Mosh 分享如何用 Claude Code 零代码开发项目，揭秘 AI 如何改变软件开发流程和工程师角色。
Code with Mosh 创始人 Mosh Hamedani 花费 3 个月时间，使用 Claude Code 完全不手写代码构建了一个项目。AI 带来的效率提升约为 2 倍，将原本需要 5-6 个月的项目缩短至 3 个月。Mosh 指出软件开发瓶颈不是打字速度，而是项目早期的未知因素和隐性需求。工程师角色从代码生产者转变为技术决策的评审者和把关人。
完全不懂代码，一天 Vibe Coding 上线一个 App？ 这种一眼就是卖课的标题，完全曲解了 Vibe Coding，更曲解和贬低了生产级项目！ @moshhamedani 作为 Code ...
完全不懂代码，一天 Vibe Coding 上线一个 App？ 这种一眼就是卖课的标题，完全曲解了 Vibe Coding，更曲解和贬低了生产级项目！ @moshhamedani 作为 Code with Mosh 创始人、知名编程教育者，实际经历 3 个月，手写代码量为零，完全由 Claude Code 构建了他的项目，实际体感 AI 带来的效率提升大约是 2 倍。 第一层：软件开发的瓶颈从来不是打字速度 "为什么不是一天或一周？因为构建严肃的软件从来都不只是敲代码。" 项目早期的未知因素、测试和反馈中浮现的隐性需求——这些探索性工作占据了大部分时间。 AI 消除的是"实现"环节的摩擦（语法、查文档、写样板代码），但无法消除"发现该做什么"这个本质复杂度。 第二层：工程师的工作内容在迁移，而非消失 他描述自己的新角色非常具体：不再逐行审查代码，而是在更高层面审查——方案、架构、设计决策。 他不断质疑 AI、与其辩论、推动更简单更健壮的方案。他把这直接定义为"这就是软件工程"。 这是整篇帖子最关键的一句话：工程师从"代码生产者"转变为"技术决策的评审者和把关人"。 第三层：基本功的价值反而被重新定义，而不是贬值 他列出的清单值得注意：关注点分离、耦合与内聚、抽象、数据建模、API 设计、错误处理、安全、性能、测试——这些全是判断力层面的知识，没有一个是语法或 API 记忆。 他给出的检验标准也很实操：当 AI 提出一个架构、增加一层抽象、引入一个依赖时，你要能问出"这真的必要吗？有没有更简单的做法？能扩展吗？安全吗？六个月后会不会变成维护噩梦？" Mosh @moshhamedani I built my last project entirely with Claude Code. I didn’t write a single line of code by hand. It still took me nearly 3 months. Why so long and not a day or week? Because building serious software has never been just about typing code. Early on, there are tons of unknowns. As you build, test, and get feedback, you discover problems and requirements you couldn’t have anticipated. I spent a lot of time reviewing Claude’s work, not line by line, but at a higher level: the approach, architecture, and design decisions. I constantly challenged it, debated its choices, and pushed for simpler, more robust solutions. That’s software engineering. AI helped me bring my vision to life without spending countless hours typing code, wrestling with syntax, searching through documentation, and figuring out the mechanics of how to implement things. Without AI, building this app would have taken me 5-6 months if not longer. As I’ve said over and over, software engineering is not going away but AI is changing _how_ we build software. The tools are changing and the engineer’s role is evolving. And this is why knowing how to build good software still matters. You may not need to remember every API or piece of syntax anymore, but you still need to understand things like separation of concerns, coupling and cohesion, abstractions, data modeling, API design, error handling, security, performance, and testing. When AI proposes an architecture, creates an abstraction, adds another layer, or introduces a dependency, you need enough engineering knowledge to ask: Is this actually necessary? Is there a simpler way? Will this scale? Is this secure? Will this become a maintenance nightmare six months from now? AI can write the code. You still need to know what good code and good software look like. 🔗 View Quoted Tweet 💬 1 🔄 0 ❤️ 2 👀 393 📊 1 ⚡