A publishing system is not a writing system

The pipeline worked. The writing hadn’t.

I wired up a Make scenario that could take a single row in Google Sheets and turn it into a ready-to-go Ghost draft. Slug, excerpt, meta description, publish date, even scheduled social posts. All automatic.

On paper, it looked like a writing system. In practice, it was a logistics pipeline that happened to move words.

It could not see that a draft about a Withings weigh-in said “significant progress” instead of “2.3 kg.” It could not know that Tuur’s batting average was .818 because it had never watched him hit. It had no concept of what mattered. It only knew where to put things.

That is the gap: automation can move text, but it cannot supply observation or judgment. A publishing system is not a writing system.

Automation treats writing as assembly

Most automation tools treat content like parts on a conveyor belt. A title goes here. A slug is derived there. An excerpt is trimmed. A social caption is stitched together from variables.

If you look only at the outputs, it is easy to believe you have automated your writing. The drafts show up on schedule. The formatting is consistent. Everything lands in the right place.

The problem is that writing is not assembly. Writing is archaeology.

The real work is digging through a week of experience and finding the fragment that matters: the wrong lineup, the plausible AI answer that was subtly wrong, the rain-soaked practice that changed how you coach. Those are not fields in a database. They are things you had to notice, doubt, and interpret.

Automation cannot dig. It can only polish what you hand it. If you hand it sand, it polishes sand.

My publishing system was elegant precisely because it hid the fact that the writing had not happened yet. The pipeline ran smoothly. The drafts appeared. The posts published. The missing part was the only one that mattered: a human actually looking at the week and deciding what was true and worth saying.

What the pipeline will never see

When I went back through the automated archive with an editorial eye, the gaps were obvious. They had been invisible to the system because they are not the kind of thing a system can see.

I wrote about this earlier in Why my posts now start with five bullets.

I found posts with three consecutive rhetorical questions. They sounded energetic, but they said nothing. The automation was happy to ship them because a paragraph is a paragraph.

I found a conclusion that contradicted the second paragraph. The structure was fine. The HTML was valid. The logic was broken. No webhook will ever raise a flag for “argues against itself.”

There was a personal insight that was statistically impossible. The pipeline did not blink. It does not know statistics. It does not know me. It does not know the difference between a plausible sentence and a true one.

The system had moved all of these from draft to published without friction because it had no concept of truth. It had no concept of me.

A publishing system is only as honest as the human who stops it.

The first mile is writing, not wiring

So I changed the way I use the automation.

The first mile is now five bullets in Notes, typed on my phone while walking to Ideebv. A phrase from practice. A number from a weigh-in. A question I actually had, not one generated to fill space. That is the writing. That is the archaeology pass where I decide what fragment is worth keeping.

Only after that do I hand anything to the system.

The automation handles the translation from raw thought to Ghost markdown, the image sizing, the scheduled send. It does the things that are boring and repeatable.

In other words, it is a courier, not an author.

That distinction matters. When the system tries to be the author, the blog drifts toward a content farm: regular, polished, and hollow. When the system is the courier, it does exactly what it should: remove the excuses between “I noticed something” and “this is published.”

More on this in my article Fluent emptiness and the missing Tuesday morning.

Where the friction belongs

The system works best when I am the bottleneck.

If the pipeline ever runs smoothly without me, that is not a success metric. It is a warning sign. It means I have become optional. It means the posts can go from idea-shaped noise to public URL with no one asking “is this real?”

A healthy publishing system should require my judgment at least once. At some point in the chain, I should have to verify that the observation actually happened, the number is accurate, and the lesson is mine to write about.

I am not trying to eliminate friction. I am trying to put the friction in the right place.

The writing should be hard. It should feel slightly uncomfortable to pin down what you really think, to name the real number instead of a vague phrase, to admit the lineup you chose was wrong.

The publishing should be easy. Moving text from Notes to Ghost, resizing images, generating a slug from a title: none of that needs soul. That is the part where automation shines.

When those two get swapped, you end up smoothing the only part that needed resistance and overcomplicating the part that should be boring.

Building systems that keep you in the loop

The question is not how much of your writing you can automate. That is how you end up optimizing the wrong thing.

A better question is where you want to guarantee that a human has to care.

Read also Shooting With Restraint: iPhone, Lightroom, And A Component Mindset.

For me, that means:

• The week’s fragments start in a place that is easy enough to use while walking.
• The system cannot publish anything I have not seen in a form close to final.
• Numbers and claims get a manual pass, even if everything else is scripted.

Everything around that can be as automated as I can make it. I am happy to let the pipeline handle dates, tags, social scheduling, and formatting. It is a logistics pipeline, not a perceptual one.

The temptation with a smooth system is to feed it more and more sand and trust it to make glass. It never will. All it can do is move what you give it, faster and with fewer typos.

The uncomfortable part is still the same: noticing what actually happened, deciding what it means, and being specific enough that “significant progress” turns into “2.3 kg.” No scenario will do that for you.

A publishing system is not a writing system. Treat it like a courier, and it will make your work easier. Treat it like an author, and it will quietly ship a version of you that was never there.