For the last 6 months, I was in flow state.
Every project I touched - from small side hustles to my Final Year Bachelor’s Project - was built with AI assistance. I’d open Cursor, fire up Claude Sonnet, and let the magic happen.
Why waste time thinking about architecture when the AI can just write the code?
Why argue over whether to use PostgreSQL or MongoDB when Claude can decide?
I was shipping features faster than ever. My GitHub looked like a construction site at 3 AM - messy, but moving.
I thought I was becoming a 10x developer.
Turns out, I was becoming something else entirely: a dangerous junior developer with a fake sense of expertise.
The Confession I Didn’t Want to Make
Here’s the truth I avoided for months:
I let Claude or Cursor choose my tech stack. Every. Single. Time.
"Hey Claude, build me a task manager with auth"
"Cursor, add real-time notifications"
"What database should I use for this?" → AI decides
And you know what? It worked. The code compiled. The app ran.
But then something strange happened.
I was presenting my final year project to my supervisor. He asked a simple question:
"Why did you choose WebSockets over Server-Sent Events for this specific feature?"
Silence.
I didn't know. Claude chose WebSockets. I just accepted it.
Then he asked:
"What happens if 10,000 users hit this endpoint at the same time? Walk me through the queueing system."
More silence.
I realized something terrifying: I had never actually understood how any of my systems worked internally.
I was just an AI prompt engineer. Not a software engineer.
The Video That Broke My Brain (In a Good Way)
I was doom-scrolling YouTube when I stumbled across this:
👉 Why you suck at system design (and how to fix it)
(Watch it. I'll wait.)
The video didn't teach me a new framework or a fancy library. It did something worse (better).
It showed me how little I actually knew about the fundamentals.
The video made me realize:
AI is brilliant at syntax but terrible at context
AI will happily suggest a microservices architecture for a to-do list app
AI doesn't know your team size, your deadline, or your actual traffic patterns
AI optimizes for "it works now" — not for "it works at scale"
I had been treating AI like an architect. But it's not an architect.
It's an extremely fast typist who has read a lot of Stack Overflow posts.
The Painful Pivot
I stopped coding for 2 weeks.
Instead, I went back to basics:
Drew system design diagrams on paper (yes, actual paper)
Studied CAP theorem, load balancers, message queues, caching strategies
Asked "why" 100 times for every architectural decision
Read engineering blogs from Stripe, Discord, and Netflix
Then I went back to my final year project.
And I redesigned everything from scratch.
Not because the AI code was broken — but because I finally understood why certain choices were wrong for my specific use case.
I kept using AI. But differently.
Now I tell Claude: "I'm using PostgreSQL because I need strong consistency for transactions. Here's my schema. Help me optimize this specific query."
Not: "Build me a social media app."
The Lesson I Wish I Learned Earlier
AI is not your senior engineer.
AI is your rubber duck that can type really fast.
If you don't understand:
How your database handles concurrency
Why your API is rate-limited
What happens when your cache misses
How your message queue guarantees delivery
...then you don't actually know what you've built. You're just hosting AI's best guess.
What I Do Now (And What You Should Steal)
Here's my new workflow. It's slower. But it's mine.
Before (AI-driven) After (Human-driven)
"Claude, pick a stack" "I'll pick the stack based on my constraints"
"AI, write everything" "I design the system, AI writes the boilerplate"
No diagrams Draw first, code second
Fix errors by reprompting Fix errors by understanding the root cause
Never read the docs Read docs, ask AI to summarize
My golden rule now: Never ask AI to make an architectural decision that you couldn't defend in a job interview.
If you can't explain why Redis is better than Memcached for your use case, don't let AI choose for you.
The Hard Truth for Students (Like Me)
We're in a weird generation of developers.
We have superpowers (AI) but also new ways to shoot ourselves in the foot.
Your final year project isn't just about getting a grade. It's about proving you understand systems well enough to build reliable software.
If your AI builds it, and you can't explain it... did you really build it?
(I'll let you sit with that one.)
Where I'm At Now
I still use Cursor daily. I still love Claude.
But now I'm the architect. AI is my pair programmer who's really good at remembering syntax.
My code is slower to write. But it's mine. I understand every line. I know why every database is indexed that way. I can draw the entire data flow on a whiteboard.
And honestly? That feels better than shipping fast.
Fast is temporary. Understanding is forever.
Your Turn
Have you ever let AI overstep and make decisions it shouldn't have?
Or are you still in the "Claude knows best" phase?
Drop a comment below. Let's save each other from building beautifully broken systems.
If you found this useful, share it with a friend who's letting AI design their database schema. They might need to hear this.
P.S. Watch that video I linked. Seriously. It changed everything for me.