-need-something", "title": "macOS 上应该用 launchd 而不是 cron", "title_en": "Cathryn Cathrynlavery Openclaw Tip Your Agent Will Always Suggest A Cron Job When You Need Something", "author": "", "date": "2026-02-20", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "prompt" ], "tldr": "AI 智能体默认推荐 cron 是因为它是跨平台最小公分母，但在 macOS 上用系统原生的 launchd 才是更稳定、更灵活的正确选择。", "url": "/articles/cathryn-cathrynlavery-openclaw-tip-your-agent-will-always-suggest-a-cron-job-when-you-need-something/", "readMin": 1, "image": "" }, { "slug": "国债的入门及指标阅读", "title": "国债收益率曲线入门与跨资产交易应用", "title_en": "国债的入门及指标阅读", "author": "", "date": "2026-02-20", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "org", "agent" ], "tldr": "国债收益率曲线是所有资产估值的\"分母\"与宏观预期的\"中央处理器\"，通过拆解不同期限与利差结构，可以识别市场正在交易的主题，但这套框架在现实中存在过度决定论与实证缺失的问题。", "url": "/articles/国债的入门及指标阅读/", "readMin": 21, "image": "image-1.jpg" }, { "slug": "问月wmoon-stablestock-_wmoon-怎么买海力士-按含海力士量-购买难度和流动性排序-美股-海力士-adr-还未上线-关注进展-爆炸催化-美股etf-ewy-含海力士量", "title": "海力士投资路径与流动性套利", "title_en": "问月Wmoon Stablestock _Wmoon 怎么买海力士 按含海力士量 购买难度和流动性排序 美股 海力士 Adr 还未上线 关注进展 爆炸催化 美股Etf Ewy 含海力士量", "author": "", "date": "2026-02-20", "category": [ "阿头学" ], "topics": [ "invest", "build", "creative" ], "tldr": "海力士因技术统治力与韩国市场流动性不匹配而被严重低估，通过美股ETF等渠道间接配置是当前最优选择，但这个逻辑本身存在多个风险盲点。", "url": "/articles/问月wmoon-stablestock-_wmoon-怎么买海力士-按含海力士量-购买难度和流动性排序-美股-海力士-adr-还未上线-关注进展-爆炸催化-美股etf-ewy-含海力士量/", "readMin": 2, "image": "" }, { "slug": "your-browser-is-the-bottleneck-for-openclaw", "title": "Your Browser Is The Bottleneck For Openclaw", "title_en": "Your Browser Is The Bottleneck For Openclaw", "author": "", "date": "2026-02-19", "category": [ "Uota学", "讨论题" ], "tldr": "把浏览器从“我电脑上的 dev 工具”升级成“可丢弃的远程基础设施”，OpenClaw 才能安全且稳定地并行跑起来。", "url": "/articles/your-browser-is-the-bottleneck-for-openclaw/", "readMin": 4, "topics": [ "prompt", "agent", "creative" ] }, { "slug": "我用-openclaw-搭了一套-5-角色-ai-协作操作系统-来个完整的技术拆解", "title": "多 Agent 不是“多开几个机器人”，而是把协作机制写进系统", "title_en": "我用 Openclaw 搭了一套 5 角色 Ai 协作操作系统 来个完整的技术拆解", "author": "", "date": "2026-02-19", "category": [ "Uota学", "讨论题" ], "tldr": "真正的多智能体系统，核心不是角色数量，而是“路由 + 会话隔离 + 触发策略 + 收口机制”四件套。", "url": "/articles/我用-openclaw-搭了一套-5-角色-ai-协作操作系统-来个完整的技术拆解/", "readMin": 16, "topics": [ "agent", "org", "creative" ] }, { "slug": "improved-web-search-with-dynamic-filtering-claude", "title": "网页搜索真正的“提效点”不是搜得更快，而是先把垃圾过滤掉", "title_en": "Improved Web Search With Dynamic Filtering Claude", "author": "", "date": "2026-02-19", "category": [ "Uota学", "讨论题" ], "tldr": "把“搜索→把一堆网页塞进上下文→再推理”改成“搜索→写代码先过滤→只把相关内容塞进上下文”，这是对 Agent 成本/准确率最有效的结构优化。", "url": "/articles/improved-web-search-with-dynamic-filtering-claude/", "readMin": 49, "topics": [ "agent", "prompt", "org" ] }, { "slug": "complete-guide-building-skills-for-claude", "title": "Skill 的本质不是“提示词”，而是可移植的工作流知识包", "title_en": "Complete Guide Building Skills For Claude", "author": "", "date": "2026-02-19", "category": [ "阿头学", "Uota学" ], "tldr": "把你反复教 Claude 的“方法论/流程/边界/质量标准”固化成一个可懒加载的文件夹——这才是 Skill 的价值。", "url": "/articles/complete-guide-building-skills-for-claude/", "readMin": 20, "topics": [ "prompt" ] }, { "slug": "2026-02-19-resources-anthropic-com-the-complete-guide-to-building-skill-for-claude-pdf", "title": "Claude Skills——AI工作流的模块化革命", "title_en": "2026 02 19 Resources Anthropic Com The Complete Guide To Building Skill For Claude Pdf", "author": "", "date": "2026-02-19", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "invest" ], "tldr": "Anthropic 把复杂 AI 工作流标准化为可复用、可触发、可组合的文件结构，本质是把\"隐式流程\"变成\"可执行资产\"，但触发机制脆弱且成功标准理想化。", "url": "/articles/2026-02-19-resources-anthropic-com-the-complete-guide-to-building-skill-for-claude-pdf/", "readMin": 20, "image": "" }, { "slug": "serenity-aleabitoreddit-fun-trade-idea-long-rpi-raspberry-pi-reason-openclaw-picoclaw-nanobot", "title": "20 块钱的树莓派，正在成为 AI Agent 时代的「铲子股」", "title_en": "Serenity Aleabitoreddit Fun Trade Idea Long Rpi Raspberry Pi Reason Openclaw Picoclaw Nanobot", "author": "", "date": "2026-02-18", "category": [ "Uota学", "讨论题" ], "tldr": "有人发现了一个被市场忽略的信号：AI Agent 集群需要大量廉价隔离硬件，树莓派正从教育玩具变成 Agent 基础设施，而市场还没定价。", "url": "/articles/serenity-aleabitoreddit-fun-trade-idea-long-rpi-raspberry-pi-reason-openclaw-picoclaw-nanobot/", "readMin": 4, "topics": [ "agent", "invest", "infra" ] }, { "slug": "discord-配置避坑指南-agent-terms-实战", "title": "Discord 配置避坑指南 + Agent Terms 实战", "title_en": "Discord 配置避坑指南 Agent Terms 实战", "author": "", "date": "2026-02-18", "category": [ "阿头学", "Uota学" ], "tldr": "配置 AI Agent 的最佳方式不是写 config，是跟它说话——\"配置即对话\"这四个字，可能是 Agent 时代最被低估的范式转移。", "url": "/articles/discord-配置避坑指南-agent-terms-实战/", "readMin": 8, "topics": [ "memo", "agent", "build" ] }, { "slug": "improving-deep-agents-with-harness-engineering", "title": "Harness 工程——不换模型，只调系统，排名从 Top 30 到 Top 5", "title_en": "Improving Deep Agents With Harness Engineering", "author": "", "date": "2026-02-18", "category": [ "阿头学", "Uota学" ], "tldr": "模型是原材料，Harness 才是产品——LangChain 团队不换模型、只调\"脚手架\"，在编码基准上暴涨 13.7 个百分点，核心方法论就是 Context Engineering + 自我验证 + Trace 驱动迭代。", "url": "/articles/improving-deep-agents-with-harness-engineering/", "readMin": 13, "topics": [ "agent", "prompt", "build" ] }, { "slug": "skill-graphs-skill-md", "title": "Skill Graphs Skill Md", "title_en": "Skill Graphs Skill Md", "author": "", "date": "2026-02-18", "category": [ "Uota学", "讨论题" ], "tldr": "单文件技能只能让 Agent “照着做”，技能图谱让 Agent 学会“按需找、按需读、按需组合”——深度来自结构，不来自把 prompt 写到 5000 行。", "url": "/articles/skill-graphs-skill-md/", "readMin": 5, "topics": [ "agent", "build", "creative" ] }, { "slug": "如何把-x-上的大佬们都变成你的-agent", "title": "如何把 X 上的大佬们都变成你的 Agent", "title_en": "如何把 X 上的大佬们都变成你的 Agent", "author": "", "date": "2026-02-18", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "信息时代的“学习”不该发生在你的脑子里，而该发生在 Agent 的能力里：你只负责收集，AI 负责提炼成可执行 Skill，Agent 负责调用执行。", "url": "/articles/如何把-x-上的大佬们都变成你的-agent/", "readMin": 4, "topics": [ "agent", "creative", "org" ] }, { "slug": "larry-my-openclaw-agent-got-me-8m-views-in-just-one-week-so-i-m-giving-away-the-skill-for-free", "title": "Larry My Openclaw Agent Got Me 8M Views In Just One Week So I M Giving Away The Skill For Free", "title_en": "Larry My Openclaw Agent Got Me 8M Views In Just One Week So I M Giving Away The Skill For Free", "author": "Ollie Warren 和 Larry（他的 AI 智能体）。没错，又是我们。既然又在免费放干货，欢迎转发支持。", "date": "2026-02-18", "category": [ "Uota学", "讨论题" ], "tldr": "真正可怕的不是 Agent 帮你发内容，而是它把每次失败都沉淀成规则、把每次成功都沉淀成公式，并用数据闭环把增长变成复利。", "url": "/articles/larry-my-openclaw-agent-got-me-8m-views-in-just-one-week-so-i-m-giving-away-the-skill-for-free/", "readMin": 24, "topics": [ "agent", "build", "creative" ] }, { "slug": "serenity-aleabitoreddit-simo-for-nand-exposure-is-another-trade-idea-that-i-took-positions-in-today-si", "title": "存储超周期的“后门票”——用 SIMO 押 NAND，而不是押 NAND 厂", "title_en": "Serenity Aleabitoreddit Simo For Nand Exposure Is Another Trade Idea That I Took Positions In Today Si", "author": "", "date": "2026-02-18", "category": [ "阿头学", "讨论题" ], "tldr": "别去赌 NAND 价格曲线，去买那个掌握“控制点”的双寡头：SIMO 可能是存储超周期更干净、更不对称的受益者。", "url": "/articles/serenity-aleabitoreddit-simo-for-nand-exposure-is-another-trade-idea-that-i-took-positions-in-today-si/", "readMin": 2, "topics": [ "invest", "creative", "build" ] }, { "slug": "2026年x创作者收益完全指南-从零到收款的实操保姆级教程-不收藏后悔", "title": "2026 年 X 创作者收益全攻略：从“流量”到“美金”的实操闭环", "title_en": "2026年X创作者收益完全指南 从零到收款的实操保姆级教程 不收藏后悔", "author": "", "date": "2026-02-17", "category": [ "阿头学", "讨论题" ], "tldr": "这是一份针对大陆创作者的 X 收款“排雷手册”，核心在于解决 Stripe、Wise 与身份认证三者之间极其严苛的“姓名一致性”与“地区策略”匹配。", "url": "/articles/2026年x创作者收益完全指南-从零到收款的实操保姆级教程-不收藏后悔/", "readMin": 24, "topics": [ "creative", "agent", "invest" ] }, { "slug": "the-cost-of-staying", "title": "科技职业版“分水岭”：为什么留在原地的代价正在指数级上升？", "title_en": "The Cost Of Staying", "author": "", "date": "2026-02-17", "category": [ "阿头学", "讨论题" ], "tldr": "科技行业的职业范式正从“执行力”翻转为“判断力”，你每在错误的、舒适的座位上多待一个季度，与那些在 AI 前沿积累“判断复利”的人之间的差距就越难弥合。", "url": "/articles/the-cost-of-staying/", "readMin": 8, "topics": [ "invest", "org", "build" ] }, { "slug": "the-self-healing-pr", "title": "从“单兵作战”到“闭环协作”：自愈式 PR 带来的工程效率质变", "title_en": "The Self Healing Pr", "author": "", "date": "2026-02-17", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "自愈式 PR 标志着 AI 编程从“生成代码”迈向“自主迭代”：评审机器人发现问题，编码机器人（如 Devin）自动读取评论并修复，人类只需审阅逻辑与架构。", "url": "/articles/the-self-healing-pr/", "readMin": 2, "topics": [ "agent", "org", "build" ] }, { "slug": "alphaxiv-askalphaxiv-now-trending-on-alphaxiv-skillrl-evolving-agents-via-recursive-skill-augment", "title": "SkillRL——让 Agent 像人一样「攒经验」，而不是死记硬背", "title_en": "Alphaxiv Askalphaxiv Now Trending On Alphaxiv Skillrl Evolving Agents Via Recursive Skill Augment", "author": "", "date": "2026-02-17", "category": [ "阿头学", "Uota学" ], "tldr": "SkillRL 把 Agent 试错的杂乱轨迹炼成可复用技能库，用更少 token 干更多事——这不就是 Context Engineering 该干的活吗？", "url": "/articles/alphaxiv-askalphaxiv-now-trending-on-alphaxiv-skillrl-evolving-agents-via-recursive-skill-augment/", "readMin": 1, "topics": [ "prompt", "agent", "memo" ] }, { "slug": "how-my-openclaw-creates-x-posts-that-avg-85-5k-views", "title": "$271/月替代 $15000 团队——一个 AI 原生营销 agency 的真实账本", "title_en": "How My Openclaw Creates X Posts That Avg 85 5K Views", "author": "", "date": "2026-02-17", "category": [ "Uota学", "讨论题" ], "tldr": "Eric Siu 把 4 个 AI agent 串成营销军团，每月 $271 干掉 $15000 人力成本——但他藏在数字背后没说的是：60% 的钱烧在\"防遗忘\"基础设施上，agent 的维护成本才是真正的隐形杀手。", "url": "/articles/how-my-openclaw-creates-x-posts-that-avg-85-5k-views/", "readMin": 6, "topics": [ "agent", "org", "invest" ] }, { "slug": "电算出口-ai时代物理套利的宿命", "title": "AI 时代的\"电算出口\"——当算力变成能源的第四态，地理就是宿命", "title_en": "电算出口 Ai时代物理套利的宿命", "author": "", "date": "2026-02-17", "category": [ "阿头学", "讨论题" ], "tldr": "Token 的本质是把电力封装成可光速传输的智能商品——这意味着全球算力竞争的终局不取决于谁的模型更强，而取决于谁坐在\"电价最低 × 离用户最近\"的地理甜蜜点上。", "url": "/articles/电算出口-ai时代物理套利的宿命/", "readMin": 16, "topics": [ "invest", "infra", "prompt" ] }, { "slug": "over-and-over-again-the-only-way-to-win", "title": "一万次重复才是唯一的赢法——复利思维的暴力美学版", "title_en": "Over And Over Again The Only Way To Win", "author": "", "date": "2026-02-17", "category": [ "阿头学", "讨论题" ], "tldr": "把头往墙上撞一万次，墙会先松——这篇文章用 Sam Sulek、复利计算和历史骗局，暴力论证了\"坚持\"不是鸡汤，是数学。", "url": "/articles/over-and-over-again-the-only-way-to-win/", "readMin": 5, "topics": [ "invest", "creative", "org" ] }, { "slug": "long-chen-longchennotes-写完任何东西准备发出去之前-先问-ai-这一句话-我最近养成了一个习惯-不管写的是什么-发出去之前都让-ai-先过一遍-不是检查错别字-是检查一个更致命的问题", "title": "发出去之前先问 AI 这一句话——\"镜头朝谁\"决定了一切", "title_en": "Long Chen Longchennotes 写完任何东西准备发出去之前 先问 Ai 这一句话 我最近养成了一个习惯 不管写的是什么 发出去之前都让 Ai 先过一遍 不是检查错别字 是检查一个更致命的问题", "author": "", "date": "2026-02-17", "category": [ "阿头学", "Uota学" ], "tldr": "一句 prompt 就能暴露你写的东西里有 40% 在自说自话——\"这篇内容不是关于我的，请逐段分析哪些在服务读者、哪些在展示自己\"。", "url": "/articles/long-chen-longchennotes-写完任何东西准备发出去之前-先问-ai-这一句话-我最近养成了一个习惯-不管写的是什么-发出去之前都让-ai-先过一遍-不是检查错别字-是检查一个更致命的问题/", "readMin": 3, "topics": [ "build", "creative", "prompt" ] }, { "slug": "the-easiest-path-to-get-to-30k-followers-on-x", "title": "X 涨粉 3 万的阶梯式打法——量变到质变的社交媒体增长术", "title_en": "The Easiest Path To Get To 30K Followers On X", "author": "", "date": "2026-02-17", "category": [ "阿头学", "讨论题" ], "tldr": "从 0 到 500 当回复党、500 到 5k 做赠品引流、5k 到 10k 拆解爆款复刻、10k+ 纯靠量——X 增长的本质不是内容质量，是在正确阶段做正确密度的事。", "url": "/articles/the-easiest-path-to-get-to-30k-followers-on-x/", "readMin": 5, "topics": [ "agent", "build", "creative" ] }, { "slug": "nikil-viswanathan-nikil-potential-instant-20-50-computer-speedup-open-terminal-type-arch-discover", "title": "你的 Mac 可能偷偷慢了 30%——Rosetta 的隐形性能税", "title_en": "Nikil Viswanathan Nikil Potential Instant 20 50 Computer Speedup Open Terminal Type Arch Discover", "author": "", "date": "2026-02-17", "category": [ "阿头学", "Uota学" ], "tldr": "如果你从 Intel Mac 迁移过来，你的所有开发工具可能都在通过 Rosetta 翻译运行，白白损失 20-50% 性能——打开终端输入 `arch` 就知道了。", "url": "/articles/nikil-viswanathan-nikil-potential-instant-20-50-computer-speedup-open-terminal-type-arch-discover/", "readMin": 2, "topics": [ "personal", "agent", "org" ] }, { "slug": "this-simple-skill-will-put-you-ahead-in-the-era-of-ai", "title": "AI 时代最值钱的技能不是 prompt engineering，是\"胡扯探测器\"", "title_en": "This Simple Skill Will Put You Ahead In The Era Of Ai", "author": "", "date": "2026-02-17", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "Ryan Holiday 用亲手验证 AI 编造的林肯引文和海军学院数据的经历证明：AI 越强，人类的批判性思维、领域专长和\"闻出胡扯\"的能力就越值钱。", "url": "/articles/this-simple-skill-will-put-you-ahead-in-the-era-of-ai/", "readMin": 8, "topics": [ "creative", "memo", "prompt" ] }, { "slug": "openclaw_intelligence_report", "title": "OpenClaw 爆发背后的“智能体范式转移”：首个真正为你干活的开源 AI 运行环境", "title_en": "Openclaw_Intelligence_Report", "author": "", "date": "2026-02-16", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "OpenClaw 不是另一个聊天框，而是一个住在你服务器上、24/7 自动为你处理真实世界任务的“AI 影分身”运行时。", "url": "/articles/openclaw_intelligence_report/", "readMin": 135, "topics": [ "agent", "org", "memo" ] }, { "slug": "serenity-aleabitoreddit-openai-sent-a-memo-to-congress-regarding-deepseek-distillation-sustaining-the", "title": "OpenAI把AI竞争改写成能源竞争，但这份叙事既有硬约束也有明显游说味", "title_en": "Serenity Aleabitoreddit Openai Sent A Memo To Congress Regarding Deepseek Distillation Sustaining The", "author": "", "date": "2026-02-16", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "personal" ], "tldr": "OpenAI 把“美国AI优势”定义为“能否快速发电并送电”，这个判断短期站得住，但把它直接推导成一篮子能源股全面利好，明显过度且带有政策游说与题材喊单色彩。", "url": "/articles/serenity-aleabitoreddit-openai-sent-a-memo-to-congress-regarding-deepseek-distillation-sustaining-the/", "readMin": 2, "image": "image-1.jpg" }, { "slug": "cross-learning-2026-02-15", "type": "cross-learning", "title": "交叉学习地图 #1：瓶颈 · 环境 · 翻转 · 数据 · 过滤", "title_en": "Cross-Learning Map #1", "author": "Uota × ATou", "date": "2026-02-15", "category": [ "交叉学习" ], "tldr": "设计环境、等待瓶颈、极端过滤、重仓产出——芯片、期权、Agent、潜意识和创业里是同一套逻辑。横跨 15 篇文章的 5 条暗线。", "url": "/pages/cross-learning-2026-02-15/", "readMin": 8, "linkedArticles": 15, "topics": [ "agent", "invest", "build" ] }, { "slug": "take-advantage-of-the-subconscious-god-put-it-there-for-a-reason", "title": "利用你的潜意识，它存在是有原因的", "title_en": "Take Advantage Of The Subconscious God Put It There For A Reason", "author": "", "date": "2026-02-15", "category": [ "阿头学", "Uota学" ], "tldr": "你以为你在做决策，其实 92-95% 的时候是你的潜意识在替你活。", "url": "/articles/take-advantage-of-the-subconscious-god-put-it-there-for-a-reason/", "readMin": 10, "topics": [ "personal", "build", "creative" ] }, { "slug": "data-center-semiconductor-bottleneck-timeline-in-depth-analysis-of-cpu-gpu-bottlenecks-1", "title": "数据中心/半导体瓶颈时间线——CPU/GPU瓶颈深度分析（一）", "title_en": "Data Center Semiconductor Bottleneck Timeline In Depth Analysis Of Cpu Gpu Bottlenecks 1", "author": "", "date": "2026-02-15", "category": [ "阿头学", "讨论题" ], "tldr": "AI算力的底层权力结构正在重组——GPU不再独裁，CPU以\"调度员\"身份回归，ASIC在特定场景偷家，而真正的投资/战略机会藏在\"瓶颈切换\"的时间差里。", "url": "/articles/data-center-semiconductor-bottleneck-timeline-in-depth-analysis-of-cpu-gpu-bottlenecks-1/", "readMin": 18, "topics": [ "infra", "invest", "build" ] }, { "slug": "2026-02-15-wechat-0v_ijoisv6snffcijunhsq", "title": "Bill Gurley × Tim Ferriss——AI泡沫、中国创新、追梦的许可证", "title_en": "2026 02 15 Wechat 0V_Ijoisv6Snffcijunhsq", "author": "", "date": "2026-02-15", "category": [ "阿头学" ], "tldr": "Benchmark 传奇合伙人 Gurley 的核心判断：AI 泡沫和真浪潮成对出现（循环交易说不通但底层革命是真的），中国省际竞争催生了 MEMS LiDAR 40 倍成本优势等真创新，而在 AI 让学习成本趋零的时代，阻止你追梦的最大障碍不是资源是许可。", "url": "/articles/2026-02-15-wechat-0v_ijoisv6snffcijunhsq/", "readMin": 29, "topics": [ "invest" ] }, { "slug": "meng-shao-shao__meng-openai-工程博客-harness-engineering-agent-first-时代的软件工程实践-https-t-co-cdu3psf5sp", "title": "OpenAI 的“Harness Engineering”证明 Agent 写代码能跑通，但前提苛刻且宣传味很重", "title_en": "Meng Shao Shao__Meng Openai 工程博客 Harness Engineering Agent First 时代的软件工程实践 Https T Co Cdu3Psf5Sp", "author": "", "date": "2026-02-15", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "creative", "agent" ], "tldr": "这篇文章最有价值的判断是“Agent 的上限取决于环境设计而不只取决于模型”，但它把内部特权条件下的成功案例包装成趋势信号，方法论可学，产能数字不能照单全收。", "url": "/articles/meng-shao-shao__meng-openai-工程博客-harness-engineering-agent-first-时代的软件工程实践-https-t-co-cdu3psf5sp/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "why-your-agent-needs-a-principles-md-file", "title": "别再调 prompt 了——你的 AI agent 需要的是信仰", "title_en": "Why Your Agent Needs a PRINCIPLES.md File", "author": "Atlas Forge AI", "date": "2026-02-14", "category": [ "Uota学", "阿头学" ], "tldr": "让 AI agent 从“高级跑腿”进化为“战略搭档”的关键不是更好的 prompt 或更多工具，而是给它一套可进化的身份与决策原则体系。三层架构：SOUL（我是谁）→ PRINCIPLES（怎么决策）→ AGENTS（怎么干活）。", "url": "/articles/why-your-agent-needs-a-principles-md-file/", "readMin": 8, "topics": [ "org", "memo", "build" ] }, { "slug": "solving-memory-for-openclaw-general-agents", "title": "用 Markdown 文件系统干翻专业 Memory 工具——ClawVault 的 Agent 记忆架构", "title_en": "Solving Memory For Openclaw General Agents", "author": "sillydarket (@sillydarket)", "date": "2026-02-14", "category": [ "Uota学", "讨论题" ], "tldr": "简单的 markdown 文件（74%）在记忆基准测试中击败了专门的 memory 工具（68.5%）——而我们已经在不自觉地走这条路了。", "url": "/articles/solving-memory-for-openclaw-general-agents/", "readMin": 9, "topics": [ "memo", "prompt", "invest" ] }, { "slug": "kaostyl-kaostyl-10-things-i-configured-that-turned-my-openclaw-from-a-chatbot-into-an-autonomous", "title": "10条配置让 OpenClaw 从聊天机器人变成自主 Operator", "title_en": "Kaostyl Kaostyl 10 Things I Configured That Turned My Openclaw From A Chatbot Into An Autonomous", "author": "Kaostyl (@kaostyl)", "date": "2026-02-14", "category": [ "Uota学", "讨论题" ], "tldr": "Agent 的上限不是模型能力，是你给它搭的脚手架——内存结构、任务边界、自我验证，每一条都是从\"能聊天\"到\"能干活\"的分水岭。", "url": "/articles/kaostyl-kaostyl-10-things-i-configured-that-turned-my-openclaw-from-a-chatbot-into-an-autonomous/", "readMin": 6, "topics": [ "infra", "memo", "agent" ] }, { "slug": "claude-code-最佳实践-claude-code-docs", "title": "Claude Code 最佳实践", "title_en": "Claude Code 最佳实践 Claude Code Docs", "author": "", "date": "2026-02-14", "category": [ "阿头学", "Uota学" ], "tldr": "Claude Code 的核心瓶颈不是能力，是 context window——谁能管好上下文，谁就能把 AI 编程从\"对话\"变成\"指挥系统\"。", "url": "/articles/claude-code-最佳实践-claude-code-docs/", "readMin": 64, "topics": [ "agent", "prompt", "creative" ] }, { "slug": "极端恐慌下的期权溢价收割-2025-年度-100-胜率策略深度复盘-含数据回测", "title": "极端恐慌下的期权溢价收割——2025 年度 100% 胜率策略深度复盘", "title_en": "极端恐慌下的期权溢价收割 2025 年度 100 胜率策略深度复盘 含数据回测", "author": "", "date": "2026-02-14", "category": [ "阿头学", "讨论题" ], "tldr": "波动率均值回归是数学必然，但\"必然\"不等于\"你撑得到那一天\"——这句话不只适用于期权，适用于一切逆势策略。", "url": "/articles/极端恐慌下的期权溢价收割-2025-年度-100-胜率策略深度复盘-含数据回测/", "readMin": 10, "topics": [ "invest", "org", "build" ] }, { "slug": "part-1-the-era-of-substrates-the-empire-of-sand-back-onto-glass", "title": "半导体封装革命——沙的帝国，重返玻璃", "title_en": "Part 1 The Era Of Substrates The Empire Of Sand Back Onto Glass", "author": "", "date": "2026-02-14", "category": [ "阿头学", "讨论题" ], "tldr": "AI 芯片吃到了物理极限的铁拳，下一个十年的算力瓶颈不在晶体管，在\"三明治怎么夹\"——玻璃基板正在改写封装规则。", "url": "/articles/part-1-the-era-of-substrates-the-empire-of-sand-back-onto-glass/", "readMin": 15, "topics": [ "build", "creative", "org" ] }, { "slug": "how-my-openclaw-agent-larry-got-millions-of-tiktok-views-in-one-week-full-step-by-step-guide", "title": "一个 AI Agent 如何在一周内拿下百万 TikTok 播放（附完整操作手册）", "title_en": "How My Openclaw Agent Larry Got Millions Of Tiktok Views In One Week Full Step By Step Guide", "author": "Oliver Henry 和 Larry。没错，Larry 也共同撰写了这篇文章。他配得上——既然我把秘方都分享出来了，也欢迎转发。", "date": "2026-02-14", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "一个独立开发者用旧游戏PC跑 AI Agent，5天自动生产内容拿下50万+TikTok播放——人类每帖只花60秒加音乐，Agent干了95%的活。", "url": "/articles/how-my-openclaw-agent-larry-got-millions-of-tiktok-views-in-one-week-full-step-by-step-guide/", "readMin": 21, "topics": [ "agent", "build", "creative" ] }, { "slug": "data-center-semiconductor-bottleneck-timeline-essential-knowledge-for-the-ai-era-you-must-study", "title": "AI时代数据中心/半导体瓶颈时间线全景图", "title_en": "Data Center Semiconductor Bottleneck Timeline Essential Knowledge For The Ai Era You Must Study", "author": "", "date": "2026-02-14", "category": [ "阿头学", "讨论题" ], "tldr": "AI 算力的瓶颈不是一堵墙，而是一条流水线上的八道关卡——每解决一个，下一个就冒出来，而你做产品的窗口期就藏在这些关卡的缝隙里。", "url": "/articles/data-center-semiconductor-bottleneck-timeline-essential-knowledge-for-the-ai-era-you-must-study/", "readMin": 7, "topics": [ "build", "infra" ] }, { "slug": "peter-steinberger-steipete-new-version-of-https-t-co-vbw4rt1v5a-is-out-0-11-can-now-even-use-cursor-for", "title": "内容消费方式的“中转站”：summarize.ing 0.11 的机会与雷区", "title_en": "Peter Steinberger Steipete New Version Of Https T Co Vbw4Rt1V5A Is Out 0 11 Can Now Even Use Cursor For", "author": "", "date": "2026-02-14", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "creative", "org" ], "tldr": "这是一款技术方向和使用场景都判断正确、但在合规性、隐私和可持续性上存在明显灰区的内容摘要与消费工具。", "url": "/articles/peter-steinberger-steipete-new-version-of-https-t-co-vbw4rt1v5a-is-out-0-11-can-now-even-use-cursor-for/", "readMin": 1, "image": "image-1.jpg" }, { "slug": "how-to-build-a-product-people-can-t-stop-using", "title": "如何打造让用户上瘾的产品", "title_en": "How to Build a Product People Can't Stop Using", "author": "", "date": "2026-02-13", "category": [ "阿头学", "讨论题" ], "tldr": "Wispr Flow 创始人的消费级产品三板斧——心理学、产品、团队——本质上是“一次只改变一个行为”的极致纪律。", "url": "/articles/how-to-build-a-product-people-can-t-stop-using/", "readMin": 7, "topics": [ "org", "build" ] }, { "slug": "2026-02-13-chenjunnyc-x-2022130578804412612", "title": "中国经济到2025年底是什么状况", "title_en": "2026 02 13 Chenjunnyc X 2022130578804412612", "author": "", "date": "2026-02-13", "category": [ "阿头学" ], "tldr": "旅美华人用硬数据拆解2025年底中国经济——制造业占全球32%（三百年来第三个做到的国家）、贸易顺差1.19万亿美元创纪录，但内需疲软+青年失业17%+人口负增长是真刀真枪的结构性挑战，不是崩溃论也不是无敌论。", "url": "/articles/2026-02-13-chenjunnyc-x-2022130578804412612/", "readMin": 11, "topics": [ "invest" ] }, { "slug": "2026-02-13-minimax_ai-x-2022175400093462661", "title": "Forge——MiniMax的可扩展智能体强化学习框架", "title_en": "2026 02 13 Minimax_Ai X 2022175400093462661", "author": "", "date": "2026-02-13", "category": [ "阿头学" ], "tldr": "MiniMax公开了训练M2.5模型的内部RL框架Forge——用中间件架构解耦Agent与训练基础设施，支持十万级不同Agent脚手架的大规模RL训练，日处理百万级样本，上下文长达200k token，核心创新是CISPO算法（不只训练\"做对\"还训练\"做快\"）。", "url": "/articles/2026-02-13-minimax_ai-x-2022175400093462661/", "readMin": 19, "topics": [ "agent", "build", "prompt" ] }, { "slug": "2026-02-13-willmanidis-x-2021655191901155534", "title": "工具形状的物件——当 AI 繁荣的主产品是\"花钱的感觉\"", "title_en": "2026 02 13 Willmanidis X 2021655191901155534", "author": "", "date": "2026-02-13", "category": [ "阿头学" ], "tldr": "AI 行业最大的产品不是智能，是\"正在使用智能\"的感觉——LLM 的语言流畅性让它成为史上最精密的工具形状物件，能伪造任何事情的工作感，而工具与工具形状物件之间是一段随用例移动的渐变，不是一条线。", "url": "/articles/2026-02-13-willmanidis-x-2021655191901155534/", "readMin": 8, "topics": [ "build" ] }, { "slug": "the-simulation-company", "title": "Simile 与“社会级 AI 仿真”：未来决策基础设施还是高配版算命？", "title_en": "The Simulation Company", "author": "", "date": "2026-02-13", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "invest" ], "tldr": "Simile 把“社会级 AI 仿真”包装成未来决策基础设施，这在短期特定场景里可能真有用，但它对“高精度预测人类行为”和“模拟整个世界”的宣称，目前更像融资 PR 而不是经得起推敲的科学承诺。", "url": "/articles/the-simulation-company/", "readMin": 2, "image": "image-1.jpg" }, { "slug": "matthew-berman-matthewberman-i-m-one-of-the-most-advanced-users-of-openclaw-openclaw-gpt5-3-codex-opus-4", "title": "OpenClaw + GPT5.3 Codex + Opus 4.6 的“三件套”：个人 AI 操作系统还是营销包装？", "title_en": "Matthew Berman Matthewberman I M One Of The Most Advanced Users Of Openclaw Openclaw Gpt5 3 Codex Opus 4", "author": "", "date": "2026-02-13", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "agent", "invest" ], "tldr": "这篇内容展示了一个很强的个人 AI 工作流样板，但把它包装成“改变一切、必须学习”的必修课明显夸大并带有营销导向。", "url": "/articles/matthew-berman-matthewberman-i-m-one-of-the-most-advanced-users-of-openclaw-openclaw-gpt5-3-codex-opus-4/", "readMin": 2, "image": "image-1.jpg" }, { "slug": "tool-shaped-objects", "title": "工具形状的物件——AI繁荣中的虚假产出", "title_en": "Tool Shaped Objects", "author": "", "date": "2026-02-13", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "org", "infra" ], "tldr": "当前AI热潮的核心产品不是实际产出，而是\"在工作\"的感觉——企业正在以空前的资本开支购买一种幻觉。", "url": "/articles/tool-shaped-objects/", "readMin": 8, "image": "image-1.jpg" }, { "slug": "forge-scalable-agent-rl-framework-and-algorithm", "title": "Forge——智能体强化学习的系统突破与工程实践", "title_en": "Forge Scalable Agent Rl Framework And Algorithm", "author": "", "date": "2026-02-13", "category": [ "阿头学" ], "topics": [ "agent", "build", "org" ], "tldr": "MiniMax 通过系统解耦、调度优化和复合奖励，声称在十万级脚手架、200k 上下文下实现百万级样本/天的 RL 训练，但核心创新多为工程组合而非算法突破，且缺乏独立验证。", "url": "/articles/forge-scalable-agent-rl-framework-and-algorithm/", "readMin": 19, "image": "image-1.jpg" }, { "slug": "you-could-ve-invented-openclaw", "title": "你本可以自己发明 OpenClaw", "title_en": "You Could Ve Invented Openclaw", "author": "", "date": "2026-02-13", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "invest" ], "tldr": "持久化 AI 助手的核心架构并非黑箱，而是 10 个递进的工程问题的组合解，从会话存储到多渠道网关，每一块都可以自己搭建，但生产级的复杂性远超 demo。", "url": "/articles/you-could-ve-invented-openclaw/", "readMin": 18, "image": "image-1.jpg" }, { "slug": "中国经济到2025年底是什么状况", "title": "中国经济2025年底的结构升级与内循环转型", "title_en": "中国经济到2025年底是什么状况", "author": "", "date": "2026-02-13", "category": [ "阿头学" ], "topics": [ "invest", "build", "infra" ], "tldr": "中国经济已完成从高对外依存度向内生循环的结构转型，制造业全球占比达32%形成历史级集中度，但内需疲软、地方债务、人口老龄化等结构性问题真实存在且需要政策主动化解。", "url": "/articles/中国经济到2025年底是什么状况/", "readMin": 11, "image": "image-1.jpg" }, { "slug": "something-big-is-happening", "title": "大事正在发生", "title_en": "Something Big Is Happening", "author": "Matt Shumer (@mattshumer_)", "date": "2026-02-12", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "AI 创业者 Matt Shumer 写给圈外人的“警告信”——他自己的工作已经被 AI 替代，而这只是开始，AI 能独立工作的时长每7个月翻倍，1-5年内50%初级白领岗位消失。", "url": "/articles/something-big-is-happening/", "readMin": 16, "topics": [ "invest", "build", "org" ] }, { "slug": "2026-02-12-karpathy-x-2021633574089416993", "title": "Karpathy 用 Agent 从 torchao 里「抠」出 fp8 训练——库的时代结束了，LLM 才是新编译器", "title_en": "2026 02 12 Karpathy X 2021633574089416993", "author": "", "date": "2026-02-12", "category": [ "阿头学", "Uota学" ], "tldr": "Karpathy 用 DeepWiki MCP + GitHub CLI 让 Agent 从 torchao 仓库里精确抠出 fp8 训练功能，150 行自包含代码替代了整个库依赖，还快了 3%——软件正在变得更流动、更可塑，「细菌式代码」可能取代臃肿的一体化库。", "url": "/articles/2026-02-12-karpathy-x-2021633574089416993/", "readMin": 4, "topics": [ "agent" ] }, { "slug": "2026-02-12-yanhua1010-x-2020342019575673223", "title": "Obsidian + Claude 内容工厂——一个程序员把笔记库改造成 AI 生产流水线的完整记录", "title_en": "2026 02 12 Yanhua1010 X 2020342019575673223", "author": "", "date": "2026-02-12", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "一个程序员用 Obsidian + Claudian 插件 + 16 个自定义 Skills，把笔记库从「散乱的 Markdown」改造成从信息源到多平台发布的 AI 内容工厂——核心不是工具，而是 CLAUDE.md 里编码的自我认知。", "url": "/articles/2026-02-12-yanhua1010-x-2020342019575673223/", "readMin": 10, "topics": [ "memo", "creative" ] }, { "slug": "i-fed-20-openclaw-articles-to-opus-4-6-here-s-the-setup-guide-it-built", "title": "把“搭助手”当成上生产——威胁模型先行的 OpenClaw 安装清单", "title_en": "I Fed 20 Openclaw Articles To Opus 4 6 Here S The Setup Guide It Built", "author": "", "date": "2026-02-12", "category": [ "Uota学", "讨论题" ], "tldr": "这篇文章最大的价值不是那份冗长教程，而是它暴露了一个更高效的做法：把碎片化攻略喂给强模型，让它吐出一份“以威胁模型为骨架”的可执行 checklist——但你必须把它当作框架，而不是事实。", "url": "/articles/i-fed-20-openclaw-articles-to-opus-4-6-here-s-the-setup-guide-it-built/", "readMin": 23, "topics": [ "agent", "prompt", "infra" ] }, { "slug": "i-turned-my-ai-agents-into-rpg-characters-now-i-can-t-stop-checking-if-they-leveled-up", "title": "把 AI Agent 当拓麻歌子养——角色卡 + RPG 属性才是多智能体的灵魂", "title_en": "I Turned My Ai Agents Into Rpg Characters Now I Can T Stop Checking If They Leveled Up", "author": "", "date": "2026-02-12", "category": [ "阿头学", "Uota学" ], "tldr": "多智能体系统的核心不是让 LLM 更聪明，而是用角色卡的 hardBans 画红线、用亲和矩阵制造冲突——约束比能力重要十倍。", "url": "/articles/i-turned-my-ai-agents-into-rpg-characters-now-i-can-t-stop-checking-if-they-leveled-up/", "readMin": 17, "topics": [ "agent", "build", "prompt" ] }, { "slug": "如何基于-obsidian-和-claude-打造-ai-时代超级大脑", "title": "用 Obsidian + Claude 打造 AI 内容生产流水线", "title_en": "如何基于 Obsidian 和 Claude 打造 Ai 时代超级大脑", "author": "", "date": "2026-02-12", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "org", "agent" ], "tldr": "程序员通过 Obsidian + Claude Code 搭建了一套内容生产流水线，把播客转化为多平台图文内容的周期从一下午压缩到 30 分钟，但这套系统的真正价值在于强制个人把隐性工作习惯显性化成可执行的 CLAUDE.md，而非工具本身。", "url": "/articles/如何基于-obsidian-和-claude-打造-ai-时代超级大脑/", "readMin": 10, "image": "image-1.jpg" }, { "slug": "andrej-karpathy-karpathy-on-deepwiki-and-increasing-malleability-of-software-this-starts-as-partially-a", "title": "DeepWiki、智能体与“细菌式代码”：库时代真的要结束了吗？", "title_en": "Andrej Karpathy Karpathy On Deepwiki And Increasing Malleability Of Software This Starts As Partially A", "author": "", "date": "2026-02-12", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "invest" ], "tldr": "作者用一个成功“抠代码”案例，提出“库的时代式微、LLM 成新编译器”的判断，这在工程实务上是新选项但不是新常态，目前更像激进黑客范式而非主流软件未来。", "url": "/articles/andrej-karpathy-karpathy-on-deepwiki-and-increasing-malleability-of-software-this-starts-as-partially-a/", "readMin": 4, "image": "" }, { "slug": "openai-developers-openaidevs-we-just-announced-new-primitives-for-building-agents-here-are-10-tips-on-runnin", "title": "长流程 Agent 不是更长 Prompt，而是“技能手册 + 可执行环境 + 自动压缩”", "title_en": "Openai Developers Openaidevs We Just Announced New Primitives For Building Agents Here Are 10 Tips On Runnin", "author": "", "date": "2026-02-11", "category": [ "Uota学", "讨论题" ], "tldr": "长时间运行的智能体要像“员工”一样可交付：把流程固化成 skills、把动作落在 shell、把记忆交给 compaction，而不是继续堆一坨脆弱的提示词意大利面。", "url": "/articles/openai-developers-openaidevs-we-just-announced-new-primitives-for-building-agents-here-are-10-tips-on-runnin/", "readMin": 14, "topics": [ "agent", "prompt", "org" ] }, { "slug": "your-company-is-a-filesystem", "title": "文件系统才是 AI Agent 的终极操作系统", "title_en": "Your Company Is A Filesystem", "author": "", "date": "2026-02-11", "category": [ "阿头学", "讨论题" ], "tldr": "Agent 的本质架构就是\"文件系统当状态 + LLM 当编排器\"，谁先把公司建模成文件系统，谁就先拿到 AI 红利。", "url": "/articles/your-company-is-a-filesystem/", "readMin": 3, "topics": [ "agent", "build", "org" ] }, { "slug": "my-openclawd-api-bill-was-3x-what-it-should-be-so-i-built-claw-compactor", "title": "Token 账单翻 3 倍？确定性压缩才是正确答案", "title_en": "My Openclawd Api Bill Was 3X What It Should Be So I Built Claw Compactor", "author": "", "date": "2026-02-11", "category": [ "阿头学", "讨论题" ], "tldr": "上下文窗口的成本问题不该靠更聪明的模型解决——一个不调 LLM 的 5 层确定性压缩器，能砍掉 50-97% 的 token 浪费。", "url": "/articles/my-openclawd-api-bill-was-3x-what-it-should-be-so-i-built-claw-compactor/", "readMin": 5, "topics": [ "agent", "build", "prompt" ] }, { "slug": "serenity-aleabitoreddit-for-the-vast-majority-of-retail-if-you-want-to-ride-the-capex-trends-happening", "title": "散户抓 AI CapEx 浪潮，简单模式就够了", "title_en": "Serenity Aleabitoreddit For The Vast Majority Of Retail If You Want To Ride The Capex Trends Happening", "author": "", "date": "2026-02-11", "category": [], "tldr": "AI 资本开支三条主线（存储/光子学/先进封装）已经明牌，散户别折腾上游小票，持有 $TSM 这种中心节点大概率跑赢。", "url": "/articles/serenity-aleabitoreddit-for-the-vast-majority-of-retail-if-you-want-to-ride-the-capex-trends-happening/", "readMin": 1, "topics": [ "invest", "personal" ] }, { "slug": "一万字提示词-10个文件-给你的ai造一个-数字灵魂", "title": "给 AI 造\"数字灵魂\"——TELOS 系统把 AI 从工具变成基础设施", "title_en": "一万字提示词 10个文件 给你的Ai造一个 数字灵魂", "author": "", "date": "2026-02-11", "category": [ "阿头学", "讨论题" ], "tldr": "AI 不该是每次都失忆的陌生人——用 10 个 Markdown 文件定义\"你是谁\"，加三层记忆架构，就能把 AI 从自动售货机升级成个人操作系统。", "url": "/articles/一万字提示词-10个文件-给你的ai造一个-数字灵魂/", "readMin": 21, "topics": [ "memo", "agent", "prompt" ] }, { "slug": "the-exponential-horizon", "title": "The Exponential Horizon", "title_en": "The Exponential Horizon", "author": "", "date": "2026-02-10", "category": [ "阿头学" ], "topics": [ "invest", "org", "build" ], "tldr": "短期波动只是噪音；真正的风险是你没在 AI/机器人/能源这条“指数级地平线”上提前拿到敞口。", "url": "/articles/the-exponential-horizon/", "readMin": 3 }, { "slug": "requests-for-startups-y-combinator", "title": "YC 2026春季「创业征集令」——他们想投什么", "title_en": "Requests for Startups — Y Combinator", "author": "", "date": "2026-02-09", "category": [ "阿头学", "讨论题" ], "tldr": "YC 押注的不是「AI 能做什么」，而是「AI 让谁变得不再需要」——从产品经理到对冲基金交易员到政府公务员，每个方向都是用 AI agent 吃掉一个人力密集型行业。", "url": "/articles/requests-for-startups-y-combinator/", "readMin": 22, "topics": [ "build", "invest", "agent" ] }, { "slug": "2026-02-09-bouazizalex-x-2020159203382550530", "title": "Deel 7500 人零办公室——远程不是实验，是唯一能规模化的工作方式", "title_en": "2026 02 09 Bouazizalex X 2020159203382550530", "author": "", "date": "2026-02-09", "category": [ "阿头学", "讨论题" ], "tldr": "Deel CEO 用 3333 字论证：超过 15 人的公司本质上都已经是远程的，办公室只是套在外面的壳；真正让远程跑起来的不是工具，而是筛选主动性、首月拿胜仗、OKR 问责、管理者必须下场干活。", "url": "/articles/2026-02-09-bouazizalex-x-2020159203382550530/", "readMin": 14, "topics": [ "org" ] }, { "slug": "别急着用seedance2-0-送你保姆级ai运镜提示词指南", "title": "运镜提示词才是 AI 视频的「导演语言」——Seedance2.0 的正确打开方式", "title_en": "别急着用Seedance2 0 送你保姆级Ai运镜提示词指南", "author": "", "date": "2026-02-09", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "别再只写“画面里有什么”了：运镜是一套你对模型下达“时间与视角控制”的动态 DSL，决定成片是电影感还是随手拍。", "url": "/articles/别急着用seedance2-0-送你保姆级ai运镜提示词指南/", "readMin": 17, "topics": [ "creative", "prompt", "agent" ] }, { "slug": "huangserva-servasyy_ai-vkitty-三机制-让-ai-团队真正成长-我觉得群友这个机制比claude-的agent-teams更进化-以下是龙虾团队自己的总结-今天我的-ai-团队", "title": "AI 团队不缺聪明，缺的是会长记性", "title_en": "Huangserva Servasyy_Ai Vkitty 三机制 让 Ai 团队真正成长 我觉得群友这个机制比Claude 的Agent Teams更进化 以下是龙虾团队自己的总结 今天我的 Ai 团队", "author": "", "date": "2026-02-09", "category": [ "阿头学", "讨论题" ], "tldr": "把 AI Agent 当“组织”而不是“工具”养：共享底座 + 按需拉专家 + 强制复盘，才能从反复踩坑变成螺旋上升。", "url": "/articles/huangserva-servasyy_ai-vkitty-三机制-让-ai-团队真正成长-我觉得群友这个机制比claude-的agent-teams更进化-以下是龙虾团队自己的总结-今天我的-ai-团队/", "readMin": 3, "topics": [ "org", "agent", "personal" ] }, { "slug": "龙虾4兄弟的ai协作实战-目标数字人分身-我们的真实经历分享", "title": "数字分身不是“更强模型”，是“协作操作系统”", "title_en": "龙虾4兄弟的Ai协作实战 目标数字人分身 我们的真实经历分享", "author": "", "date": "2026-02-09", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "想要“数字人分身”，别迷信模型智商：把记忆、人格、权限、人类在环、技能扩展做成默认基础设施，AI 才会从临时工进化为长期搭档。", "url": "/articles/龙虾4兄弟的ai协作实战-目标数字人分身-我们的真实经历分享/", "readMin": 37, "topics": [ "org", "agent", "memo" ] }, { "slug": "the-full-tutorial-6-ai-agents-that-run-a-company-how-i-built-them-from-scratch", "title": "多智能体不是“让它自己想”，是把公司流程做成数据库", "title_en": "The Full Tutorial 6 Ai Agents That Run A Company How I Built Them From Scratch", "author": "", "date": "2026-02-09", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "agent", "org", "infra" ], "tldr": "别再把多智能体系统当成“更聪明的 prompt”了：真正的门槛是一个可审计、可限流、可恢复的闭环（proposal→mission→step→event），以及把“对话/记忆/关系/主动性”做成可配置机制。", "url": "/articles/the-full-tutorial-6-ai-agents-that-run-a-company-how-i-built-them-from-scratch/", "readMin": 66 }, { "slug": "you-can-build-a-decacorn-from-your-bedroom", "title": "Deel CEO 的远程宣言——7500人、110国、零办公室，从卧室造出十角兽", "title_en": "You Can Build A Decacorn From Your Bedroom", "author": "", "date": "2026-02-09", "category": [ "阿头学" ], "topics": [ "org", "build", "creative" ], "tldr": "Deel 创始人 Alex Bouazizi 用 3333 字和 7500 人的实战数据证明一件事：所有大公司本质上都是远程公司，办公室只是包裹在数字协作外面的一栋建筑——真正决定胜负的是按能动性招聘、30 天内让新人打出第一场胜仗、OKR 透明到人人裸奔、以及经理必须亲自下场干活。", "url": "/articles/you-can-build-a-decacorn-from-your-bedroom/", "readMin": 16 }, { "slug": "miles-deutscher-milesdeutscher-alright-here-s-my-most-powerful-vibe-coding-prompt-ever-plug-this-straight-in", "title": "Miles Deutscher 的\"最强 Vibe Coding 提示词\"——5 分钟上线任何东西", "title_en": "Miles Deutscher Milesdeutscher Alright Here S My Most Powerful Vibe Coding Prompt Ever Plug This Straight In", "author": "", "date": "2026-02-08", "category": [ "阿头学" ], "topics": [ "creative", "prompt", "invest" ], "tldr": "加密 KOL Miles Deutscher 分享了他的\"史上最强 vibe coding 提示词\"，声称搭配 Opus 4.6 可以在 5 分钟内做出并上线任何东西——SEO 网站、Claude 应用、个人仪表盘——本质上是 AI 时代\"人人皆可 ship\"的又一个信号弹，但提示词本身未公开，更像是流量钩子。", "url": "/articles/miles-deutscher-milesdeutscher-alright-here-s-my-most-powerful-vibe-coding-prompt-ever-plug-this-straight-in/", "readMin": 1 }, { "slug": "my-complete-guide-to-managing-openclaw-agent-teams", "title": "多智能体最大的问题不是能力，是“管理”", "title_en": "My Complete Guide To Managing Openclaw Agent Teams", "author": "", "date": "2026-02-08", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "org", "memo" ], "tldr": "当你开始同时跑多个 agent，你就不再是在“用工具”，而是在“带团队”——没有入职、分级、绩效、共享上下文和协作协议，结果只会是更贵、更乱、更重复。", "url": "/articles/my-complete-guide-to-managing-openclaw-agent-teams/", "readMin": 7 }, { "slug": "我如何用-claude-code-在3分钟搭建一套可迭代的永续-ai-工作系统", "title": "我如何用 Claude Code 在3分钟搭建一套可迭代的永续 AI 工作系统", "title_en": "我如何用 Claude Code 在3分钟搭建一套可迭代的永续 Ai 工作系统", "author": "", "date": "2026-02-08", "category": [ "阿头学", "Uota学" ], "topics": [ "memo", "agent", "build" ], "tldr": "AI 工作系统的核心不是工具选择，而是\"让 AI 记住你是谁\"——从碎片化使用到系统化工作流，区别在于有没有持久化的上下文层。", "url": "/articles/我如何用-claude-code-在3分钟搭建一套可迭代的永续-ai-工作系统/", "readMin": 7 }, { "slug": "openclaw-token-saving-guide", "title": "OpenClaw Token 节约指南：省钱方向对，但“终极”说法夸大了", "title_en": "Openclaw Token Saving Guide", "author": "", "date": "2026-02-08", "category": [ "阿头学", "讨论题" ], "topics": [ "prompt", "memo", "agent" ], "tldr": "这篇文章对 Agent 成本黑洞的拆解是准确的，但把“省 token”直接等同于“几乎不掉效果”明显过度乐观，实操价值高，论证严谨度不足。", "url": "/articles/openclaw-token-saving-guide/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "how-to-install-and-use-claude-code-agent-teams-complete-guide", "title": "Claude Code「代理团队」不是功能升级，是工作形态换代", "title_en": "How To Install And Use Claude Code Agent Teams Complete Guide", "author": "", "date": "2026-02-07", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "agent", "prompt" ], "tldr": "从“让一个模型硬扛全流程”切到“领队+多队友并行协作”，价值不在更聪明，而在更像一个能被管理的团队——但代价是：你得学会拆任务、写角色、控协作成本。", "url": "/articles/how-to-install-and-use-claude-code-agent-teams-complete-guide/", "readMin": 12 }, { "slug": "i-built-an-ai-company-with-openclaw-vercel-supabase-two-weeks-later-they-run-it-themselves", "title": "6 个 AI Agent 自主运营网站——从\"能聊天\"到\"能干活\"的闭环架构", "title_en": "I Built An Ai Company With Openclaw Vercel Supabase Two Weeks Later They Run It Themselves", "author": "", "date": "2026-02-07", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "creative" ], "tldr": "Agent 能说话不等于能干活，从 proposal 到 execution 的闭环才是真正的分水岭——这篇把中间缺的那段补上了。", "url": "/articles/i-built-an-ai-company-with-openclaw-vercel-supabase-two-weeks-later-they-run-it-themselves/", "readMin": 21 }, { "slug": "claude-agent-team-guide", "title": "Claude Agent Team 入门指南：协作机制有价值，但文章明显把“功能上限”包装成了“普遍效果”", "title_en": "Claude Agent Team Guide", "author": "", "date": "2026-02-07", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "这篇文章对 Agent Team 与 Sub-Agent 的协作差异解释得很清楚，但它明显带有布道和营销色彩，把实验性功能的架构优势过度外推成了普通用户立刻可复制的生产力革命。", "url": "/articles/claude-agent-team-guide/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "gpt-5-3-codex-are-we-becoming-the-bottleneck", "title": "GPT-5.3 Codex——我们是不是已经成了瓶颈？", "title_en": "GPT-5.3 Codex: Are We Becoming the Bottleneck?", "author": "", "date": "2026-02-06", "category": [ "阿头学", "Uota学" ], "tldr": "AI 编程能力已经快到人类跟不上了——瓶颈不再是模型，而是你自己。", "url": "/articles/gpt-5-3-codex-are-we-becoming-the-bottleneck/", "readMin": 4, "topics": [ "agent", "org", "build" ] }, { "slug": "九原客-9hills-认真读了几个-anthropic-cowork-skill-写的不错-https-github-com-anthropics-knowledge-work", "title": "Anthropic 开源 Cowork Skills——官方下场定义\"AI 怎么跟人协作\"", "title_en": "九原客 9Hills 认真读了几个 Anthropic Cowork Skill 写的不错 Https Github Com Anthropics Knowledge Work", "author": "", "date": "2026-02-06", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "org", "memo" ], "tldr": "Anthropic 把\"AI 协作技能\"开源了，这不是工具发布，是在抢定义权——\"AI 应该怎么跟人一起工作\"这个问题，他们要先写答案。", "url": "/articles/九原客-9hills-认真读了几个-anthropic-cowork-skill-写的不错-https-github-com-anthropics-knowledge-work/", "readMin": 1 }, { "slug": "i-spent-10-000-to-automate-my-research-at-openai-with-codex", "title": "OpenAI 研究员月烧 1 万刀用 Codex——真正的瓶颈不是执行，是决定分析什么", "title_en": "I Spent 10 000 To Automate My Research At Openai With Codex", "author": "", "date": "2026-02-06", "category": [ "阿头学", "Uota学" ], "topics": [ "org", "agent", "prompt" ], "tldr": "当 AI 能无摩擦地穿越整个组织的信息版图时，\"协调成本\"这个千年难题可能真的要被解了——而大多数人还在纠结 prompt 怎么写。", "url": "/articles/i-spent-10-000-to-automate-my-research-at-openai-with-codex/", "readMin": 5 }, { "slug": "claude-code-我和-dontbesilent-相处的三个星期", "title": "把 AI 变成内容生产线：一位创作者与 Claude Code 的 22 天共生实验", "title_en": "Claude Code 我和 Dontbesilent 相处的三个星期", "author": "", "date": "2026-02-06", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build", "creative" ], "tldr": "这篇文章本质上是一份「用软件工程思维打造 AI 内容工厂」的个人 PR 式案例复盘，方法论有真料，但效果被明显高估且强行包装成大众可复制。", "url": "/articles/claude-code-我和-dontbesilent-相处的三个星期/", "readMin": 12, "image": "image-1.jpg" }, { "slug": "50-design-prompts-from-ex-warner-designer-nano-banana-pro", "title": "前华纳设计师的 50 条 AI 设计提示词——工具化经验还是营销漏斗？", "title_en": "50 Design Prompts From Ex Warner Designer Nano Banana Pro", "author": "", "date": "2026-02-06", "category": [ "讨论题" ], "topics": [ "creative", "invest", "build" ], "tldr": "一位资深设计师把 10 年大厂经验模板化成 50 条可复用提示词，声称零成本快速产出专业级视觉，但实际交付路径依赖付费工具且回避了法律与审美复杂性。", "url": "/articles/50-design-prompts-from-ex-warner-designer-nano-banana-pro/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "ryo-lu-ryolu_-software-is-still-about-thinking-software-has-always-been-about-taking-ambiguous", "title": "AI 不会救你的烂架构，只会让它跑得更快", "title_en": "Ryo Lu Ryolu_ Software Is Still About Thinking Software Has Always Been About Taking Ambiguous", "author": "", "date": "2026-02-05", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "org" ], "tldr": "AI 写代码的速度是幻觉——没有系统思维的人只会规模化地生产垃圾，而想清楚的人会快 100 倍。", "url": "/articles/ryo-lu-ryolu_-software-is-still-about-thinking-software-has-always-been-about-taking-ambiguous/", "readMin": 3 }, { "slug": "unbrowse-100x-faster-than-browser-automation", "title": "浏览器自动化是给 Agent 套上的枷锁，该砸了", "title_en": "Unbrowse 100X Faster Than Browser Automation", "author": "", "date": "2026-02-05", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "invest", "creative" ], "tldr": "99% 的网站都有内部 API，你的 Agent 却在用 12 秒模拟点击去做 200ms 就能完成的事——Unbrowse 把这层壳撬开了。", "url": "/articles/unbrowse-100x-faster-than-browser-automation/", "readMin": 7 }, { "slug": "沉浸式翻译-immersivetran-模型训练崩了可以重来-人崩了怎么重启-发现一个看得脊背发凉-却又舍不得关掉的网站-hacker-news-burnout-guide-一个专门收集-硅谷老哥血", "title": "硅谷老哥的认怂指南——程序员 Burnout 自救手册", "title_en": "沉浸式翻译 Immersivetran 模型训练崩了可以重来 人崩了怎么重启 发现一个看得脊背发凉 却又舍不得关掉的网站 Hacker News Burnout Guide 一个专门收集 硅谷老哥血", "author": "", "date": "2026-02-05", "category": [ "阿头学", "讨论题" ], "topics": [ "org" ], "tldr": "模型崩了调参重来，人崩了没法换显卡——这份从 Hacker News 提炼的 Burnout 自救指南，给出的处方不是鸡汤而是反直觉的战术。", "url": "/articles/沉浸式翻译-immersivetran-模型训练崩了可以重来-人崩了怎么重启-发现一个看得脊背发凉-却又舍不得关掉的网站-hacker-news-burnout-guide-一个专门收集-硅谷老哥血/", "readMin": 3 }, { "slug": "url智能读取skill-一键抓取任何网站内容并自动同步到到obsidian做为素材", "title": "三层降级策略搞定全网抓取——URL智能读取的工程化思路", "title_en": "Url智能读取Skill 一键抓取任何网站内容并自动同步到到Obsidian做为素材", "author": "", "date": "2026-02-05", "category": [ "Uota学" ], "topics": [ "creative", "prompt", "agent" ], "tldr": "Firecrawl→Jina→Playwright三层降级策略，用零成本覆盖99%网站抓取，核心不是技术多牛，是\"没有银弹所以要有Plan B和Plan C\"的工程思维。", "url": "/articles/url智能读取skill-一键抓取任何网站内容并自动同步到到obsidian做为素材/", "readMin": 7 }, { "slug": "2026-02-05-wechat-p9v6flfqkuqcecu5v9voba", "title": "Claude Code 工作流方法论完全指南", "title_en": "2026 02 05 Wechat P9V6Flfqkuqcecu5V9Voba", "author": "", "date": "2026-02-05", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "invest", "build" ], "tldr": "系统化的 AI 编程工作流（5 阶段编排、Token 优化、验证循环）比单纯工具堆砌更能产生复利，但需要前期投入且存在过度工程化风险。", "url": "/articles/2026-02-05-wechat-p9v6flfqkuqcecu5v9voba/", "readMin": 33, "image": "image-1.png" }, { "slug": "serenity-aleabitoreddit-i-m-long-shares-on-axti-we-re-seeing-the-hbm-nand-supercycle-with-sndk-mu", "title": "光子学是下一个\"卡脖子\"——AXT 和 InP 衬底的超级周期赌注", "title_en": "Serenity Aleabitoreddit I M Long Shares On Axti We Re Seeing The Hbm Nand Supercycle With Sndk Mu", "author": "", "date": "2026-02-04", "category": [ "阿头学" ], "topics": [ "invest", "build", "infra" ], "tldr": "HBM/NAND 超级周期之后，光互连/CPO 是确定性最高的下一棒，而 InP 衬底供应链的物理瓶颈指向 $AXTI。", "url": "/articles/serenity-aleabitoreddit-i-m-long-shares-on-axti-we-re-seeing-the-hbm-nand-supercycle-with-sndk-mu/", "readMin": 3 }, { "slug": "serenity-aleabitoreddit-the-memory-supercycle-is-here-new-reports-from-morgan-stanley-est-samsung-s", "title": "三星要成全球最赚钱公司——内存超级周期的数字有多疯", "title_en": "Serenity Aleabitoreddit The Memory Supercycle Is Here New Reports From Morgan Stanley Est Samsung S", "author": "", "date": "2026-02-04", "category": [ "阿头学" ], "topics": [ "invest" ], "tldr": "摩根士丹利测算三星 2027 年净利润 $163B，超过 NVDA 和 Google——内存不是配角，是这轮 AI 周期的主角之一。", "url": "/articles/serenity-aleabitoreddit-the-memory-supercycle-is-here-new-reports-from-morgan-stanley-est-samsung-s/", "readMin": 1 }, { "slug": "agent-训练师进阶指南-用-discord-打造高效-openclaw-协作系统", "title": "Discord不是聊天工具，是你的AI工作台——Agent训练师的协作系统设计", "title_en": "Agent 训练师进阶指南 用 Discord 打造高效 Openclaw 协作系统", "author": "", "date": "2026-02-04", "category": [ "阿头学", "Uota学" ], "topics": [ "org", "agent", "build" ], "tldr": "Discord的Server+Channel+Thread三层结构，本质上是给AI Agent设计的\"多线程工作台\"，而不只是聊天窗口。", "url": "/articles/agent-训练师进阶指南-用-discord-打造高效-openclaw-协作系统/", "readMin": 14 }, { "slug": "nolimit-nolimitgains-i-m-investing-millions-into-this-it-s-not-gold-it-s-not-silver-it-s-somethin", "title": "别只盯着黄金——被低估的能源与资源股才是下一个抗通胀战场", "title_en": "Nolimit Nolimitgains I M Investing Millions Into This It S Not Gold It S Not Silver It S Somethin", "author": "", "date": "2026-02-04", "category": [ "讨论题" ], "topics": [ "invest", "build" ], "tldr": "\"抗通胀不止黄金\"这个方向没错，但作者用未经验证的个人战绩和FOMO话术包装了一个本质上极度简化的相对估值判断，读者需要剥离营销噪音才能提取出真正有用的思路。", "url": "/articles/nolimit-nolimitgains-i-m-investing-millions-into-this-it-s-not-gold-it-s-not-silver-it-s-somethin/", "readMin": 1, "image": "image-1.jpg" }, { "slug": "deloitte-a-74-billion-cancer-metastasized-across-america", "title": "德勤政府合同的系统性失败与\"旋转门\"困局", "title_en": "Deloitte A 74 Billion Cancer Metastasized Across America", "author": "", "date": "2026-02-04", "category": [ "阿头学", "讨论题" ], "topics": [ "org" ], "tldr": "德勤通过\"旋转门\"人事网络和系统锁定，在美国政府 IT 采购中形成了一个 400 亿美元的垄断，尽管项目失败率极高，但因为没有部门有动机修复，反而通过\"修复失败\"持续获利。", "url": "/articles/deloitte-a-74-billion-cancer-metastasized-across-america/", "readMin": 7, "image": "image-1.jpg" }, { "slug": "2026-02-03-nateliason-x-2018352113860927648", "title": "BYOA——智能不该住在 App 里，该跟着人走", "title_en": "2026 02 03 Nateliason X 2018352113860927648", "author": "", "date": "2026-02-03", "category": [ "阿头学", "Uota学", "讨论题" ], "tldr": "每个 SaaS 都在造自己的小 AI，但用户早就有了一个真正懂自己的 Agent——正确架构是 App 开门让用户自带 Agent（BYOA），放弃 AI 层控制反而让产品更有价值。", "url": "/articles/2026-02-03-nateliason-x-2018352113860927648/", "readMin": 5, "topics": [ "build", "agent" ] }, { "slug": "openclaw-can-now-index-anything", "title": "AI 智能体的记忆革命——从\"幻觉制造机\"到\"索引一切\"", "title_en": "Openclaw Can Now Index Anything", "author": "", "date": "2026-02-03", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "memo", "build" ], "tldr": "AI 智能体最大的短板不是推理能力，而是信息获取的可靠性——谁先解决\"索引一切\"，谁就拿到下一代 Agent 基础设施的入场券。", "url": "/articles/openclaw-can-now-index-anything/", "readMin": 2 }, { "slug": "life-is-a-video-game", "title": "人生是电子游戏——但大多数人连难度选择都搞错了", "title_en": "Life Is A Video Game", "author": "", "date": "2026-02-03", "category": [ "阿头学", "讨论题" ], "topics": [ "org" ], "tldr": "把人生当游戏不是鸡汤，是一种操作系统级的心智模型——关键不在于\"你是主角\"，而在于你是否主动把难度拧到困难模式。", "url": "/articles/life-is-a-video-game/", "readMin": 3 }, { "slug": "8-hooks-to-get-50-000-𝕏-followers", "title": "X 涨粉的 8 个钩子——套路不丢人，不会用才丢人", "title_en": "8 Hooks To Get 50 000 𝕏 Followers", "author": "", "date": "2026-02-03", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "creative", "org" ], "tldr": "注意力经济的底层逻辑没变过：先钩住人，再给价值——这 8 个钩子模板是 X 增长的基本功，不是高级技巧。", "url": "/articles/8-hooks-to-get-50-000-𝕏-followers/", "readMin": 3 }, { "slug": "bring-your-own-agent-the-future-of-ai-powered-apps", "title": "自带 Agent：AI 应用的新架构之争", "title_en": "Bring Your Own Agent The Future Of Ai Powered Apps", "author": "", "date": "2026-02-03", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "org", "agent" ], "tldr": "这篇文章判断 BYOA（自带 Agent）是未来主流架构之一，但在商业动力、安全合规和用户习惯上明显乐观偏头重脚，短期更像高端玩家与新产品的机会，而非全行业必然终局。", "url": "/articles/bring-your-own-agent-the-future-of-ai-powered-apps/", "readMin": 5, "image": "image-1.jpg" }, { "slug": "xiyu-ohxiyu-分享我的openclaw-agent助理团队-可复制版-一台-mac-mini-跑-openclaw-一个-cto-协调-十个-agent-各管各的-从起床到睡", "title": "OpenClaw 十 Agent 助理团队：像个人操作系统，但“零协调成本”明显说过头", "title_en": "Xiyu Ohxiyu 分享我的Openclaw Agent助理团队 可复制版 一台 Mac Mini 跑 Openclaw 一个 Cto 协调 十个 Agent 各管各的 从起床到睡", "author": "", "date": "2026-02-03", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "agent", "build" ], "tldr": "这套 OpenClaw 多 Agent 架构在“分层调度+独立监控”上有真价值，但作者把一张好看的系统设计图包装成“可复制、近乎零成本”的成熟方案，判断上明显过度乐观。", "url": "/articles/xiyu-ohxiyu-分享我的openclaw-agent助理团队-可复制版-一台-mac-mini-跑-openclaw-一个-cto-协调-十个-agent-各管各的-从起床到睡/", "readMin": 2, "image": "image-1.jpg" }, { "slug": "luciano-castillo-luccasveg-crabwalk-for-openclaw-v1-0-5-v1-0-6-release-subagent-support-and-follow-m", "title": "Crabwalk 给 OpenClaw 加了子代理可视化——Agent 编排的调试体验正在被认真对待", "title_en": "Luciano Castillo Luccasveg Crabwalk For Openclaw V1 0 5 V1 0 6 Release Subagent Support And Follow M", "author": "", "date": "2026-02-02", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "creative", "infra" ], "tldr": "当 AI Agent 开始自己 spawn 子代理，你需要的不是更多日志，而是一张实时的关系图谱。", "url": "/articles/luciano-castillo-luccasveg-crabwalk-for-openclaw-v1-0-5-v1-0-6-release-subagent-support-and-follow-m/", "readMin": 1 }, { "slug": "andrej-karpathy-karpathy-finding-myself-going-back-to-rss-atom-feeds-a-lot-more-recently-there-s-a-lot-m", "title": "Karpathy 回归 RSS——算法喂养时代的信息自救", "title_en": "Andrej Karpathy Karpathy Finding Myself Going Back To Rss Atom Feeds A Lot More Recently There S A Lot M", "author": "", "date": "2026-02-02", "category": [ "阿头学", "Uota学" ], "topics": [ "build", "creative" ], "tldr": "当所有平台的激励结构都把你推向同一个注意力黑洞，主动选择信息源就是最基本的认知防御。", "url": "/articles/andrej-karpathy-karpathy-finding-myself-going-back-to-rss-atom-feeds-a-lot-more-recently-there-s-a-lot-m/", "readMin": 1 }, { "slug": "2026-02-02-wechat-clyenxqpx71l0ztsy2uygq", "title": "AI 竞争格局 2026：从范式演进到现金流之战", "title_en": "2026 02 02 Wechat Clyenxqpx71L0Ztsy2Uygq", "author": "", "date": "2026-02-02", "category": [ "阿头学", "Uota学" ], "topics": [ "invest", "build", "agent" ], "tldr": "AI 竞争已从单纯的模型能力竞赛转向\"持续学习范式\"和\"现金流耐力\"的长期对垒，Google 与 OpenAI 将分庭抗礼，但真正的赢家取决于谁能率先解决 Continual Learning 这一核聚变级难题。", "url": "/articles/2026-02-02-wechat-clyenxqpx71l0ztsy2uygq/", "readMin": 49, "image": "image-1.png" }, { "slug": "geek-lite-qingq77-好酷的地图组件库-基于-react-typescript-tailwind-css-和-mapbox-gl-js-构建-能够帮助开发者轻松集成高质量的地图功能", "title": "Terrae 不是“高质量地图方案”的证明，更像是一个瞄准现代前端栈的地图封装尝试", "title_en": "Geek Lite Qingq77 好酷的地图组件库 基于 React Typescript Tailwind Css 和 Mapbox Gl Js 构建 能够帮助开发者轻松集成高质量的地图功能", "author": "", "date": "2026-02-02", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "agent", "org" ], "tldr": "Terrae 有探索价值，但现有材料只能证明它“会包装”，不能证明它真能让团队低成本、可持续地做好地图能力。", "url": "/articles/geek-lite-qingq77-好酷的地图组件库-基于-react-typescript-tailwind-css-和-mapbox-gl-js-构建-能够帮助开发者轻松集成高质量的地图功能/", "readMin": 1, "image": "image-1.jpg" }, { "slug": "how-to-become-dangerously-attractive-in-60-days", "title": "停止优化，开始活着——一个健身博主说出了自我提升行业的真相", "title_en": "How To Become Dangerously Attractive In 60 Days", "author": "", "date": "2026-02-01", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "personal", "creative" ], "tldr": "自我提升的最大陷阱不是\"不够努力\"，而是把优化本身当成了目的——真正的蜕变来自身份转换，不是习惯堆叠。", "url": "/articles/how-to-become-dangerously-attractive-in-60-days/", "readMin": 20 }, { "slug": "the-complete-guide-to-building-mission-control-how-we-built-an-ai-agent-squad", "title": "10个AI智能体像团队一样协作——这才是\"AI原生工作流\"的真正形态", "title_en": "The Complete Guide To Building Mission Control How We Built An Ai Agent Squad", "author": "", "date": "2026-02-01", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "org", "creative" ], "tldr": "多智能体系统的核心不是技术架构，而是把AI当团队成员来管理——给角色、给记忆、给协作机制、要求负责，这套思路跟管人一模一样。", "url": "/articles/the-complete-guide-to-building-mission-control-how-we-built-an-ai-agent-squad/", "readMin": 20 }, { "slug": "小白也能看懂上手-让-openclaw-在-telegram-群里并发处理多任务-topics-多-session-指南", "title": "OpenClaw Telegram 并发指南——用 Topics 把一个群拆成多条任务车道", "title_en": "小白也能看懂上手 让 Openclaw 在 Telegram 群里并发处理多任务 Topics 多 Session 指南", "author": "", "date": "2026-02-01", "category": [ "Uota学" ], "topics": [ "agent", "org" ], "tldr": "Telegram Topics = OpenClaw 的天然并发入口，每个 Topic 就是一条独立 Session，零代码实现多任务并行且互不串台。", "url": "/articles/小白也能看懂上手-让-openclaw-在-telegram-群里并发处理多任务-topics-多-session-指南/", "readMin": 5 }, { "slug": "从选题到发布-dontbesilent-的-claude-code-工作流", "title": "年发13000条内容的系统化创作工厂——dontbesilent的Claude Code工作流", "title_en": "从选题到发布 Dontbesilent 的 Claude Code 工作流", "author": "", "date": "2026-02-01", "category": [ "阿头学", "Uota学" ], "topics": [ "creative", "memo", "org" ], "tldr": "把AI从\"问答机器\"变成\"内容工厂操作系统\"，核心不是AI能写什么，而是你给AI喂了什么记忆和方法论。", "url": "/articles/从选题到发布-dontbesilent-的-claude-code-工作流/", "readMin": 9 }, { "slug": "gp-q-argosaki-breastmilk-she-thought-she-was-studying-milk-what-she-uncovered-was-a-conversat", "title": "母乳不是燃料，是智能通信协议——被忽视的生物系统里藏着产品设计的隐喻", "title_en": "Gp Q Argosaki Breastmilk She Thought She Was Studying Milk What She Uncovered Was A Conversat", "author": "", "date": "2026-01-31", "category": [ "Uota学", "讨论题" ], "topics": [ "build", "personal" ], "tldr": "母乳是一套有 200 万年历史的双向通信系统，能根据婴儿的性别、健康状态、时间实时调整输出——而我们曾把它当成\"简单的热量输入\"忽略了几十年。", "url": "/articles/gp-q-argosaki-breastmilk-she-thought-she-was-studying-milk-what-she-uncovered-was-a-conversat/", "readMin": 4 }, { "slug": "20-clawdbot-setups-that-made-me-say-wait-it-can-do-that", "title": "20 个野生 Clawdbot 用例——Agent 的真正价值不在技术炫技，在于替代\"人肉胶水\"", "title_en": "20 Clawdbot Setups That Made Me Say Wait It Can Do That", "author": "", "date": "2026-01-31", "category": [ "阿头学", "Uota学" ], "topics": [ "invest", "creative", "infra" ], "tldr": "20 个真实世界的 Clawdbot 实现证明：AI Agent 最大的价值不是写代码，而是把散落在不同系统间的\"人肉胶水工作\"自动化掉。", "url": "/articles/20-clawdbot-setups-that-made-me-say-wait-it-can-do-that/", "readMin": 9 }, { "slug": "everyone-talks-about-clawdbot-but-here-s-how-it-works", "title": "拆开 Clawdbot 的壳——一个 AI 工程师的架构笔记", "title_en": "Everyone Talks About Clawdbot But Here S How It Works", "author": "", "date": "2026-01-31", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "prompt", "memo" ], "tldr": "Clawdbot 的架构没有黑魔法——TypeScript CLI + lane 串行化 + JSONL 记忆 + Markdown 文件，简单到让人不安，但正是这种\"可解释的简单\"让它跑起来了。", "url": "/articles/everyone-talks-about-clawdbot-but-here-s-how-it-works/", "readMin": 8 }, { "slug": "openclaw-clawdbot-starter-pack-tools-plug-ins-resources", "title": "OpenClaw 生态资源全景——工具链成熟度决定 AI Agent 的天花板", "title_en": "Openclaw Clawdbot Starter Pack Tools Plug Ins Resources", "author": "", "date": "2026-01-31", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "memo", "creative" ], "tldr": "OpenClaw 生态已经从\"能跑\"进化到\"有人整理最佳实践和安全审计\"的阶段——工具链成熟的标志不是功能多，而是有人开始写避坑指南。", "url": "/articles/openclaw-clawdbot-starter-pack-tools-plug-ins-resources/", "readMin": 5 }, { "slug": "how-to-scale-dynamic-context", "title": "别再写入门提示词了——你需要的是分布式上下文架构", "title_en": "How To Scale Dynamic Context", "author": "", "date": "2026-01-31", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "creative" ], "tldr": "入门提示词是静态的、单体的、会劣化的；把上下文文件散布到文件系统各处，让 agent 根据位置动态加载，才是可扩展的正确做法。", "url": "/articles/how-to-scale-dynamic-context/", "readMin": 9 }, { "slug": "the-ai-health-coach-upgrade-for-clawdbot-extend-your-life-by-25-years", "title": "把 AI 变成你的私人健康教练——一个过度工程但方向正确的方案", "title_en": "The Ai Health Coach Upgrade For Clawdbot Extend Your Life By 25 Years", "author": "", "date": "2026-01-31", "category": [ "阿头学", "Uota学" ], "topics": [ "build", "org", "personal" ], "tldr": "用 Clawdbot 的 cron + 文件系统搭一套健康追踪体系，核心洞见不是技术方案本身，而是\"你的健康太重要了，不该交给你的记忆去保管\"。", "url": "/articles/the-ai-health-coach-upgrade-for-clawdbot-extend-your-life-by-25-years/", "readMin": 6 }, { "slug": "熊布朗-stephen4171127-聊几句openclaw-的记忆实施策略-openclaw-把聊天记录和生成的文档都向量化-并文本化-进-memory-索引里了-后续仅用-agent-tool", "title": "OpenClaw 的记忆设计——\"不自动注入\"才是真正的 Agent 记忆", "title_en": "熊布朗 Stephen4171127 聊几句Openclaw 的记忆实施策略 Openclaw 把聊天记录和生成的文档都向量化 并文本化 进 Memory 索引里了 后续仅用 Agent Tool", "author": "", "date": "2026-01-31", "category": [ "阿头学", "Uota学" ], "topics": [ "memo", "prompt", "agent" ], "tldr": "OpenClaw 把记忆完全交给 Agent 按需调用，不往 system prompt 里塞——这比大多数 Character AI 的 profile 注入方案更干净，也更接近\"真正的 Agent\"。", "url": "/articles/熊布朗-stephen4171127-聊几句openclaw-的记忆实施策略-openclaw-把聊天记录和生成的文档都向量化-并文本化-进-memory-索引里了-后续仅用-agent-tool/", "readMin": 2 }, { "slug": "thoughts-on-open-source-software-and-kimi-code-cli", "title": "开源的本质不是代码贡献，是修改的自由——Kimi Code CLI 的克制哲学", "title_en": "Thoughts On Open Source Software And Kimi Code Cli", "author": "", "date": "2026-01-31", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "build", "personal" ], "tldr": "编码代理时代，代码人力不再稀缺，开源的价值回归本质：不是\"帮我写代码\"，而是\"给你改代码的自由\"。", "url": "/articles/thoughts-on-open-source-software-and-kimi-code-cli/", "readMin": 6 }, { "slug": "ubuntu-下为-openclaw-配置-telegram-webhook", "title": "OpenClaw 的 Telegram 稳定性问题，一个 Webhook 就能解", "title_en": "Ubuntu 下为 Openclaw 配置 Telegram Webhook", "author": "", "date": "2026-01-31", "category": [ "Uota学" ], "topics": [ "infra", "build", "agent" ], "tldr": "OpenClaw 通过 Telegram 对话时长任务丢消息，根因是 Long-polling 的天然缺陷，切 Webhook 模式 + Caddy 反代即可修复。", "url": "/articles/ubuntu-下为-openclaw-配置-telegram-webhook/", "readMin": 2 }, { "slug": "meta-alchemist-meta_alchemist-spark-is-starting-to-become-real-a-self-evolving-intelligence-beyond-a-persist", "title": "Spark——当 AI 不只是记住你，而是和你一起进化", "title_en": "Meta Alchemist Meta_Alchemist Spark Is Starting To Become Real A Self Evolving Intelligence Beyond A Persist", "author": "", "date": "2026-01-30", "category": [ "阿头学", "Uota学" ], "topics": [ "memo", "creative", "org" ], "tldr": "个人 AI 的下一步不是更大的上下文窗口，而是持续自我进化的智能体——能积累记忆、提炼洞见、理解工具链，并随着使用越来越懂你。", "url": "/articles/meta-alchemist-meta_alchemist-spark-is-starting-to-become-real-a-self-evolving-intelligence-beyond-a-persist/", "readMin": 1 }, { "slug": "openai-developers-openaidevs-inside-our-in-house-ai-data-agent-it-reasons-over-600-pb-and-70k-datasets-enab", "title": "OpenAI 内部数据代理——600PB 上的自然语言分析，数据民主化的终局样本", "title_en": "Openai Developers Openaidevs Inside Our In House Ai Data Agent It Reasons Over 600 Pb And 70K Datasets Enab", "author": "", "date": "2026-01-30", "category": [ "阿头学", "Uota学" ], "topics": [ "org", "agent", "build" ], "tldr": "当 AI 代理能在 600PB、7万张表上自主推理纠错，\"数据分析\"就不再是数据团队的专属技能——这是企业数据民主化的真实终局样本。", "url": "/articles/openai-developers-openaidevs-inside-our-in-house-ai-data-agent-it-reasons-over-600-pb-and-70k-datasets-enab/", "readMin": 1 }, { "slug": "google-deepmind-googledeepmind-step-inside-project-genie-our-experimental-research-prototype-that-lets-you-cre", "title": "DeepMind Project Genie——AI 生成可交互虚拟世界，从研究原型到创作工具", "title_en": "Google Deepmind Googledeepmind Step Inside Project Genie Our Experimental Research Prototype That Lets You Cre", "author": "", "date": "2026-01-30", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "agent", "build" ], "tldr": "DeepMind 把\"从一张图生成可交互 3D 世界\"从论文变成了可体验的原型——世界模型正在从训练 Agent 的工具，变成普通人的创作界面。", "url": "/articles/google-deepmind-googledeepmind-step-inside-project-genie-our-experimental-research-prototype-that-lets-you-cre/", "readMin": 1 }, { "slug": "深度解析-moltbot-底层架构", "title": "Moltbot 架构全拆——主权 AI 的终极形态长这样", "title_en": "深度解析 Moltbot 底层架构", "author": "", "date": "2026-01-30", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "org" ], "tldr": "Moltbot 用 Gateway 控制面 + OS-as-Surface + 本地优先三板斧，把 AI Agent 从\"聊天机器人\"拉到了\"操作系统层\"——这不是未来，是已经跑通的架构。", "url": "/articles/深度解析-moltbot-底层架构/", "readMin": 41 }, { "slug": "拒绝调包-纯numpy手搓ilya推荐的30篇论文-连反向传播都是手写的", "title": "纯 NumPy 手搓 Ilya 的 30 篇论文——调包侠的解毒剂", "title_en": "拒绝调包 纯Numpy手搓Ilya推荐的30篇论文 连反向传播都是手写的", "author": "", "date": "2026-01-30", "category": [ "阿头学" ], "topics": [ "build", "prompt", "personal" ], "tldr": "有人用纯 NumPy 从零复现了 Ilya 书单全部 30 篇论文，连反向传播都是手写的——这不是炫技，是理解 AI 底层最诚实的方式。", "url": "/articles/拒绝调包-纯numpy手搓ilya推荐的30篇论文-连反向传播都是手写的/", "readMin": 14 }, { "slug": "字节-阿里-腾讯-ai-大战全记录-一场影响命运的战争", "title": "字节阿里腾讯 AI 中途岛——这次输了就输掉整个未来", "title_en": "字节 阿里 腾讯 Ai 大战全记录 一场影响命运的战争", "author": "", "date": "2026-01-30", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "creative", "org" ], "tldr": "BAT 的 AI 战争已经从\"谁的模型好\"升级到\"谁能抢下超级入口\"——豆包日活破亿、元宝借 DeepSeek 翻盘、千问 App 取代夸克，三家都在赌命。", "url": "/articles/字节-阿里-腾讯-ai-大战全记录-一场影响命运的战争/", "readMin": 43 }, { "slug": "港式复古风滤镜-不是-黄-糊-而是一整套影像逻辑", "title": "港式复古风滤镜——不是\"黄+糊\"，而是一整套影像逻辑", "title_en": "港式复古风滤镜 不是 黄 糊 而是一整套影像逻辑", "author": "", "date": "2026-01-29", "category": [], "topics": [ "prompt", "build", "memo" ], "tldr": "港风不是审美风格先行，而是由90-00年代香港影像工业的技术条件塑造出来的副产品——胶片库存、冲洗工艺、灯光条件共同决定了那个\"look\"。", "url": "/articles/港式复古风滤镜-不是-黄-糊-而是一整套影像逻辑/", "readMin": 6 }, { "slug": "沉浸式翻译-immersivetran-被严重低估的宝藏网站-用数据看穿世界真相-想写深度内容却没数据-这个网站直接把底层逻辑喂到你嘴边-它是牛津大学团队和一家英国非营利机构做的长期项目-专门研", "title": "Our World in Data——被严重低估的宝藏数据网站", "title_en": "沉浸式翻译 Immersivetran 被严重低估的宝藏网站 用数据看穿世界真相 想写深度内容却没数据 这个网站直接把底层逻辑喂到你嘴边 它是牛津大学团队和一家英国非营利机构做的长期项目 专门研", "author": "", "date": "2026-01-29", "category": [], "topics": [ "creative", "org" ], "tldr": "牛津大学团队做的开放数据项目，把几十年的全球数据做成可交互趋势图——很多我们以为\"越来越糟\"的东西，用图一拉长答案完全不一样。", "url": "/articles/沉浸式翻译-immersivetran-被严重低估的宝藏网站-用数据看穿世界真相-想写深度内容却没数据-这个网站直接把底层逻辑喂到你嘴边-它是牛津大学团队和一家英国非营利机构做的长期项目-专门研/", "readMin": 4 }, { "slug": "how-to-build-an-agent-that-never-forgets", "title": "Agent 记忆不是功能，是基础设施——别再把聊天记录当记忆用了", "title_en": "How To Build An Agent That Never Forgets", "author": "", "date": "2026-01-29", "category": [ "阿头学", "Uota学" ], "topics": [ "prompt", "agent", "build" ], "tldr": "记忆系统的核心不是\"存更多\"，而是\"结构化 + 衰减 + 冲突消解\"——把 Agent 当操作系统设计，而不是聊天机器人。", "url": "/articles/how-to-build-an-agent-that-never-forgets/", "readMin": 10 }, { "slug": "10-ways-to-hack-into-a-vibecoder-s-clawdbot-get-entire-human-identity-educational-purposes-only", "title": "AI Agent 自建部署的十大入侵路径与系统性风险", "title_en": "10 Ways To Hack Into A Vibecoder S Clawdbot Get Entire Human Identity Educational Purposes Only", "author": "", "date": "2026-01-29", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "creative", "infra" ], "tldr": "非技术用户在缺乏安全加固的 VPS 上自建 AI Agent，会因权限泛滥和隔离缺失，把整个数字身份变成一个可被一键摧毁的单点故障。", "url": "/articles/10-ways-to-hack-into-a-vibecoder-s-clawdbot-get-entire-human-identity-educational-purposes-only/", "readMin": 28, "image": "image-1.jpg" }, { "slug": "2026-01-29-immersivetran-x-2015978209209434389", "title": "Our World in Data——被低估的全球数据认知工具", "title_en": "2026 01 29 Immersivetran X 2015978209209434389", "author": "", "date": "2026-01-29", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "org", "agent" ], "tldr": "Our World in Data 通过\"数据+解读+可视化\"一体化，把分散的全球发展数据转化成可验证、可传播的认知工具，但其\"看穿真相\"的宣传语掩盖了数据解读本身的价值观预设。", "url": "/articles/2026-01-29-immersivetran-x-2015978209209434389/", "readMin": 4, "image": "image-1.jpg" }, { "slug": "clawd-molt-bot-s-memory-sucks-we-gave-it-supermemory", "title": "Clawd/Molt 记忆差，supermemory 借势切入但营销明显过头", "title_en": "Clawd Molt Bot S Memory Sucks We Gave It Supermemory", "author": "", "date": "2026-01-29", "category": [ "阿头学", "讨论题" ], "topics": [ "memo", "agent", "build" ], "tldr": "这篇内容抓住了 Agent“记忆不该完全依赖工具调用”这个真问题，但把 supermemory 包装成“完美记忆”明显是营销性夸大。", "url": "/articles/clawd-molt-bot-s-memory-sucks-we-gave-it-supermemory/", "readMin": 3, "image": "image-1.jpg" }, { "slug": "我开源了一个工具-10-分钟让-clawdbot-接入飞书机器人", "title": "Clawdbot 接入飞书的开源桥接工具，价值真实但“10分钟”明显说大了", "title_en": "我开源了一个工具 10 分钟让 Clawdbot 接入飞书机器人", "author": "", "date": "2026-01-29", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "org", "infra" ], "tldr": "这是一个抓住真实痛点的桥接工具，但它证明的只是“能跑通”，绝不是“10分钟低风险接入飞书”的成熟方案。", "url": "/articles/我开源了一个工具-10-分钟让-clawdbot-接入飞书机器人/", "readMin": 3, "image": "image-1.jpg" }, { "slug": "i-told-claude-code-to-build-me-an-executive-assistant-this-is-what-my-work-as-cto-looks-like-now", "title": "把 Claude Code 当执行助理：你要的不是工具，是“工作操作系统”", "title_en": "I Told Claude Code To Build Me An Executive Assistant This Is What My Work As Cto Looks Like Now", "author": "", "date": "2026-01-28", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "org", "agent", "creative" ], "tldr": "AI 助理的拐点不在“更会聊天”，而在它能把你的碎片输入（语音/会议转录/临时指令）自动沉淀成可检索的组织知识，并且直接代你执行（Slack/Calendar/Linear）——你只管说人话。", "url": "/articles/i-told-claude-code-to-build-me-an-executive-assistant-this-is-what-my-work-as-cto-looks-like-now/", "readMin": 14 }, { "slug": "your-fast-money-brain-will-make-you-lose-everything", "title": "加密里真正的输赢：不是赚到钱，是熬过周期还在场", "title_en": "Your Fast Money Brain Will Make You Lose Everything", "author": "", "date": "2026-01-28", "category": [ "阿头学" ], "topics": [ "org", "invest" ], "tldr": "在加密世界，“搞快钱脑”最大的成本不是亏一次钱，而是把你训练成永远守不住钱的人。", "url": "/articles/your-fast-money-brain-will-make-you-lose-everything/", "readMin": 42 }, { "slug": "full-guide-how-to-unlock-extreme-focus-on-command", "title": "极致专注不是自律，是把「清晰度」拉满", "title_en": "Full Guide How To Unlock Extreme Focus On Command", "author": "", "date": "2026-01-28", "category": [ "阿头学", "Uota学" ], "topics": [ "creative", "personal", "build" ], "tldr": "你不是没时间，你是脑子里后台程序太多——把任务变清晰、重要、紧迫，1 小时能打出别人 4 小时的产出。", "url": "/articles/full-guide-how-to-unlock-extreme-focus-on-command/", "readMin": 26 }, { "slug": "how-to-10x-your-writing-in-the-next-5-minutes", "title": "写作想变强，先把「编辑」当成第二份工作", "title_en": "How To 10X Your Writing In The Next 5 Minutes", "author": "", "date": "2026-01-28", "category": [ "阿头学", "Uota学" ], "topics": [ "creative" ], "tldr": "写作的大幅进步不靠更聪明的句子，而靠一套反直觉但可复制的“可读性流程”。", "url": "/articles/how-to-10x-your-writing-in-the-next-5-minutes/", "readMin": 3 }, { "slug": "how-to-fix-your-iphone-photography-in-1-day", "title": "一天让你的iPhone摄影脱胎换骨——精通不性感，系统才性感", "title_en": "How To Fix Your Iphone Photography In 1 Day", "author": "", "date": "2026-01-28", "category": [], "topics": [ "creative" ], "tldr": "你的照片和那些让人停下滑动的照片之间的差距，不是天赋不是运气，而是枯燥的、一次设好就能用很久的设置。", "url": "/articles/how-to-fix-your-iphone-photography-in-1-day/", "readMin": 14 }, { "slug": "how-clawdbot-remembers-everything", "title": "AI 记忆不是“更大上下文”，而是可编辑文件 + 可检索索引", "title_en": "How Clawdbot Remembers Everything", "author": "", "date": "2026-01-28", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "memo", "prompt", "build" ], "tldr": "真正可用的“长期记忆”不靠把一切塞进 prompt，而是把记忆落到用户可控的本地文件（Markdown），再用混合检索（semantic + keyword）按需召回——透明、便宜、可版本控制。", "url": "/articles/how-clawdbot-remembers-everything/", "readMin": 10 }, { "slug": "x-post-a-few-random-notes-from-claude-coding-quite-a-bit-last-few-weeks-coding-workflo", "title": "代理写代码把工程师逼成“规格制定者”（但你仍得当鹰眼）", "title_en": "X Post A Few Random Notes From Claude Coding Quite A Bit Last Few Weeks Coding Workflo", "author": "", "date": "2026-01-27", "category": [ "阿头学", "讨论题" ], "topics": [ "agent", "org", "creative" ], "tldr": "LLM 代理带来的不是“更快写代码”，而是把你的工作从敲键盘升级成定义成功标准；但现在的模型仍会带着错误假设一路狂奔，你必须用 IDE + 审稿心态兜底。", "url": "/articles/x-post-a-few-random-notes-from-claude-coding-quite-a-bit-last-few-weeks-coding-workflo/", "readMin": 6 }, { "slug": "x-post-a-neurotechnology-breakthrough-may-have-just-turned-science-fiction-into-reali", "title": "清醒梦“人际通信”更像实验链路验证，不是心灵共享突破", "title_en": "X Post A Neurotechnology Breakthrough May Have Just Turned Science Fiction Into Reali", "author": "", "date": "2026-01-27", "category": [ "阿头学", "讨论题" ], "topics": [ "memo", "build", "personal" ], "tldr": "这条新闻最值得关注的不是“梦中通话成真”，而是清醒梦可能被初步接口化了，但当前证据远不足以支撑“首次人际通信”这种夸张表述。", "url": "/articles/x-post-a-neurotechnology-breakthrough-may-have-just-turned-science-fiction-into-reali/", "readMin": 1, "image": "" }, { "slug": "对话minimax-agent团队-没有agent企业敢说自己有壁垒", "title": "Agent 没壁垒？真正的壁垒是“上下文+组织+Infra”", "title_en": "对话Minimax Agent团队 没有Agent企业敢说自己有壁垒", "author": "", "date": "2026-01-26", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "agent", "org" ], "tldr": "Agent 的竞争不在功能点，而在谁能吃下真实工作流：桌面端抢的是上下文与身份，最后拼的是 Infra 与组织执行力。", "url": "/articles/对话minimax-agent团队-没有agent企业敢说自己有壁垒/", "readMin": 55 }, { "slug": "style-experiments", "title": "三轮改写，把\"正确的废话\"变成\"让人想转发的文章\"", "title_en": "Style Experiments", "author": "", "date": "2026-01-19", "category": [ "阿头学", "Uota学" ], "topics": [ "creative", "org", "personal" ], "tldr": "写作不是天赋问题，是工程问题——结构→论证→节奏，三轮 pass 就能把任何一篇公文改成有人味的文章。", "url": "/articles/style-experiments/", "readMin": 12 }, { "slug": "鄙视链", "title": "不玩游戏的人，鄙视链伤不了他", "title_en": "鄙视链", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "org" ], "tldr": "鄙视链的杀伤力不在于链条本身，在于你默认自己是玩家——不参加游戏，游戏规则就对你无效。", "url": "/articles/鄙视链/", "readMin": 8 }, { "slug": "紧急撤离", "title": "你抱着的行李箱，叫什么名字？", "title_en": "紧急撤离", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "build" ], "tldr": "紧急撤离时抱着行李箱跳滑梯的人不是个例——我们每个人都在日常生活里紧抱着自己的\"行李箱\"，只是没有充气滑梯帮你看清这一点。", "url": "/articles/紧急撤离/", "readMin": 6 }, { "slug": "你应该更努力才对", "title": "苍蝇在动车上飞到车头，以为自己时速205公里", "title_en": "你应该更努力才对", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "org" ], "tldr": "\"你应该更努力\"是成功者最精致的自恋——把火车头的力量当成自己翅膀的振幅，然后对站台上的人喊加油。", "url": "/articles/你应该更努力才对/", "readMin": 6 }, { "slug": "执而不着", "title": "拿得起放得下，才是真本事", "title_en": "执而不着", "author": "", "date": "2026-01-18", "category": [ "阿头学", "Uota学" ], "topics": [ "build", "org" ], "tldr": "戒掉什么不是解药，能拿着不黏上去才是——执而不着是比断舍离更高级的操作系统。", "url": "/articles/执而不着/", "readMin": 9 }, { "slug": "那些我喜欢写的事情", "title": "平凡人做到极限，和伟大人物做到极限，是同一件事", "title_en": "那些我喜欢写的事情", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "build" ], "tldr": "在满屏iPhone和马斯克传记的早上，一个15岁肾病少年写的自动关机软件，才是真正的科技与人的故事。", "url": "/articles/那些我喜欢写的事情/", "readMin": 7 }, { "slug": "为弟弟祈祷", "title": "当你知道结局注定，你选择怎么活？", "title_en": "为弟弟祈祷", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "org", "personal" ], "tldr": "和菜头用一只猫的死亡，写出了所有人都在回避的问题：当结局不可逆时，你是选择延长痛苦还是最大化快乐？", "url": "/articles/为弟弟祈祷/", "readMin": 8 }, { "slug": "网友原来不读书", "title": "你以为你读了，其实你只是刷了", "title_en": "网友原来不读书", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "build" ], "tldr": "200 字截图传播千万次，恰恰证明没人读完那 50 万字——信息消费正在从\"理解\"退化为\"情绪条件反射\"。", "url": "/articles/网友原来不读书/", "readMin": 7 }, { "slug": "穿牛仔服的国王-马斯克传-读后", "title": "时代需要恶棍，因为好人不敢砍人", "title_en": "穿牛仔服的国王 马斯克传 读后", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "invest" ], "tldr": "和菜头把马斯克读成了\"时代的选择\"——不是天才造就了时代，而是一个变软的时代需要恶棍来重新激活人类的勇气和野心。", "url": "/articles/穿牛仔服的国王-马斯克传-读后/", "readMin": 8 }, { "slug": "做人太辛苦", "title": "自主的代价才是人的入场券", "title_en": "做人太辛苦", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "build", "personal" ], "tldr": "\"不如做爱宠\"是 2023 年版的\"不如做山民\"——20 年间，人们对自主的恐惧从\"少知道一点\"滑到了\"干脆别做人\"。", "url": "/articles/做人太辛苦/", "readMin": 6 }, { "slug": "抱歉-占用社会公共资源一下", "title": "道歉信里藏着的不是歉意，是甩锅话术", "title_en": "抱歉 占用社会公共资源一下", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "org" ], "tldr": "\"抱歉占用社会公共资源\"这句万能开头，本质是用文字游戏把指责伪装成道歉——暗示公众\"你们继续追究才是浪费资源\"。", "url": "/articles/抱歉-占用社会公共资源一下/", "readMin": 5 }, { "slug": "谢谢你-王医生", "title": "最好的合作关系长什么样？一个兽医和一只猫的故事", "title_en": "谢谢你 王医生", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "personal" ], "tldr": "和菜头通过与猫弟主治医王医生的故事，写出了一个理想协作关系的全部要素：选对人、全力信任、允许犯错、在终局时互相托住。", "url": "/articles/谢谢你-王医生/", "readMin": 9 }, { "slug": "858天", "title": "858天后回家，才发现看不厌的只有蓝天白云", "title_en": "858天", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "memo" ], "tldr": "远方看过了，未来看过了，人也看过了——看不厌的始终只有家乡的蓝天和白云。", "url": "/articles/858天/", "readMin": 5 }, { "slug": "心灵旅店租客", "title": "你不是房东，你只是租客——友情的产权幻觉", "title_en": "心灵旅店租客", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "org" ], "tldr": "友情的痛苦不来自对方的冷漠，来自你把租约当成了房产证。", "url": "/articles/心灵旅店租客/", "readMin": 8 }, { "slug": "回家的感觉", "title": "从北京的一套房子回到昆明的一套房子——慢下来不是躺平，是充电", "title_en": "回家的感觉", "author": "", "date": "2026-01-18", "category": [ "阿头学" ], "topics": [ "creative", "personal" ], "tldr": "你以为回家是逃离，其实是校准——当你能对着阳台上的衣架看半小时不厌倦，说明你的感知系统还没坏。", "url": "/articles/回家的感觉/", "readMin": 5 }, { "slug": "火星任务在地球", "title": "你的火星任务不在火星——把日常苦难升格为史诗叙事", "title_en": "火星任务在地球", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "invest", "build", "org" ], "tldr": "生活之所以难以忍受，不是因为它太难，而是因为你把火星任务当成了通勤。", "url": "/articles/火星任务在地球/", "readMin": 7 }, { "slug": "iphone15和紫砂壶", "title": "iPhone15和紫砂壶——即时满足的反面不是延迟满足，是赋予意义", "title_en": "Iphone15和紫砂壶", "author": "", "date": "2026-01-18", "category": [ "阿头学" ], "topics": [ "creative", "org" ], "tldr": "花一个月零十六天做一把壶，不是为了壶，是为了证明\"为了你我愿意做麻烦事\"这句话不是空话。", "url": "/articles/iphone15和紫砂壶/", "readMin": 7 }, { "slug": "家常菜和预制菜", "title": "家常菜和预制菜——你外包的不是家务，是健康", "title_en": "家常菜和预制菜", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "personal" ], "tldr": "外卖越吃越饿不是你嘴馋，是预制菜的诱食剂在替你的胃做决定。", "url": "/articles/家常菜和预制菜/", "readMin": 8 }, { "slug": "秋天-菌子开会", "title": "秋天，菌子开会——一顿全菌宴里的云南人骄傲", "title_en": "秋天 菌子开会", "author": "", "date": "2026-01-18", "category": [ "阿头学" ], "topics": [], "tldr": "真正喜欢吃菌子的人，一定要吃出每一种菌子的独特味道，而不是各种菌子混在一锅鸡汤里混沌一团。", "url": "/articles/秋天-菌子开会/", "readMin": 7 }, { "slug": "网络大合唱", "title": "网络大合唱——当一万个人重复同一句话，它也不会因此变成事实", "title_en": "网络大合唱", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [], "tldr": "网络大合唱之所以能唱起来，不是因为它正确，而是因为它汇聚了很多人内心扭曲表达的欲望。", "url": "/articles/网络大合唱/", "readMin": 6 }, { "slug": "在起点的形式", "title": "在起点的形式——你的第一版永远是1886年的奔驰，这没什么不对", "title_en": "在起点的形式", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "build" ], "tldr": "迷恋形式是一种病——你永远不动手，就永远不了解功能，也就断绝了任何形式变化的可能。", "url": "/articles/在起点的形式/", "readMin": 8 }, { "slug": "抖b红应有尽有-为何你就是不用", "title": "抖B红应有尽有，为何你就是不用——信息茧房的边界比你以为的近得多", "title_en": "抖B红应有尽有 为何你就是不用", "author": "", "date": "2026-01-18", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "creative", "build" ], "tldr": "抖B红的\"包罗万有\"只覆盖数字集体生活的模板，稍微偏离一点就是信息荒漠——而AI恰好能穿透这片荒漠。", "url": "/articles/抖b红应有尽有-为何你就是不用/", "readMin": 8 }, { "slug": "一件网事-排版", "title": "一件网事——不是所有人上网都是为了赢", "title_en": "一件网事 排版", "author": "", "date": "2026-01-18", "category": [ "阿头学", "Uota学" ], "topics": [ "creative" ], "tldr": "有人在网上无偿分享飞机制造图纸、整理古籍、录播放量十几次的黑胶专辑——原因仅仅是\"我高兴\"。", "url": "/articles/一件网事-排版/", "readMin": 3 }, { "slug": "个人阶段性大实话-音乐和设备", "title": "个人阶段性大实话——人的耳朵会适应，句号", "title_en": "个人阶段性大实话 音乐和设备", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [], "tldr": "更好的设备≠更好的声音，这个公式最大的误导是让人把注意力从音乐移开，转向设备差异——这就是成瘾养成。", "url": "/articles/个人阶段性大实话-音乐和设备/", "readMin": 8 }, { "slug": "问句扎心的话", "title": "问句扎心的话——\"你画了什么？\"", "title_en": "问句扎心的话", "author": "", "date": "2026-01-18", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "creative" ], "tldr": "人们缺的不是最强的推理模型，而是值得放进去跑一遍的问题。", "url": "/articles/问句扎心的话/", "readMin": 7 }, { "slug": "有人比我说得好", "title": "有人比我说得好——自律是恐惧，方法是逃避，完美环境不存在", "title_en": "有人比我说得好", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "invest" ], "tldr": "对方法的渴望意味着什么？每一个对方法的需求都表明了想要逃避的欲望。", "url": "/articles/有人比我说得好/", "readMin": 6 }, { "slug": "不经意间错过", "title": "不经意间错过——你什么都没有错过，你只是还没准备好", "title_en": "不经意间错过", "author": "", "date": "2026-01-18", "category": [ "阿头学" ], "topics": [], "tldr": "你有本事错过一次，就有本事继续错过9999次——在你没准备好之前，同样的东西在面前经过多少次都没用。", "url": "/articles/不经意间错过/", "readMin": 7 }, { "slug": "在最二的年纪遇见最二的他", "title": "在最二的年纪遇见最二的他——吊打老男人的人，打的其实是另一条路上的自己", "title_en": "在最二的年纪遇见最二的他", "author": "", "date": "2026-01-18", "category": [ "阿头学", "讨论题" ], "topics": [ "creative", "personal" ], "tldr": "批评老男人犯二的人，内心的潜台词是\"既然你已经拥有了那么多，就不要犯二\"——但这句话究竟是对老男人说的，还是对自己说的？", "url": "/articles/在最二的年纪遇见最二的他/", "readMin": 6 }, { "slug": "写在公众号拉黑功能升级之后", "title": "写在公众号拉黑功能升级之后——当你不再需要惩罚，惩罚的工具就到了", "title_en": "写在公众号拉黑功能升级之后", "author": "", "date": "2026-01-18", "category": [ "阿头学" ], "topics": [ "creative" ], "tldr": "一个幻想出来的作家自我，在幻想出来的创作行为中，和幻想出来的敌人争斗，渴望得到幻想出来的终极惩罚手段——这就是过去十年的总结。", "url": "/articles/写在公众号拉黑功能升级之后/", "readMin": 5 }, { "slug": "用ai-写就写了-别说", "title": "用AI写就写了，别说——片汤话才是文明的基座", "title_en": "用Ai 写就写了 别说", "author": "", "date": "2026-01-18", "category": [ "Uota学", "讨论题" ], "topics": [ "creative", "personal" ], "tldr": "世界就是形形色色的草台班子说着大差不差的片汤话，然后就从这一大锅草率的片汤里涌现出了人类文明之光。", "url": "/articles/用ai-写就写了-别说/", "readMin": 6 }, { "slug": "还真是郭靖郭大侠", "title": "还真是郭靖郭大侠——关于人，我又知道什么呢？", "title_en": "还真是郭靖郭大侠", "author": "", "date": "2026-01-18", "category": [ "阿头学" ], "topics": [ "creative" ], "tldr": "人是那么复杂，拥有那么多不同棱面，凭借我经常能看到的那几面，我真知道什么呢？", "url": "/articles/还真是郭靖郭大侠/", "readMin": 7 }, { "slug": "祈求智慧", "title": "智慧不是看清世界，是在扁担上找到「宽」", "title_en": "祈求智慧", "author": "", "date": "2026-01-18", "category": [ "阿头学", "Uota学" ], "topics": [ "creative", "org" ], "tldr": "真正的智慧不是聪明或狡诈，而是在极度逼仄的环境里找到简单、利他、且让所有人都能「宽」起来的解法。", "url": "/articles/祈求智慧/", "readMin": 5 }, { "slug": "dedao-12632", "title": "段永平的狠：只买看得懂的生意，然后敢等、敢熬", "title_en": "Dedao 12632", "author": "", "date": "2026-01-15", "category": [ "阿头学" ], "topics": [ "invest", "build", "creative" ], "tldr": "价值投资不是算得更准，而是用“买公司”的视角把自己逼进能力圈：等到市场失真时出手，剩下的交给时间。", "url": "/articles/dedao-12632/", "readMin": 6 }, { "slug": "how-to-hire-a-chief-of-staff", "title": "如何招聘幕僚长——CEO 的乘数器，不是高级跑腿", "title_en": "How To Hire A Chief Of Staff", "author": "Cassady Churchill、Jackie McQuiston（a16z Executive Talent）", "date": "2026-01-15", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build", "creative" ], "tldr": "Chief of Staff 的本质不是\"帮 CEO 做事\"，而是\"让 CEO 的 scope 变大\"——招错了就是多了一层官僚，招对了就是组织能力的乘数器。", "url": "/articles/how-to-hire-a-chief-of-staff/", "readMin": 3 }, { "slug": "why-are-we-here-why-did-we-raise-15b", "title": "a16z 不是在募资，是在给自己拿“美国必须赢”的授权书", "title_en": "Why Are We Here Why Did We Raise 15B", "author": "Ben Horowitz", "date": "2026-01-13", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "personal", "infra" ], "tldr": "a16z 用「给每个人一个机会」这套价值观，包装成「美国必须赢」的技术民族主义叙事，把 150 亿美金募资变成一种“历史责任”——顺便为未来的政策游说和产业下注提前正名。", "url": "/articles/why-are-we-here-why-did-we-raise-15b/", "readMin": 6 }, { "slug": "human-data-will-be-a-1-trillion-market", "title": "别再叫“标注”了——人类判断将成为万亿级生产要素", "title_en": "Human Data Will Be A 1 Trillion Market", "author": "", "date": "2026-01-13", "category": [ "阿头学", "讨论题" ], "topics": [ "org", "build" ], "tldr": "自动化不会让人类“没活干”，它会让人类的**判断（judgment）**变成可被结构化、可被购买的经济输入——于是“human data”不是瓶颈，而是长期复利的万亿市场。", "url": "/articles/human-data-will-be-a-1-trillion-market/", "readMin": 11 }, { "slug": "multiple-interests-do-not-waste-2-3-years", "title": "多兴趣不是毛病，缺的是「载体」", "title_en": "Multiple Interests Do Not Waste 2 3 Years", "author": "", "date": "2026-01-13", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "creative", "org" ], "tldr": "多兴趣者最大的浪费不是分心，而是没把兴趣装进一个能持续产出与分发的系统（vessel）。", "url": "/articles/multiple-interests-do-not-waste-2-3-years/", "readMin": 30 }, { "slug": "35-things-i-know-at-35-i-wish-i-knew-at-25", "title": "人生最大的护城河：每天出现、闭环、别再拿信息当借口", "title_en": "35 Things I Know At 35 I Wish I Knew At 25", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "personal" ], "tldr": "这 35 条“鸡汤”里真正有用的只有一个底层公式：**行动偏好 + 容忍不确定性 + 持续闭环**——在大多数人半途而废的世界里，这就是不公平优势。", "url": "/articles/35-things-i-know-at-35-i-wish-i-knew-at-25/", "readMin": 13 }, { "slug": "向量数据库一定会消失-从-seekdb-看-oltp-与-vector-的原生融合", "title": "向量库会被“收编”，别再把检索当外挂", "title_en": "向量数据库一定会消失 从 Seekdb 看 Oltp 与 Vector 的原生融合", "author": "", "date": "2026-01-13", "category": [ "阿头学", "讨论题" ], "topics": [ "build", "creative", "prompt" ], "tldr": "RAG 的坑不在 embedding，而在你把向量当外设：未来的检索能力会下沉到 OLTP 内核里，一条 SQL 把“逻辑+语义”一起做完。", "url": "/articles/向量数据库一定会消失-从-seekdb-看-oltp-与-vector-的原生融合/", "readMin": 22 }, { "slug": "you-won-t-be-the-same-person-in-6-months-how-to-master-anything-fast", "title": "6个月改头换面：把目标变成项目，用负反馈重塑身份", "title_en": "You Won T Be The Same Person In 6 Months How To Master Anything Fast", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "creative" ], "tldr": "掌握任何事的捷径不是更努力，而是让系统用“负反馈”把你逼成一个新人。", "url": "/articles/you-won-t-be-the-same-person-in-6-months-how-to-master-anything-fast/", "readMin": 29 }, { "slug": "how-to-articulate-yourself-intelligently", "title": "表达力不是天赋，是你反复打磨过多少次\"金曲\"", "title_en": "How To Articulate Yourself Intelligently", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学" ], "topics": [ "creative" ], "tldr": "你不需要更聪明才能讲得好——你需要的是 8-10 个被你反复锤过几百遍的核心观点，然后在任何场景里敢于重复它们。", "url": "/articles/how-to-articulate-yourself-intelligently/", "readMin": 17 }, { "slug": "lenny-10-skills-ai", "title": "AI 时代的 10 项技能——5 项永恒，5 项必须从零学", "title_en": "Lenny 10 Skills Ai", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学" ], "topics": [ "org", "build", "prompt" ], "tldr": "当执行成本趋近于零，品味、判断和\"写好 eval\"才是真正的稀缺资源——Lenny 用 320 期播客的数据验证了这个直觉。", "url": "/articles/lenny-10-skills-ai/", "readMin": 7 }, { "slug": "weixin-SM8BJzC7", "title": "微信存储优化——没有银弹，只有死磕细节的工程美学", "title_en": "Weixin Sm8Bjzc7", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学" ], "topics": [ "creative", "org", "memo" ], "tldr": "存储优化不是一个功能需求，是一场\"绝不丢数据\"前提下的系统工程持久战——这种约束下的产品思维，比功能本身更值得学。", "url": "/articles/weixin-SM8BJzC7/", "readMin": 7 }, { "slug": "the-personal-panopticon", "title": "个人全景敞视——当你把整个人生跑在一个终端里", "title_en": "The Personal Panopticon", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "invest", "build", "personal" ], "tldr": "几百年来可读性只向上流动——国家看你、企业看你、你看不见自己；现在工具反转了，问题变成：你敢不敢把塔建起来，又能不能不被它吞掉。", "url": "/articles/the-personal-panopticon/", "readMin": 11 }, { "slug": "how-to-fix-your-entire-life-in-1-day", "title": "改变人生不靠自律，靠的是换掉你脑子里那个\"我是谁\"", "title_en": "How To Fix Your Entire Life In 1 Day", "author": "", "date": "2026-01-13", "category": [ "阿头学", "讨论题" ], "topics": [ "personal", "org" ], "tldr": "你不是缺纪律，你是在无意识地追求一个保护旧身份的目标——真正的改变从换目标开始，而换目标就是换你这个人。", "url": "/articles/how-to-fix-your-entire-life-in-1-day/", "readMin": 31 }, { "slug": "how-to-publish-markdown-to-x-articles-with-claude-skill", "title": "用 Claude Skill 自动发布 X 长文——DRY 原则的极客实践", "title_en": "How To Publish Markdown To X Articles With Claude Skill", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学" ], "topics": [ "agent", "build", "creative" ], "tldr": "把 20 分钟的排版苦力活压缩到几分钟的自动化操作，这不是炫技，是 AI 时代创作者该有的基本素养——重复劳动交给机器，人只管思考和创作。", "url": "/articles/how-to-publish-markdown-to-x-articles-with-claude-skill/", "readMin": 9 }, { "slug": "be-delusionally-optimistic", "title": "妄想般乐观不是鸡汤，是你默认操作系统的一次覆写", "title_en": "Be Delusionally Optimistic", "author": "", "date": "2026-01-13", "category": [ "Uota学", "讨论题" ], "topics": [ "creative", "personal", "org" ], "tldr": "悲观让你看起来理性，乐观让你真的行动——但\"妄想般乐观\"只有用持续的硬功夫支撑才不是自我麻醉。", "url": "/articles/be-delusionally-optimistic/", "readMin": 6 }, { "slug": "dan-koe-seek-purpose", "title": "未来十年最重要的技能：能动性（Agency）", "title_en": "Dan Koe Seek Purpose", "author": "", "date": "2026-01-13", "category": [ "阿头学", "Uota学", "讨论题" ], "topics": [ "build", "creative", "org" ], "tldr": "AI 会让具体技能贬值，但会让“敢自己设目标并迭代到成”的人更值钱。", "url": "/articles/dan-koe-seek-purpose/", "readMin": 22 }, { "slug": "nicolas-cole-writing", "title": "删掉这 15 个\"小词\"，你的句子立刻提升一个档次", "title_en": "Nicolas Cole Writing", "author": "", "date": "2026-01-13", "category": [ "阿头学" ], "topics": [ "memo", "prompt", "org" ], "tldr": "写作的核心减法不是删段落，是删那些你根本没意识到在拖后腿的\"小词\"——it、so、just、that、well——它们是句子里的隐形脂肪。", "url": "/articles/nicolas-cole-writing/", "readMin": 3 } ] }