I was blown away by Amir Abboud’s anecdote: they suspected that All Pairs Max Flow on undirected graphs must admit a subcubic algo since known FGC tricks to reduce APSP to UAPMF failed — they found a problem ripe for better algos guided by the library of tools developed in FGC.
SETH might be false but fine-grained reductions and equivalences live on. (Any one of our hypotheses could end up dying once the right technique is developed. In a sense, we hope for this, once we have proven equivalences.)