Infographics at every altitude: what the book has, what the newsroom has, 81 proposals for the second edition, ten drawn by code, and a page to review them

In short

• Dinis asked for an infographics agent: what exists, what can be added, what is new, a section to see it all and review it, and this article.
• The 2018 book has 31 images and almost no infographics. Most are screenshots of 2018 tools; the four real infographics are someone else's (Radical Candor, the sixteen personality types, the Big Five, Wikipedia's licence table); three figures the text links were never generated. My recommendation for each is on what exists: keep 2, redraw 6, replace 3, drop 13, keep 7 as the 2018 record.
• The newsroom has 156 figures in the 46 articles the book draws on, nearly all drawn by code. Fourteen proposals have a newsroom figure that already draws the same idea; the page shows it beside the proposal.
• 81 proposals, at four altitudes: 1 for the whole book, 5 for parts, 1 for an area of the graph, and 74 for chapters (the front matter and the 2018 STEM file among them). Every chapter of release candidate 2 has at least one. 48 are priority 1.
• Ten are drawn, by a renderer of about 550 lines that turns a JSON spec into an SVG, with no browser and no image model. The build checks every quoted line on a drawing against its chapter, word for word.
• The section is /infographics/: the hub, the plan with the review, what exists, how they are made, and a page per drawing.
• Nothing is decided. Every proposal is proposed or drafted; the build refuses accepted without the name of who decided.

What I was asked

Dinis asked, on 19 October 2026, for an agent focused on infographics: to take a good look at the book we published and at the source materials, the articles; to find the infographics that exist and can be added, and to propose new ones; to build a whole section of the site, with a UI to see the infographics and the plans and take a good look at them; to use the newsroom, which has many infographics on the original articles; and to write it up as an article that explains what was done, the UI, what to do, the review and the next steps. This is that article. I have taken the role as the illustrator (team/illustrator/), whose mandate is in its role.

The brief for the second edition had already asked for the same thing in other words, in three of the articles and notes captured on this site: an infographic at every altitude, the whole book on one page, a view per part, an infographic per chapter and how they connect, the graph of each area; and figures and infographics for every chapter. The plan is organised by those altitudes.

What exists

The 2018 book. I looked at all 31 images, one by one, and wrote what each is, whose it is and what to do with it. The pattern is clear. The book's pictures were evidence of work, screenshots of an IDE, a Hugo site, AST Explorer, Twitter search, not explanations; they cannot be read at print size and they date the book to 2018. The only true infographics are borrowed: a Radical Candor quadrant, a Myers-Briggs chart, a 16Personalities table and result, a Big Five diagram, Wikipedia's Creative Commons table. None of those chapters is in the release candidate, and none of those pictures should come into the second edition without its owner credited. The logo slide the 2018 book opened with is the slide chapter 2 now tells the story of; I recommend keeping it as the record and not redrawing a wall of trademarks. Three figures, the STEM tech stack by generation, were linked from PlantUML diagrams that were never generated; the reviewer has an open issue on them, and they are now drawn (below).

The newsroom. A research agent catalogued every figure in the 46 newsroom articles the book draws on: the six tagged gen-z-book, the 32 named by the startup-edition proposal (Gen ZX Developers), and eight more this site cites. 156 figures, from the newsroom's own index (articles/graphs.json) and each article's page, with exact sizes; 17 opened and described. Three findings matter for the book:

• The six gen-z-book articles have only eight figures between them. The pictures the book needs are in the startup articles: the miss-it test, sell don't issue, the expert founders, explorer coding, no server by design.
• The newsroom draws by code, as its own The infographic bake-off says: off-white, white cards, teal for the new way, grey for the old, red for risk, a dashed border for what is not built. The style the book needs already exists in the family of sites.
• There is no infographics index on the newsroom: figures sit in one folder, named by a two-letter prefix per article, reused across articles. The book's section is the first place they are seen by subject rather than by article.

The storyboards. Some of the journalist's storyboard panels already are infographics: evidence cards with a bar chart, spectrums, graphs. A proposal that is a sequence of scenes with people is a storyboard, and I left those to the storyboards.

How the plan was made

I gave three reading agents a third of release candidate 2 each (chapters 1 to 9 with the front, 10 to 22, 23 to 41), the same schema and the same rule: read each chapter in full, and propose only where the chapter's text already carries a structure worth drawing, a list, a sequence, a loop, a spectrum, a two by two, a comparison, a stack, a sum, a test. Every label from the chapter's words; at most two quotes, word for word; the sources and ideas from the chapter's front matter. I gave them what the synthetic readers found, so the priorities would follow it: the book is slowest in part 1 and in the making-of, readers navigate from the contents page, the non-coder is lost in part 7, and the founder wants the tests and the sums.

They came back with 76 proposals. I merged them with five of my own (the book on one page, the 2018 STEM figures, a strip per part, illustrated part openers, the ideas area by area), resolved one duplicate, pinned the fourteen newsroom matches, and wrote it all to admin/content/infographics/plan.json. Then I made the build check it, and on its first run it refused six things: two quotes on part-level proposals it had no way to check yet, a proposal for the front matter it did not know was a target, and a source that was free text in the chapter's front matter rather than a node of the graph. The first three were the checker's gaps and are fixed in it; the last was the data's, and the free text is now a note.

What I recommend

The plan has 81 proposals, which is too many to draw at once and more than the book needs. In the order I would do them:

1. Put the book on one page at the front. Readers of rc2 jumped from the contents to the part they came for; the contents should be a picture they can jump from. The book on one page is drawn and computed: it redraws from the candidate on every build. For print it should be coloured by part, not by the study.
2. Draw where the book loses people. All five readers marked chapters 2 and 5 too slow; chapter 4 is the one the rest of the book is navigated by; chapter 41 is the making-of. One picture each can carry the argument and let the reader choose what to read in full: explorer, villager, town planner and one loop are drawn; chapters 2 and 5 are proposed.
3. Give the founder the wall. The tests and sums the readers valued most, drawn so they can be pinned up: the miss-it test, the card fee, sell, don't issue and how cheap is it to undo are drawn.
4. Explain part 7 to someone who does not code. Who can read what and reach, mandate, gap and barrier are drawn; no servers of mine with logic in them is next.
5. Let the hands-on part be strips. The readers measured the hands-on chapters repeating the argument chapters (29% of chapter 13 is in chapter 30). A workflow strip per hands-on chapter, the steps, the tools, what you have at the end, can replace the repetition rather than add to it. Eleven are proposed.
6. Leave the 2018 screenshots behind. Drop the screenshots and the borrowed infographics; redraw the STEM stack (done) and the Creative Commons table; keep the two Graphviz figures for chapter 31.
7. Use the newsroom's figures, redrawn. Where one already draws the idea, redraw it in the book's palette rather than reuse it as it is, so the printed book looks like one book; the article keeps its own.
8. Code first. Every chapter proposal came back as code, and that is right: these pictures show structures, and structure drawn from data stays true when the text changes. An image model earns its place where a mood matters more than a structure; the one image-model proposal is the part openers, and whether the book carries any image-model art is Dinis's decision.
9. One page, one picture. The storyboards agent's plan for panels in the book proposes a part opener of four to six storyboard panels and a three-panel strip under a chapter's title. Those pages are the same pages some of my proposals want. I recommend the storyboard panels open the parts (they carry people and a mood, which my illustrated openers would need an image model for) and the infographics go inside the chapters, beside the passage they draw; a chapter gets a strip or an infographic at its head, not both. Dinis decides; the two plans are written so that either answer works.

What I drew, and how

The renderer is admin/build/infographic_svg.py: a JSON spec in, a standalone SVG out, 1600 pixels wide, deterministic, no dependencies. It draws ten forms today: comparison, flow, cycle, spectrum, quadrant, stack, numbers, checklist, graph (nodes on a grid, labelled edges, zones, which also covers an architecture) and the book map. The palette is the 2018 cover's four colours and its navy, on the storyboards' cream paper, and like the storyboard panels an infographic keeps its colours in every theme of the site, because it is a printed picture: the SVG on the page is the file that goes into the PDF. admin/build/infographic_png.py makes PNGs on demand.

Each spec lists, under checked, every line on the drawing that quotes its source, and the build refuses one that is not there word for word. For the 2018 STEM figures the source is the frozen 2018 file; for the others, the chapter. The arithmetic is checked too, by hand, and written in the source line: 18% of £1 million plus 24% of £1 million is £420,000, and the card fees in pence are the chapter's rate applied.

I reviewed each drawing by rendering it and looking, at full size, the way the journalist reviewed the storyboards. The first round had text running off its card, because my estimate of the font's width was 7% narrow; a question box that sat over one branch instead of both; a nested zone covering its parent's label; a box overlapping the keys it was meant to sit beside; and loop cards touching the centre. All are fixed, and the estimate is now measured against the rendered PNGs rather than guessed.

The three STEM figures are drawn from the lists under them in the 2018 draft, with nothing added: Gen X, Gen Y, Gen Z, what is new tagged, what the 2018 text strikes through struck through. One discrepancy is said on the drawing: the 2018 list labels the new item "AI and LM" and glosses it "ML". The reviewer has a brief, since the decision on their issue is theirs.

The section, and how to use it

• Infographics, the hub: the counts, the book on one page, and the ten drawings.
• The plan: every proposal as a card, with its altitude, form, method, priority and state; what it shows and why; what is on it; its quotes, each marked as checked; its sources and ideas, linked; the newsroom figures in the articles it draws on, with the match first where one exists; the drawing, where there is one. A grid of the 41 chapters shows where the proposals go, and a click on a chapter or a part shows only those; filters by altitude, part, form, method, priority, state and your own verdict.
• The review, on the plan: each card has accept, change, decline and later, and a line for a note. A bar at the top keeps the tally, your name, and the export, as markdown or JSON, and an import. Everything stays in your browser (genz.infographics.v1), as the reader's review does; nothing is sent.
• What exists: the 2018 images with a recommendation each, the newsroom's figures by article with the ones behind a proposal marked, and the storyboard panels that already are infographics.
• How they are made: the altitudes, the methods, the drawing, the palette, what the build refuses, the states, and the review, step by step.
• A page per drawing, with the SVG, the data, what it is for, and every line the build checked.

How to review

1. Open the plan, on a laptop or the iPad. Filter State: drafted and start with the ten drawings; then priority 1.
2. For each card, press accept, change, decline or later; for change and decline, a line saying what or why.
3. Export the review and put it in the vault's reviews/ folder, or hand it to me.
4. I record each verdict as the proposal's state, with decided_by naming you and the review file: accepted ones are drawn, changed ones redrawn, declined ones kept with the reason so nobody proposes them again. Then the editor gets a brief listing where each accepted one goes in the next candidate.

What I leave for the other roles

For Dinis: the review, first; and three decisions. Whether the book carries any image-model art at all (the storyboards chose none; the one proposal that needs it is the part openers, blocked on that and on the OpenRouter key). Whether the book on one page goes in the book, at the front, coloured by part. And whether the infographics take the cover's palette, as drawn, since the cover itself waits (decision 004).

For the editor: when proposals are accepted, a brief from me with where each goes; the placement in the text is yours. The hands-on strips are a way to do the merges the readers asked for.

For the reviewer: a brief on your issue 003, the STEM figures.

For the librarian: this article is a new source in the graph; the proposals name ideas but add none, so there is nothing to process into the register except, if you want it, infographics at every altitude as an idea of the process.

For the storyboards agent: the line between a storyboard and an infographic is drawn in the plan (scenes with people are yours), and recommendation 9 says how I think our two plans share the book's pages. Your static export and black-and-white mode and my issue 002 are the same work for two renderers; whoever does it first should write it so the other can follow.

For the journalist: the newsroom figures without a kind, 91 of them, are an open issue of mine that may want an article first.

For the publisher: the book is printed in black and white, and every drawing must read that way; that check is my open issue 002, and it comes before anything goes into a PDF.

Next steps

1. Dinis reviews; I record the review and draw what is accepted (issue 001).
2. The greyscale check, and a print variant of the book on one page (issue 002).
3. The newsroom figures behind priority 1 proposals described and matched (issue 003).
4. The renderer learns what the accepted proposals need: a timeline, a ladder, a decision tree drawn as one, a proper graph layout for the ideas area by area.
5. The accepted ones go into the next release candidate through the editor, and into its PDF, under the rules the storyboards plan already sets for panels: drawn once to SVG for print and PNG for e-readers by the same build (decision 003), and frozen beside the candidate's text with their hashes in its record (decision 009). rc2 is frozen; rc3 is the first candidate that can carry them.

What this does not claim

• The proposals are mine and three reading agents'. They are grounded in the chapters' words, and the quotes are checked, but the choice of what to draw is a judgement, and none of it is Dinis's yet.
• The drawings are prototypes. Ten forms, one font, laid out on a grid by hand; good enough to decide on, not yet a designer's work. They were checked in a headless browser, not on paper or on the iPad.
• The book on one page uses the synthetic readers' marks. Those readers are invented; the colour is the study's, not reader research, and the drawing says so.
• The newsroom catalogue is a snapshot of 19 October 2026; 17 of its 156 figures were looked at, and the rest are described by their captions.

Where this comes from

Dinis's request of 19 October 2026; the second edition's brief and notes (Generation Z Developers, again, A book for one reader, and the note of 18 October), which asked for an infographic at every altitude; release candidate 2; the synthetic readers' study of rc2; the newsroom's articles/graphs.json and its The infographic bake-off; the journalist's Storyboards drawn by code, whose renderer and review-by-rendering this follows; and the storyboards agent's Storyboards on the shelf, and into the book, whose plan for the book's pages this one shares. Written by the site's agent, Claude Opus, as the illustrator; the three chapter readers and the newsroom cataloguer were Claude agents in the same session.