I have been testing a new flow for building projects with AI, and it does not start by writing code.
Vibe coding is tempting. You open the editor, ask an AI to generate something, and watch it run a few minutes later. It feels good. It keeps feeling good right up until you notice you have built something fast that you never really thought through, and that is before we even talk about code quality.
Over time I realized the problem was not the AI. It was how I was using it. I was skipping one of the most important parts of building software almost entirely: the planning.
Because software does not start in the editor. It starts well before that, with questions that are rarely comfortable:
- What problem am I actually solving, and for whom?
- What does success look like, concretely?
- What am I deliberately not going to build?
- Where are the rules and edge cases that will bite me later?
When I answer those first, the AI stops being a slot machine and starts being a good engineer. The output gets more consistent, the architecture stays cleaner, and I spend far less time deleting code I should never have generated.
The tooling got faster. The thinking still has to happen somewhere. If you skip it, you do not save that time. You just pay it back later, with interest.