Almost any work on compilers will involve good working familiarity with graphs
Work with concurrency will tend to involve linked data structures, like treiber stacks, or modeling computations as directed acyclic graphs of data dependencies
Working in a memory constrained environment tends to involve data structures and algorithms tailored to that domain, etc.
Now it's totally true that many algorithms questions just reduce to brain teasers, or are so esoteric and interview-specialized (Find out if this linked list has a cycle in O(1) space!) that they only serve to stoke the egos of the interviewer. But I wouldn't say that using algorithms or implementing a data structure is a skill that only exists to facilitate interviewing - and being able to explain your thought process as you work is really important when making contributions to other teams' code. I try to focus my interview questions on miniature versions of problems I've actually had to solve in the course of my job. For example, the futex state machine I wrote here:
If you subtract the interrupt and timeout handling, you have a real problem with a simple, approachable (for someone with some concurrency knowledge) solution, and you only need to know about two futex calls which have simple behavior and are easy to explain in limited time. I wouldn't ask this question of someone looking for a web development position; but then again, I don't interview those candidates because I know nothing about web development. But it's a nice question for someone who listed concurrency on their resume.