返回列表
🧠 阿头学 · 🪞 Uota学

AI 时代的 10 项技能——5 项永恒，5 项必须从零学

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。

2026-01-13
阅读简报
双语对照
完整翻译
原文
讨论归档

核心观点

  • 品味和判断是 AI 时代的终极瓶颈 当 AI 能无限生成选项时，能从中挑出对的那个才是真本事。Lenny 用"exposure hours"来量化品味的培养——这不是天赋，是投入时间的函数。对 ATou 来说，这直接验证了"审美即竞争力"的判断。
  • "Builder"身份比任何职能标签都值钱 "Dissolve role boundaries and call ourselves builders"——这句话在 20 人团队里尤其致命。Neta 的特种作战阵型本质上就是在要求每个人都是 builder，不是某个职能的螺丝钉。
  • 写 Eval 是 AI 能力的真正天花板 "AI is almost capped by how good we are at evals"——这条被严重低估了。大多数人还在纠结 prompt 写得好不好，但真正决定 AI 产出上限的是你能不能定义"什么算好"。这是 Context Engineer 的核心技能。
  • 管理 AI Agent = 管理人的技能迁移 "Used to be people, but now it's basically AI models"——管理技能直接迁移到 Agent 协作。设目标、给反馈、划边界，这套东西对 Uota 的架构设计有直接启发。
  • 战略思维的杠杆在执行成本下降时暴涨 当执行变便宜，想错方向的代价反而更大——因为你能更快地在错误方向上跑得更远。这是 ATou 2026 战略（海外产品/增长/品牌）需要反复校准的原因。

跟我们的关联

  • 👤ATou：品味 + 判断 + 战略思维是 ATou Gallup Top 5（学习/搜集/理念/前瞻/思维）的天然延伸。意味着 ATou 的天赋组合在 AI 时代反而更值钱了——但前提是把"搜集"转化为"判断"，而不是停留在信息囤积。下一步：刻意练习从"收集 10 个选项"到"5 秒内砍到 2 个"的决策速度。
  • 🪞Uota：写 Eval 和管理 Agent 直接关联 Uota 的架构。当前 Uota 的 self-review gate 已经在做 eval 的雏形，但还不够系统化。下一步：把 eval 从"三审自检"升级为可量化的质量基准。
  • 🧠Neta：20 人团队 = 每个人必须是 builder。招人和评估标准应该从"你会什么技能"转向"你能不能端到端交付一个东西"。

讨论引子

1. Neta 团队里，谁已经是 builder、谁还是"某个职能的人"？如果要求每个人都能端到端交付，会淘汰谁？ 2. ATou 作为 Context Engineer，"写 eval"这件事在日常工作中占多大比重？如果答案是"很少"，是不是说明最关键的杠杆点反而被忽略了？ 3. 品味靠"exposure hours"培养——但 ATou 的 exposure 是否足够多元？还是一直在同一个信息茧房里转？

Lenny Rachitsky on X: "I asked Claude Cowork to identify the 10 most important skills for thriving in the age of AI, based on my 320 podcast conversations.

I asked Claude Cowork to identify the 10 most important skills for thriving in the age of AI, based on my 320 podcast conversations. Impressed with the results. Part 1: Timeless Skills (become more valuable)1. Taste and judgment — The bottleneck when AI generates unlimited options. Develop through "exposure hours." —

  1. Curiosity — The meta-skill that enables all other learning.

says it's what he'd prioritize for children in an AI world. 3. Becoming a cross-functional "builder" — "Dissolve role boundaries and call ourselves builders." —

  1. Clear communication and storytelling — As execution is automated, articulation becomes your primary output. 5. Strategic thinking — "The leverage of getting strategy right goes up when execution costs go down." Part 2: AI-native skills (must develop)1. Writing evals — "AI is almost capped by how good we are at evals." —

  2. Prompting and context engineering — "Great prompters are great writers." 3. AI fluency through constant use — You can't understand AI by reading about it. Cancel your meetings and play with every AI product. 4. Understanding systems under the hood — Paradoxically, fundamentals become MORE valuable as AI abstracts them away. 5. Working with AI Agents as teammates — Management skills transfer directly. "Used to be people, but now it's basically AI models." —

相关笔记

🧭 主题 MOC

  • [[AI MOC|AI]]：(MOC) 整理 AI 时代 10 项关键技能，包含「evals」与「AI Agents」协作等 AI-native 能力。

🎯 Timeless skills

  • [[30 Wiki/33 商业_创业/how to start google|练技术]]：(Wiki) 用持续「build stuff」与 Jobs 学书法的例子说明「品味/判断」来自长期兴趣驱动的投入。
  • [[40 Library/41 读书笔记/乔布斯的魔力演讲/2022-11-14-08-36-13|说清楚]]：(乔布斯的魔力演讲) 把复杂技术讲成可理解语言，直接对应 AI 自动化后更稀缺的「清晰沟通/叙事」。
  • [[40 Library/41 读书笔记/亚马逊逆向工作法/2022-12-22-10-35-29|目标倒推]]：(亚马逊逆向工作法) 强调从目标与用户需要出发再选手段，呼应执行成本下降后更重要的「战略思考」。
  • [[40 Library/41 读书笔记/系统之美/2025-04-03-10-12-09|修补系统]]：(系统之美) 区分“理解系统”与“动手修补”，提醒 AI 抽象之上仍要掌握「系统/基本功」才能有效用 Agent。

⚙️ AI-native skills

  • [[30 Wiki/36 AI_Industry/2023-12-29-07-45-23|意图交互]]：(Wiki) 交互从“命令”走向“意图”要求更强的「上下文工程/需求拆解」与迭代写作能力。
  • [[30 Wiki/36 AI_Industry/2023-12-28-14-37-04|嵌入工作流]]：(Wiki) 把 LLM 当作可嵌入「工作流」的能力而非聊天壳，更贴近 prompting/context engineering 的实战落点。
  • [[40 Library/41 读书笔记/亚马逊逆向工作法/2022-12-14-19-45-40|指标迭代]]：(亚马逊逆向工作法) 指标体系需随阶段迭代，可类比写好「evals」要持续校准评估口径与目标。
  • [[00 Inbox/Flomo_Import/2025-05-23-10-13-57|最后1%]]：(Flomo) “99% 易、1% 难”提示 AI 产出上限常卡在「评估/测试」与细节正确性。
  • [[00 Inbox/Flomo_Import/2021-05-22-09-31-22|管理=赋能]]：(Flomo) 把管理定义为让身边的人变强，可迁移为带 AI 「Agents」：设目标/反馈/边界而不是替它做。
  • [[40 Library/42 人物_传记/Pieter_Levels/2024-09-16-20-49-20|先上线]]：(Pieter_Levels) 用“落地页+收款+迭代”展示跨职能「Builder」的最小闭环，与 AI 时代“角色边界溶解”同频。

⚔️ L2 对立

  • [[30 Wiki/36 AI_Industry/2024-04-07-19-43-21|开新世界]]：(Wiki) 把 AI 优先用于打开「可能性」而非提效，强调探索与审美协作的路线。
  • [[30 Wiki/36 AI_Industry/2023-05-22-15-05-53|降本提效]]：(Wiki) 把 AI 定位为降低「生产成本」并鼓励“成为写 Prompt 的人”，代表提效优先路线。

I asked Claude Cowork to identify the 10 most important skills for thriving in the age of AI, based on my 320 podcast conversations. Impressed with the results. Part 1: Timeless Skills (become more valuable)1. Taste and judgment — The bottleneck when AI generates unlimited options. Develop through "exposure hours." —

I asked Claude Cowork to identify the 10 most important skills for thriving in the age of AI, based on my 320 podcast conversations. Impressed with the results. Part 1: Timeless Skills (become more valuable)1. Taste and judgment — The bottleneck when AI generates unlimited options. Develop through "exposure hours." —

  1. Curiosity — The meta-skill that enables all other learning.
  1. Curiosity — The meta-skill that enables all other learning.

says it's what he'd prioritize for children in an AI world. 3. Becoming a cross-functional "builder" — "Dissolve role boundaries and call ourselves builders." —

says it's what he'd prioritize for children in an AI world. 3. Becoming a cross-functional "builder" — "Dissolve role boundaries and call ourselves builders." —

  1. Clear communication and storytelling — As execution is automated, articulation becomes your primary output. 5. Strategic thinking — "The leverage of getting strategy right goes up when execution costs go down." Part 2: AI-native skills (must develop)1. Writing evals — "AI is almost capped by how good we are at evals." —
  1. Clear communication and storytelling — As execution is automated, articulation becomes your primary output. 5. Strategic thinking — "The leverage of getting strategy right goes up when execution costs go down." Part 2: AI-native skills (must develop)1. Writing evals — "AI is almost capped by how good we are at evals." —
  1. Prompting and context engineering — "Great prompters are great writers." 3. AI fluency through constant use — You can't understand AI by reading about it. Cancel your meetings and play with every AI product. 4. Understanding systems under the hood — Paradoxically, fundamentals become MORE valuable as AI abstracts them away. 5. Working with AI Agents as teammates — Management skills transfer directly. "Used to be people, but now it's basically AI models." —
  1. Prompting and context engineering — "Great prompters are great writers." 3. AI fluency through constant use — You can't understand AI by reading about it. Cancel your meetings and play with every AI product. 4. Understanding systems under the hood — Paradoxically, fundamentals become MORE valuable as AI abstracts them away. 5. Working with AI Agents as teammates — Management skills transfer directly. "Used to be people, but now it's basically AI models." —

相关笔记

相关笔记

🧭 主题 MOC

  • [[AI MOC|AI]]：(MOC) 盘点 AI 时代的关键能力，覆盖「Evals」「Prompt」「Agents」等主题。

🧭 主题 MOC

  • [[AI MOC|AI]]：(MOC) 整理 AI 时代 10 项关键技能，包含「evals」与「AI Agents」协作等 AI-native 能力。

🎯 Timeless Skills：判断/表达/战略

  • [[40 Library/41 读书笔记/10x_is_easier_than_2x/2024-04-16-23-50-51|注意力瓶颈]]：(10x_is_easier_than_2x) 选项无限时真正稀缺的是「注意力/判断」，10x 的聚焦训练可作为「品味」的实践抓手。
  • [[00 Inbox/Flomo_Import/2021-05-22-10-47-38|呈现即抵达]]：(Flomo) 提醒「清晰沟通」里“信息呈现”常比信息本身更关键，叙事与表达是可被训练的主输出。
  • [[40 Library/41 读书笔记/亚马逊逆向工作法/2022-12-22-10-35-29|逆向工作]]：(亚马逊逆向工作法) 用“从目标/用户出发”校准「战略思维」：先定方向再决定执行细节。

🎯 Timeless skills

  • [[30 Wiki/33 商业_创业/how to start google|练技术]]：(Wiki) 用持续「build stuff」与 Jobs 学书法的例子说明「品味/判断」来自长期兴趣驱动的投入。
  • [[40 Library/41 读书笔记/乔布斯的魔力演讲/2022-11-14-08-36-13|说清楚]]：(乔布斯的魔力演讲) 把复杂技术讲成可理解语言，直接对应 AI 自动化后更稀缺的「清晰沟通/叙事」。
  • [[40 Library/41 读书笔记/亚马逊逆向工作法/2022-12-22-10-35-29|目标倒推]]：(亚马逊逆向工作法) 强调从目标与用户需要出发再选手段，呼应执行成本下降后更重要的「战略思考」。
  • [[40 Library/41 读书笔记/系统之美/2025-04-03-10-12-09|修补系统]]：(系统之美) 区分“理解系统”与“动手修补”，提醒 AI 抽象之上仍要掌握「系统/基本功」才能有效用 Agent。

⚙️ AI-native Skills：Evals / Prompt / Agents

  • [[30 Wiki/31 AI_Prompts/长文反馈专家|长文反馈]]：(Wiki) 以结构/逻辑/语言的 checklist 做「评估」与迭代，把“写 evals”落到可复用的标准与反馈回路。
  • [[30 Wiki/31 AI_Prompts/网站规划|建站规划]]：(Wiki) 把需求访谈→站点地图→技术约束写成 prompt 流程，示范跨职能「builder」如何用 AI 从想法到交付。
  • [[30 Wiki/36 AI_Industry/2023-12-28-23-04-54|当老板]]：(Wiki) 把模型升级为「Agent」并用多版本+反馈推进交付，等价于练习“与 AI 同事协作/管理”。
  • [[30 Wiki/36 AI_Industry/2024-02-24-10-11-48|可感知性]]：(Wiki) 反向提醒「prompt/上下文工程」的边界：纯文字意图难被感知，需要示例/反馈降低对齐成本。

⚙️ AI-native skills

  • [[30 Wiki/36 AI_Industry/2023-12-29-07-45-23|意图交互]]：(Wiki) 交互从“命令”走向“意图”要求更强的「上下文工程/需求拆解」与迭代写作能力。
  • [[30 Wiki/36 AI_Industry/2023-12-28-14-37-04|嵌入工作流]]：(Wiki) 把 LLM 当作可嵌入「工作流」的能力而非聊天壳，更贴近 prompting/context engineering 的实战落点。
  • [[40 Library/41 读书笔记/亚马逊逆向工作法/2022-12-14-19-45-40|指标迭代]]：(亚马逊逆向工作法) 指标体系需随阶段迭代，可类比写好「evals」要持续校准评估口径与目标。
  • [[00 Inbox/Flomo_Import/2025-05-23-10-13-57|最后1%]]：(Flomo) “99% 易、1% 难”提示 AI 产出上限常卡在「评估/测试」与细节正确性。
  • [[00 Inbox/Flomo_Import/2021-05-22-09-31-22|管理=赋能]]：(Flomo) 把管理定义为让身边的人变强，可迁移为带 AI 「Agents」：设目标/反馈/边界而不是替它做。
  • [[40 Library/42 人物_传记/Pieter_Levels/2024-09-16-20-49-20|先上线]]：(Pieter_Levels) 用“落地页+收款+迭代”展示跨职能「Builder」的最小闭环，与 AI 时代“角色边界溶解”同频。

🔄 底层与系统：fundamentals 更值钱

  • [[40 Library/41 读书笔记/系统之美/2025-04-03-10-12-09|修补系统]]：(系统之美) 强调理解「系统」与动手修补是两码事，呼应“懂底层”在 AI 抽象时代更值钱。

⚔️ L2 对立

  • [[30 Wiki/36 AI_Industry/2024-04-07-19-43-21|开新世界]]：(Wiki) 把 AI 优先用于打开「可能性」而非提效，强调探索与审美协作的路线。
  • [[30 Wiki/36 AI_Industry/2023-05-22-15-05-53|降本提效]]：(Wiki) 把 AI 定位为降低「生产成本」并鼓励“成为写 Prompt 的人”，代表提效优先路线。

⚔️ L2 对立：用 AI 开“可能性” vs 用 AI 做“提效”

  • [[30 Wiki/36 AI_Industry/2024-04-07-19-43-21|可能性优先]]：(Wiki) 把 AI 用在「可能性/探索」而非纯提效，连接「好奇心」这类元技能的增长路径。
  • [[30 Wiki/36 AI_Industry/2023-05-22-15-05-53|降本提效]]：(Wiki) 代表“AI for productivity”路线：尽量让工作可被 AI 加速/替代，把人从执行挪到「调度/写 prompt」。

Lenny Rachitsky on X: "I asked Claude Cowork to identify the 10 most important skills for thriving in the age of AI, based on my 320 podcast conversations.

  • Source: https://x.com/lennysan/status/2010884315794849901?s=20
  • Mirror: https://r.jina.ai/https://x.com/lennysan/status/2010884315794849901?s=20
  • Published: Tue, 13 Jan 2026 06:55:15 GMT
  • Saved: 2026-01-13

Content

I asked Claude Cowork to identify the 10 most important skills for thriving in the age of AI, based on my 320 podcast conversations. Impressed with the results. Part 1: Timeless Skills (become more valuable)1. Taste and judgment — The bottleneck when AI generates unlimited options. Develop through "exposure hours." —

  1. Curiosity — The meta-skill that enables all other learning.

says it's what he'd prioritize for children in an AI world. 3. Becoming a cross-functional "builder" — "Dissolve role boundaries and call ourselves builders." —

  1. Clear communication and storytelling — As execution is automated, articulation becomes your primary output. 5. Strategic thinking — "The leverage of getting strategy right goes up when execution costs go down." Part 2: AI-native skills (must develop)1. Writing evals — "AI is almost capped by how good we are at evals." —

  2. Prompting and context engineering — "Great prompters are great writers." 3. AI fluency through constant use — You can't understand AI by reading about it. Cancel your meetings and play with every AI product. 4. Understanding systems under the hood — Paradoxically, fundamentals become MORE valuable as AI abstracts them away. 5. Working with AI Agents as teammates — Management skills transfer directly. "Used to be people, but now it's basically AI models." —

相关笔记

🧭 主题 MOC

  • [[AI MOC|AI]]：(MOC) 盘点 AI 时代的关键能力，覆盖「Evals」「Prompt」「Agents」等主题。

🎯 Timeless Skills：判断/表达/战略

  • [[40 Library/41 读书笔记/10x_is_easier_than_2x/2024-04-16-23-50-51|注意力瓶颈]]：(10x_is_easier_than_2x) 选项无限时真正稀缺的是「注意力/判断」，10x 的聚焦训练可作为「品味」的实践抓手。
  • [[00 Inbox/Flomo_Import/2021-05-22-10-47-38|呈现即抵达]]：(Flomo) 提醒「清晰沟通」里“信息呈现”常比信息本身更关键，叙事与表达是可被训练的主输出。
  • [[40 Library/41 读书笔记/亚马逊逆向工作法/2022-12-22-10-35-29|逆向工作]]：(亚马逊逆向工作法) 用“从目标/用户出发”校准「战略思维」：先定方向再决定执行细节。

⚙️ AI-native Skills：Evals / Prompt / Agents

  • [[30 Wiki/31 AI_Prompts/长文反馈专家|长文反馈]]：(Wiki) 以结构/逻辑/语言的 checklist 做「评估」与迭代，把“写 evals”落到可复用的标准与反馈回路。
  • [[30 Wiki/31 AI_Prompts/网站规划|建站规划]]：(Wiki) 把需求访谈→站点地图→技术约束写成 prompt 流程，示范跨职能「builder」如何用 AI 从想法到交付。
  • [[30 Wiki/36 AI_Industry/2023-12-28-23-04-54|当老板]]：(Wiki) 把模型升级为「Agent」并用多版本+反馈推进交付，等价于练习“与 AI 同事协作/管理”。
  • [[30 Wiki/36 AI_Industry/2024-02-24-10-11-48|可感知性]]：(Wiki) 反向提醒「prompt/上下文工程」的边界：纯文字意图难被感知，需要示例/反馈降低对齐成本。

🔄 底层与系统：fundamentals 更值钱

  • [[40 Library/41 读书笔记/系统之美/2025-04-03-10-12-09|修补系统]]：(系统之美) 强调理解「系统」与动手修补是两码事，呼应“懂底层”在 AI 抽象时代更值钱。

⚔️ L2 对立：用 AI 开“可能性” vs 用 AI 做“提效”

  • [[30 Wiki/36 AI_Industry/2024-04-07-19-43-21|可能性优先]]：(Wiki) 把 AI 用在「可能性/探索」而非纯提效，连接「好奇心」这类元技能的增长路径。
  • [[30 Wiki/36 AI_Industry/2023-05-22-15-05-53|降本提效]]：(Wiki) 代表“AI for productivity”路线：尽量让工作可被 AI 加速/替代，把人从执行挪到「调度/写 prompt」。

📋 讨论归档

讨论进行中…