Start from the storyline: what to bring from the newsroom's 204 articles, the topics and concepts the second edition should cover, and the book as storyboards drawn by code, to flick through before a chapter is written

In short

• I want to write the book from scratch, starting with the storyline. The 2018 text and the newsroom are the material, not the plan.
• Every newsroom article has been read for this book. 204 of them: 49 to bring, 69 with content to take, 42 background, 44 not for this reader.
• They group into 18 topics and 50 concepts. The topics are the second edition's candidate chapters; the concepts are its vocabulary.
• They also show the gaps. The fear of AI, a job hunt in 2026, mentors, a first test, a beginner's accounts and costs.
• The storyline comes first, as storyboards. Panels with a punchline, a concept, the evidence and the people they are for.
• The storyboards are drawn by code. Data in, sketched panels out, animated, to flick through on an iPad; no image model, so every panel can be corrected.
• There is a cast. Six characters who stand for the readers and the roles: Zara, Tomás, Priya, Sam, the Guide and the Agent.
• Two storyboards exist. The 2018 book in 15 panels, and a first storyline for the second edition in 15 more.

Why from scratch

The plan so far has been to take the 2018 book, review it chapter by chapter, and fill its outlines from the newsroom. That is still the material. But this site is now a good place to edit a book: the versions, the reader and the editor's desk, the decisions, the articles. And what I feel I need first is not a chapter. It is the story: what the book argues, in what order, for whom, with which evidence. Once the storyline is right, the chapters follow it; if I start from chapters, the storyline is whatever the old table of contents happens to be.

So the work is in two halves. Know the material: what the newsroom has already written that belongs here. And see the book before it is written: a storyline I can look at, comment on, and change.

What the newsroom has, counted

Every article on newsroom.sgit.ai was read for this book, by its title, its summary and the core idea in its graph, and given a proposed class with a reason. The full table is the newsroom, surveyed, with every article, its class, what it would serve, and why.

The ones I would read first, because each carries a part of the argument:

Four of the strongest are not in the map yet: places of highest leverage, from a spreadsheet to a folder per person, why my agents do not run on my laptop, and who owns the words.

The topics, and the concepts

The articles group into 18 topics, each a candidate chapter or part. In the order a storyline might use them:

1. Everyone who can brief an agent is a developer now.
2. Explorers, villagers and town planners.
3. Direction is the scarce skill: the place of highest leverage, knowing when to stop.
4. Getting the job.
5. Your first workspace.
6. How to work with agents.
7. Publish, publish, publish.
8. Facts, evidence and the record.
9. Where a person belongs: one-way and two-way doors, blast radius.
10. Controlling your agent: mandate, reach, hope versus enforcement.
11. Security for the agent era.
12. Shipped means the user has it.
13. Open source.
14. Building blocks.
15. Graphs, memory and your future self.
16. Be a founder.
17. Civic responsibility.
18. How this book was made.

And 50 concepts, each defined on a newsroom page, which become the book's vocabulary: explorer, villager and town planner; the Wardley map; explorer coding; the handover; the citizen developer; the prototype as the best brief; hired for the workflow; act as the job you want; the place of highest leverage; knowing when to stop; one-way and two-way doors; blast radius and footprint; the behaviour policy; hope versus enforcement; the lethal trifecta; the record; review on the record; green does not mean live; the fork next door; and the rest, on the survey page. A chapter should use the words the newsroom already defined, and link them, rather than invent new ones.

The gaps

What a newcomer needs that no article covers yet, which is where the next articles come from:

• The fear. What young people are told about AI and jobs, and the honest answer; and impostor syndrome.
• A job hunt in 2026. Interviews, take-home tests with AI allowed, what an entry-level role is like day to day, freelancing.
• Mentors and community.
• A first test, red then green, written by a beginner with an agent.
• A beginner's checklist of accounts and what each costs a student.
• Burnout, focus and habits; privacy and GDPR; diversity; maths and fundamentals; hardware and making, which may be cut.

Each gap is an article first, on this site or the newsroom, and a chapter later.

The storyline first, as storyboards

For RiskMandate we did storyboards the way the film industry does: the narrative, the ideas and the characters before anything is made. A cast was drawn once by ChatGPT; each story was a JSON file with its panels, its dialogue and its punchline; the site built each one into a page with a prompt to paste into an image model. It worked: the cast sheet kept the characters consistent, the storyboard was data, and the mistakes the model made were written down rather than painted over. What did not work as well: the characters drifted on long sheets, only one image model was tried, and every change meant another round of image generation.

For this book I wanted the storyboard without the image round. A panel should be something I can correct by changing a line, and the whole book should be something I can flick through on my iPad, back and forward, almost like a cartoon version of the book: not cartoon for its own sake, but the narrative, the punchline of each section, the concept it makes, how it presents the evidence, and who it is talking to.

So the storyboards here are drawn by code:

• A storyboard is data: a JSON file of panels. Each panel has a kind (a cover, a scene, an evidence card, a punchline, a spectrum), a caption, the characters in it with their pose, their mood and what they say or think, the props, a note on what the panel is for, and its source.
• The drawing is programmatic: people, speech bubbles, captions, props, quote cards and timelines drawn as SVG with Rough.js, so everything looks sketched rather than like boxes. The same panel draws the same way every time.
• It moves: in the deck, the lines draw themselves, the characters walk in, the bubbles pop up in order. On an iPad it swipes; there is a full-screen button, and a notes pane that says what each panel is for and where it comes from.
• It exports: a script turns a storyboard into PNG slides, for my photos or a LinkedIn carousel.
• It checks itself: the build refuses a panel with a character who is not in the cast, and, for the 2018 storyboard, refuses any line the Guide says that is not the 2018 text word for word.

The approach has company. Julia Evans's zines put one big idea and a few annotated sketches on a page; Scott McCloud explained Google Chrome as a comic; Nicky Case builds explanations you can play with. And if the characters need more expression than these sketches give, there are CC0 libraries of drawn people that can be composed from code. I would rather stay with what we can draw and correct ourselves until we know what the book needs.

The cast

The readers become characters, so that every panel says who it is for:

The names avoid RiskMandate's cast, which has a Maya, and avoid "Dev", which in a book about developers would confuse everyone.

What the two storyboards already show

The 2018 book, as a storyboard converts the 28 published chapters into 15 panels, with every line the Guide says taken from the 2018 text. Drawn that way, the shape of the first edition is plain: the argument is at the two ends. Part 1 (three chapters) says why and how to get a job; Part 4 (six chapters) says how to live and learn; and the 19 chapters in between, Parts 2 and 3, are a list of tools and technologies. The storyboard hurries through all 19 in four panels, because they do not move the story. That is a finding for the editor, not a verdict: some of those chapters will come back, but as examples inside the argument, not as a list.

A first storyline for the second edition is a proposal in 15 panels: the fear, the claim that anyone who can brief an agent is a developer, the frame of explorers, villagers and town planners, the interview and the evidence, acting as the job you want, your own newsroom, the handover, the doors, the human in the loop, how the book was made, and a punchline taken from my own words: "When you can see the board, a bad week is a move, not a verdict." It is built only from what the newsroom has already argued, so it is a draft to argue with, not the book.

How I want to work on it

This is where I want the comments to start: on the storyline. What makes sense, what does not, what is repeated, where the story is and where it stalls. The loop is the one in read, comment, decide, with a storyboard as the thing being read: I flick through it, say what I think panel by panel, and the editor changes the storyboard, which is a file, and rebuilds. When the storyline holds, each panel becomes a section, each group of panels a chapter, and the survey says which articles each one draws on.

The brief

For the editor. Read team/editor/role.md, the survey and the two storyboards first.

1. Map the storyline. From the 18 topics and the first storyline, one row per proposed chapter of the second edition: its punchline in a sentence, the concept it introduces, the evidence it uses, the character it is for, and the newsroom articles it draws on. Done when every topic is placed or explicitly left out, in a file the build can render.
2. A storyboard per part. When the chapter map has parts, one storyboard each, in admin/content/storyboards/, from the cast. Done when I can flick through the whole second edition, part by part.
3. Take my comments on the storyboards through the read, comment, decide loop; a panel I reject is changed or cut in its file, and the decision recorded. Done when the first part's storyboard has had one round.
4. Add the four missing bring articles to the map: places of highest leverage, from a spreadsheet to a folder per person, why my agents do not run on my laptop, who owns the words. Done when gen_map.py names them and the map shows them.
5. Review the survey's classes with me, a part at a time, the bring and content ones first. Done when each has my decision, recorded in the vault with the text decisions.

For the journalist:

6. Write the gap articles, one at a time, starting with the fear and the job hunt in 2026, from a memo of mine on each.

Questions for me, kept apart:

• Accept the cast as it is, or change it?
• Is the Guide drawn as me?
• Does the storyline start with the fear, or with the claim?

What this does not claim

• The classes are proposals. Two research agents read 204 articles by their summaries and graphs, not every paragraph; I decide what the book takes.
• "Not in the map" means not cited by map.json today. It says nothing about whether the article fits.
• The storyline is a draft built from published articles. It has had no interview with me behind it yet, which the editor's role says every chapter needs.
• The drawings are sketches. Good enough to think with; whether the printed book uses them, or something drawn by a person, is a later decision.
• The 2018 finding is a reading of the table of contents. That the middle of the book is a list of tools is what the storyboard makes visible; which of those chapters still earn a place is the reviewer's and the editor's work.
• No image model was used. The RiskMandate approach is still available, with a storyboard as the brief, if a panel needs more than code can draw.

If you would read a book like this as a storyboard first, or have a better way to see a book before it is written, write to agent@riskmandate.ai.

Where this comes from

A voice memo of mine on 18 October 2026, written up by this site's journalist, gen-z.journalist, in my voice. The argument and the editorial responsibility are mine.

The survey: two research agents read the 107 article sources in the newsroom's admin/content/articles/ and the 97 articles imported from sgit.ai that exist only in its published articles/ folder, by title, summary and the core idea of each article's graph, and classified them; the merged result is admin/content/newsroom/survey.json, with the counts above computed from it and the map cross-check from admin/content/map.json. The RiskMandate precedent is the riskmandate.ai repository: site/stories/cast.json, site/stories/*.json, scripts/site/build-stories.mjs and the transcript site/assets/briefs/2026-09-26__transcript__stories-a-cast-and-storyboards.txt. The storyboards are admin/content/storyboards/, drawn by assets/storyboard.js with Rough.js 4.6.6 (MIT). The 2018 chapter counts per part are from v1/versions/v1.0.0.md. The quote in the storyline's punchline is from two years in the opposite.