I suppose that depends on whether you go from kana to kanji or the other way around during your reviews. If you use kana (or audio) for questions, it becomes pretty natural to write out what you "hear". I currently go from kanji to reading, though, and I like to write out the more difficult parts of a sentence.

Personally I'm going kanji -> kana. Later, I will use other methods to train output (real output, not writing what I hear which I consider to be pure mimicing).

Going Kanji to kana is similar to the division of labour in Heisig.  You could insist on being able to produce the kanji from audio, the kanji from kana, and be able to recognise the reading from the kanji for each word you learn.  However I don't feel the need to learn every single thing about a word at one time.

So I also go kanji to kana only, it's easier but also faster.  Which means I get a large passive vocabulary to use with my immersion environment.  At the moment I don't need to be able to write every word I come across, the priority is to quickly increase my passive vocabulary so I can understand better what people are saying to me.  I don't need to write letters, all I need to do is be able to read and listen.  The more I can read, the more input I get.  The more words I can hear, the more input I get.  More input leads to greater all around skill.

Passive vocabulary needs to be larger than active vocabulary.  It's a major beginner fallacy to think that you need to know everything about every word.  I want to be able to understand the NHK news, but in daily life there is only a small subset of words that I realistically need to be able to produce.  And I'm with Tobberoth in thinking that being able to do dictation is not real production anyway.  It would be better to write emails, letters, essays, get on lang-8, etc.