Home / Articles / 143 notes in a day: how my pen on a printed release candidate became a record, decisions with their provenance, a review hub and a plan for the next candidate, read by agents

143 notes in a day: how my pen on a printed release candidate became a record, decisions with their provenance, a review hub and a plan for the next candidate, read by agents

By Dinis Cruz · 2026-10-31 · article v1.0.0 · newsroom v0.14.29 · gen-z-bookbooksreviewhandwritingpaperagentsprovenancearticle

Abstract: On 19 October I printed release candidate 2 of my book, Generation ZX Founders, 168 A4 pages, spiral bound, and read the first 19 pages in a café with a pen. I photographed the pages in four batches and sent them to the book's team of agents. By the end of the day my 143 handwritten notes had become a record: each note typed word for word, with where it sits on the page and what it means, and each page drawn three ways, my photograph with numbered callouts, the page redrawn with my marks in place, and a sketch of the page after the change. Where the agents could not read my hand, 11 times, they did not guess; they asked, and I answered by voice. Four of the notes were decisions, and the book's decisions page records them with their provenance: decision 016, a release candidate is a release, sold now, made in pen on page 4; decision 017, it is a new book, not a second edition, made in pen on page 9, where I circled "new book"; decision 018, every candidate has two editions, the release and the review, from a note that says we need two; and the title itself grew in the margins. My pen asked for a lot: the voice above all, 32 notes on it; meeting the reader where they are; drawing what the text only describes; facts marked apart from opinions; and QR codes in the margins. When I asked whether all the pages were being tracked, they were not, so the agent built a review hub where every note has its own address. I wrote about the habit in Reading with a pen again. This is what the agents actually did with the notes, and what it shows: a handwritten review can be as traceable as a pull request.

In short

• 19 pages, 143 handwritten notes, one day: the first part of a review of my book's release candidate.
• Every note became a record: word for word, where it sits, what it means, drawn three ways.
• Uncertainty was asked, not guessed: 11 readings went back to me as questions, all answered.
• Four notes were decisions, recorded with their provenance: made in pen, on page 4.
• The voice led: 32 notes, from "too harsh voice" to "weird, not me".
• A missing view was built on request: a review hub where every note has its own address.
• A handwritten review can be as traceable as a pull request.

A café, a pen and 168 pages

On 19 October I printed release candidate 2 of my book, Generation ZX Founders, 168 A4 pages, spiral bound, and sat in a café with a pen. I read the first 19 pages and marked them as I would any book I care about: underlines, boxes, arrows and notes in the margin. I photographed the pages in four batches, out of order in one case, and sent them to the team of agents that works on the book.

In Reading with a pen again I wrote about the habit: reading has three moments, and the agents now do the third, processing what you marked. That article was written that day, when the review had reached 83 notes. This one is about what the agents did with all 143, from the book site's own four-part record.

Reading the hand

The agents did three things with every page.

• Every note typed word for word, with where it sits and what it means, into a data file for the batch. My misspellings were kept as written. Every drawing is generated from that record, so a correction is one line.
• Every page drawn three ways: my photograph with numbered callouts, so I can check the reading; the page redrawn with my marks in place, for anyone else; and a sketch of the page after the change, for the book's editor agent.
• Uncertainty asked, not guessed. Eleven readings were uncertain across the four batches: one, then four, two and four. Each went back to me as a question, and I answered, mostly by voice. It was "startups", not stand-ups. "We need two" meant two PDFs. In one case I corrected the agent's grouping: a line it had attached to note 124 belonged to note 125.

My handwriting is not easy. That the agents read 143 notes and asked about only 11 is the part I still find remarkable.

What the pen asked for

The themes grew batch by batch, from five on the first three pages to thirteen by page 19. In the record, the largest are:

• The voice is mine, 32 notes. "too harsh voice". "Don't mention my daughters". "weird, not me". "DEMEANING". One was a fact, not a tone: "they didn't, agents are a 2026 thing".
• Meet the reader where they are, 21 notes. "reader has no idea what a Wardley map is (by page 15)". Tell readers on a screen they "should stop" and print it. The licence marked "VERY important".
• Keep the making-of out of the reader's way, 15, and topics first, details under them, 14.
• Show the book, do not list it, 13: "Graph representation of book"; and on the graph chapter, "visualise this" or "show this visually" nine times.
• Fact or opinion, visibly: "we need a visual way to reference to the reader facts or opinion"; "find data to support this".
• The page has room for more: "what about using QR codes here?", and later, a QR code on every page, which "will drive revenue".

And the pen said what works, too: six ticks on one batch alone, "exactly", "these are great :)", and "this needs to be a common theme across the book". On page 9, after eight pages of asking for more: "Finally we tell the reader why they should read this book :)".

When a margin note is a decision

Some notes were not edits. They were decisions about the book, and the book's decisions page records them with where they were made:

• 016: a release candidate is a release, shipped on Leanpub and sold now. Recorded as made in pen, on page 4.
• 017: it is a new book, not a second edition. Made in pen, on page 9, where I circled "new book".
• 018: every release candidate has two editions, the release and the review. From note 51: "But I really like this kind of box, but it means we need two". The review edition gets boxes with ids, so a photographed page can be read back answer by answer.
• 019: the title, Generation ZX Founders (GenAI Edition), which grew in the margins of pages 4, 8 and 10, and was decided by memo.

That is the part I find most useful. A decision made with a pen on paper is usually lost, or remembered as a vague intention. Here it has an id, a date, the page it was made on, and the note that made it.

The page asks for room

Two of the requests changed what a page of the book can be. The fact-or-opinion notes led to a margin system that marks a claim as fact, with its data, or as my opinion. And the QR-code notes led to codes in the margin that work: the agents' sketch has seven, each resolving to the book's site through a short address, so a printed page links to what the paper cannot hold. That idea has since grown into the augmented book.

The view I asked for

Partway through the fourth batch I asked whether all the pages so far were being tracked. They were not, so the agent built one: a review hub with all 19 pages and 143 notes, which can be filtered by theme and chapter, where every note has its own address, so a single note can be linked from a decision, a brief or a reply. The build now refuses a batch with no article, or notes that are not numbered one to the last. A view that was missing became a rule that keeps notes from being lost.

What it shows

A handwritten review is usually the least traceable kind there is: marks on paper that the author reads once and forgets. Here it became as traceable as a pull request: every comment recorded, every reading checked, every decision linked to the note that made it, and every page drawn as it would look after the change. The pen stays the best way I know to think while I read. The agents make what it writes count.

Where it stands

The review stopped at page 19 of 168. The agents planned the changes, chapter 1 paragraph by paragraph and chapters 3 and 4 swapped among them, but there is no third candidate yet: my focus moved to the book of the talk, which is using the same method. When I come back to this book, the record is waiting, note by note.

What this does not claim

• 19 of 168 pages were reviewed; the changes are planned, not shipped, and there is no third candidate.
• The counts are from the book site's record, read on 31 October 2026.
• One hand was read. Eleven uncertain readings in 143 is the only measure of accuracy, and it is mine.
• My interest: it is my book, and the agents work for me.

Where this comes from

• The book site's four-part record of the review, 19 October 2026: Dinis's pen, made readable, pages 4 to 9, pages 10 to 13 and pages 14 to 19; the decisions page; paper that talks both ways; and the review hub. The figures are the book site's own page drawings and photograph.
• My messages of 31 October 2026 asking for this article.
• Related: reading with a pen again, the augmented book, the book of the talk and its companion, Generation Z Developers (AI edition).

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