Turn a Book Into an Agent—an AI Coach That Stays With You Until You Actually Do It
The best thing about reading is that it lets you reach across time and befriend the smartest people who ever lived.
I spent half a year and more than ten thousand yuan of compute building a skill: it turns a book into an agent. Not a search box where you ask and it pulls an answer out of the book, but something that carries the book's methods, speaks in the author's own voice, and coaches you step by step until you actually do it.
This episode I use Tiny Habits (the Fogg Behavior Model) to build a habit coach that walked me to a tiny habit I could start that same night. Swap in a biography or The Almanack of Naval Ravikant, and it becomes a conversation with that person.
The best thing about reading is that it lets you reach across time and befriend the smartest people who ever lived—understand how they think, and talk with them. I spent half a year and more than ten thousand yuan of compute on a skill that pushes this one step further: turning a book into an agent that will talk back.
Not a “chat with your book” search box
Most of the “chat with your book” tools out there are search boxes at heart: you ask, they retrieve a passage and read it back. This one is different. It isn’t retrieval-based context augmentation. It’s a real coach, and it carries more than the theory in the book—it carries the author’s own language and turns of phrase, so it sounds like the author talking to you.
Take Tiny Habits: a habit coach
The demo this time uses Tiny Habits. The agent it produced behaves like the author, walking you step by step through building a tiny habit of your own. The critical part: every sentence it says has a line in the book behind it. It isn’t making things up. It’s like talking to the author himself.
Tiny Habits is a methodology handbook, so the agent also becomes a well-informed advisor that helps you work out a solution.
A real conversation: from “I want to sleep earlier” to one action I could take that night
Ever since I started working with AI, I can’t bring myself to go to bed—sleeping always feels like wasted time. So I asked it: how do I use a tiny habit to get my sleep schedule under control?
It didn’t recite theory at me. It asked questions. First it took the guilt of “failed again” off my shoulders. Then it worked out what I really wanted to change. From my answers it learned I’d set alarms before and they hadn’t worked—so reminders alone are useless. You have to find the real psychological motive, make the behavior easier to do, and then use an action so small it cannot fail as the switch.
The method we settled on is one I still use: the phone lives in the study, I go to the bedroom without it, and even the charger stays in the study. Every time I pull it off, I give myself a small, deliberate celebration—“yes, phone’s in the study again today.” That’s how a behavior slowly turns into a habit.
Not just handbooks: biographies and The Almanack of Naval Ravikant can become “that person”
It isn’t limited to how-to books. Biographies work too. Drop in a famous person’s life story, or a book like The Almanack of Naval Ravikant, and you get the feeling of a long conversation with a first-rate mind—talking with Naval about what happiness is, and how to hold real non-attachment to outcomes while still wanting things.
How it stays faithful to the book (and doesn’t invent)
The process goes like this. You hand the skill the book’s original text. It reads the whole thing, then converts the book into conversation on a couple of fronts: it pulls out the author’s expression fingerprint, and it maps the structure of the methodology. When it answers, it adds nothing of its own invention. When it needs to quote, it goes back to the source through retrieval augmentation and quotes verbatim.
Copyright and commercial use
Copyright matters here. Running this skill on a book or e-book you have already purchased, to understand it better and create value of your own, is fine. If you want to use it commercially, you have to secure the rights to that book first.
I think a platform like Dedao, which publishes books with a lot of well-known figures, is perfectly positioned for this—turning its contracted authors’ knowledge into agents. I’m working with friends in publishing on exactly this kind of commercial agent: help a well-known figure publish a book, and bring that book to market as an agent at the same time. A book has historically sold for a few dozen yuan, once. As an agent it can earn the author and the publisher recurring income for years, and it helps readers understand the material better.
The whole process, packaged as a skill
I packaged this “book into agent” production process as a reusable skill (book-to-skill). Get the e-book you legally own ready and it walks you through the whole thing: parse the source text → build source grounding (verbatim + line numbers) → extract the capability core → proof line by line against the real book → audit coverage → scenario acceptance. Every step has guardrails against the traps, and it all comes down to one line: stay faithful to the book, invent nothing.
Source: EP0043_audio.mp3 · ASR model gemini-2.5-pro (chunked parallel) · full text of the original recording
[00:00] I spent more than half a year and over ten thousand yuan of compute just to build one skill: turning a book into a person. Reading is one of the great joys in life—you get to reach across time and space, make friends with the smartest people there are, understand how they think, talk with them. And now, with this skill, AI can put that author right in front of us to have a conversation. So how does it do that? Let me walk you through the whole idea. First off, it’s not a search box. It’s not doing retrieval-style—
[00:25] context augmentation. It’s genuinely like a coach: not just the theory in the book, but the author’s own way of speaking, their way of putting things. The one I’m showing you today I built with the book The Fogg Behavior Model. An agent like this walks with you the way the author would, step by step, like a coach, helping you build one tiny habit of your own. And every single sentence the agent says is backed by actual text in the book. It’s like the author himself is chatting with you.
[00:51] Of course the Fogg Behavior Model is just one example. It’s a methodology handbook, so it can turn into a consultant who really knows the field and helps us put together solutions. Now, there’s copyright involved here. So—assuming you’ve already bought the book or the ebook—we can use this skill for ourselves, to understand the content better, or to create some value with the knowledge in it. But if you want to use it commercially, you absolutely have to get the rights to the book.
[01:16] Take Dedao—they help a lot of well-known people publish books. Honestly I think Dedao is a perfect fit for a feature like this: turn the knowledge of the authors they’ve signed into their own agents, turn the expert playbooks in Dedao Chengbang into an agent. I’m actually working with a friend in publishing on exactly this kind of commercial agent. We can help a well-known figure publish a book, and then turn that book into an agent and take it to market too. And the upside of doing that—
[01:41] a book normally only sells for a few dozen yuan, something in that range, and it’s a one-time purchase. But with a capability like this, the agent can keep generating revenue—long-term business income for multiple parties, the author and the publisher—and it helps readers understand the book better too. Let’s keep going. The whole process is: you hand the original manuscript text to the skill, and it reads the whole thing through and turns it into conversation, working on a few different fronts—
[02:07] like I said earlier: building the author’s voice fingerprint, then building the structure of the methodology. When it answers, it doesn’t make up extra content. And when it needs to quote the original text, it can use retrieval augmentation to find the passage and cite it. Let’s look at a real conversation I had with the Fogg Behavior Model agent after it was built. So, ever since I started using AI, I really hate going to sleep. Sleeping feels like a huge waste of time.
[02:32] So I got talking with it about how I could set up a tiny habit that gets me on a more regular sleep schedule. You can see the agent starts by asking questions to take the guilt off me, then works out what exact change I’m after. From my answers it also learns that I’ve set alarms before and they didn’t work—so reminders are useless here. It needs to find the real psychological motivation, and then to build better conditions for—
[02:58] getting it done. And it uses one really small thing as the on-switch that kicks the action off. So after we finished talking I came away with a method, and I’m using it now: every day I use my phone in the study, and when I go back to the bedroom, I don’t bring the phone in with me. I left the charger in the study too. And that really does get me to bed on—to sleep on time, I mean. It also reminded me that every time I pull it off,
[03:23] every day, every single time I do it, I should give myself a small celebration—something with a bit of ritual to it, something memorable. Like, yes! I put my phone in the study again today, I didn’t carry it into the bedroom. That’s how it reinforces the habit. And of course, it’s not just handbooks. We can drop in a biography of a well-known person too. You can see the publishing tool we built lets you feed in a biography directly, or a book like The Almanack of Naval Ravikant, and get that feeling of really going deep with top-tier—
[03:49] wisdom. Talk with Naval in depth about what happiness is, and how to tell apart—how to tell apart the experiences that desire brings. Discuss how you can want something and still hold a genuine non-attachment to the outcome. That’s a wrap for today. Talk to me down in the comments. See you next time. Bye-bye.