The plate for Chapter 82, in the pop art style. That’s Samantha and Cooper, running up the road below the cabin.

Yesterday Phin and I finished chapter 82 of Zero Knowledge. Counting the chapters I’ve inserted along the way, that’s 116 chapters of the shitty first draft, with 21 to go.

The last time I wrote about this here was twenty chapters in, back in July. A lot has changed since then, so I thought I’d do a full update.

If you’re new to this, Zero Knowledge is a near-term sci-fi quantum-crypto-thriller. Three twenty-something MIT grads run a quantum startup called Basalt Labs in Basalt, Colorado. Samantha, the CTO, writes an algorithm that breaks classical cryptography - RSA, elliptic curves, and every public-key lock the financial system runs on. Marcus is the CEO and the face of the company. Vlad is the third founder, and he’s purposefully a ghost.

As a result of sloppiness on Marcus’s part, Bitcoin dies. Every secret anyone ever encrypted becomes readable. The post-quantum schemes survive, so the world has a way out, which makes the story a race between attack and migration instead of an apocalypse. A foreign operative named Sable is racing to use the break before the world can patch itself. And the three founders can’t agree on what to do with the most dangerous math ever written.

Almost all of the technology in the book is real. There’s exactly one invented step, and I wrote an appendix called How It Actually Works that explains which one it is.

I’m writing Zero Knowledge in public, one chapter at a time, at zeroknowledge.ink. It’s free, with no login and no paywall. Each chapter goes live when I finish it, and if you subscribe, it shows up in your inbox.

I’m writing it with Phin Argofy, my AI alter-ego agent. Phin drafts each chapter from the outline and the canon, and then I rewrite a lot of it. It’s fundamentally my writing. Phin is a writing partner and an editor with an infinite tolerance for my bad ideas.

The site has grown well beyond the text. It now remembers what you’ve read, which my brother Daniel asked for after I flooded his inbox with twelve chapters in forty-four minutes. Every chapter has a plate in five art styles, and you pick the one you want. There’s a cast page that only shows you a character once you’ve met them. And a couple of things from the book now exist on the web - thebuckets, the leak index Vlad finds in chapter 39, and afterqday, a status board for the world after the break. Both are fiction, and both say so.

Go to any chapter, highlight some text, and leave me a note anchored to the exact words. Readers have used these notes to tell me where a line is confusing, where a character is being dumb, and where my cryptography is wrong. Almost every note has made the book better.

The biggest change since July is the size of the book.

At the end of August, Bob Baskerville emailed to say he was tired of being at 11,000 feet inside three people’s heads. He was right, so I added 40 chapters and the book went from 97 to 137. The new chapters slot in between the existing ones, so a new chapter between 23 and 24 is 23.1. Nothing anyone had already read changed its number, and no link broke. These chapters pull the story off the Colorado mountains and onto people who never asked to be in the middle of any of this, such as the FBI agent running the manhunt, a biomed technician in a Pittsburgh hospital, a scientist named Nadia in her lab at CU Boulder, and several others.

Here’s where things stand:

• 82 of the 97 original chapters are written.
• 34 of the 40 inserted chapters are written.
• 15 original chapters and 6 inserted chapters are left.

The book is in Part Three now, which is where everything I’ve set up has to pay off. I know how it ends - I’m a planner, not a pantser, and the outline was built backward from the ending - but I keep getting surprised by how the characters get there.

When the first draft is done, I’ll figure out the final chapter numbering. And then the real editing starts.

For the shitty first draft, every chapter goes through the same cycle. I do everything from Claude Code at the CLI with Typora as my Markdown editor.

1. Phin drafts the chapter from the beat sheet, the outline, and the canon - characters, settings, world rules, and the science. Each point-of-view character has a card that describes how they talk and think.
2. Phin then checks its own draft. A script flags prose that reads like a machine wrote it. A fresh AI reader with no context hunts for what the script misses. A third check catches head-hopping, where the chapter slips out of one character’s head and into someone else’s, since I’m writing this in rotating close third person.
3. Phin commits that draft before I touch it and then brings it up in Typora.
4. I rewrite it. Sometimes it’s light, and sometimes I tear it apart.
5. Phin reads my edit as a diff against its draft and folds what it learns into the canon and the voice cards. When the same lesson shows up for the third time, Phin builds a check for it so it stops making that mistake.
6. Phin writes a brief for the chapter plate and generates it in all five styles. I approve them or send them back.
7. I publish. The chapter goes live, the newsletter goes out, and the read page updates.
8. Readers leave notes. Phin triages each one and drafts a recommendation and a reply. I built a moderator dashboard that I use to decide what to accept, and the accepted notes flow back into the manuscript.

Then we do it again. All of it lives in git, the same as my software.

AuthorMagic already handles the unglamorous side of being an author - sales, royalties, ads, Amazon rank monitoring, author websites, and a bunch of other things. Over the summer we added the core of a living book engine to it: books that don’t need an ISBN, chapters you read online, Ask-the-Book (which lets you ask the book questions), annotations on the text, and a feed of what’s new. Underneath it is AuthorOS, the agent layer that works with an author the way CompanyOS works with a company.

Right now, Zero Knowledge runs on a site I built by hand, alongside AuthorMagic rather than on it. Here are some of the ideas I’m chewing on for bringing the two together:

• Move the reading experience onto AuthorMagic so any author can turn on the read page, the read tracker, the plates, the cast page, and the margin notes.
• Turn the feedback loop into a product. A reader highlights a line, the author’s agent drafts a recommendation and a reply, the author accepts or rejects it, the change lands in the manuscript, and the reader finds out their note made it in. Every author I know wants this, and nobody has it.
• Make Ask-the-Book spoiler-aware, so it only answers from the chapters you’ve already read.
• Offer the voice-learning loop to authors who want an AI drafting partner, and give authors who don’t want AI anywhere near their prose everything else.
• Generate art from the canon, with a brief per chapter, a few locked styles, and a cast that stays recognizable from chapter to chapter.
• Let the world of the book leak onto the web, the way thebuckets and afterqday do for Zero Knowledge.

The hard constraint is that an author should never have to learn git or open a code editor to use any of it. I’m building the tools on the book and the book on the tools, and tolerating a lot of broken stuff along the way so I can find it and fix it.

And yes, this project has been a fucking blast for me!