Mathematics had concepts of convergence/bounds of infinite sequences since antiquity. Big-O is just mathematical notation that you can use for whatever you want. Big-O is useful because most of the time we want to reduce time complexity for algorithms when inputs get large, where constant and lesser terms usually become insignificant.
Some people measure steps, but we call it "time complexity" because they're essentially the same thing assuming that time taken for each step does not depend on `n`.
There are occasionally philosophical implications when you prove a complexity bound, so in those cases making stuff run faster isn't the goal, but unless you're deep into theoretical computer science, usually making code faster is your goal when you have to deal with time complexity.
FWIW, for loops isn't actually a primitive of CS or even programming. Of course you can achieve Turing completeness with C-style for loops, but that tweaking for loops isn't usually how people think about reducing time complexity...
Some people measure steps, but we call it "time complexity" because they're essentially the same thing assuming that time taken for each step does not depend on `n`.
There are occasionally philosophical implications when you prove a complexity bound, so in those cases making stuff run faster isn't the goal, but unless you're deep into theoretical computer science, usually making code faster is your goal when you have to deal with time complexity.
FWIW, for loops isn't actually a primitive of CS or even programming. Of course you can achieve Turing completeness with C-style for loops, but that tweaking for loops isn't usually how people think about reducing time complexity...