Vibecoding How I Dived Into AI Programming and What I Found
Last year I decided to try it. People around me were building working apps without writing a single line of code manually, and they didn’t call themselves developers. It’s called vibecoding, and over the past few months it’s consumed more of my attention than anything else in tech in recent years.
I have nine years of custom software development behind me, so code isn’t foreign to me. But vibecoding still surprised me, not because of what it can do, but how quickly it changes the way you think about what’s even possible to build.
How It Started
My first attempt was modest. I opened Google AI Studio, described what I wanted in plain language, and watched code appear. The goal: a mobile app for tracking car-related expenses. Fuel, service, insurance, all in one place.
The result wasn’t perfect, but it worked. That was enough to understand what vibecoding is about. It’s not about perfect code. It’s about having a result in an hour instead of a week.
Where It Really Clicked
I quickly moved to a more ambitious project, Antigravity, a web app built entirely through Gemini. No manual coding. This was the moment it all made sense.
My development background helped me understand what was happening under the hood. But I realized I didn’t actually need to, what matters is a clear picture of what the app should do, not how it works technically.
A macOS App for 3D Printer Filament
Then I went further. A native macOS and Windows app, inventory management for my 3D printer filaments. Connects to the printer, deducts material after each print, alerts me when stocks run low, tracks costs.
I’ve shelved it for now, but every project teaches you something. That might be vibecoding’s biggest advantage, you iterate fast, try things, and fail cheaply.
What Actually Works
Don’t start coding right away. Before you open Claude Code or Cursor, chat with AI about your app. Walk through the intent, use cases, edge cases. The resulting conversation gives you context to hand off to the coding environment.
Preparation pays off more here than in traditional development. Bad requirements cost hours in vibecoding, not weeks, but still unnecessary. Better context means fewer steps backward.
Use planning mode. Before coding, switch to plan mode. AI looks at your requirements from a technical angle. Want user registration? It proposes an auth solution. Need portable data? It designs a database structure.
Write in plain language. “After clicking the button, show a form, and send a confirmation email after submission” works better than “implement a POST endpoint at /api/submit”.
💡 Think of it as having your own developer available 24/7. Come with a request, explain the situation and intent, the same way you’d approach a real developer meeting. Better context, better output.
Claude Code
Vibecoding tools aren’t all equal. After trying several, Claude Code is where I landed. The interaction feels natural, it holds context well, and results are consistent.
Security
Are vibecoded apps secure? I’m not a security expert, but my take: the average vibecoding project will have roughly the same potential vulnerabilities as a traditionally built project on a limited budget.
How I handle it: after finishing a working version, I run a security audit through AI. Simple prompt, go through the code as an ethical hacker and find as many vulnerabilities as possible. Suggest fixes. I decide what to address.
Vibecoding Changes the Balance of Power
A personalized app used to require a designer, developer, project manager, and a serious budget. You’d see results in weeks. Today, minimal budget and a few hours gets you comparable results.
Want your own website? A mobile app? A tool that saves hours every week? All of this is now within reach. The question is just how you use it.