I view reliance on Leetcode-style exercises as an attempt at putting a hard number on a notoriously subjective process, whether because a company is cargo-culting big tech, HR wants to eliminate subjectivity as a defense against lawsuits, or because interviewers can't agree how to run things and it ends up being the least-hated solution.

My team's interview process has a couple algorithm questions but they're fairly basic (to ensure you can still code), the main portion is about API design and system architecture. It's not very objective, but I like it since it is directly relevant to our work developing middleware systems and navigating competing interests in the process.