Plus, and this is often ignored today, big pictures are not entirely mutually exclusive. To use a comp sci analogy let's look at search algorithms:
Use cases are more similar to deterministic partial solution searches (backtracking depth first search, breadth first search) for which the more interesting problems are intractable. Additionally, such search algorithms are very self contained, and it is usually quite difficult to combine them, making these algorithms a one time tool.
On the other hand, big pictures are more similar to stochastic solution spaces searches, which won't get you the best solution, but they'll get you a pretty good one much, much more quickly. And since solution space searches all interface with the same space, they can be mixed and matched depending on what is most useful for the problem domain. For example, with genetic algorithms, once you can make your problem domain fit a standard GA representation, the whole of GA theory can be applied.
Aldous Huxley, for instance, both in his fiction and non-fiction, and I see no great gap between them other than the mechanisms employed, has profoundly shaped my thought with his writing.
This doesn't mean that fiction is useless for learning about the world. It just means that fiction does not contain the tools necessary to decide if something is true (or accurate, or possible). Fiction may spark an idea or make an argument, but you must go outside of fiction to see if those ideas have merit.