Until we have some absolute killer -- and I really mean killer -- APIs for image recognition and AI, their use is going to be rather limited. It is telling that the vast majority of demos that they have used to demonstrate Hololens would actually be more compelling as VR experiences. Navigating the solar system, the human body, or the silly shooting game they used to demo it a few years ago are all fine uses of the technology, but they don't in any way meaningfully interact with the world around you -- which is ultimately the value add with AR as opposed to VR.

When I used it, all I could get as a developer was a mesh with the surfaces that the Hololens had recognized in the environment. That is really good for properly occluding objects and allowing things to appear in real space; it is insufficient if your goal is to actually meaningfully interact with the world. I don't just need to know there's a wall there; knowing that it has a TV in the middle of it is important if I want to build an experience to accompany a live sports broadcast. This technology will really shine when we can, with incredible speak and accuracy, put a Hololens on and instantly know that its in a living room, and that the object in front is a coffee table, and there's a picture frame on it, etc.

We do have a lot of this capability already available, but to me the key leap that is missing is a way of accessing and using all of that in a coherent and easy way as a dev.