Ask HN：AI 写代码比我好，我该如何保持自己的身份？
4 分•作者: jdw64•大约 1 个月前
如今，几乎所有人都在使用 AI 辅助编程，对吧？我是一名自由职业者，整个市场现在都基于 AI 辅助编码速度来设定截止日期。如果你不使用 AI，根本无法按时完成任务。因此，我在今年年初（大约从 GPT-5.3 开始）完全转向了 AI 驱动的编码。
并不是说 AI 编码不好玩。它很有趣。但一开始，AI 还有很多地方做得不够好。社区的焦点是“工程化”，附加 MCP（模型控制点）或深度提示工程。在那段时间里，我实际上构建并使用了不少定制的 AI 工具。
过去，你不仅要说明你的意图和方向，还要提供详细的分步说明才能达到目标。但最近，即使你省略了详细说明，Claude 和 GPT-5.6 也能完美处理。你甚至不必纠结于使用哪个模型。“图工程”目前很流行，但似乎模型已经原生学会了这些模式——如果你和它们建立一个高层计划，它们会自动将其执行策略嵌入其中。每当我尝试学习和应用一种新的 AI 工程技术时，下一个模型更新就已经开箱即用了，这让我的努力感觉毫无意义。
真正的问题是，AI 现在太优秀了。它应用抽象数据类型（ADT）模式、不可变性、Option/Maybe、Result 模式、函数式编程模式、单子链式操作，以及像 MVC、MVVM、MVI 和六边形架构这样的架构模式，其效果远超我本人。这极大地打击了我的积极性。
我花了这么多年时间记忆 GoF（Gang of Four）设计模式，将它们慢慢应用到工作中，并将其调整成我自己的风格——这一切现在感觉毫无意义。
老实说，绝大多数自由职业工作都围绕着简单的 CRUD（增删改查）应用程序。当然，还有领域知识需要记忆，但就 CRUD 的结构化、使用函数式技术进行错误和值处理，或者编写边界测试而言，AI 都比我更细致、更强大。
甚至感觉 AI 在领域建模方面也做得更好。我试图催眠自己，相信我仍然是必需的，但现实是，我不在大型科技公司工作。我是一名程序员，主要在现有框架之上组装业务逻辑，而事实是，AI 比我做得更好。而且更快。我认为我再也无法否认这一点了。
顶尖开发者可能仍然比 AI 强，但对于我们绝大多数人，包括我自己来说，这已不再是现实。要写出 AI 生成的那种代码，我过去需要打开文档放在第二台显示器上，深入思考，并不断搜索更好的方法。AI 只是瞬间生成，这让过去的努力显得过时。
当 AI 写出的代码比我写得更好时，我该如何保持我的动力和身份认同？我特别想听听其他自由职业开发者的看法。随着市场截止日期因为 AI 而越来越紧迫，你们是如何应对这种转变的？
再说一遍，并不是说 AI 的“随心所欲编码”不好玩。它很有趣。但抛开这一点不谈，我正在失去竞争优势的严峻现实正让我越来越沮丧。
查看原文
Nowadays, almost everyone codes with AI, right? I am a freelancer, and the entire market is now setting deadlines based on AI-assisted coding speeds. If you don't use AI, you simply cannot meet the deadlines. Because of this, I completely transitioned to AI-driven coding at the beginning of this year (starting around GPT-5.3).<p>It's not that AI coding isn't fun. It is. But in the beginning, there were still many things AI couldn't do well. The community focus was on "Harness Engineering," attaching MCPs, or deep Prompt Engineering. I actually built and used quite a few custom AI tools during that phase.<p>In the past, you had to provide not just your intent and direction, but also detailed, step-by-step instructions to reach that goal. Lately, though, even if you skip the detailed instructions, both Claude and GPT-5.6 handle it perfectly. You don't even have to agonize over which model to use. "Graph engineering" is currently trending, but it seems the models have already learned those patterns natively—if you establish a high-level plan with them, they automatically embed the execution policies into it. Whenever I try to learn and apply a new AI engineering technique, the next model update already does it out of the box, making the effort feel pointless.<p>The real problem is that AI is now too good. It applies ADT patterns, immutability, Option/Maybe, Result patterns, functional programming patterns, monadic chaining, and architectural structures like MVC, MVVM, MVI, and Hexagonal architectures far better than I can. It's heavily demotivating.<p>All those years I spent memorizing GoF patterns, slowly applying them to my work, and tweaking them into my own style—it all feels meaningless now.<p>Honestly, the vast majority of freelance work revolves around simple CRUD apps. Sure, there is domain knowledge to memorize, but when it comes to structuring CRUD, using functional techniques for error and value handling, or writing boundary tests, AI is just far more meticulous and capable than I am.<p>It even feels like AI is better at domain modeling. I try to hypnotize myself into believing I'm still necessary, but realistically, I don't work in Big Tech. I'm a programmer who mostly assembles business logic on top of existing frameworks, and the truth is, AI does my job better. And faster. I don't think I can deny it anymore.<p>Top-tier developers might still outperform AI, but for the vast majority of us, including myself, that is no longer the reality. To write the kind of code AI generates, I used to have to keep documentation open on a second monitor, think deeply, and constantly Google better approaches. AI just generates it instantly, making my past efforts feel obsolete.<p>When AI writes better code than I do, how am I supposed to maintain my motivation and identity? I would especially love to hear from other freelance developers. With market deadlines getting tighter and tighter because of AI, how are you handling this transition?<p>Again, it's not that AI "vibe coding" isn't fun. It is. But entirely separate from that, the stark realization that I am losing my competitive edge is making me increasingly depressed.