Do you see further research paths for metacompilers [1] to reduce code and enable customizable user interfaces?
With containers and hypervisors now in desktop OSes (Windows, Mac OS, Linux), could an open-source research OS (e.g. KSWorld) be packaged for developers and end-users who want to test your team's experimental UIs?
Is there long-term value in "private machine learning" where some data and algos are focused on user/owner interests, with "public machine learning" providing variably-trusted signals to user-owned algos for intelligence augmentation?