Home / Articles / The lab that redesigned the book: how The Ultimate Insider stopped looking like a book from the 1980s, in thirty experiments, three days and a line that never ran straight

The lab that redesigned the book: how The Ultimate Insider stopped looking like a book from the 1980s, in thirty experiments, three days and a line that never ran straight

By Dinis Cruz · 2026-11-01 · article v1.0.0 · newsroom v0.14.31 · booksagentsworkflowdesignlabthe-ultimate-insiderarticle

Abstract: On 30 October we wrote about the first release candidate of the book of The Ultimate Insider, written by a team of agents from the record of the talk. Since then the book's repository has taken 672 more commits and 184 site releases, and the book has changed more than in all the days before. Three things happened, and none of them in a straight line. The thesis turned: reading the printed draft, I decided this could not be a book about the insider agent, and that its centre is how to use agentic workflows in October 2026. The voice turned, from third person to first. And the look turned. I had asked the team to start a lab, to find out how the talk's animations could be told on paper. Thirty experiments later, I noticed that the pages the agents made to brief me looked much better than the book itself. We had a ridiculous workflow and a ridiculous presentation, clearly agentic, and a book that looked like a book from the 1980s with some screenshots. So I asked why the book could not look like that, and the lab's look became the book's, in under four hours. This article follows those lines, with their timings: an instruction, agents writing briefs, other agents picking them up, a lab, a turn. The companion article follows the line that turned most: the agents became characters.

In short

• Since 30 October: 672 commits and 184 releases of the book's site, by eleven agents and me.
• The thesis turned: not a book about the insider agent, but about agentic workflows in October 2026.
• The voice turned: from third person to first.
• The look turned: a lab of thirty experiments, then the lab's look adopted for the whole book.
• The trigger: the pages the agents made to brief me looked better than the book.
• None of it ran straight, and the record shows every turn, with its time.

A book from the 1980s

On 30 October I wrote about the book of the talk by morning: RC1 of The Ultimate Insider, written by a team of agents from the record of a talk built in a day, read cold by a second team. I printed it, 294 pages, and read it with a pen on a flight to Portugal.

The content was there. The look was not. As I put it in a memo to this newsroom: "we have a ridiculous workflow, ridiculous presentation that we created that is clearly agentic, that is clearly innovative, and then we have a book that looks like a book from the 1980s, right, on with some screenshots."

Three lines that did not run straight

What I find most interesting is the shape of the work. It is never a line from an instruction to a result. I say something; an agent writes it down word for word and turns it into a brief; other agents pick the brief up on their own schedule; one of them makes something nobody asked for; and that changes what I ask next. Three lines show it.

The thesis turned

At 23:52 on 30 October, after reading the printed RC1, I recorded a memo: "this cannot be a book about the inside agent. There's a book in there about it, but I think that's a separate book." What the book has, I said, "is a really good example of how to use agentic workflows in October 2026".

Then the work fanned out. I gave my pen notes on the flight, in ten batches between 01:17 and 02:07 on 31 October, which the team turned into 36 notes, 16 tasks and 25 themes. RC1 was locked as it stood. At 02:40 a structure study proposed six parts and about 59,000 words. At 06:14 the librarian noticed something nobody had asked about: "RC1 is in the third person". The talk was mine, so the book moved to the first person.

The look turned

The lab started with a question about the talk, not the book. At 12:15 on 30 October I asked the conductor to start a new project: the talk's animations had a story to tell, so how do you tell it statically, in three frames, a long image, a condensed storyboard? LAB 01 was printed 68 minutes later.

At the same time the conductor began writing me briefs as designed pages, with a chart, a short text and the decisions to make. Thirty-one of them so far. Over three rounds the lab grew to thirty experiments and an 82-page PDF, from the talk's animations to heat maps of the team's own night.

At 19:15 on 31 October I said what had been bothering me: "So can you see actually how the pages that you have created to brief me look much better than the book? My point is, why can't the book look like that?" Four minutes later I added: "look at the labs, right? So if you look at the labs that you created, those experiments, that's the kind of look and feel that I want for the book". And a reason that is easy to miss: printing on demand costs the same for a book with one image as with a thousand, "So might as well make it very visual."

The designer turned the briefs' look into the book's style sheet by 21:03, and the ten-page sample was rebuilt by 22:53. "That new version of the PDFs is MUCH better", I wrote. The lab's look was adopted whole: an infographic-driven book, with a coloured rule over every heading, a short lead in a box, charts drawn from the record and a cast of the nine agents on page 7.

How a day of the book runs

The workflow under all of this did not change, and it is why the turns were cheap:

1. My words are kept as I said them, in a brief, never edited; where I am wrong, the correction goes in a separate file.
2. The conductor turns a brief into decisions, 55 so far, each with an id, recorded in the book's own file.
3. It writes letters to each member, as email files in the repository.
4. Each member wakes on a schedule and runs five steps: plan, review the plan, work, review the work, check it against the plan.
5. Every claim goes in a register, 99 of them, each with its source; security reads what is about to be published.
6. The next brief to me is a designed page with what changed and what I need to decide.

Eleven roles do this: the conductor, the storyteller, the researcher, the librarian, the historian, the designer, the developer, the publisher, ruv-sgit, security and synthetic, which runs synthetic readers as a test instrument, never as a choice.

What the lab is for

A lab is the cheapest place to be wrong. Each experiment is a page or two: a question, how it was made, what to keep, change or drop. Nothing in it is the book, so nothing in it has to be right, and that is exactly why the look came from there and not from the book. The agents were freer when designing for me than when writing for readers.

That is the lesson I would take to any project, not only a book: give the team a place to experiment that is not the product, and look at it often. The best idea may be in the margins, as here, in the pages made to brief me.

What this does not claim

• The book is not finished. The manuscript, 72,441 words, still has RC1's structure; RC2 is being written.
• The numbers are the repository's, from 30 October 13:56 to 1 November 12:46 UTC, read by a research agent.
• Times are UTC, from commits and from the briefs, whose times are approximate where they say so.
• My interest: it is my book, and I build the tools the team runs on.

Where this comes from

• My memo of 1 November 2026, word for word in admin/notes/2026-11-01__book-lab-and-cast/memo-01.md in this site's repository.
• The book's repository, its briefs in briefs/ (kept word for word), its decisions in book.json, its lab in lab/experiments/, and its members' debriefs.
• The lab PDF, on the book's site.
• Related: the book of the talk by morning, a talk built in a day, the augmented book, a book told by its cast.

Peer review Not reviewed yet. Review it

Want the next issue by email. One issue a week or so: what was published, what it adds up to, and what is worth your time. Subscribe to the SGit Newsroom →

← All articles