I'm Not a Developer. But I Direct an AI to Build Apps and Run Them on My Server

I can't write code from scratch. I work in planning. Yet right now my server runs several apps I built by 'telling an AI what I want' in plain words: a personal inbox, a travel food map, a vocabulary app, blog automation. They run — until one day they quietly break. This blog is about that process, and above all, an honest log of where I broke.

1. What 'Vibe Coding' Even Means

Nothing grand. I describe what I want in ordinary language: "when I send a photo, read the text, save it, and make it searchable later." The AI writes the code. I run it, and when it fails I paste in "here's the error." It fixes it. Repeat that loop and eventually a working thing exists.

The point is that I can start even without understanding the code. But there's a price: not knowing the code, I also don't know why it broke — at first. Closing that gap is what this journey is really about.

2. So What Did I Build?

  • A personal inbox — throw memos, receipts, links at it via chat; find them later by search.
  • A travel food map — I tried to pull restaurants from videos onto a map (and hit a legal wall).
  • A vocabulary app — a web app for drilling beginner English/Japanese words as cards.
  • Blog automation — the pipeline this very post is published through.

No fancy tech. To others these look modest. But me — a non-developer — building and running things that actually work, alone — was a big deal to me.

3. The Hard Part Isn't 'Building' — It's 'Keeping It Alive'

Building was the easy part; the AI helps. The hard part comes after. Running once on my screen and running continuously on a server when no one is watching are completely different problems. And I learned that difference entirely by breaking things.

4. Logs of Where I Broke (stories I'll unpack one by one)

  • My blog was silently dead for 4 days. One hidden error stopped the whole publishing flow, and nobody told me. I noticed four days later. Lesson: silent failure is the scariest kind.
  • A restaurant got pinned in the wrong region. I let it 'fill in something' when it couldn't find a location — and one shop landed hundreds of km from where it belonged. Lesson: 'just fill something in' creates quiet lies.
  • I moved a few app settings and the whole site threw errors. I thought it was a trivial change. Lesson: there's no such thing as a 'trivial change.'
  • I upgraded something and felt proud — turns out that code was never running. I did fix it, but it was dead code executed nowhere. Lesson: 'fixed' and 'in effect' are not the same.
  • My vocabulary app's AI suddenly went dark. A heavy job I was running elsewhere had eaten up the free usage. Lesson: your own things steal resources from each other.

5. What This Blog Will Be

The world has plenty of technical tutorials. What's rare is a non-developer's honest log of flailing and failing. Success stories overflow; failure stories get deleted. I'll do the opposite. Less about what I built, more about where and how I fell over — because that's where the learning was, and it's what someone like me needs to fall over less next time.

You can build things now even without coding. But between 'can build' and 'can maintain' lies a wide river. This series is a non-developer crossing that river bare-handed.

6. Who This Is For

  • People who can't code but have something they want to build.
  • People who tried building with AI, kept hitting breakage, and got discouraged.
  • People who want the real 'how it goes wrong,' not someone else's highlight reel.

7. Next Up

The first two entries: (1) "My notes kept vanishing, so I had an AI build me an inbox" — bringing scattered notes into one place. (2) "I tried to build a travel map from YouTube food videos and stopped at a legal wall" — learning what you technically can but legally shouldn't do. A build story and a don't-build story. Both real.