When "Can't Code" Is No Longer a Barrier 當不會寫程式不再是障礙

English Version (中文版本在下方)

A year ago, I decided to build my own personal website.

The reason was simple: I wanted to practice vibe-coding. I’d always found this approach of thinking-while-building and rapid prototyping really appealing, and creating my own blog felt like a good practice project. Building it myself seemed more educational than using an existing platform.

I used ChatGPT to complete the basic infrastructure: copied a Hugo framework template, deployed to GitHub, purchased a domain on Netlify, completed the integration. Sounds impressive, right? But in reality, I was just a copy-paste worker following AI instructions.

After the infrastructure was done, I faced a problem: now what?

How do I customize the layout? How do I remove the template elements I don’t need? How do I optimize the design? Most frustrating of all, I had 20 articles on Medium. How do I migrate them over?

I had no idea.

Ironically, I wanted to use this project to practice vibe-coding, but I couldn’t even write code, let alone vibe with it. The project sat untouched for a year.

Then last month, I solved all these problems in one day using Claude Code. Literally, one day.

The result? My blog is now live at hankwhy.com, fully customized with all 20 articles migrated and running exactly how I wanted.

A live demo

The turning point came from a founder friend. He’s a mentor I really respect, and one day he told me he wanted to demo a tool for me.

He opened Claude Code and started vibe coding right in front of me. I watched him chat while writing code, and in less than half an hour, he built a mockup app. Not a concept mockup, but an actual working app.

The most amazing part? He said: you can download Claude Code right now and try this app.

After I downloaded it, I could actually open, modify, and play with that app. The immediacy kind of shocked me. My impression of coding had always been that it’s a lengthy process, but this felt like prototyping at a speed that seemed almost unreal.

I started getting curious: what else can this tool do?

Starting simple

After downloading Claude Code, I didn’t immediately jump into the website project. I started with something simpler: cleaning up my computer.

My computer was running slow, cluttered with junk files, but I wasn’t sure how to clean it. I just asked Claude Code directly: help me clean up my computer and make it run faster.

Then it started scanning, finding duplicate files, cleaning temp files, optimizing settings. Throughout the process, it would confirm with me which files could be deleted and which should be kept. I just typed what I needed, and it executed.

This experience made me realize something: these technical operations might not be hard for engineers, but for someone like me who can’t code, just not knowing what commands to run or which settings to change creates a roadblock. Now I could just type what I wanted and it would happen.

The feeling was kind of magical. I knew these tasks weren’t inherently complex, but when you can’t do them yourself, even simple things become barriers.

After cleaning my computer, I thought: could this tool help me continue that year-old website project?

Reopening the dusty project

I opened up the website project folder that had been sitting there for a year.

Claude Code automatically read the entire Hugo project and started explaining what each folder was for, how files connected to each other, how the template worked.

With ChatGPT in the past, I had to struggle to describe my problem. But Claude Code directly saw my project. It’s a subtle feeling, like finally having someone who can look directly at your notebook instead of you having to describe what your notebook looks like.

I started making requests.

Customizing the design

I wanted to remove certain template elements, adjust colors, change the layout. These were all black boxes to me. I knew what result I wanted, but had no idea which file to edit or which line to change.

I told Claude Code directly: I want to remove the sidebar from the homepage, make the article card corners less rounded, adjust the title font size.

It found the corresponding CSS files, pointed out which classes to modify, explained how those classes affected the layout, suggested values to change. After making the changes, I refreshed the browser and saw the results.

More importantly, it explained why these changes worked. I started understanding how these things operated. Not just following instructions, but actually understanding why.

This aha moment was kind of unbelievable: these technical edits could happen so fast, I just typed what I needed. I know these probably aren’t hard for engineers, but for someone like me who can’t code, being able to directly realize ideas like this still feels amazing.

Migrating 20 articles

The most painful part was content migration. I had 20 articles on Medium, each needing format conversion, image path adjustments, link fixes, metadata setup. Just thinking about manually processing 20 articles made me want to give up.

I told Claude Code: I want to move my Medium articles to this website, is there a way to batch process them?

It started building a workflow, converting formats while preserving the original formatting, quotes, code blocks. It would also point out which parts needed manual checking, like specially formatted quotes or images, rather than blindly automating everything.

Within one day, all 20 articles were migrated.

I sat at my computer, kind of dazed. This project had sat for a year. I thought it would take several weeks to complete, but it was done in a day.

What did this feel like?

If I had to describe it, using ChatGPT in the past felt like consulting an expert. You ask a question, get an answer, and the conversation ends. You might understand the answer or you might not.

Using Claude Code felt more like pair programming with an engineer. They’re looking at your screen, your code, and thinking with you. They explain why, show you how, help you understand the logic.

This collaborative feeling was something I’d never experienced before. And ironically, the vibe-coding I wanted to practice a year ago? I was finally actually doing it.

A bigger question

This experience got me thinking: if AI enables every non-technical person to code, what would the world look like?

How many good startup ideas die at the barrier of founders not being able to code? When technical execution is no longer an obstacle, truly valuable ideas will surface. Those ideas might come from educators, social workers, artists, people from any field, not just engineers.

Being able to code has long been a form of power. It determines who can build products, who can control platforms, who can realize ideas. When this barrier lowers, how will power redistribute? Will there be more diverse voices, more varied products?

If everyone can code through AI, what becomes the new scarce capability?

My guess: taste, judgment, asking the right questions.

AI can help you execute, but it can’t decide for you what to build, why to build it, who to build it for. People who can clearly define problems, have unique perspectives, know what’s good will become more valuable. Technical execution will become widespread, but creativity and judgment won’t.

Imagine when all the designers, writers, artists, educators, social workers who were previously blocked by technical barriers can realize their ideas. How much interesting stuff will get created?

This reminds me of what Naval Ravikant calls permissionless leverage. In the past, learning to code took years. Now, you need clear goals, effective communication, basic judgment. The technical execution part, AI can help with.

This isn’t about replacing engineers. Engineers’ value won’t disappear. They’ll probably focus more on truly complex problems that need deep thinking. This is about enabling more people to become builders, shortening the distance between ideas and execution.

From can’t to can direct

Back to my personal website project. A year ago, I was stuck because I couldn’t code. A day later, I completed it because I learned how to direct AI to code.

These are two completely different abilities.

The former requires months to years of learning. The latter requires clear goals, effective communication, basic judgment. These can all be cultivated.

If you also have a long-abandoned technical project, maybe now is the time to reopen it. Not because you suddenly became an engineer, but because the tools changed, the way of collaborating changed, what’s possible changed.

Technology should be a tool for realizing ideas, not a barrier. When AI can handle the how, you can focus on the what and the why.

Those two questions are what we should really spend our time thinking about.

---

中文版本

一年前，我決定建立自己的個人網站。

起因很簡單：我想練習 vibe-coding。我一直覺得這種邊想邊做、快速 prototype 的方式很吸引人，而建一個自己的 blog 感覺是個不錯的練習專案。比起用現成平台，自己建可以學更多東西。

我用 ChatGPT 完成了基礎建設：複製了 Hugo 框架的模板、部署到 GitHub、在 Netlify 購買域名、完成整合。聽起來好像很厲害？但實際上我就是個照著 AI 指令操作的複製貼上工人。

基礎建設完成後，我面對的問題是：現在要怎麼辦？

怎麼客製化版面？怎麼把模板裡我不需要的東西拿掉？怎麼優化設計？最頭痛的是，我在 Medium 上有 20 篇文章，要怎麼搬過來？

我完全不知道。

諷刺的是，我本來想用這個專案練習 vibe-coding，結果連 code 都寫不了，還談什麼 vibe。專案就這樣放了一年。

然後上個月，我用 Claude Code 在一天之內解決了所有問題。真的，一天。

結果？我的 blog 現在在 hankwhy.com 上線了，完全客製化、20 篇文章全部遷移完成，完全按照我想要的方式運作。

一場 live demo

事情的轉折是一位創業者朋友。他是我很敬重的 mentor，某天突然跟我說要 demo 一個工具給我看。

他打開 Claude Code，然後就在我面前開始 vibe coding。我看著他邊聊天邊寫 code，不到半小時就做出了一個 mockup app。不是那種概念圖，是真的可以跑的 app。

最神奇的是，他說：你現在就可以下載 Claude Code 試試看這個 app。

我下載之後，真的可以直接打開、修改、玩那個 app。這種即時性讓我有點震驚。過去我對寫程式的印象都是很漫長的過程，但這個感覺像是在做 prototype，速度快到有點不真實。

我開始好奇：這個工具還能做什麼？

先從簡單的開始

下載 Claude Code 之後，我沒有馬上跳進網站專案。我先從更簡單的事情開始：清理我的電腦。

我的電腦跑得有點慢，堆了很多垃圾檔案，但我不太確定要怎麼清。我就直接問 Claude Code：幫我清理電腦、讓它跑快一點。

然後它就開始掃描、找出重複檔案、清理暫存、優化設定。過程中會跟我確認哪些檔案可以刪、哪些要保留。我只是打字說我的需求，它就執行了。

這個經驗讓我意識到一件事：這些技術操作可能對工程師來說不難，但對我這種不會 code 的人來說，光是不知道要下什麼指令、改哪個設定，就會卡住。現在我只要打字說我要什麼，就可以完成。

這種感覺有點神奇。我知道這些事情本身不複雜，但當你做不到的時候，再簡單的事情都是障礙。

清理完電腦後，我想：會不會那個卡關一年的網站專案，也可以用這個工具繼續下去？

重新打開塵封的專案

我打開了放了一年的網站專案資料夾。

Claude Code 自動讀取了整個 Hugo 專案，然後開始跟我解釋每個資料夾是幹嘛的、檔案之間怎麼連結、模板怎麼運作。

過去用 ChatGPT，我要費力描述我的問題。但 Claude Code 直接看到我的專案。這種感覺很微妙，就像終於有人能直接看著你的草稿本，你不用再用文字描述草稿本長什麼樣子。

我開始試著提需求。

客製化設計

我想移除模板裡的某些元素、調整配色、改排版。這些對我來說都是黑盒子，我知道我要什麼結果，但完全不知道要改哪個檔案、改哪一行。

我直接跟 Claude Code 說：我想把首頁的側邊欄拿掉、把文章卡片的圓角改小一點、調整標題的字體大小。

它找到對應的 CSS 檔案、指出要改的 class、解釋這個 class 怎麼影響版面、建議修改的數值。改完之後，我重新整理瀏覽器，就看到變化了。

更重要的是，它會解釋為什麼要這樣改。我開始理解這些東西是怎麼運作的。不只是照做，是真的懂為什麼。

這個 aha moment 讓我有點不敢相信：這些技術編輯竟然可以這麼快完成，我只是打字說我的需求。我知道這些對工程師來說可能不難，但對我這種不會 code 的人來說，能夠這樣直接實現想法，還是覺得很神奇。

遷移 20 篇文章

最讓我頭痛的是內容遷移。我在 Medium 上有 20 篇文章，每一篇都要轉格式、調整圖片路徑、修正連結、設定 metadata。光想到要手動處理 20 篇我就想放棄。

我跟 Claude Code 說：我想把 Medium 上的文章搬到這個網站，有沒有辦法批次處理？

它開始建立流程，轉換格式的同時保留原本的排版、引用、程式碼區塊。它還會指出哪些地方需要我人工檢查，像是特殊格式的引用或圖片，不是盲目地全自動化。

一天之內，20 篇文章全部遷移完成。

我坐在電腦前，有點恍惚。這個專案放了一年，我以為會需要好幾個禮拜才能完成，結果一天就搞定了。

這感覺像什麼？

如果要我形容，過去用 ChatGPT 像是在請教一個專家。你問問題、得到答案，對話就結束了。你可能懂答案也可能不懂。

用 Claude Code 更像是在跟工程師 pair programming。對方看著你的螢幕、你的 code，然後跟你一起思考。會解釋為什麼、show you how、讓你理解邏輯。

這種協作感是我過去沒體驗過的。而且諷刺的是，我一年前想要練習的 vibe-coding，現在終於真的在做了。

一個更大的問題

這個經驗讓我開始想：如果 AI 讓每個非技術人都能寫程式了，這個世界會變成什麼樣子？

有多少好的創業點子死在創辦人不會寫程式這個門檻上？當技術執行不再是障礙，真正有價值的想法會浮現出來。那些想法可能來自教育工作者、社會工作者、藝術家、任何領域的人，而不只是工程師。

會寫程式長期以來是一種權力。它決定了誰能建產品、誰能掌控平台、誰能實現想法。當這個門檻降低，權力會怎麼重新分配？會不會有更多元的聲音、更多樣的產品？

如果人人都能透過 AI 寫程式，那什麼會變成新的稀缺能力？

我猜是品味、判斷力、問對的問題。

AI 可以幫你執行，但它沒辦法替你決定要做什麼、為什麼做、做給誰。那些能清晰定義問題、有獨特視角、知道什麼是好的人，會變得更有價值。技術執行力會變得普及，但創意和判斷力不會。

想像一下，當所有設計師、寫作者、藝術家、教育者、社會工作者這些過去被技術門檻擋在外面的人，都能實現他們的想法時，會有多少有趣的東西被創造出來？

這讓我想起 Naval Ravikant 說的 permissionless leverage，無需許可的槓桿。過去你要寫程式，需要花數年學習。現在，你需要的是清楚的目標、有效的溝通、基本的判斷力。技術執行的部分，AI 可以協助。

這不是要取代工程師。工程師的價值不會消失，反而可能更專注在真正複雜、需要深度思考的問題上。這是讓更多人成為 builder，讓想法和執行之間的距離縮短。

從不會到會引導

回到我的個人網站專案。一年前我卡關是因為不會寫程式。一天後我完成是因為學會了引導 AI 寫程式。

這是兩種完全不同的能力。

前者需要數月到數年的學習。後者需要的是清晰的目標、有效的溝通、基本的判斷力。這些都是可以培養的。

如果你也有一個擱置很久的技術專案，或許現在是時候重新打開它了。不是因為你突然變成工程師了，是因為工具變了、協作方式變了、可能性變了。

技術應該是實現想法的工具，不是障礙。當 AI 可以幫你處理怎麼做，你就能專注在做什麼和為什麼做。

這兩個問題，才是真正值得我們花時間思考的。