Asked Claude if it was confident in its own approach.
It said no. Then gave me a completely different and better solution.
Most people take the first answer and run with it.
Two words. Saved me hours of debugging broken code.
Always pressure test the first answer.
What's your go-to prompt for pushing back on AI output?