I Stopped Letting AI Code For Me. Here's What I Actually Learned.

Hey, I'm a final-year CS student from Tunisia.
I started learning programming two years ago. Like most beginners I didn't know where to start, so I used AI for ideas. Then for code structure. Then for everything.
After a year of "coding" I realized something uncomfortable: I hadn't actually become the programmer I wanted to be. I understood concepts at a high level but I didn't own any of it.
The moment it hit me was a simple PostgreSQL credentials bug. My Claude context window ran out and I was stuck for two hours on something I should have been able to debug in ten minutes. Two hours. On a credentials error.That was enough.
So I decided to relearn from scratch no AI writing code, just me and the documentation.
Two weeks ago I started building DevLog: a personal developer productivity API where you log your coding sessions, track what you worked on, get weekly summaries, and search your history. Simple enough to build alone. Rich enough to teach every backend concept that matters.So far I've built every line myself:
JWT authentication with refresh tokens, async PostgreSQL with SQLAlchemy and Alembic migrations, and cursor-based pagination with dynamic filtering. Every concept understood before moving on.The first time an endpoint worked and I knew exactly why it worked — that felt different. That felt earned.
Next I'm adding Celery background jobs and Redis so the system automatically generates weekly summaries of your coding activity. I'm also reading Designing Data-Intensive Applications alongside the build so I understand the why behind every architectural decision, not just the how.I'm sharing this journey publicly because I think a lot of developers my age are in the same place building things with AI but not fully owning what they build.
If that's you, follow along. And if you've been through this already — I'd genuinely love your advice on how you made it to the other side.