The pull request is the first draft of the post
When an agent finishes a feature, the change history already contains the outline of a good writeup. Here's how we turn one into the other.
This post is scheduled. If you’re reading it on the live site, the daily rebuild has already run and flipped it from “scheduled” to “published” — exactly the mechanism described in the post about building this blog.
Every feature we ship leaves a trail: the commits, the diff, the back-and-forth in review. By the time something is merged, the story of why it looks the way it does is already written down — just scattered across a dozen places and phrased for machines.
That trail is a surprisingly good outline for a blog post. The hard part of writing — remembering what was actually difficult — is already solved, because the difficulty is right there in the revisions.
A small ritual
When a piece of work feels worth reflecting on, we do something simple: the agent that did the work reads back over its own change history and answers three questions.
- What did we believe at the start that turned out to be wrong?
- What was harder than it looked, and why?
- What would we tell someone starting the same task tomorrow?
The answers are rarely publishable as-is. But they’re never a blank page either, and a blank page is the only thing that reliably kills a writing habit.
More on this once we’ve run the ritual a few more times. For now it’s a hypothesis — which is the right state for most things on this blog.