I've done some commercial CC work recently for a studio and it is REALLY, REALLY hard work. Way harder than it looks to get it right.

Some pieces I would do would have computer-generated first pass. This would be good in places where the words were very clear and the vocabulary was regular. But TV and films can use a lot of weird domain words (see e.g. sci-fi or fantasy) that the computer can't track. And the computer has serious problems with proper nouns and names.

On top of that you have to assign each phrase to a character who might or might not be visible when they are speaking (which sucks in a whole room full of people with similar voices), and you have to time it correctly.

I got fired over an argument about comma placement.