1. Never ask AI to one shot the answer.
Transgression
Lukas is vibe coding his todo list app. He has 2000 lines source code files. He encounters several bugs when building. He provides the entire file to the Claude Sonnet. Claude one shots the fix. Claude fixes the main issues, but it also makes it's own sets of bugs. Lukas ends up with more bugs than he started with.
Observance
Samantha is also coding a todo list app. She encounters a few bugs when trying to implement a function through a function. She provides the file to Claude and asks for the issues it contains. Samantha looks carefully and points out the misidentified issues. After she provides feedback, and she gives a go ahead, Claude provides the fix. The fix works in the first try.
2. Don't use AI as a way to skip work.
Use AI for discourse, understand concepts. And then use your own critical thinking because AI can and will hallucinate.
Do the difficult part yourself. Try to author the initial version yourself. Have your originality. Use AI to upgrade your existing style. Use it more as a conversational partner that allows you to express your thought clearly. Use it to retrieve information, but do not trust it. Verify it yourself.
You must get your hands dirty yourself, if you depend on /aI you might be useless without it.