How this page works

The short version

I run a program that pulls my Github event feed, picks out interesting things that I did since last time, updates the page and publishes it to a couple of places (this site, my employer’s internal CMS, etc).

The longer version

I wrote some tools. One to pull the events feed and spit out events that might be interesting, and one to take a markdown file and convert and publish it to each of the targets I’m interested in.

The rest is a Claude Code project. I gave it an initial markdown template, instructions for how to run the tools, and guidance for how I want it to summarise the work and include it in the output. Some of that is writing style, some of it is things to ignore, and the big ones are indicators for what the “current focus” areas are, and how to manage the pull request tables. All the vague texty parts, basically.

I read every change that it makes before it is published. Occasionally, I’ll point out something it missed, or ask it remove something it should have skipped. Less often, I’ll request an edit or reword, and even less often, I’ll provide that edit myself.

The purpose of the review is to avoid obvious mistakes or confusion, not carefully vet every single item. Or put another way, It’s built for speed, not for accuracy.

Why AI at all?

Mainly because it can get the job done at all. I’ve been meaning to set up something like this for three years, since I started working on OpenZFS full time. I haven’t so far though, because reading and writing and summarising are tasks that I am very slow at. Whereas this dumb thing has taken about a day total to set up and a few minutes each day to keep up to date.

Feedback welcome

I’m still kicking the tyres on AI tools, trying to filter out the hype and understand the places where they might work for me, and the places where they definitely will not. If you’ve got something interesting to say on this, feel free to say hello!