The takeaway is that an obfuscator that produces a "black-box" version of any program cannot exist, but one that can produce equal but indistinguishable obfuscated versions of any one program does exist.

I find that the article's most intriguing point is the link between homomorphic encryption and indistinguishable obfuscation.