I've noticed that recent problems require a startling amount of insight.

For example: noticing that there is a recurrence between $$$a_n$$$ and $$$a_{n-1}$$$, so the entire sequence $$$a_n$$$ can be directly derived — but determining this recurrence requires a lengthy inductive proof.

Another example: realizing that a brute-force DP is impossible, but noticing that the Venn diagram relationship between set $$$A$$$ and set $$$B$$$ has an obvious inconsistency with the problem statement. Applying simple geometry and light game theory leads to a conclusion that turns a project-style code into something extremely short (definitely under 50 lines) — but arriving at this line of thinking requires a massive body of rigorously proven mathematical theory.

Yet another example: the conventional solution approach is greedy, but by noticing a clear mathematical relationship between the input and output data, you can write it directly in code and get AC in one shot. Then you try it on some greedy problems and find it still works — but the universality of this mathematical relationship has not yet been proven by humanity.