I wanted control over my books, and I got admin.
That is the quiet bargain behind indie publishing: you choose the story, the cover, the release date and the price, then discover that each choice creates another small job. The novelist gradually acquires a second occupation as the manager of a tiny publishing company.
I run mine across multiple pen names, and on a normal week I might work on a manuscript, check an Amazon ad, plan a newsletter, maintain a website, update launch tasks and inspect sales. None of these jobs is especially difficult, but together they can eat the week.
The switching is worse than the work.
You open a manuscript, remember that an ad needs checking, open the advertising dashboard and notice that sales were odd yesterday, and now you are in KDP; twenty minutes later, you are comparing two dates in a spreadsheet while the unfinished chapter waits in another window.
By lunch, you have been busy for hours and written nothing.
Most conversations about AI concentrate on generating prose, the one part of the job most writers want to keep for themselves. Sales reports, newsletters, ad checks, metadata and half-remembered experiments get far less attention.
I don’t need a machine to have my ideas for me, but I would happily give it the Tuesday afternoon spent working out whether a campaign became expensive or sales merely had a bad day.
A chatbot still needs a manager
An author can already hand a KDP export to ChatGPT and ask for an analysis, and the same chatbot can draft a newsletter or suggest ad copy. All of that is useful, but the author still has to remember what needs to happen, gather the files, explain the context and ask the right question, while the tool waits for someone to give it work.
That saves labour inside a task, but it does little about the mental burden of running the business.
For the past few weeks I have been running something different: AI employees with defined jobs. Plural, because I am testing two of them on the same work, Viktor and Grok Bot, and building a third of my own. Their responsibility is the marketing and admin around the books, while anything that touches my money, my readers, my reputation or my published work still comes back to me.
The distinction is simple: a chatbot says, “What do you need?”, while an AI employee says, “I noticed this, and here is what I think deserves your attention.”
What I no longer do myself
Ads first, because that is where an author’s attention leaks fastest. Every Monday one of my agents pulls Amazon and Meta performance and compares the last seven days with the seven before. It can read the numbers, but it cannot change a bid or a budget, so the worst outcome of a bad analysis is a bad paragraph in my Monday report.
Sales come from KDP exports. Because KDP has no public sales API, I still download the file myself, which keeps that step stubbornly manual. After that, both agents read the export knowing my series order and last week’s ad numbers, so I no longer explain my catalogue every time I want an answer.
Newsletters slip when the manuscript is going well, which is exactly when an author wants to stay inside the book. Viktor checks whether mine went out on schedule and helps me prepare the issues themselves, while I decide what goes in and when it is sent.
Content research runs twice a week for The Indie Lab: a pack of five to eight angles with a shortlist on top, and a rule against bringing back an idea I have already shortlisted unless something new has happened. The agent drafts no article and publishes nothing.
The last routine has nothing to do with marketing, and it exists to protect the reason for all the others. On Fridays one agent compares my calendar with what actually happened to the manuscript and tells me when a call or a meeting fell inside a writing block. Twice a week it looks at the folder for the book I am drafting and stays silent unless a chapter changed, or a writing session passed with nothing to show for it.
I don’t want an autonomous publisher
There is an obvious way for this idea to become stupid: give an AI access to advertising accounts, email tools, the author website and publishing platforms, then let it change budgets or contact readers without supervision. It would remove work in the same sense that a kitchen fire removes the need to cook dinner.
The sequence that works is more restrained: observe, analyse, recommend, prepare and ask for approval. In my setup the ad data comes in read-only and drafts stay drafts. Nothing goes out, and no money moves, until I say so.
The other requirement is memory. My agents know my pen names, the series order, the experiments I have already run and the rules I have set, because if I had to explain my catalogue from scratch every Monday, I would have acquired a needy intern with an excellent vocabulary.
It still needs supervision. Publishing data does not always arrive neatly, the KDP export is still my job, and one of my agents talks to me through Slack, which is a faintly ridiculous thing to install when the whole company is one person and a cat. Last month one agent’s scheduled checks went quiet for weeks, and I only noticed when I asked it what it had been doing. Managing the AI is a small job, but it is a job.
The big labs have noticed
At the end of September OpenAI launched Dots, always-on agents inside ChatGPT that keep working between conversations and come back when a decision needs a human. Grok Bot, one of the two I am testing, comes from xAI. The idea behind this post is now a shipping product from the biggest AI companies. European authors get a catch with it: Dots on ChatGPT Pro is not available to subscribers in Europe, so from Poland I cannot test it on a personal plan.
The business created by the books
I am not trying to escape the commercial side of publishing, because an indie author who ignores sales, costs, cash flow and reader acquisition is running a hobby with unusually elaborate paperwork.
I want my attention where it changes the outcome. Choosing what to publish, setting release timing, deciding how much money I can risk and understanding why readers continue through a series are my decisions. Copying numbers between systems and remembering that the newsletter is due are jobs I would hand to staff, if I had any.
Most authors don’t. The awkward middle group has enough books and revenue to create real operational work, but not enough money for an assistant and an advertising specialist, which is where an AI employee earns its keep.
Nobody can promise that an AI employee will sell a book, and anyone who does has probably found a more reliable business in selling promises to writers. My agents hand me something smaller: what happened, what needs a decision, what can be ignored and what can wait, so the marketing keeps moving while I work on the next book.
I set up Viktor for clients, so if you want to test it, send me a message.
This is fantastic. I've been doing the same. Using agents to track my operations and work in the background. I'm in the US and Dots has been the most helpful item on that operations end.
I enjoyed this discussion and found it useful. I have one novel, with a collection of short stories in the wings, meaning my publishing world is small, but the issues of admin tasks and switching between tasks and all the time and attention that are required resonate for me just the same. I’m overwhelmed imagining a large backlist balanced with multiple active projects!