My favorite things are the backgrounds (which really surprised me; before reading them I thought the very idea of additional backgrounds was silly and superfluous, but I really ended up liking what they did with them in this book), the bladesinger, the racial restrictions on class options, and the way some of the text hinted at things to come. I like the bladesinger and I think it was very well done - some have complained about it's lack of 'tankiness,' but I think that such complaints miss the point. A bladesinger isn't a tank, it's a skirmisher, and works really well in that regard.
Like Garnuk above, I like the wording of the racial restrictions. I think that restricting access to class and other options adds a valuable layer to the game and to worldbuilding, and I liked how in the SCAG, the designers played to that idea while still making it clear that individual game tables can and will vary dramatically from that assumed norm. But then, this book was written by Steve Kenson, who knows a thing or six about game design.
Lastly, I like the way they basically told us that the 5e interpretation of Al-Qadim's Sha'ir was a sorcerer, without ever saying so directly. I actually think that a warlock pact fits the sha'ir idea better, but the text on page 14 is a great example of how this book manages to pack an enormous amount of ideas and information into its relatively short runtime. Though again, this is Steve Kenson we're talking about.