作者 ｜ 木子 " 我们会开始看到 "软件工程师"这个头衔慢慢消失。 可能会变成 builder、product manager ，或者头衔还保留，但只是一个遗留符号。 因为大家做的工作不再只是写代码：软件工程师还会写 spec、还会跟用户沟通。" 放出这话的，正是 Claude Code 的创始人 Boris Cherny 。 他最近在 Y Combinator 的一场圆桌访谈中，一人对阵四位 YC 高管，几乎句句都带着点"重锤感"。 在他看来， 编程正在被"解决" 。 在 Anthropic，很多人已经 70%–100% 用 Claude 写代码 ，IDE 的存在感正在下降。写代码这件事，正在从"核心能 力"变成"默认能力"。 而另一边， 模型能力会指数增长 ，今天的"勉强可用"，六个月后可能原生支持，如果只围绕当前模型做 PMF，很快会被下一代能力抹平： "在 Anthropic ，我们一直有一个核心理念： 我们不是为'今天的模型'做产品，而是为'六个月后的模型'做产品。 " 用 Boris 的话来说，就是： "我以前在 Meta 负责代码质量，也负责跨多个产品线的代码库质量。当时我们做"提升生产力"， ...

Core Insights - The article discusses the evolution of infrastructure providers like OpenAI and Anthropic, which are transitioning from merely supplying foundational AI capabilities to directly competing in the consumer application space with products like Sora2 and Claude Teams [1][2][3] - It emphasizes the strategic challenge for startups in this environment, suggesting that they should focus on creating defensible business models by cultivating "walled gardens" of proprietary data [2][3] Group 1: Infrastructure Providers and Competition - Infrastructure providers are now competing directly with startups by offering consumer-facing applications, moving beyond their initial role as mere suppliers of AI capabilities [1] - Companies like OpenAI and Anthropic are developing products that not only provide APIs but also complete productivity suites for enterprises, intensifying competition in the AI landscape [1][2] Group 2: The Concept of Walled Gardens - The article introduces the idea of "walled gardens" as areas where data access is restricted and proprietary, creating a competitive moat for companies that can cultivate such data [2][3] - High-quality, exclusive data is seen as a more sustainable competitive advantage than the models themselves, as the race for model scale and computational power will eventually converge [3] Group 3: Case Studies of Data Moats - VLex, a legal software company, has built a comprehensive legal database by acquiring and digitizing fragmented legal documents, establishing a strong data moat that supports its AI legal research tools [5][6] - OpenEvidence has developed a high-trust medical research database, allowing it to provide evidence-based answers to clinical questions, thus creating a superior user experience compared to general models [7] Group 4: Potential Areas for New Walled Gardens - The article identifies several sectors ripe for the creation of new data walled gardens, including: 1. Supply Chain and Logistics: Integrating proprietary trade data for predictive management [8][9] 2. Local and Municipal Government Records: Systematizing data for real estate and infrastructure developers [11][12] 3. Frontier Science: Aggregating research data to accelerate innovation [14][15] 4. Cultural and Creative Archives: Digitizing and structuring cultural resources for AI training [17] 5. Vertical Industry Processes: Targeting specialized data in overlooked markets [19][20] 6. Climate and Environmental Data: Creating a proprietary climate data repository for compliance and risk assessment [22][23] Group 5: Importance of Data Moats - The article concludes that while model companies will dominate in scale and computational resources, there exists an opportunity in fragmented, sensitive, or hard-to-access data areas where trust and data ownership are paramount [24] - Building a new data moat requires significant upfront investment and meticulous groundwork, but once established, it becomes nearly impossible to replicate, providing a lasting competitive edge in the AI landscape [24]