There's this great quote from Donald Knuth I found at some point about why he does not use e-mail which goes something like this:

"E-mail is great for people who like to be on top of things, but that's not me - I like to get to the bottom of things"

I'm frequently reminded of this when thinking about how we use AI.

It's too easy - and the first and major pitfall of AI - to just use it to stay on top of things, to not engage your mind in a world of TMI - Too Much Information.

In the Danish school system we're learning the hard way what happens when the students don't use AI to get smarter, but just to skip their homework - the grades are cratering. And let's be honest we've all seen this at work too.

To get out of the habbit of not diving in - I made an AI tool whose sole purpose is diving in. I named it Aedificium, it's named after the library (building) in Umberto Eco's The Name of The Rose.

What is it? It's a github repo with a small webserver baked in which talks to your Claude Code using the - still experimental - channel functionality. The point of the website is to build knowledge libraries - you can add PDFs, write notes and build a lexicon of terms or concepts you want to understand. Since I'm a math nerd the website speaks LaTeX as well - all info lands as files in your github - pdfs using github large files and notes as markdown files. The github sync is automatic.

As a companion to this there's also the scriptorium - where you can write and run code that illuminates the ideas you have.

There's a million plays on this idea and maybe Notebook LM or Claude itself is your thing for this need - I wanted something I can keep for myself and own in a repo that is mine; not embedded into some platform that could go away tomorrow.

How to use this

1. Fork the repos - probably make yours private
2. Hook it up with your Claude
3. Start diving