I think voice as input is ideal. Text or voice as output should be the rarer case. We are at the “Google and get a list of links” phase; it was a massive improvement over a directory of categorized websites, but not it’s final form