You're not bad at your job. Your prompts just aren't built for the model you're using.
Vague prompts burn tokens on back-and-forth clarification instead of working code.
The same prompt that works great in Claude falls flat in Cursor or Codex every model expects something different.
Switching tools mid-project means re-explaining context from scratch, every single time.