On the friction of publishing.

Most company blogs are graveyards. Three posts from 2022. A half-written draft. The marketing lead left, and the queue stopped. The about page still claims weekly publishing.

The reason isn't will. The reason is friction.

Every post is a stack of small decisions. Pick a topic. Find a quiet hour. Draft. Edit. Find an image. Write the meta tags. Push the deploy. Each step is small. Stack the friction high enough and publishing becomes a budget item that loses to the next budget item.

What we removed

We build automation for the exact shape of that problem — not for blogs specifically, but for any process that decays because each step has its own resistance. So when we set up our own publishing surface, we set it up the way we'd set one up for a client.

Templates carry the structure. A model fills the variables. A single prompt becomes a deployed page in the time it takes to push a commit.

The writing is still ours. The voice is ours. The judgment about what to publish is ours. The friction we removed is the friction that has nothing to do with the writing. The rituals around it. Choosing the template. Formatting the metadata. Picking the date. Counting the words. Pushing to production. The parts that don't decide quality but always decide whether a post happens at all.

What it doesn't do

The pipeline doesn't generate the post. We do. The model handles formatting — applying the template, computing the date, slugging the title, writing the meta tags, counting the words for the reading-time estimate. It doesn't pick what to write about. It doesn't decide whether something is worth saying. It doesn't argue about the lede. Those are still arguments we have with ourselves.

This is the line. Everything above the line is judgment. Everything below the line is mechanical. Automation belongs below the line. When it creeps above the line, the work gets duller in a way readers can feel even when they can't name what's missing.

The useful test

There's a test we apply before automating anything. Whether the automation removes a decision a human should make, or only the ritual around it.

Good automation removes the ritual. Bad automation removes the decision.

Most of the loud claims about AI at work are people building the second kind and calling it the first. The lead-generation tool that writes the cold email and chooses the recipient. The product roadmap generator. The customer-service script that no longer reads the customer's message before replying. Each one removes the part of the work that was actually the work.

The version worth building is dull. It saves twelve minutes per post. It saves a meeting. It saves the third email of the day asking where the spec lives. None of it changes what gets decided. All of it changes whether the deciding happens at all.

Why we publish here

Two reasons.

The first is selfish. We need a place to think out loud — to articulate positions we hold but haven't yet written down, and to stop ourselves from holding ones we can't articulate.

The second is demonstration. The dispatch you're reading was produced through the same pipeline we'd hand a client. Templated. Automated. Quiet. The medium is part of the message. If we built it badly, you'd notice. So far you haven't.

This is the first one. There will be more. Slowly.