pull down to refresh

Hi, I have experienced non predictable results using AI.
Claude code max has one shot complex features and totally missed easy ones.
For example: He one shot a browser extension. While totally missing the landing page requirements for that extension.
After 6 months, my conclusion would be that I often loose more time by prompting and yelling at the model to do the right thing than I would if I had coded it myself. - Than could be a prompting skill issue.
I tend, and that's personal, to become lazier while using AI and could be a bad discipline of myself: using it for everything.
Currently, I am in an in between:
- I really like using AI for reviewing my code.
- I like using it for righting tests
- I also tend to prefer it to act like a coding tutor / learning mode, so that I still have to implement things myself. It's more like a safe guard and copilot. But I still right most of the code myself. For some really little piece of code, or to generate boilerplate, I delegate the work to AI.
What are you experience ?
GENESIS
Salut ! :)