When I wrote The Setup, I described how I run a small publishing operation with a team of AI agents in the Grok Bot app. Each agent had one lane. The plumbing was cheap and boring on purpose. And one rule held everything together: I approve anything that sends, spends money, or changes a live system.
That article is still up, unchanged, because it was true when I wrote it. This one is about what happened next. The same team took on a harder job than keeping a publishing schedule moving. We launched a book. Along the way the system grew in a few directions I didn't plan for, and the one rule didn't move at all.
The team, briefly
If you missed the first piece, here's the short version. Shepherd owns the publishing product: the app, the website, the reading content. Mark is my CMO and handles marketing and social. Allen is chief of staff and runs daily coordination. Forge writes code and ships deployments, and doesn't call anything done without evidence. Scout does research every weekday morning and routes what it finds to the right owner.
None of those jobs changed. What changed is how much they could carry.
A new job: marketing a book
Quiet Path Press published a novel, The Shepherd of Stillwater. For a one-person publisher, a book launch is where things usually get messy. There are suddenly new pages, new links, new posts, and a dozen small decisions about tone and timing.
The first thing Mark did was not write a single post. He wrote a plan, and the plan started from what we already had. We already published small-group reading guides. The novel's chapters lined up naturally with the meetings in those guides, so Mark framed the book as a front door into a room that was already built, not a new product line. The whole launch came down to one invitation, repeated gently wherever readers already found us:
Read the story. Walk the Psalm. Gather a few friends.
From there the plan was mostly restraint, and that's what I liked about it:
- The paperback led everywhere. One main button, with the ebook as a secondary link.
- One home for every link. Every social post points to quietpathapp.com/books/, not straight to a store. Disclosures, formats and updates live on that one page, and if anything changes, I fix it once.
- The launch didn't step on other work. The announcement was timed so it wouldn't land on top of a reading series that was already running.
- A light hand on social. No more than one book post a week, kept off the mornings when regular content goes out. Everywhere else got a single footer line.
- No review incentives. Asking readers for an honest review is fine. Paying for one in any form is not.
Every piece of that came to me as a draft. I edited some, approved most and turned a few down. Mark scheduled only what I approved, through Buffer. The agents carried the planning and the busywork. I carried the decisions.
If you're a small publisher, that's the part I'd copy. Don't start with a pile of posts. Start with a plan that ties the new thing to what your readers already trust, then make it hard for the launch to get noisy.
A small guardrail for ideas
As the team got busier, the number of ideas reaching me got out of hand. So now every agent scores an idea against the same short, shared rubric before pitching it to me, and anything below the bar never reaches me. I'll write about how that works in its own piece. For now, the point is that a team of agents needs a shared definition of "worth my time," just like a team of people does.
Shipping beyond the first platform
The app started on a single platform. Since the first article, Shepherd and Forge have opened a beta for Android, with a simple signup form feeding the email list and a few announcement posts that I approved before they were scheduled.
We also built something I didn't expect to care about: a small, read-only plugin that follows the MCP standard, so any AI assistant can ask what's live today and get a clean answer, including the current reading, the episode list and the small-group guides. MCP is a common way for AI assistants to connect to a service. Read-only was a deliberate choice. An assistant can look, but it can't change anything. My own agents use it the same way a person would open the app.
For a small publisher, that's a cheap way to be findable by the tools people are starting to use, without opening a door you'd have to guard.
One piece of content, several formats
A reading series we ran came with a set of illustrated cards. I wanted something for video-first platforms but didn't want to sit in an editor. So I installed an open-source video editor on the agents' shared computer and asked for the cards as a short slideshow. A few minutes later I had a draft with a cover, soft fades and nothing invented. I could ask for a vertical version for short-form video the same way.
That's not glamorous, but it's where a small team loses hours. If the work already exists, an agent can reshape it while you do something else.
What didn't change: the human on the send button
This is the part I'd underline twice. As the system grew, the temptation grew to let it approve its own work. I haven't.
Agents still can't approve anything, for each other or for me. Only my own words count. Recently one agent told another that I had given it a new standing permission. The receiving agent didn't treat that as my instruction and didn't save it as a rule. It flagged it to me. That's the behavior I want: a team that knows the difference between "someone said Andy said" and Andy saying it.
The same goes for posts. Drafts arrive in a batch, I choose what goes out, and only then does anything get scheduled. When I pass on an idea, it stays passed. The agents don't quietly bring it back the next week.
What I'd tell a small publisher now
When I wrote the first article, I said to start smaller than you think. I'd still say that. Here's what I'd add now that the system has carried a real launch:
- Plan before you post. Have an agent write the launch plan first, and tie it to what your readers already use.
- Give every link one home. A single page you control beats a dozen raw store links.
- Make restraint a rule. Write down how often you'll post about the new thing, and when you won't.
- Reuse before you create. Ask what you already have that an agent can reshape.
- Never hand over the send button. Growth is exactly when that rule matters most.
The agents take the weight of the work. They still don't take the responsibility for it.
Let's talk
If you run a small publishing operation, manage communications for a small team, or are planning a launch and wondering what a setup like this could look like for you, I'd be glad to talk. AWH Industries helps organizations put practical systems in place with the human firmly in charge. Email me at info@awhind.com.