Naval 和 Nivei 讨论 AI 驱动的编码工具（如 Claude Code）如何让非程序员用自然语言构建应用，导致应用海啸，但同时也聚焦于最佳和最细分的解决方案。
Naval and Nivei discuss how AI-powered coding tools like Claude Code enable non-coders to build applications using natural language, leading to a tsunami of apps but also a focus on the best and most niche solutions.
要点 · TL;DR
氛围编码将自然语言变成编程工具，让非程序员也能创建应用。 Vibe coding turns natural language into a programming tool, making app creation accessible to non-programmers.
AI 在分布内表现出色，但在分布外失败；人类的优势在于对抗性场景。 AI excels in-distribution but fails out-of-distribution; human edge lies in adversarial scenarios.
最好的应用赢得全部市场份额；平庸的应用没有需求。 The best app wins all market share; average apps have no demand.
核心观点 · Key points
AI 是抽象堆栈中的新一层，通过自然语言使编程变得人人可用。 AI is a new layer in the abstraction stack, making programming accessible via natural language.
氛围编码让非程序员能用英语作为编程语言，从想法直接构建应用。 Vibe coding lets non-programmers build apps from ideas using English as a programming language.
每个类别中最好的应用几乎赢得全部市场份额；平庸没有需求。 The best app in a category wins almost all market share; there is no demand for average.
AI 是企业家不知疲倦、无我的盟友，而非对他们工作的威胁。 AI is a tireless, egoless ally for entrepreneurs, not a threat to their jobs.
AI 擅长其训练分布内的任务，但在分布外问题上表现不佳。 AI excels at tasks within its training distribution but struggles with out-of-distribution problems.
反共识 · Contrarian takes
超级智能早已存在：计算器在算术上超越人类。 Superintelligence already exists: a calculator outperforms humans in arithmetic.
智力的唯一真正考验是从生活中得到你想要的；AI 失败因为它无所欲求。 The only true test of intelligence is getting what you want out of life; AI fails because it wants nothing.
不要学习提示工程；让 AI 适应你，因为它学习的速度比你适应它更快。 Don't learn prompt engineering; let AI adapt to you, as it learns faster than you can adapt.
AI 不会取代程序员；它让程序员更有杠杆，去取代其他所有人。 AI won't replace programmers; it makes programmers more leveraged to replace everyone else.
要担心的是与 AI 结合的不良人类，而非 AI 本身的对齐问题。 Worry about unaligned humans with AI, not unaligned AI itself.
本期章节 · Chapters（共 23）
开场与录制风格Introduction and recording style
Naval 当前工作与哲学Naval's current work and philosophy
氛围编程与新产品管理Vibe coding and the new product management
对中型企业及编程本质的影响Impact on Medium-Sized Firms and the Nature of Coding
传统软件工程未死Traditional Software Engineering Is Not Dead
赢家通吃市场与做到最好Winner-take-all markets and being the best
AI 的谄媚与个性化AI obsequiousness and personalization
程序员用 AI 取代他人Programmers replacing others with AI
创业者与 AI 代理Entrepreneurs and AI agency
AI 作为探索者、科学家和艺术家的盟友AI as an Ally for Explorers, Scientists, and Artists
AI 在压缩域与现实中的运作AI operates in compressed domain vs. reality
超级智能与计算器Superintelligence and calculators
AI 作为学习工具AI as a learning tool
智能的定义Definition of intelligence
AI 未通过智能测试AI fails the test of intelligence
人类在对抗场景中的优势Human edge in adversarial scenarios
AI 作为用户友好界面AI as a User-Friendly Interface
早期采用者优势Early Adopter Advantage
注意事项：幻觉与偏见Caveats: Hallucination and Bias
AI 作为学习工具AI as a Learning Tool
使用思考模型与为智能付费Using Thinking Models and Paying for Intelligence