I’ve had interviews where I implemented a small feature on an existing application. It didn’t take any longer than your typical hour-long algorithmic implementation interview. Given that we don’t have post-hire data on any of these practices, my hypothesis is the one that most closely resembles the actual work would give the best signal as to whether that person would be able to do the job or not.
I agree that style of interview can be better. I think FAANG companies don't use this style because it's not selective enough and it requires special questions for every role.