← All writing

    The agent team that wrote a book — and became a toolkit

    4 min readValto Loikkanen

    Writing Personal Ecosystem did not start with a manuscript. It started with a pile of notes, a decade of thinking, and no realistic way to turn that into a ~50,000-word, illustrated, bilingual nonfiction book by working alone, one prompt at a time. You can already read the introduction or browse the chapter summaries if you want the book itself before the story of how it got made.

    So instead of prompting a general assistant chapter by chapter, we built a team.

    The biggest source was years of ChatGPT conversations

    Before there was any agent team, there was raw material. The single biggest source for the book wasn't a fresh round of interviews or new writing — it was years of my own ChatGPT conversations, going back to shortly after ChatGPT itself launched. Thousands of exchanges where ideas about data ownership, personal AI infrastructure, and what became the PIOS framework got worked out in real time, long before any of it was organized into a book.

    I pulled that entire history out via ChatGPT's own data takeout, and fed it to the agent team as primary source material — the same way you'd hand a biographer your journals. That's also the origin of one of the twelve agent roles: source inventory exists specifically because raw chat exports are enormous, messy, and full of both gold and dead ends, and someone has to go through them first.

    A team, not a prompt

    Twelve specialized AI agent roles, each with a narrow, well-tested brief. One role does nothing but inventory the source material. Two others independently propose a structure for the book — on purpose, so a third role can compare their disagreements rather than trust a single first draft. Separate roles handle developmental editing, line editing, a fresh-eyes read with no memory of earlier drafts, commercial positioning, and illustration.

    None of this happened in one long chat. It was 17 defined workflow specifications, run phase by phase, with checkpoints where a human (Valto) reviewed and decided before the next phase started. The agents did the grind — reading, drafting, comparing, revising. The judgment calls stayed human.

    What the system actually had to produce

    A finished nonfiction book is not just a manuscript. This one needed:

    • A ~50,000-word manuscript, in English and Finnish
    • A full illustrated interior, in two locked, reusable illustration styles kept consistent across dozens of images
    • An EPUB and a print-ready PDF
    • A multi-chapter audiobook, produced from the same manuscript file

    Getting each of these right on its own is a solved problem. Getting all of them to stay consistent with each other, across two languages, without redoing work by hand every time the manuscript changed — that's what the agent team and its production scripts were actually for.

    The system turned out to be the reusable part

    Partway through, it became obvious that the specific content of Personal Ecosystem — the ideas about owning your own data, personal AI infrastructure, the PIOS framework — was only half of what we'd built. The other half was a general-purpose book production system that didn't care what the book was about, or what raw material it started from.

    The book itself is still waiting on its ISBN before it can go on sale. The system that wrote it doesn't have to wait.

    The Agentic Book Toolkit

    We generalized the agent team, the workflows, the illustration styles, and the working production scripts into a standalone package: the Agentic Book Toolkit. It's not a prompt collection — it's the same 12 agent roles, 17 workflow specs, and the actual scripts that build EPUB, print PDF, and audiobook, plus a worked example and an honest account of what went wrong along the way and how we caught it.

    If your own raw material is a chat export, a decade of notes, old blog posts, or anything else — the same source-inventory approach applies. You bring whatever you have; the toolkit's pricing and licenses explain the rest.

    Setup is deliberately hands-off. You unzip the package, open your own AI coding agent (Claude Code, Codex, or similar), and say: "Read START-HERE.md and set this up for me." Your agent installs what's missing, asks what you want, and walks you through the rest — including getting your own API keys, which aren't included.

    It's available in three licenses: Self-Use, Self-Use + an Assisted Start session with Valto, and Agency/Commercial for producing books for clients — full terms are on the toolkit page.


    Ready to see what this system actually produced? Start with the Personal Ecosystem introduction or the chapter summaries. Ready to run the same system on your own book? Go to the Agentic Book Toolkit for setup and licenses.

    Personal AI coaching for owner-managers

    1:1, vendor-neutral, attached to the work you are already doing. See how the coaching is structured.

    Personal AI Coach