I use both, done some complex things in OpenFL as POC, with Feathers as UI, it's great. For now, the Harman SDK and Unity are more polished for "releases", only reason I am holding back for doing more complex commercial things in OpenFL/Feathers.
I know Unreal as well, so what my point is, there is nothing Starling right now can't give you with as good as feeling as the others if Harman gets AAB done in time. AIR is, for the large part a massively stable piece of software compared to other multi platform options I have investigated. I have spent at least 2 years figuring out what I am doing for the next 10.
If Harman keeps going, I would like to invest in Distriq's ANEs as well just to help thing keep going while at the same time deving in Unity and Unreal (also bare metal C++) as well. Best of both worlds for me, not all eggs in one basket etc.
So as said above, AAB is the only looming thing Harman needs to get implemented and it's complicated, so hope they get it right. 🙂 The rest is how fast can you realize your dream, just because AIR is not cool, it' is still fast to dev with for certain use cases.