However you do it, start simple, because you can always get more complicated later. You might have a cool idea in your head -- and then realize that coding it or putting it in the game is harder than expected. That might force you come up with an alternative, so stay flexible with your vision of what the game is going to be like.

In my own experience, I wish I had spent just a little more time plotting things out on paper before writing code, but at the end of the day you really have to just jump in and try things to see what's possible/feasible and what isn't. And the strange thing is, sometimes you have the skills to do something (like create a cool model or animation in Blender) and you still end up having to scrap it because it physically increases the size of your game too much, or causes lag, or conflicts with other parts of the game.

For me, I grabbed a short section of the story and tried to make it into a demo. The structure was pretty simple at first, there were the basic input system and skill tree, and that's all. Then I started to work on the aesthetic for a bit, and then I go back to programming again. Just like that, block by block.

My first project was too ambitious, I made a lot of systems first, but some of them are actually not so useful, so time waster (or not, lesson was learned :P)