There's no fundamental reason that command line applications can't work with input from voice recognition.
Voice recognition, as far as I know, can't make use-mention distinctions. There's no way to "quote" a voice command. "Siri, tomorrow tell me the result of 'Siri, what is today's date?'"
The key problem with command line applications is they lack discoverability. If you're sitting at a blank screen and a blinking "_", there's no way of knowing which commands are available. To me, this is a big feature of WIMP systems. You can visually scan the menu items.
Auto-complete in CLIs gives you much of that power back. Start typing and see what results start coming up. That's why most input boxes on the web these days—even search engines!—auto-complete as you type.
I can't imagine an equivalent user experience for voice that wouldn't be maddening. "Siri, call m–" "Mom? Want me to call your mom?" "-o" "Oh, you mean MOMA? The museum? Want me to call that?"
(Discussion on who coined this phrase at http://www.greenend.org.uk/rjk/misc/nipple.html)
Edit: ah, his 2001 quote seems to validate my frustrations :)
GUI are considered more intuitively because options in general are visible as menus and buttons. On a CLI adding parameters to a command can get complicated if you are halfway and forgot if it was '-l' or '-L'.
New tab New window
What could anyone infer from the terms 'tab' or 'window' without some prior knowledge, these interfaces succeed because the cost of failure is low and the level of feedback and reward is high, so we persevere, an intuitive interface would require a whole new conceptual model of how we interact with data. No, I don't have one to show you, I am fairly certain that existing approaches will appear draconian to our children's children though :)