Guy (micro)blogs that his imaginary friend now has a data structure that apparently mimics a calendar and possibly exposes an ical feed.
Good for them, but why does the internet need to know that?
I've been pondering a conceptually similar approach for keeping slop out of important git repos. The agent has its own slop repo that it can slop up as much as it likes for R&D, and then I transcribe what works into my clean repo (ensuring I understand it, in the process).
One could probably also give you words sampled from /dev/urandom and it might spark an actually interesting thought, but that doesn't give urandom any merit.
FWIW, I've been using a similar workflow as you describe, though the lines have been blurry at times.
But I can see that not everyone is bothered by that as much as I am. Not that I would have any mandate to do so anyway.
In general I'd like more services to maintain a wall between me and agent, but then again, we can also let the agents build it.