Interesting, wonder how it handles much more complex setups. I find a big part of taking a design from Photoshop and getting it running on iOS/Android is often more about thinking about what the constraints are, i.e. does an element sit at an absolute distance from a screen edge, does it sit a specific distance from another element, does it expand to fit an area, you obviously have to do this to ensure it works at different screen resolutions. These would certainly be hard decisions for AI to get right on a relatively complex screen, but then again, maybe with enough training it could actually do really well and solve these problems in a completely different way to how a human would. There is also stuff like considering whether any of the information is dynamic (text being received from a server) in which case elements needs to be able to adapt to different sizes etc. again hard for AI to have any clue from this from a photoshop image only.