I would say the most useful part of studying algorithms is knowing how to analyze them and understanding the meaning of asymptotic analysis. Understanding that asymptotically fast algorithms might be slow for the problem sizes you are dealing with is the most important thing an algorithms course can teach.
This. I've been marking undergrad assignments, and you would not believe how many sets and sorted sets I've seen implemented by linear scan over an unsorted array (or linear scan over a sorted array with linear time insert, and a few other, worse things). This lack of understanding of algorithms leads to slow, unscalable solutions, and even writing business middleware you at least need to know which standard library collections to use for good performance.