From working with someone elses' Jetson K1 & TX, the old Jetsons were a pain in the rear to re-deploy when re-imaging/upgrading the OS and its been a good move the Jetson NX bridges the gap between Jetson Nano and the AGX for a small footprint with plenty of features and ease of OS upgrades/reimaging via SD card to attract more developers. AGX has useful encoding abilities which I've ran a few experiments on a first gen model(16GB RAM model) but there are things about certain Ubuntu-specific stuff either works or broken on the Jetson platform even though they're included with the base Ubuntu image from nVidia.
Work wise I have 3 early Jetson Nanos, actually find them more reliable may it be passive cooling or active cooling vs the unpredictable nature of other SBC boards(serious thermal throttling without active cooling or extremely picky power delivery even if you're underclocking it to keep the thermals extremely low). Jetson NX looks good, however it'll be interesting if there will be 3rd parties making aftermarket large passive heatsinks as there are use-cases the fan isn't ideal.
Beyond AI/DL/FR, there are plenty of applications where the Jetson Nano/NX are great for the encoding ability if you're using a mixed SBC setup for shared tasks--Jetson's CUDA side of compute/encoding fills that gap. I've ran a mixed SBC setup of dual Raspberry Pi 3B/4 with a Jetson Nano or a dual Raspberry Pi 3B, Adapteva Parallella and a Jetson Nano.
Based on the nVidia Embedded Developer site, their hardware roadmap is more focused on keeping a much more open direction of what to expect may it be for mainstream or commercial/industrial usage. It'll be interesting how the expected 2021 Nano Next specs will shift towards based on all the other $99/$149 SBCs which popped up with dedicated AI compute core(s).