> AI processing

This one is super interesting to me - clearly Snapchat, Google, Apple, and I assume Samsung have proved that this is viable.

While some features (e.g. portrait mode) exist to work around limitations that many lenses don't have (ability to create a shallow DoF) other ones are more interesting, i.e. Snapchat filters.

Still, I worry that these are too difficult to reason about as compared to something like the exposure triangle, and the imperfections of things like portrait mode can be very obvious. Perhaps these are better left to post-processing tools?