The feature is available only for Denver, Las Vegas, Miami, New York City, and the San Francisco Bay Area, but it will be extended as soon as Google has more street-level imagery.
And here's a short demo from Google:
Philipp Lenssen from Google Blogoscoped thinks "this looks & feels amazing, albeit there's potential privacy issues due to the level of detail (you can make out individual faces, license plates and so on)". Now all we need is some real sound and a way to integrate information about different places.