I’m the same. Often the first step is a time-boxed exploration, just trying to make the key pieces work in any way to encounter major blockers as early as possible. No planning, no design, not following any best practices, often all in a single file. Then from there, either refactor/rewrite or just use it as input for planning.

Of course, it requires some discipline to not just yolo the prototype into production when that’s not appropriate.