I think this is too strong an opinion to hold. I think there is great value in being able to put problems you face into some sort of detailed taxonomy, and either know directly or be able to find the correct/optimal solution for those problems. To be able to do that, you do have to be exposed to what exists in the body of computer science knowledge. But while I disagree with anti-leetcode people who just claim googling is enough, I do believe it's enough to know that something is a graph/tree/optimisation/whatever problem and google a valid implementation. Having an intuition for performance and complexity is also very useful, but not knowing the exact complexity of an algorithm isn't a blocker to good work.

I think it's essential to have a good map in your mind. I don't think it's essential to have a photographic memory of every route through that map.