One of the biggest upgrades to my AI workflow has been spending more time planning before I start building. I'll usually have a back-and-forth conversation with ChatGPT or Claude to discuss my concept, pressure-test ideas, uncover edge cases, and shape the project into something that actually makes sense. Once the idea is solid, I turn it into an mvp_prd.md (Minimum Viable Product Product Requirements Document), which is a living source of truth document that captures the project's goals, scope, research, assumptions, and roadmap. From there, I convert the mvp_prd.md into plan.md, which becomes the actionable task list I work through while building. A few extra minutes of planning can save hours (or days) of rebuilding later. How do you ideate your projects before you start coding?