Theoretically, I agree these are a great way of generating some signal where there's not enough to be found by other methods. But, in practice, what I've found is that most of these types of exercises are way overscoped, and even those which are not put a hugely disproportionate time burden on the candidate with minimal return for most of them. In other words, nobody wants to take the time to properly review these things.

I've had much better experiences when people give me code and ask me to review it rather than when they give me a "spec" and ask me to write code based on it. One big red flag for the latter type of assignment is the phrase "production ready," as that can mean so many different things it's a borderline meaningless criterion.

I could really go on and on about this, but I'm going to stop here.