Essays • Complexity and systems • • 13 min read

Topics First

My blog's menu was a list of formats, not subjects. How I rebuilt it around six topics in one late conversation with an agent, my prompts included, and measured the result on the pages a reader gets.

A dim library with six clear aisles, each lit by one warm lamp at its entrance

The menu said INQUIRY. Behind it, a row of cards. Essays, The Library, Journal, Field Notes, Thoughts. It was late and I was on that page looking at those cards, and something about them was wrong.

They are formats. They say how long a piece is and how it got written. This site started with writing, and essays were the first thing on it, so the menu grew out of that. But what I come looking for on a site is a subject. AI and engineering, or the decans, or a book, or the films. The site was organized around how I make things. A reader comes in some other way.

I did what I do now when a thought like that arrives. I typed it to the agent as it came, spelling included.

one thing i just noticed is there isnt an easy way to find corenrstone and other page types. i was here and saw different type categories and wondering is there a better layout for the blog than how we have just all the article under categories. shouldnt we be leading with our more powerful hubs and cornerstone and funnels and do this more smartly, think about human UI/UX

Two screenshots of the page went with it. What came back was a proposal, written up as a decision record, and the first thing it showed me was that the site already knew its topics.

Several maps of the same place

It knew them in several places, and the places did not agree.

Where What it called them
Start Here five territories
The homepage four ways in, cut differently
The link graph seven clusters, from a rule over categories and tags
A tag file sixteen tags pointing at hub pages

None of them was wrong. I made each one at a different time for a different reason, and I had written or approved every one. I just had never looked at them side by side, which is odd for someone who had already written up how to map a domain and never pointed the method at his own front door.

And the best pieces had no home in any of them. Some essays I mark as cornerstones, the ones that frame a whole subject, the ones I would hand you first. On the essays page a cornerstone looked like every other card, sorted by date. There was no way to ask the site for the best of anything.

The numbers pointed the same way. The hubs and the named ideas drew readers and search. The format listings mostly did not.

Page Kind Views, 90 days Search, July
Projects hub 137 15
Complexence named idea 56 8
Films hub 48 24
What Is People of the Stars named idea 42 53
Situational Governance named idea 38 9
AI Development hub 11 23
Coherent Complexity named idea 12 23
Inquiry format index 47 7
Essays format listing 20 12
Journal format listing 10 3
Thoughts format listing 4 0

Page views over ninety days and search impressions for July, as of the night of the rebuild. Small numbers, and the view counter undercounts, so a direction, not a proof.

One file

The fix started as a file. Each topic in it has a name, a hub page, its cornerstones in the order I would read them, its series, and where it leads, a book, an album, a tool.

Topic Hub Start with
AI and engineering AI Development How AI Is Changing Software Engineering
Complexity and systems a new hub Making Complexity Visible
Time and the decans The decan journal What Is Decanal Journaling?
Books The Library its "Where to start" shelf
Making things Projects Why I Build Creative Technology
Philosophy and the notebooks a new hub Living with Antifragility

Everything that shows a reader a topic now reads that one file. The menu, a new topics page, the homepage, Start Here, the hubs, the footer, every post. Which topic a post belongs to comes from the same rule the link graph already used, so what you see and what I measure cannot drift apart. The cornerstones were already flagged once in each post's front matter, so that flag became the only definition.

How it got built

The proposal had three phases. I said yes to each and looked at the result before the next one. That was my whole part, apart from the noticing. The agent wrote the file, the pages, the tests and the deploys. I read what it put on the screen and typed back what I saw.

What I typed What it turned into
yes absolutely every post names its topic and points up to its cornerstone; the essays page opens with the cornerstones
yes push and go for phase 2 now one hub layout, and two new hubs for the topics that had none
yes and yes the menu, the topics page, the homepage doors
pull this from my freewriting: Philosophy hub promise: it reuses a single line from Start Here. A paragraph of your own would make that hub stronger. the philosophy hub opens with a sentence from my own notebook
sorry laptop ran out of battery nothing lost; it picked up where it was
anything to do here otherwise - have tests been run and checks made on the new site controls and layout? seo checks also? the checks, two gaps found and fixed

The fourth line is the one I care about. After each phase the agent handed back a short review of its own work, what held up and what was approximate, and one of its notes was that the new philosophy hub opened on a line borrowed from Start Here, and that a paragraph of my own would make it stronger. I did not write a new paragraph. I asked it to go into my freewriting and find one. The hub now opens with a sentence from a June entry about trying to breathe and make nice scribbles while a thought arrives to ruin the experience, quoted the way I wrote it. I like that kind of borrowing. The agent built the room and I put something of mine on the wall.

The laptop line is in there because it happened. The battery died in the middle of a deploy, and when the machine came back the work was where I had left it. The whole sitting felt like that, a conversation with pauses in it.

What a reader gets

The menu is Start Here, Topics, Works, About. Topics opens a panel of them, each with the piece to start with. On a phone it goes to a page that does the same. The formats did not disappear. They moved down to a browse-by-format line in the panel and the footer, which is where a format belongs. A way to browse, not a front door.

Every hub has the same shape, so you learn it once. The promise of the topic. Start with these, the cornerstones, numbered. The series. Where it leads. Then everything in the topic, grouped by format. Two topics had no hub at all, so they got one, Complexity and systems and Philosophy and the notebooks.

Every post points up to its topic, named beside the category, and at the end it hands you that topic's cornerstone and a way into everything else in it. Arrive on an old field note from a search and there is now a road from there to the best of what it belongs to.

The essays page opens with the cornerstones, grouped by topic, before the long grid by date, and each cornerstone card wears a mark.

Measured on what a reader gets

I track which posts are linked from hubs and cornerstones, because those links tell a reader, and a search engine, which pieces I stand behind. The first measurement used the link graph behind the live map of this blog, the one you can walk in the Atlas, and it was wrong. It counted every link, including the related-posts block generated on every page, so it called posts curated when only a machine had linked them.

So I measured the built pages instead, the HTML a crawler actually receives once the navigation and the generated blocks are stripped out. A post counts as curated when a hub or a cornerstone links to it in its own words. Listing pages do not count. They link everything.

The field notes went from almost none hanging from anything a person chose to every one of them, and the journal the same.

Curated before After
Field notes 3 of 60 60 of 60
Journal 10 of 31 31 of 31
Essays 53 of 63 59 of 63

Counted on the built pages the day the rebuild shipped, in early October.

The deploy runs that measurement now and warns about any new post that hangs from nothing.

Books is the exception, on purpose. The library lists every review on its shelves, which does not count, so it got a short "Where to start" at the top, the reviews readers and search find most, picked from the numbers, not from my taste.

Whose numbers these are

The agent made that table. Later the same day, reading drafts it had written for me, I typed this.

one thing all the numbers, 28 essays and this many here and this and that. that is what agent mode helps me with but that is not me. im terrible with dates, counts and memorizing in general of details like these. they are handy but im not sure it feels authentic in writing to lead with these. and it might be better to use visuals and tables and diagrams which i really enjoy for that and to keep the writing more in my style which is more prose like.

So the counts live in tables now, here and everywhere I write. The agent keeps them. I keep the prose. It is a small rule and it has changed how every piece on this site gets made, including this one.

And even with all the SEO work and the analytics, I still prefer to make this about the content, my artistic expression and my mind's outlet. The measuring is there so the writing can be found. It is not what the site is for.

What I left alone

No address changed. Every post, category page and listing still answers where it always did, and the old format index stays on as the place to browse by format. The homepage did not grow. I had cut it back once already, and the topic doors replaced the old cards instead of adding a section.

The topic names, the hubs, the order of the cornerstones: judgment calls, and some of them are wrong. Fine. They are wrong in one file now, where I can see them, instead of scattered across the site.

Still moving

The decision record has four phases and the fourth is still open. It says measure. The first three went out in one sitting and the fourth takes months, because readers have to find the doors before the doors can tell me anything.

One thing I did not plan on. When I looked at which of my pages a search engine had kept over the autumn, it was the hubs and the named ideas, the pages that everything now points up to. The dated posts under them mostly fell out of view. I do not know what that proves yet, but the pages built to be a front door are the ones that stayed findable from outside.

A second reader from outside saw the other side of it. Every page now carries the topics panel as text, and to a crawler that is the same block on every page, with the risk that a short entry looks like mostly menu. So I measured it on the built pages.

Page kind The piece itself, as a share of the page's words
A freewriting entry 77%
An essay 87%

Enough of the page is still the writing. I will keep watching the short ones.

Things have moved since that night. The decan topic's hub was the Decan Log page, and a few weeks later it became the journal itself, because that is where a reader of the decans actually wants to land. The feeds still go by format, essays and journals and reviews, while the menu goes by topic, so a reader who wants only AI and engineering in a feed reader cannot have it yet. The hubs and the homepage grew a Look around block, a different handful each visit, drawn from what readers have found and from what nobody has seen yet. The attachment count has already slipped, since newer field notes came in faster than anyone chose to link them, and the measurement is what told me. And the same afternoon, still in the same conversation, I asked for a role that would own a post from the idea to a month after it is live, with the checks, the deploy, the sitemap, the atlas and the dashboards all part of the same motion. That role is how this post got here.

This is the legibility work again, pointed at my own front door. Is it legible yet? Ask someone who has never been here. Anything said is said by an observer, and I am the wrong observer for my own menu.

Related reading: Making Complexity Visible, and Every Check Passed for the measurement that caught the link graph counting the wrong thing.