It’s a great tool for knowledge sharing, ramp up and debugging. Definitely not something that needs to happen on every story. Stuck on something or working on a weird bug? Get someone on a call and walk them through it. New team member or old susbsytem not many people understand? Pair the less knowledgeable person up with an SME for the first couple tasks so they can pick the SME’s brain while they work and get valuable context that might be lost in code or the story description.
It also doesn’t need to drag on. I find 30 minutes is best because as you approach an hour+ attention is hard to maintain. Get on the same page, learn a few things and once your making progress move to async communication.
Pair programming is a tool and only valuable if you know how and when to use it.
Knowing what and when to abstract can be hard to define precisely. Over abstraction has a cost. So does under abstraction. I have seen, writen and refactored terrible examples of both. Anecdotally, flattening an over abstracted hierarchy feels like less work and usually has better test coverage to validate correctness after refactoring than abstracting under abstracted code (spaghetti code, linear code, brand it how you will). Be aware of both extremes and try to find the balance.