Nine Months of Vibe-Coding: What I Learned
Nine Months of Vibe-Coding: What I Learned
TL;DR: You don't need to learn to code. You need to learn to direct code. Here's what nine months of stumbling around with AI tools taught me.
A Little Background
Vibe-coding is a gateway drug into development.
I don't consider myself technical. I've taken dev classes, tried virtually every learn-to-code tool, and distinctly remember deciding to stop trying. Nothing stuck.
Then over the holiday break, I experimented with Nat Eliason's Build Your Own Apps and Cursor.
Nine months later, I've built more projects than I can count. I describe the process as stumbling around an unfamiliar house with a flickering flashlight. Sometimes it lights the way. Often you bump into walls.
I'm only scratching the surface. But that surface area keeps expanding.
What I Actually Built
Through trial, error, and restarts, I've ended up with two things: a framework of tools to scaffold apps quickly, and a mental model of how things fit together.
- Websites with Node.js, Tailwind, shadcn—deployed to Netlify, Vercel, Digital Ocean
- A personal website replacement (Notion as CMS, no more $200/year service)
- Internal tools for partner management, sales prospecting, email outreach
- Mobile apps with Expo using AI APIs, deployed via TestFlight
Each project started with "I wonder if I could..." or "I spend five hours a week on this..." Some took hours. Some took days.
Three Patterns That Emerged
1. Curiosity has never been cheaper.
AI is a patient partner. Trying several options has low cost and low effort.
2. You can get 80% very quickly. The last 20% takes forever.
It's all polish and details. AI needs direction. The ability to write clear Product Request Documents (PRDs) is critical. To go fast, you need clarity.
3. Build the first version as quickly as possible.
Create a PRD, build the thing, then revisit the PRD. Having something you can use pushes you to learn faster than theorizing.
The Gap Most People Miss
There's a tension between the promise of vibe-coding and the reality.
The promise: describe what you want, get working code. The reality: you still need to know what "working" looks like.
I lead partnerships at Gadget. I have engineers to lean on when stuck. I've also explored Replit, v0, Lovable, and others.
What I learned: the skill isn't coding. It's knowing when the code is right.
The So What
The best builders aren't the ones who write the most code. They're the ones who ship fastest, learn most, and iterate relentlessly.
AI lets you do all three.
But direction beats implementation. Knowing what you want is harder than getting it.
Framework: Experience → Reflect → Create. Repeat.
