In Two Hours, I Built an App Without Writing a Single Line of Code

Lately I've added a lot of information sources. They come from all kinds of platforms and take different forms.

For reading, for example, I now follow lots of blogs, X accounts, and their teams' or personal websites and WeChat Official Accounts. For video, I've followed lots of YouTubers and Bilibili uploaders too. Given all that, I had a small need: I wanted an easier way to get to my information sources.

Last night, while chatting with DC in "Between the Lines," I mentioned that I wanted an app to manage all my information sources. Vibe Coding is red-hot right now, and I'm interested in the field, so I said: "Why don't I just build what I need with Claude Code?"

DC said: "Sure."

Then, without a second thought, I climbed out of bed, sat down at my computer, opened Claude, and fell into thought: create an app? Oh my god. I studied Java and Python in college, but I've worked in healthcare for four years now. What code do I still know? Writing "Hello World" is about my limit. I'd long heard Vibe Coding was powerful, but could it really let someone who can't read or write a line of code build an app from nothing but a need?

Luckily, I've built a project before without knowing anything. In the past, I went from 0 to 100 building a personal Hugo blog under AI's guidance. So this time, even though Vibe Coding was completely new to me, I had two heavy hitters, Gemini Pro and Claude Pro, and that gave me a lot of confidence.

I immediately said to Claude, my favorite lately: "Claude, I want to use Claude Code to build an app that manages my information sources. Do you think that's doable?"

It all started with that one sentence. Claude immediately backed my idea, and started guiding me through what core features the app should have and what kind of app I actually wanted.

So Claude and I talked through my needs and ideas for a while. Then I got started. I told it: "I think we've talked enough. We can start now. But I have no programming background, so I need you to teach me step by step, hand in hand, and have me pick up some basic programming along the way."

Claude then asked whether I was on Windows or macOS and whether I had a Claude Pro subscription. After those two questions, it gave me a detailed tutorial, from opening the Terminal to creating the project.

• Open the Terminal and get to know the command console
• Install Claude Code
• Log in to my account
• Create a project folder and learn three basic commands: cd, mkdir, and pwd
• Install Homebrew, Node.js, and Rust development tools
• Use Claude Code to create the project: send it my requirements and it writes the code
• Run the program

Along the way, I ran into plenty of small problems with details, including but not limited to:

• Claude Code wasn't installed properly
• I couldn't tell whether a prompt was asking if I should keep installing
• After the code started running, I found the model was wrong and didn't know how to switch (stop the task, then enter claude --model claude-opus-4-6)
• I didn't know how to switch the current task (Ctrl+C)

All of these got solved through asking again and again. If I were a programmer and someone kept asking me questions like these, I'd lose patience for sure.

But Claude answered me in full detail every time, and kept teaching me programming along the way.

In the end:

• 18:50: I started
• 19:19: I logged in to Claude Code
• 19:50: the first version was out
• 20:32: all the basic features worked

In under two hours, I built an information source manager. It isn't perfect, but it works fine.

I have to admit, compared with when ChatGPT first came out a couple of years ago, AI's progress is almost miraculous. I looked it up and found that the concept of Vibe Coding appeared in the first quarter of 2025. Before that, back in 2024, which wasn't long ago, who would have dared to think, "I have a need and an idea, so I'll write an app myself"?

I wouldn't have. In under two hours, I created an app out of my original idea, even though I can't read a line of code or write one.

AI has lowered the pressure of starting to create. The magic of it, I think, is that as long as you have an idea, even the simplest sentence, like "I want to go a day without eating" or "I want to write down the name of every plant outside," anything, you can tell AI, and through exchange after exchange, you push your own thinking deeper.

This helps us understand what's in our own hearts better, so we can distill what has real universal value and use it to create something of our own.

I finished the basic features I wanted in my information source manager. I can freely add, edit, and delete sources. I can sort them into categories and manage them easily. Clicking a source opens it in the browser, and can jump straight to my default browser. After doing all that, I talked with AI again.

This is a way of strengthening deep thinking. I'd met my original needs, but I asked the AI: if I wanted to expand the app's features further, where could I start?

The AI gave me plenty of possibilities, of course, including but not limited to whether to clean up the text, whether to support marking and saving, whether to support export, whether to support text highlighting, and a pile of other odds and ends. This isn't being led around by the nose by AI. The AI did everything before this. But on this one thing, I have to use my head.

Because AI can accomplish what I want to achieve, but I have to know what I want. It's a process of restraint. Every decision I make adds or removes something from this app.

This is what the user of AI has to do when creating with it: decide.

AI lowers the barrier to starting to create, but it puts more demands on your ability to decide. Deciding is a combined skill.

Many things bear on a person's decisions. Are you hungry? Are you in a good mood? How well do you understand the product? What's your taste? What's your state of mind? All of it affects your decisions.

Today, I'm still talking with AI about my information source manager. Compared with last night, I've settled on a few new features.

• Add streaming translation
• Add automatic saving of copied text into flash notes and my note card library
• Support exporting the note card library as Markdown
• Support automatically importing my YouTube following list
• Support automatically importing my Bilibili following list

Finally, let me share how I communicate my needs to AI and do the coding.

• Settle the requirements first. I talk through the overall needs with the AI in detail, then list them all together.
• Go step by step. I tell the AI we'll do it one step at a time, and test the related features after each step. Only when the test is fine do we move on to the next.
• When there's anything you don't understand, just throw a screenshot at the AI. It understands 1000% better than I do.

Okay, back to tinkering!