I'd personally settle for a voice-output-only Glass (maybe with chording for text input to avoid the "talking to yourself crazy person" problem, and with a camera/microphone built in for ambient data gathering). Hardware we could have built for the last 10 years, and really unobtrusive. You'd need better UI and software than with video, since information needs to be more closely tailored for a lower bitrate channel,…
You don't even _need_ Glass for that -- just an unobtrusive bluetooth headset, and a good backend.
It should support something more reliable than SCO and an input touchpad like Glass. It would also make sense to have some of the initial NLP processing in the device, or even a full ARM processor like the Glass running arbitrary applications.
You would be able to make calls through an offloaded cell phone using a more advanced profile than SCO HS.
If I could remove the device and place it on the car visor to use as a noise-cancelling speakerphone like the Motorola Bluetooth unit I have, even better.