How do you balance clarity and speed when prompting AI to write code?

I keep running into the same pattern with AI coding tools: I type a quick “starter prompt,” get something that looks promising for a moment, and then, inevitably, it collapses into messy code and outputs I never wanted in the first place.

I’ve seen this happen to others too. The tool isn’t the problem. The problem is the prompt. Or rather, the lack of structure, clarity, and intention behind it.

So I’m curious:

How do you plan your prompts when working with AI for code generation?
How much context and detail do you include up front?
Do you start small and iterate, or do you specify the entire mental model before generating anything?
What habits or prompting frameworks have actually helped you get clean, reliable code?

Would love to hear what’s worked (or failed) for you!