I Asked AI How to Use It
I spent today asking an AI how to best work with it. Instead of collecting advice from blog posts and social media (where everyone shares their strategies), I figured I'd go straight to the source and ask the model itself.
I shared my current workflow, the pre-planning approach I described in my last post, and asked: what am I missing? How can I improve this?
The response wasn't about magic prompts or new techniques. It was practical, direct feedback on how to think about the interaction. Here are some of the interesting parts:
Don't describe code structure — point me to it. I can read code faster than I can parse your description of it.
Tell me what NOT to change.
Verify, don't trust. I'm good at generating plausible code. Plausible ≠ correct. Always run it. My confidence in my output has zero correlation with its correctness.
You can read the full write-up on GitHub.