My AI tells me what to read

Drowning in reading material

Keeping up with AI is almost a full-time job in itself. More appears every day than you can follow, and most newsletters are a wall of headlines with no distinction. What's gold to me and what's noise sit side by side.

The problem wasn't that I read too little, but that I read the wrong things. A deep piece that touched exactly on one of my projects could stay unread, while I spent time on something that looked fun but got me nowhere. I wanted a filter that knows what I'm working on.

A catcher and a sorter

The router works in two steps that run independently. The first catches incoming reading material and stages it. The second reads that pile twice a week and turns it into advice. Keeping them separate keeps things calm: the catcher can take in mail all day without the sorter getting in the way.

Along the way I learned one thing bigger than the rest: cleaning up before the AI sees anything does more than a smarter prompt afterward. The moment I took the clutter out of the mails before the model looked at them, the advice got noticeably sharper. And the connection that pulls the mail in I deliberately set up in a more stable way than the standard Google login, which dropped out now and then. An unattended system has to keep running without me looking after it.

What a reading list looks like

This is a rebuilt example of what's ready twice a week: a handful of pieces, sorted by what they mean to me, with the "why" attached and an honest pile that didn't make it.

Rebuilt example, based on a real recommendation. The structure is exactly what the router delivers each time.

The advice lands automatically in my knowledge base and in my mail.

Where it chafes

No system is ever finished. When a lot of mail arrives at once, one can occasionally drop out of the queue: roughly one in ten during such a burst. I accepted that on purpose rather than putting a heavier machine behind it, because for a reading list twice a week, simplicity outweighs completeness.

And the advice is only as good as the taste that goes into it. The router knows what I've told it about my projects. If my focus shifts, I have to adjust that, or it keeps advising for last month's Sjoerd. It's an assistant I steer, not an oracle that knows better.

Set up your own reading router

You don't need the automation right away. The heart of this system is the instruction you give your AI, and you can use that today. Gather a handful of articles you have lying around, and let an AI pre-sort them for you with the prompt below.

You are my reading router. I'll give you a list of articles (title, link, short summary).

Weigh each article against the projects I'm working on right now:
[for example: building a new website, learning photography, a data-analysis course]

Then give me a short reading list, sorted into three groups: high, medium and low. For each piece, include:
- a title and the link
- one sentence on why it's worth my time right now
- an estimated reading time

Put everything that doesn't fit separately under "skipped", with a brief reason for each.

Be strict: better a short list that hits the mark than a long one that includes everything.

Part of a bigger system

The Knowledge Router is one of the streams that feed my AI assistant Kai, and the advice lands in the same knowledge base as my second brain. Together they make sure what I read doesn't evaporate, but sticks somewhere I can find it again.