Blog / Build with AI /
Vibe coding: what it is and when it is a bad idea
Vibe coding is building software by telling an AI what you want in plain language, running what it produces, and carrying on without carefully reading the code. You steer by results (“make the button bigger”, “this error appeared, fix it”) rather than by understanding each line.
The term was coined by AI researcher Andrej Karpathy in a post earlier this month, and it spread because it put a name on something many people were already doing. If you’ve ever pasted an error message back into a chatbot and hoped for the best, you’ve done a version of it.
How it works, with an analogy
Think of building a house with a very fast builder who never sleeps. You stand in the garden and say, “I’d like a kitchen at the back, with a window over the sink.” The builder returns in ten minutes: a kitchen. It looks right. You say, “The window’s too small.” Ten minutes later, a bigger window.
You never looked at the wiring, the pipes or what’s holding the roof up. As long as you’re only judging what you can see, that’s a wonderful way to work. The risk is everything you can’t see.
In practice the loop is:
- Describe what you want.
- The AI writes code (usually in a tool like Cursor, Replit, Bolt or Lovable, or in a chat window).
- You run it and look at the result.
- If something’s off, you describe the problem, or paste the error.
- Repeat until it looks right.
If you want to compare the tools people use for this, Lovable, Bolt and Replit compared is a good place to start, and AI app builders vs AI code editors explains the difference.
Why it matters to a normal person
It lowers the bar for making software. Someone who has never written a line of code can now build a simple calculator, a landing page or a small tool for their business. That’s a real change, and I don’t want to sneer at it. I wrote about the shift in how AI changed what one person can build.
For someone who runs websites, it means small jobs that used to need a developer, such as a custom form, a redirect checker or a bulk-edit script, are now within reach of a determined non-developer.
Where vibe coding works well
- Personal tools that only you use.
- Prototypes to test whether an idea is worth building properly.
- Learning: seeing code appear and working change is a great way to get curious about how things work.
- Throwaway scripts, like a one-off clean-up of a spreadsheet.
- Design mock-ups where the point is to look at the idea, not to ship it.
When it is a bad idea
When other people depend on the result
If customers or colleagues will rely on it, “it seemed to work” isn’t enough. Hidden bugs show up when real people do unexpected things.
When money, logins or personal data are involved
AI-written code often looks tidy and still has security holes: forms that accept anything, secrets left in public files, no checks on who’s allowed to see what. Security basics for AI-built web apps goes through the common ones, and AI-written code: five mistakes to check before you trust it is a quick read.
When you’ll need to maintain it
Code you don’t understand is hard to change. A month later, when you want a new feature, you may find the AI has built it in a way that makes every change a gamble.
When the AI is confidently wrong
AI tools sometimes invent functions or libraries that don’t exist, and announce that a task is done when it isn’t. When AI says it’s done and it isn’t and spotting invented functions cover that.
Common myths
“Vibe coding means no skill is needed.” Not quite. You still need to describe things precisely, notice when something’s wrong, and know what to ask for next. People with a little technical grounding get much further.
“It will replace developers.” For small, simple tools, it reduces the need for one. For systems that must be secure, reliable and maintained, you still need someone who understands the code. I think AI builders and developers are friends, not enemies.
“If it runs, it’s fine.” Running is the lowest bar. Safe, correct and maintainable are higher ones.
“It’s all new.” The idea of building by feedback is old. What’s new is how far a plain-language description can get you.
How to vibe code more safely
You don’t have to stop. You can keep the speed and add a small safety net:
- Keep your projects in version control, so you can go back. See git basics for non-developers.
- Ask the AI to explain what it just changed, in plain words.
- Never paste real passwords or API keys into prompts or code.
- Test on a copy, not on the live website.
- Make one change at a time.
I also keep a short list of things to settle before I ask for anything big, in my checklist before asking AI to build anything.
Key takeaways
- Vibe coding means building by describing and testing, without reading all the code.
- It’s great for personal tools, prototypes and learning.
- It’s a bad idea for anything involving money, private data or other people’s trust.
- You can keep the speed and add basic safety habits.
FAQ
Is vibe coding the same as using GitHub Copilot?
Not exactly. Copilot suggests code while you write. Vibe coding is more about handing over the writing and judging the output.
Can a complete beginner vibe code?
Yes, for small things. A beginner should start with something harmless and low-stakes.
Do I need to learn to code anyway?
It helps. Even a little understanding makes you a much better director of the AI. I’ve explored that in learning to code in the age of AI: still worth it?.
Related reading
What’s the first small thing you’d try building this way?