I would counter that many complex phenomena are best represented by graphs. There's a sort of elbow in the learning curve where you have a lot of tools but as your datasets grow in size the graphs become unmanageable and the signal:noise ratio goes way down. At this point you need to prune or chunk somehow, but it rapidly becomes more of an art than a science and reproducibility is elusive. Advanced techniques disseminate slowly and are not that intuitive, but there's still steady progress, eg deep learning models for sparsification while preserving structure.
Thinking in systems, complex emergent behaviors, networks and graphs, category theory... are all very powerful cognitive tools. They are not explored in basic education and people stumble upon these concepts in college, during masters, and get a boost in their analytical skills. They were all popular objects of popular science writing over different decades and generated some hand wavy "theory of everything" marketing for books. They are quite useful, if a bit overblown in their own little hype context.
