I don't write code anymore - I sculpt it
【原文摘要】
I’ve noticed a shift in how I code: I rarely write code myself now, as claude-code handles most of the coding work, and my role is to sculpt its outputs into more durable, polished code. Once the generated code meets my original prompt/specs and is functional, I refine it—cutting excess lines, removing bloat, ensuring components fit together and are reusable. I analyze the diffs, evaluating if each iteration moves us closer to our implementation goal.
I don’t believe current coding models can replace humans yet, as evidenced by the fact that I still need to craft prompts to get any work done. Today’s coding agents lack creativity, struggle to develop novel, bold solutions, and aren’t good at making critical judgment calls or flagging major flaws. Their strength lies in regurgitating time-tested, elegant coding abstractions that I often couldn’t write on my own—this is similar to how I’ve long relied on established abstractions (like pre-built sorting algorithms) from more experienced developers instead of building my own.
That said, these models have key limitations: they almost never remove dead code, rarely disagree with me or push back. Without human oversight to sculpt their work, the end result is bloated, unfocused code with no clear purpose or structure, like an overdone statue too heavy to stand and with no story to tell. For now, my role as the "sculptor" of the code remains essential.
【译文摘要】
我注意到自己的编程方式发生了转变：现在我很少自己写代码了，因为claude-code承担了大部分编程工作，我的角色是将它输出的内容打磨得更耐用、更精良。一旦生成的代码符合我最初的提示/规格要求且能正常运行，我就会对它进行优化——删掉冗余代码行、去除臃肿部分，确保各个组件衔接顺畅且可复用。我会分析代码差异，评估每一次迭代是否让我们更接近实现目标。
我认为当前的编程模型还无法取代人类，我仍需要编写提示词才能推进工作就证明了这一点。如今的编程智能体缺乏创造力，很难想出新颖、大胆的解决方案，也不擅长做出关键判断或指出重大缺陷。它们的优势在于能复刻那些经过时间检验的、精妙的编程抽象概念——这些概念我自己往往写不出来，这就像我长期以来一直依赖资深开发者打造的成熟抽象工具（比如现成的排序算法），而非自己从零构建一样。
话虽如此，这些模型也存在关键局限性：它们几乎不会删除无用代码，很少会反驳或质疑我。如果没有人类的监督来打磨它们的产出，最终结果会是臃肿、杂乱无章的代码，没有明确的目标或结构，就像一件雕琢过度的雕像，沉重到无法立起，也没有任何内涵可言。目前，我作为代码“雕琢者”的角色依然至关重要。
【单词表】
- durable /ˈdjʊərəbl/ 耐用的，持久的
- polished /ˈpɒlɪʃt/ 精良的，优美的，打磨过的
- specs /speks/ 规格，说明书（specifications的缩写）
- refine /rɪˈfaɪn/ 优化，改进，精炼
- excess /ɪkˈses/ 多余的，过量的
- bloat /bləʊt/ 臃肿，膨胀，冗余
- reusable /ˌriːˈjuːzəbl/ 可重复使用的
- diffs /dɪfs/ 差异（differences的缩写，此处指代码差异）
- iteration /ˌɪtəˈreɪʃn/ 迭代，重复
- implementation /ˌɪmplɪmenˈteɪʃn/ 实施，实现
- evidenced /ˈevɪdənst/ 证明，表明
- craft /krɑːft/ 精心制作，构思
- novel /ˈnɒvl/ 新颖的，新奇的
- critical /ˈkrɪtɪkl/ 关键的，批判性的
- flag /flæɡ/ 指出，标记
- flaw /flɔː/ 缺陷，瑕疵
- regurgitate /rɪˈɡɜːdʒɪteɪt/ 机械重复，照搬
- elegant /ˈelɪɡənt/ 优美的，精妙的
- abstraction /æbˈstrækʃn/ 抽象概念，抽象（编程术语）
- established /ɪˈstæblɪʃt/ 已确立的，成熟的
- algorithm /ˈælɡərɪðəm/ 算法
- limitation /ˌlɪmɪˈteɪʃn/ 局限性，限制
- oversight /ˈəʊvəsaɪt/ 监督，监管
- overdone /ˌəʊvəˈdʌn/ 做得过头的，雕琢过度的
【句子翻译】
- I’ve noticed a shift in how I code: I rarely write code myself now, as claude-code handles most of the coding work, and my role is to sculpt its outputs into more durable, polished code. 我注意到自己的编程方式发生了转变：现在我很少自己写代码了，因为claude-code承担了大部分编程工作，我的角色是将它输出的内容打磨得更耐用、更精良。
- Once the generated code meets my original prompt/specs and is functional, I refine it—cutting excess lines, removing bloat, ensuring components fit together and are reusable. 一旦生成的代码符合我最初的提示/规格要求且能正常运行，我就会对它进行优化——删掉冗余代码行、去除臃肿部分，确保各个组件衔接顺畅且可复用。
- I analyze the diffs, evaluating if each iteration moves us closer to our implementation goal. 我会分析代码差异，评估每一次迭代是否让我们更接近实现目标。
- I don’t believe current coding models can replace humans yet, as evidenced by the fact that I still need to craft prompts to get any work done. 我认为当前的编程模型还无法取代人类，我仍需要编写提示词才能推进工作就证明了这一点。
- Today’s coding agents lack creativity, struggle to develop novel, bold solutions, and aren’t good at making critical judgment calls or flagging major flaws. 如今的编程智能体缺乏创造力，很难想出新颖、大胆的解决方案，也不擅长做出关键判断或指出重大缺陷。
- Their strength lies in regurgitating time-tested, elegant coding abstractions that I often couldn’t write on my own—this is similar to how I’ve long relied on established abstractions (like pre-built sorting algorithms) from more experienced developers instead of building my own. 它们的优势在于能复刻那些经过时间检验的、精妙的编程抽象概念——这些概念我自己往往写不出来，这就像我长期以来一直依赖资深开发者打造的成熟抽象工具（比如现成的排序算法），而非自己从零构建一样。
- That said, these models have key limitations: they almost never remove dead code, rarely disagree with me or push back. 话虽如此，这些模型也存在关键局限性：它们几乎不会删除无用代码，很少会反驳或质疑我。
- Without human oversight to sculpt their work, the end result is bloated, unfocused code with no clear purpose or structure, like an overdone statue too heavy to stand and with no story to tell. 如果没有人类的监督来打磨它们的产出，最终结果会是臃肿、杂乱无章的代码，没有明确的目标或结构，就像一件雕琢过度的雕像，沉重到无法立起，也没有任何内涵可言。
- For now, my role as the "sculptor" of the code remains essential. 目前，我作为代码“雕琢者”的角色依然至关重要。
文章来源：https://www.jerpint.io/blog/2026-01-24-i-dont-write-code-anymore-i-sculpt-it/
