Context
I’ve been skeptical about AI coding assistants. For years, I thought they were just fancy autocomplete that would make developers lazier. But after watching my team struggle with repetitive tasks, I decided to give it a real shot.
What I Used
I used Claude Code and Cursor for 30 days on actual production work. Not toy projects. Real code, real deadlines, real pressure.
The Good Parts
The AI was genuinely helpful for three things:
- Boilerplate code – Writing repetitive patterns that I hated but had to do
- Explaining unfamiliar code – Reading other people’s code became way faster
- Writing tests – It generated decent test skeletons I could refine
The Bad Parts
It also made some mistakes that cost me time:
- It suggested solutions that didn’t match our architecture
- It sometimes invented APIs that don’t exist
- I got lazy and stopped thinking through problems before asking
What I Learned
The AI is a tool, not a replacement. The biggest win wasn’t the code it wrote – it was how it changed my workflow. I started spending less time on busywork and more time on actual problem-solving.
But I had to stay alert. The moments I blindly trusted the AI were the moments I introduced bugs.
Would I Keep Using It?
Yes, but differently. I use it as a thinking partner now, not a code generator. I describe my problem, look at its suggestions, and then decide. That’s the sweet spot.