A 16-year-old’s honest confession about shipping code he didn’t understand.
I built 200+ files last month.
I used Cursor. I used Claude. I used every AI tool I could find. The code worked. Features shipped. Everything looked great.
Then someone asked me to explain how my authentication middleware worked.
I couldn’t.
Not because I forgot. But because I genuinely never knew. I had accepted the AI’s output, seen it work, and moved on. Repeat 200 times.
That moment was uncomfortable in a way I hadn’t expected. I had built something real — something with actual users in 30+ countries — and I couldn’t explain the code that ran it.
This is the part of vibe coding nobody talks about.
What Vibe Coding Actually Is (Honestly)
Vibe coding — a term coined by Andrej Karpathy — is the practice of using AI to generate code at high speed, often without fully understanding every line that comes out.
It sounds irresponsible when described that way. But that’s not the full picture.
Vibe coding is genuinely powerful. It lets a 16-year-old with limited experience build a full-stack AI chatbot visited by users in 30+ countries. It democratizes software creation in a way that nothing else has.
But it has a shadow side that only shows up later.
The problem isn’t that the code doesn’t work. It usually does — at first.
The problem is what happens when it breaks.
The Moment Everything Became Clear
I was running a security check on one of my own projects. I was looking for obvious issues before a bigger launch.
vibecheck — a CLI tool I was building — flagged something in my own code:
“Note: API key shown is a fake example key from the demo file, not a real credential.”
I stared at this for a full minute.
I had written that line. Or rather — I had accepted it when an AI wrote it. It had worked perfectly in every test. It had never thrown an error. It had never caused a problem.
But it was a open door into my database.
I didn’t know what parameterized queries were. I didn’t know why f-strings in SQL were dangerous. I had been shipping this for weeks.
This is the vibe coding hangover. And it comes for everyone eventually.
The Actual Problem: We Optimized the Wrong Thing
AI coding tools have made the easy part of programming faster — writing syntax, boilerplate, and common patterns.
But they haven’t touched the hard parts:
- Understanding why code is structured a certain way
- Knowing what breaks when assumptions change
- Being able to explain your system to someone else
- Recognizing dangerous patterns before they reach production
We got faster at the part that was already getting faster. We ignored the part that actually causes production incidents.
A study found that AI-generated code has 2.74x more security vulnerabilities than human-written code. Not because AI is bad at writing code. But because AI doesn’t know your specific context, your threat model, or what “safe” means for your system.
And when you vibe code — you often don’t know these things either.
So you ship the vulnerability. Confidently. Because the code compiles and the tests pass.
The Knowledge Debt Nobody Measures
There’s a concept in software called technical debt — shortcuts taken now that you pay for later.
Vibe coding creates a different kind of debt: knowledge debt.
You ship features fast. But every feature you ship without understanding adds to a growing gap between what your system does and what you know about it.
This gap is invisible until it isn’t.
- A new team member asks how something works. You can’t explain it.
- Something breaks in production in a way you’ve never seen. You don’t know where to start debugging.
- A security researcher finds a vulnerability. You didn’t know it was possible.
The worst part? The gap compounds. The more you ship without understanding, the harder it becomes to go back and understand what you built.
What I Did About It
I built vibecheck — a CLI tool that reads AI-generated code and teaches you what it actually does.
Not a linter that tells you something is wrong.
Not a formatter that fixes your style.
A tool that explains, in plain English, what your code does — and more importantly, what it could silently do wrong.
The part I think is genuinely novel: it remembers what you’ve already learned.
If it explained “exponential backoff” to you last week, it won’t explain it again. It just notes [LEARNED — skipping] and moves on. Over time, as your knowledge grows, the explanations get shorter and the gaps get smaller.
The goal is to make itself unnecessary. A tool that succeeds when you no longer need it.
pip install vibecheck-ai-tool
vibecheck your_file.py --learn
This Is Not Anti-AI
I want to be clear about something.
I’m not arguing against AI coding tools. I use them every day. I will keep using them. They are genuinely transformative.
What I’m arguing is that speed without understanding is fragile.
The best developers I’ve seen using AI treat it like a very fast junior developer. They review every output. They ask why. They don’t accept code they can’t explain.
That’s the workflow worth building — not slower, but more deliberate. Use AI to go fast. Use understanding to go far.
A Simple Framework Going Forward
If you vibe code — and there’s nothing wrong with that — here’s what I’d suggest:
Before shipping any file: Run it through something (vibecheck, or just read it carefully) and ask yourself: “Can I explain what this does to someone else?”
When something breaks: Don’t just fix it. Understand why it broke. That’s the lesson that prevents the next break.
Once a week: Pick one file in your project that you’re least confident about. Read it. Understand it. Close the gap.
The goal isn’t to never use AI. The goal is to ensure that you — not just your AI tools — understand what you’re building.
The Uncomfortable Truth
Vibe coding will keep growing. The tools will keep getting better. The speed will keep increasing.
And the knowledge debt will keep accumulating — for everyone who doesn’t actively fight it.
The developers who win in this environment won’t be the ones who use AI the most. They’ll be the ones who use AI the fastest and understand their systems the deepest.
Those two things are not mutually exclusive. But they require intention.
I’m Afif, 16, and I build things. vibecheck is open source and free.
GitHub: github.com/apiprdt/vibecheck
pip install vibecheck-ai-tool
If this resonated — or if you disagree — I’d genuinely like to hear why.
Tags: #webdev #ai #programming #beginners #discuss
Top comments (0)