We are in an age of derived data where people are using LLMs to digest content. Blog posts are easily summarized by our LLM agents, why not offer an author-approved summary right from the start? Never mind summaries, now there's people all getting lists of key points, one sentence tldr's, various levels of explanation. The proliferation of all this derived data points quickly towards the possibility of caching it.

Why should we all be burning tokens on getting summarizes of the top 20 WSJ articles of the day? Just make a central cache of this stuff. Then you start to wonder how to cache other connections, and what schemas would be widely acceptable here? Pretty soon you've gone down the SWT rabbit hole.

Let's make an ontology for all this derived data to let people collaboratively cache it.