GODELIAN

Playbook — opinion

Vibe coding, honestly: what it's great at, and where it breaks

Vibe coding — building by describing what you want to an AI — went from a tweet to a movement. It's real, it's powerful, and it has a sharp edge nobody warns you about. Here's the honest version.

By Andrew, Founder & CTO8 min readUpdated 2026-07-12

Vibe coding went from a tweet to a movement in about a year. It's real, it's genuinely powerful — and it has a sharp edge nobody warns you about.

If you're a founder, you've heard “just vibe-code it.” Sometimes that's great advice. Sometimes it hands you a product you can't maintain and didn't know was broken. This is the honest version — where it wins, where it fails, and how to use it without regret. It pairs with AI app builder vs. custom code.

Definition

What vibe coding actually is

Vibe coding is building software by describing what you want to an AI in plain language and letting it write the code — steering with prompts and feedback instead of typing every line yourself. The term was coined by Andrej Karpathy in early 2025 to describe leaning into the flow of working with an AI, and it caught on because it captured something true: for the first time, you can go from “I want an app that does X” to a working thing without knowing how to code.

The moment

Why it exploded

Two things collided. AI models got good enough to write real, working code from a plain-English description. And a wave of tools — AI-native editors and app builders — made that loop fast and visual. Suddenly the gap between an idea and a prototype collapsed from weeks to an evening.

Vibe coding didn't just make coding faster. It changed who gets to build at all.

That's genuinely a big deal — and it's the reason a non-technical founder can now build a real first version themselves. But “can build something” and “can build something you should ship to paying customers” are two different claims.

The upside

What it's genuinely great at

  • Prototypes. Getting a working version of an idea in front of your own eyes, fast, so you can feel whether it's right.
  • Interfaces. Generating clean, working UI — pages, forms, layouts — that would have taken a designer-developer days.
  • Learning by doing. Watching an AI build, and asking it to explain, is one of the fastest ways to actually understand how software fits together.
  • Momentum. The single hardest thing in a startup is getting the first real thing to exist. Vibe coding demolishes that barrier.

Used for these, it's not a shortcut — it's just the modern way to work. Every serious builder uses AI this way now.

The catch

Where it quietly breaks

Here's the part the hype skips. AI is brilliant at the first 80% — the visible, happy-path parts. It thins out on the last 20%, which happens to be the 20% that decides whether you have a business:

  • Security. The AI may generate an app that works and is quietly wide open — missing the rules that stop one user reading everyone's data. You can't see it in your own testing.
  • Maintainability. As the app grows, AI-generated code you don't understand becomes a tangle you can't safely change — and the AI itself starts breaking things it built earlier.
  • The hard stuff. Payments, scale, data isolation, compliance — the parts that must be exactly right — are exactly where unsupervised AI code is least reliable.

The danger isn't that vibe coding produces bad code. It's that a non-technical founder can't tell when it did.

Andrew, founder of Godelian

Written by Andrew

Founder & CTO of Godelian. Fifteen years building AI systems, trained as a founder at Entrepreneur First, a published AI researcher (Nature Scientific Reports), and had a startup acquired in 2025. He builds first versions that actually get used — and helps founders build theirs.

Questions, answered

What is vibe coding?

Vibe coding is building software by describing what you want to an AI in plain language and letting it write the code — a term popularised by Andrej Karpathy in 2025. You steer with prompts and feedback rather than writing code line by line. It's remarkably powerful for getting to a working prototype fast.

Is vibe coding good or bad?

Both, depending on the job. Vibe coding is excellent for prototypes, UIs, and getting a working first version fast. It gets dangerous when a non-technical founder ships it to real users without understanding it — the AI can produce code with security holes, poor structure, or bugs you can't see. The fix isn't to avoid it; it's to own and understand what you ship.

Can you vibe code a real, production app?

You can vibe-code your way to a real first version, but production means handling the hard 20% — security, row-level data isolation, payments, scale — which is exactly where unsupervised AI code tends to fail. The responsible path is to vibe-code the 80% on a stack you own, understand what it produced, and get the critical 20% right (or bring in a professional for it).

When you hit the hard 20%

Build it with the founder.

Prototyping is the easy part. When real users, money, and data are on the line — security, scale, the parts you can't see are broken — that's the work. Godelian builds owned, production-grade first versions, typically for $15–30k in weeks.