Do you know vibe coding?
Before you answer, please open 20 PRs with AI, get burned at least once, and only then come back.
Please share this with friends who just discovered Cursor / Claude / Grok.
¯\(ツ)/¯
This is a collection of things that happen when you vibe code that may not be well recognized, especially to beginners.
Disclaimer: Some of these are simply to demonstrate the quirky parts of AI-assisted coding and by no means encourage best practices.
Q. What's the result?
You: "Add a login form"
AI: *generates a full Next.js app with auth, middleware, database, and tests*
A.
You now have 47 files you didn't ask for and a bug in the middleware.
Q. What's the result?
You: "Fix this function"
AI: *rewrites the entire file*
A.
Your git diff is now unreadable and you lost the one comment that explained the weird edge case.
Q. What's the result?
You: "Use the official SDK"
AI: import { Client } from 'super-cool-sdk-that-does-not-exist'
A.
npm install fails. The package never existed. The AI was very confident.
Q. What's the result?
You paste 3 files into the chat.
You: "Why is this failing?"
AI: *gives a perfect answer based on only the first file*
A.
The real bug was in file 3. The AI never saw it.
Q. What's the result?
You: "Make it faster"
AI: *adds caching, memoization, web workers, and a Redis layer*
A.
Your 20ms function is now 40ms and has a new dependency.
Q. What's the result?
You: "This code works in the playground"
*paste it into your project*
A.
It doesn't work. Missing imports, wrong types, and it assumed global variables that don't exist.
Q. What's the result?
You ask Claude, Grok, and ChatGPT the same question.
A.
Three different answers. All of them sound correct. One of them is.
Q. What's the result?
You: "Don't change the function signature"
AI: *changes the function signature*
A.
It also updated the call sites. Half of them are wrong.
Q. What's the result?
You: "Explain this error"
AI: "This is a common issue with async/await"
A.
The error was a missing comma. The AI never looked at the actual stack trace.
Q. What's the result?
You accept every suggestion for 2 hours.
A.
Your codebase now has 4 different ways to do the same thing and you can't explain any of them.
Q. What's the result?
You: "Write tests"
AI: *writes tests that pass*
A.
The tests test the AI's implementation, not your requirements. They will pass even when the feature is broken.
Q. What's the result?
You: "Just a small change"
AI: *refactors half the module "for clarity"*
A.
The small change is correct. The refactor introduced two new bugs.
Q. What's the result?
You: "Use the latest version of X"
AI: *uses a version from 2023*
A.
The API it used was removed last year. The AI was very confident.
Q. What's the result?
You copy AI code into production.
A.
It works until the edge case the AI never considered. Then you spend the next hour debugging code you don't understand.
Q. What's the result?
You: "Make it production ready"
AI: *adds logging, error handling, retries, and metrics*
A.
Your prototype is now 300 lines longer and still missing the one validation you actually needed.
Q. What's the result?
You ask the AI to review its own code.
A.
It finds two style issues and misses the logic bug.
Q. What's the result?
You: "Don't use any libraries"
AI: *uses three*
A.
One of them is unmaintained.
Q. What's the result?
You regenerate the same component five times.
A.
You now have five slightly different versions and no idea which one is correct.
Q. What's the result?
You: "This is taking too long, just ship it"
A.
You ship it. The bug shows up in production on Friday at 5pm.
Q. What's the result?
You finally read the code the AI wrote.
A.
You understand it less than when you started.
Released under the "I should have just written it myself" license.
