For a long time I did it the other way round. Guitar, a rough vocal, then build the track underneath it. It felt like the sensible order: get the song down, then dress it. What I kept ending up with was a vocal that had decided everything before the track had a say in anything.
Now the vocal goes on last, and the difference is not the take. It’s that by the time I sing it, the track has already told me what the vocal is for. Sometimes it wants to sit under the synths and barely surface. Sometimes it’s the only dry thing in a very wet mix and that’s the whole point. You cannot know which of those you are singing until the thing it sits in exists.
The obvious cost is that you spend weeks on an arrangement built around a melody you have not committed to. I use a scratch take for that, recorded badly and deliberately: phone quality, one pass, no fixing. Bad enough that I never fall in love with it and try to keep it.
The other thing it changes is the writing. Lines that read fine on paper turn out to have nowhere to go once there’s a track under them, and lines I would have thrown away start working because something else in the mix is holding them up. Leaving the vocal until last means the words get to answer the music instead of the other way round.
None of this is a rule. It’s just that every time I have gone back to doing it in the old order, I have ended up rewriting the vocal anyway.