You can go either way; you can take a sort of vague, weak web you've already started building and use the stakes to fortify it. You can lay out a lot of stakes to help you build the web later. You can do any hybrid thing you want.
But either way, Anki may help you memorize things, but it doesn't really help you learn things. I think it's important to understand that when using Anki, so you don't accidentally convince yourself you're "learning" when you aren't. But it can greatly accelerate the learning, to the point that the memorization+learning time << learning time, depending on the circumstances.
(For human language learning, a common Anki task, I suggest an "all of the above" approach; some pre-made cards that probably are higher quality than what you can assemble yourself, some cards you make yourself based on your own learning and reading material, some raw vocabulary cards that doesn't much matter where they come from, non-Anki sources like textbooks, a little bit of everything. In most endeavors in life that's a recipe for diffusion of effort and a lot of wasted effort but it seems to me to be the way to go for human language. A single type of Anki card on its own may let you lay down a truly glorious set of stakes, but you still won't have "learned" much of anything.)