He's talking about an optical phased array system. Once it's working, competing with a 35mm system "should" (famous last words) be a matter of scaling up the number of elements to match the lens size. The thing is, once that happens, it's got a huge advantage: it doesn't need any other lenses. At all. Any "lens" you could want is a software configuration away.

But is that true? What if I want a 50x zoom lens? How would software do what an entire assembly of precision lenses does?

It's not so much that software does what the lenses do, it's that software would configure the hardware to perform the same transfer function on the incoming photons. It's more a reconfiguration of the input than a processing of the output.

On edit: I should also point out that you don't get to escape fundamental optical limits with this stuff. You're still diffraction limited, so "lens" size still matters.

is there already a model on sale? Like lytro?

No, the technology demonstrated is only capable of imaging 8x8 pixels. Here's a link to the actual research: http://www.caltech.edu/news/ultra-thin-camera-creates-images...

Lytro doesn't quite work the same way. It doesn't capture the phase of the light, so it's a bit more limited.