Five synthetic readers of release candidate 2
Five invented readers, two from Generation Z, two from Generation X and one publisher, read release candidate 2 from the first page the way each would, said where they would put it down and whether they would buy it there, then finished it as beta readers and were interviewed. Nobody would buy it as it stands. Four of the five would buy it, or commission it, once the chapters that form the company are written; the fifth, the sceptic, would not buy it in any form he was offered, and still rated it 3 out of 5.
5ed445042a0b5499…, read on 2026-10-19.The protocol · every finding · the interviews, question by question · the brief
Will they buy it?
Not this candidate. Where they put it down, Aisha said no, Ellie and Nadia said maybe, Martin said no, and Catherine, the publisher, said she would commission it with conditions. After reading the whole book: four maybes and one no. The ratings are 2, 3, 3, 3 and 3.
One reason, from all five. The book promises on its first pages that by the end you will have started a company, and the two chapters that say how, The company itself (22) and Form your company (37), are placeholders. Every reader who came for the company went looking for those chapters; Nadia put the book down at 37 and Aisha just after it, and Ellie, who came to build, put it down at the weekend chapter that follows. Aisha, Ellie and Nadia all went to Start the company or Hands on out of order, from the contents page; nobody but Nadia read the book in order past part 2. This gap is already known: it waits on Dinis's memos (the UK, Lithuania, Portugal, the EU). The study's contribution is to show that it is not one gap among six but the one that decides the sale.
Would they see the value? Yes, in specific chapters, and the same ones come up again and again: Everyone who can brief an agent is a developer (8), Explorer coding (9), Your first workspace (13), the miss-it test and the card-fee arithmetic in Ship it and Price it (18, 19), Fund it, or do not (21) and the reach, mandate and gap model in Controlling your agent (25). All five would send something on to somebody, and every one of them named chapters, not the book. That is the shape of the problem: a good book of about 50,000 words inside a 75,000-word manuscript.
What stands between them and the purchase, after the company chapters: the hands-on part repeats the argument chapters it follows (measured: 29% of chapter 13 is in chapter 30, 15% of chapter 19 in chapter 39); there is a long road to the first thing to do (8,751 words of part 1 after chapter 1, and the first afternoon of work about 26,000 words in); the worked examples are usually the author's own products; the technical chapters lose the non-coder the frame says is a developer; and the co-writing with Claude is said only in the last chapter.
The pacing, from the readers' own marks (the grid below): all five marked chapters 2 and 5 and the making-of (41) too slow, and four marked 39, the repeat of 19. The highest interest is in chapters 1 and 15 (4.0), then 8, 13, 17, 18 and 38 (3.8); the lowest in the two placeholders (22 at 1.4, 37 at 1.2), the storyboards chapter (34) and the long game (part 8, 2.3 on average). By part, Direction and judgement, Working with agents and The frame hold readers best (3.4 to 3.6); Before you start (2.8), Getting the job (2.6) and Hands on (2.7) least. The book is slowest exactly where a reader decides whether to keep going.
The publisher's verdict: commission, with conditions. The hook (two generations, start a company, Gen ZX) and the expert founder as the reader are worth it. The conditions: the company chapters written; one lead reader; 20 to 25% cut by merging duplicates and moving the process to an appendix; the dated tables on a web page; the products disclosed as case studies; a legal read of the funding and tax passages; and the free first edition and free online edition settled in the contract.
The five readers
| Reader (invented) | Gen | Arrives with | Put it down at | Buy, there | At the end | Rating | Questions | Confusions | Findings |
|---|---|---|---|---|---|---|---|---|---|
| Aisha Bello Final-year computer science student in Manchester | Z | Should I take the job or start the company, and does this book tell me how to actually do the second? | Ch. 39: Your first price and your first ten customers 12,657 of 75,643 words | no | maybe | 2/5 | 34 | 34 | 8 |
| Ellie Marsh Account manager at a small marketing agency in Bristol | Z | Can someone like me, who has never coded, actually build and sell this, and will this book show me how without making me feel stupid? | Ch. 40: Ship a serverless product in a weekend 25,620 of 75,643 words | maybe | maybe | 3/5 | 37 | 35 | 8 |
| Nadia Rahman Twenty-two years in insurance claims operations in Leeds, now head of claims operations at a mid-sized insurer | X | At 51, with no coding, can I really build the first version of this myself, and what would it take to leave my job and run it as a company? | Ch. 37: Form your company 35,331 of 75,643 words | maybe | maybe | 3/5 | 28 | 28 | 8 |
| Martin Kowalski Former engineering director at a fintech that was acquired; eight months between roles, doing a little consulting, deciding whether to start something | X | Is this anything I have not read already, and is it actually a person's thinking or AI slop with a famous-ish name on it? | Ch. 41: How this book was made, and what you made 17,695 of 75,643 words | no | no | 3/5 | 30 | 22 | 7 |
| Catherine Doyle Publisher at Fernhill Press (invented), an independent UK publisher of business and technology books, about 30 titles a year, sold through bookshops, Amazon and corporate bulk orders | X · publisher | Is there a book here I can sell, to whom, and would I put the Fernhill name on it as it stands or after what changes? | Ch. 41: How this book was made, and what you made 19,876 of 75,643 words | yes with conditions | yes with conditions | 3/5 | 27 | 18 | 8 |
What we expected, and what happened
| Reader | Expected, before the run | What happened |
|---|---|---|
| Aisha (Gen Z, on a phone) | Puts it down in part 1 or 2; no | Skimmed to chapter 2, then jumped from the contents to Start the company and put it down at 39, a repeat of 19, after finding 22 and 37 empty. No, then maybe. The jump was not expected, and it is the pattern of the study. |
| Ellie (Gen Z, never coded) | Puts it down in the frame or the build part; maybe | Read 8 and 13 (interest 5 each), then put it down at 40, the weekend chapter, on its jargon. Maybe. As expected on the jargon; she got further than expected because chapter 8 held her. |
| Nadia (Gen X, the expert founder) | No put-down, or at 22; yes if the company chapters were filled | Read in order to 22, skipped the job chapters, put it down at 37 after finding 22 empty too. Maybe, the paperback at £19.99 once written. As expected; not expected: her first question was money, how a person with a mortgage survives leaving a salary, which the book never mentions. |
| Martin (Gen X, the sceptic) | Puts it down mid build part; no | Hunted (1, 2, 17, 24, 25, 16, 21, 41); no. As expected on repetition and products; not expected: he rated it 3, credits the honesty, and names 21 and 25 as ideas he would use tomorrow. |
| Catherine (publisher) | Commission with conditions | Commission with conditions, as expected, and the conditions match. Not expected: the late co-writing disclosure, the image rights and named people to clear, and her title advice, to drop Generation Z Developers from the front. |
What would make it better, by what it costs
Marked for Dinis: a decision only Dinis can take, or material only he can supply. The rest is the editor’s to propose to the board.
- loses the sale for Dinis Write chapters 22 and 37, at least for the UK: the registration, the bank account, what it costs, in steps, from Dinis's own experience; or, until they are written, take the promise of a registered company off the first pages.
All five name it, and it is the reason the three readers who came for the company did not buy. It waits on Dinis's memos: the gap articles of the Gen ZX brief. From aisha-1, catherine-1, catherine-2, ellie-1, martin-1, nadia-1. - loses the sale for Dinis Answer the person leaving a salary: how long to keep the job, what savings to put in, what the first three months cost, using the expert founders the book already names.
The frame's primary reader asked this first, and the book never mentions runway, salary, savings or a mortgage. One reader, but the one the frame is written for. From nadia-2. - loses the reader Merge the hands-on chapters into the argument chapters they repeat, or cut the repetition from one side: 13 and 30, 19 and 39, 7 with 32 and 33, 12 and 36, 16 and 35. Put the exercise beside the idea it practises.
Five of five noticed, and two stopped trusting the length because of it. Measured: up to 29% of a chapter is restated word for word. From aisha-2, catherine-3, ellie-5, martin-2, nadia-6. - loses the reader Get the reader doing something within fifty pages: open chapter 1 with the promise and the first step, shrink chapters 2 to 5 (8,751 words) to one or two, and bring the first workspace afternoon (13) forward, to just after 8.
Aisha, Ellie and Nadia skimmed or skipped most of part 1; Aisha and Nadia name it as where the book was slowest. From aisha-3, aisha-6, aisha-7, ellie-7, nadia-3. - loses the reader Give the non-coder a path through part 7 and the weekend chapter: define each term at first use (serverless, Worker, R2, Durable Object, repository, fork, vault), and a no-code route to a page a stranger can use in chapter 40.
Chapter 8 tells the non-coder she is a developer, and she believed it (interest 5); chapter 40 is where she put the book down. From ellie-2, ellie-3, nadia-4. - costs trust Use at least one worked example per founder chapter that is not the author's product (an insurance tool, a clinic, a dog groomer's diary), and label the product chapters (24, 25) as case studies of the author's own companies.
Five of five raised it. The check says 'always' is too strong but 'usually' holds; the disclosure helps, and readers still could not tell advice from a pitch. From aisha-4, aisha-5, catherine-4, ellie-4, martin-3, nadia-5. - costs trust for Dinis Say that the book is co-written with Claude at the front, not only in chapter 41, and move how it was made (the about note and chapter 41) to one short appendix, keeping the measure and the invitation to tell us about your company.
The brief behind the book says to meet the suspicion of AI slop head on; four of five readers said it reads mostly as a person and named the same machine marks (the repeated openings, the near-copied chapters), and the publisher says the disclosure comes 40 chapters late. From catherine-5, martin-6. - worth fixing Make a reader's edition of each candidate without the For Dinis notes, for the next study and for any reader outside the team, and replace the opening of chapter 2, which rests on an essay the author could not read.
The notes are right in a candidate (decision 009), and every reader still read through 47 of them; the paywalled essay was the first claim two readers did not believe. From catherine-6, ellie-6, martin-7, nadia-7. - costs trust In Fund it, or do not, add a section for the founder with no revenue yet, and let the worked example be one of several.
The worked example is a £10 million company; the expert founder at the start has no use for it yet, and the sceptic found the close more confident than the evidence. From martin-4, nadia-8. - worth fixing for Dinis Decide the length: the publisher and the sceptic would cut towards 50,000 to 60,000 words; decision 012 says 200 pages or more. Merging the duplicates (r3) alone removes several thousand words without losing an idea.
The first edition was liked for being short and punchy, which the brief behind this edition says it must be again. From martin-2, catherine-3. - worth fixing for Dinis Decide whether the book leads with one reader. The publisher would lead with the expert founder and move the job chapters (10 to 12) to a companion; Nadia skipped them; Aisha, the Gen Z reader, skipped them too and went to the company part.
Decision 015 makes it a book for two generations; the readers suggest the two meet in the company part, not in the job part. From the interviews. - worth fixing for Dinis Move the dated tables (programmes, credits, prices) to a web page refreshed each year, with the book pointing at it; keep the argument in print.
The publisher's condition. Decision 013 already dates the hands-on part; this is about where the dated part lives. From catherine-7.
What is strong
- The promise and the reader: every one of the five understood in one sentence what the book is for, and the publisher calls the hook (two generations, start a company, Gen ZX) worth commissioning.
- Chapter 8, Everyone who can brief an agent is a developer: interest 5 from the non-coder and the expert founder, who were not expecting to be included.
- Chapters 18 and 19: the miss-it test and the card-fee arithmetic were the most quoted material in the study (Aisha, Ellie, Nadia, Catherine).
- Chapter 9, explorer coding and the handover; chapter 13, the first workspace; chapter 15, one-way and two-way doors; chapter 21, selling shares and lending back; chapter 25, reach, mandate and gap.
- Honesty: two readers believed the book because it shows its failures (the thirty customers corrected to ten, what the agent corrected in chapter 6, the books that stalled).
- The voice: four of five said it reads mostly as a person, and the fifth, the publisher, said the voice is a person's and the build method is machine-shaped.
The book, unit by unit
Each column is a reader; each cell is what they did with that unit in the natural phase (or, faded, in the asked phase), and the number is their interest out of 5. Where the cells go pale is where the book loses people. read skim skip put down asked not opened
| Unit | Words | Aisha | Ellie | Nadia | Martin | Catherine | Interest | Pace |
|---|---|---|---|---|---|---|---|---|
| About: About this release candidate | 704 | 2 | 2 | 3 | 3 | 3 | 2.6 | 3× too slow 2× right |
| Part 1: Before you start | 67 | 2 | 3 | 3 | 2 | 2 | 2.4 | 4× right 1× too slow |
| Ch. 1: Gen ZX: who this book is for, and what it is for | 1,733 | 3 | 4 | 5 | 4 | 4 | 4.0 | 4× right 1× too slow |
| Ch. 2: You are lucky to be here | 2,136 | 2 | 3 | 2 | 2 | 3 | 2.4 | 5× too slow |
| Ch. 3: Everything is connected: graphs | 2,048 | 2 | 2 | 2 | 3 | 2 | 2.2 | 3× too slow 1× too fast 1× right |
| Ch. 4: Where do you land? | 2,122 | 3 | 4 | 3 | 4 | 3 | 3.4 | 3× right 2× too slow |
| Ch. 5: Connect the dots backwards | 2,445 | 2 | 2 | 2 | 2 | 3 | 2.2 | 5× too slow |
| Part 2: The frame | 57 | 2 | 3 | 4 | 2 | 2 | 2.6 | 5× right |
| Ch. 6: What generative AI is | 1,941 | 2 | 3 | 4 | 3 | 3 | 3.0 | 3× right 1× too slow 1× too fast |
| Ch. 7: Lenses, roles and loops | 1,945 | 2 | 3 | 4 | 3 | 4 | 3.2 | 4× right 1× too slow |
| Ch. 8: Everyone who can brief an agent is a developer | 1,735 | 2 | 5 | 5 | 3 | 4 | 3.8 | 4× right 1× too slow |
| Ch. 9: Explorer coding, not vibe coding, and the handover | 2,203 | 3 | 3 | 4 | 4 | 4 | 3.6 | 4× right 1× too slow |
| Part 3: Getting the job, or making it | 86 | 2 | 2 | 2 | 1 | 3 | 2.0 | 5× right |
| Ch. 10: Hired for the workflow | 2,179 | 3 | 3 | 2 | 2 | 3 | 2.6 | 4× right 1× too slow |
| Ch. 11: Act as the job you want | 2,203 | 3 | 3 | 2 | 3 | 2.6 | 5× right | |
| Ch. 12: Publish, publish, publish: your own newsroom | 2,206 | 2 | 3 | 3 | 3 | 2.6 | 4× right 1× too slow | |
| Part 4: Working with agents | 57 | 2 | 3 | 4 | 2 | 2 | 2.6 | 5× right |
| Ch. 13: Your first workspace | 2,106 | 3 | 5 | 5 | 3 | 3 | 3.8 | 4× right 1× too slow |
| Ch. 14: How to work with agents | 2,203 | 3 | 3 | 4 | 3 | 3.2 | 4× right 1× too slow | |
| Part 5: Direction and judgement | 59 | 2 | 3 | 2 | 3 | 2.6 | 5× right | |
| Ch. 15: Direction is the scarce skill | 2,204 | 4 | 4 | 3 | 5 | 4.0 | 5× right | |
| Ch. 16: Facts, evidence and the record | 2,205 | 2 | 3 | 4 | 3 | 3.2 | 3× right 2× too slow | |
| Part 6: Start the company | 75 | 4 | 4 | 5 | 2 | 4 | 3.8 | 5× right |
| Ch. 17: Be a founder: the expert founders | 2,250 | 3 | 4 | 4 | 4 | 4 | 3.8 | 4× right 1× too slow |
| Ch. 18: Ship it, and see if they miss it | 2,160 | 4 | 4 | 4 | 3 | 4 | 3.8 | 5× right |
| Ch. 19: Price it, then give it away | 2,169 | 4 | 4 | 3 | 3 | 4 | 3.6 | 5× right |
| Ch. 20: Sell | 2,194 | 2 | 3 | 4 | 4 | 4 | 3.4 | 4× right 1× too slow |
| Ch. 21: Fund it, or do not | 2,213 | 3 | 3 | 3 | 4 | 4 | 3.4 | 4× right 1× too slow |
| Ch. 22: The company itself: the UK, Portugal, Lithuania, the European Union | 706 | 1 | 1 | 2 | 1 | 2 | 1.4 | 4× too fast 1× right |
| Part 7: Build it without a server | 68 | 2 | 2 | 3 | 2 | 3 | 2.4 | 5× right |
| Ch. 23: Serverless, for real | 2,387 | 3 | 1 | 3 | 4 | 4 | 3.0 | 3× right 2× too fast |
| Ch. 24: Privacy by design: SG/Send, sgit, and the value of constraints | 2,118 | 2 | 1 | 3 | 3 | 3 | 2.4 | 3× too fast 2× too slow |
| Ch. 25: Controlling your agent: RiskMandate | 2,221 | 1 | 3 | 4 | 4 | 4 | 3.2 | 4× right 1× too slow |
| Ch. 26: Open source | 2,208 | 3 | 3 | 3 | 3 | 3 | 3.0 | 5× right |
| Ch. 27: Shipped means the user has it | 2,117 | 3 | 3 | 3 | 4 | 3 | 3.2 | 5× right |
| Part 8: The long game | 36 | 1 | 2 | 2 | 2 | 2 | 1.8 | 4× right 1× too slow |
| Ch. 28: Graphs, memory and your future self | 1,927 | 2 | 2 | 3 | 3 | 2 | 2.4 | 3× too slow 2× right |
| Ch. 29: Civic responsibility | 2,203 | 1 | 2 | 2 | 3 | 3 | 2.2 | 4× right 1× too slow |
| Part 9: Hands on, 2026 | 59 | 4 | 4 | 4 | 3 | 4 | 3.8 | 5× right |
| Ch. 30: An agentic workflow from zero | 1,099 | 4 | 2 | 3 | 1 | 4 | 2.8 | 2× right 2× too slow 1× too fast |
| Ch. 31: Your first graph, built by hand | 1,106 | 2 | 3 | 4 | 3 | 3 | 3.0 | 5× right |
| Ch. 32: One idea, five lenses | 1,032 | 2 | 4 | 3 | 2 | 2 | 2.6 | 3× right 2× too slow |
| Ch. 33: A tiny agent loop | 1,106 | 2 | 4 | 3 | 3 | 3 | 3.0 | 5× right |
| Ch. 34: Images, infographics and storyboards, drawn by code | 1,107 | 2 | 1 | 2 | 2 | 2 | 1.8 | 5× right |
| Ch. 35: Meeting briefs and follow-ups | 850 | 3 | 3 | 3 | 2 | 3 | 2.8 | 3× right 2× too fast |
| Ch. 36: Your own newsroom and inbox | 1,105 | 2 | 3 | 3 | 3 | 3 | 2.8 | 5× right |
| Ch. 37: Form your company | 698 | 1 | 1 | 1 | 1 | 2 | 1.2 | 4× too fast 1× right |
| Ch. 38: The credits and programmes, 2026 | 1,084 | 5 | 2 | 4 | 4 | 4 | 3.8 | 5× right |
| Ch. 39: Your first price and your first ten customers | 1,110 | 1 | 3 | 4 | 2 | 3 | 2.6 | 4× too slow 1× right |
| Ch. 40: Ship a serverless product in a weekend | 1,109 | 4 | 1 | 3 | 3 | 4 | 3.0 | 3× right 2× too fast |
| Part 10: How this book was made, and what you made | 78 | 1 | 2 | 2 | 2 | 2 | 1.8 | 4× right 1× too slow |
| Ch. 41: How this book was made, and what you made | 2,404 | 2 | 2 | 2 | 3 | 3 | 2.4 | 5× too slow |
How it was run, and what went wrong
Run on 19 October 2026, five Claude Sonnet agents in parallel, each given its persona, the protocol and the candidate split into its 52 units with the contents page, and nothing else. The runs took three to five minutes each. All five records passed the build's checks first time or after correcting quotes, and every quote on these pages is word for word in the unit it cites.
What went wrong, kept rather than hidden. Ellie's agent said that in the asked phase it read only the opening and the headings of the chapters she had not opened, not the whole text, so her asked-phase steps are lighter than the protocol says; her natural phase, where her put-down point and decision come from, was read in full. Nadia's agent filled the read and thought fields of skipped steps with "Not opened", because the build asks for them on every step; the next version of the protocol should not ask for them on a skip. Ellie's agent wrote its record with a script, which is allowed; the record is what is checked.
What the method cannot tell. These are invented readers played by one model family, reading a manuscript with notes in it. They agree with each other more than five real people would. The numbers (put-down points, ratings) are the model's, not a market's. What is evidence is the text: the quotes, the measured repetition, the searched absences.
Next time. A reader's edition without the For Dinis notes (r8); the same five on rc3, so a finding that disappears is a fix that worked; one reader outside both generations (a millennial, the reader the book says is not its first) to test what decision 015 leaves out.