I spent 3 months working on an AR project on the Magic Leap (the best funded AR startup, I think?). My conclusion was that the hardware just isn't there yet.
VR tech feels impressive because the hardware works well. AR tech, on the other hand, still feels very much like a prototype, and far from real production use. Some of the main deal-breaker issues are heat, discomfort, dizziness, and software instability (I wrote more about the specific issues I encountered with the Magic Leap in a past thread [1]). To me it felt significantly worse than the Oculus Rift Development Kit 2 (the second public release of a VR device back in 2014). Back then when I got that version because I was super hyped for VR, I thought "Woah, clearly needs more work but this is amazing". With AR, my experience thus far has sadly been "Oh, this is it?".
A lot of it is presumably because AR is much harder to do well. A very simplified way of thinking of VR is basically it's a screen close to your face that can track its own pitch/yaw/roll/small amounts of movement and redraw accordingly. This is tricky, but totally doable (and thus, it was done). Smartphones have basically been able to do most of this for years (not so much the movement, but that's also a pretty new thing in VR).
AR actually has to interact with the real world. If you have a virtual image anchored to, say, a pendulum, that's fairly non-trivial computation-wise to recalculate without prior knowledge of the pendulum's movements. It's basically object detection which ML models can do fairly well, but they do it with latency. And when the latency is in something your brain is trying to process as the real world, it's nauseating. Add on the fact that it takes a lot of energy to do inference, and that AR devices need to be standalone devices that can be taken anywhere by themselves, and you also have a serious heat problem where the compute needed to constantly analyze the real world means you just have a hot brick on your head or in your pocket.
I think it's going to take a lot more time to get it right. It feels like a hard enough problem that I think the first truly production-grade AR device will have to come from one of the big companies, possibly Microsoft with HoloLens.
Of course, I write all this with the definition of AR as "something you wear on your head that overlays your full field-of-view with virtual graphics". Other forms of AR like Snapchat filters or camera overlays are probably less challenging.
[1] https://news.ycombinator.com/item?id=20253177