Podcast: Play in new window | Download
I went back on Before the Bestseller with Alex Strathdee to talk about AI agents for authors and how to automate your book marketing. This time Alex wanted to see the AI team I built to help run Author Media.
We covered which bot to build first, how an Amazon reviews bot can tell you what readers want from your next book, and what it costs. We also talked about security and why you should praise your bot when it does good work.
You don't need to be a programmer to start. You need one job you hate doing and a bot willing to do it.
Key takeaways
- Grok Bot is the easiest entry point for authors who want AI agents without the technical complexity of Claude Code or Codex.
- The best first agent to build is a webmaster bot that handles website updates on your behalf.
- An Amazon reviews bot can synthesize reader feedback and surface insights for your next book.
- Grok Bot lives in the cloud, so your context follows you across devices and you never lose your work.
- Telling your bot when it did a good job is genuinely useful, not just polite. It helps the agent learn what success looks like.
- Be intentional about setting routine frequencies. Running checks too often will burn through your usage allowance fast.
- Security is a real consideration, but Grok Bot's architecture compares favorably to competing agentic tools on that front.
Why Thomas Umstattd Uses Every AI Tool at Once
Alex Strathdee opens the conversation by noting that his own team at Shelf Life defaults to Claude, and that many people in the book marketing space lean toward one tool or another. He assumed Thomas had a clear favorite. Thomas quickly set the record straight.
Thomas does not pick sides. He holds active subscriptions to OpenAI, Claude, Cursor, Grok, and Grok Bot simultaneously. His reasoning is straightforward: every platform has genuine strengths and genuine weaknesses, and the only honest way to talk about them is to use them all. Once you move past the flat-rate subscription tier, the costs climb steeply. A $100 Claude subscription, Thomas explains, effectively gives you around $1,000 worth of tokens at retail rates. That deal disappears the moment you start purchasing tokens directly, so staying inside the subscription model matters for anyone who plans to use these tools heavily.
For authors who are just getting started, though, Thomas recommends Grok Bot as the first stop. The reason is not that Grok Bot is the most powerful option. It is that Grok Bot removes the friction that causes most people to give up before they see any results.
The FrontPage 2000 Problem: Why Most Authors Never Get Started
Thomas draws a useful comparison to the early days of the internet. When websites first became possible for ordinary people, the technical community said the barrier was low. Learn a little HTML and you could have your own site. That was true in a narrow sense, but only about one percent of people ever did it. The friction was just enough to stop the other ninety-nine percent cold.
Claude Code is the modern equivalent of FrontPage 2000 or Dreamweaver. It makes the powerful stuff more accessible, but you still need to know what DNS is. You still have to run terminal commands to get it going. You have to know what Codex is before you can switch into it. OpenAI recently merged some of these tools to simplify the experience, and it is still largely unused by the authors who would benefit most from it.
The deeper problem with sticking to basic chat in any of these tools is a structural one. When you are chatting with an AI, the AI is telling you what to do and you are doing the work. That is the old model. Agents flip the dynamic. When you use Grok Bot, Codex, or Claude Code in agent mode, you tell the AI what to do and it does the work. That is a meaningful shift, and it is the shift that makes these tools genuinely useful for busy authors who do not want to add another skill set to their plate.
Starting With a Webmaster Bot
Thomas recommends building a webmaster bot as your very first Grok Bot agent. The logic is simple: every author hates dealing with their website, and it is one of the few areas where the bot is unambiguously better equipped than the author. You do not need to coach it, correct its voice, or explain what your audience cares about. You just tell it what you want the site to do and it figures out how to make that happen.
Grok Bot has its own built-in browser. That means it can log in to your WordPress site, navigate the dashboard, and make changes using the site's API without you touching a single setting. Thomas has published a walkthrough video on YouTube that goes through the entire process of updating his own website using Grok Bot. He also offers a free course on building an author website, and he has built a Grok Bot version of that course where you paste one long text prompt into the interface and the bot builds your entire site according to the course curriculum. It will occasionally ask you a clarifying question, such as whether you prefer one layout over another, but the heavy lifting is done for you.
The website use case is also where Grok Bot's architecture really shows its value. Because Grok Bot lives on its own server in the cloud rather than on your local machine, you can start a conversation on your laptop, pick it up on your phone, and return to it on a second computer without losing any context. That portability is something Claude Code cannot match without knowing how to use GitHub and managing your own file sync setup.
Building an Amazon Reviews Bot
The second agent Thomas recommends building is one that monitors and analyzes your Amazon reviews. Here is how he sets it up.
The bot visits Amazon, pulls all the reviews for each of your books, and saves them to a markdown file on your computer. Once a month it goes back, checks for anything new, and updates the file. You now have a searchable, organized database of every reader response your books have ever generated, maintained automatically without you lifting a finger.
Once the database exists, you can ask the bot to analyze it. Thomas gives several examples of prompts that become useful at this point:
- Tell me about my target reader based on what they say in their reviews.
- What are readers looking for in this book that they feel they did not find?
- What topics come up repeatedly that I have not addressed fully?
That last question is particularly valuable for nonfiction authors. You may have three main topics in a book and one chapter that touches lightly on a fourth. If a surprising number of reviews mention that fourth topic, the bot can surface that pattern and flag it as a potential subject for your next book. That kind of signal is easy to miss when you are reading reviews one at a time, especially because reviews are emotionally charged and psychologically difficult to process at scale.
Thomas makes a point that is worth sitting with. Most authors find it hard to read their own reviews. The most common words in any review are "I" and "my." Readers are writing about themselves and how they felt, not really about the book. A one-star review is rarely a verdict on the quality of the work. It is often a reader explaining that they came looking for something the cover or the marketing promised and did not find inside. That is actually useful information, but it is almost impossible to extract that signal when you are reading the review as a personal judgment against you.
When the bot reads those same reviews and synthesizes them into a structured report, the emotional charge disappears. What remains is the data. Thomas notes that hearing him say a negative review is not really about the book does not make reading one-star reviews feel better. Having the bot absorb them and hand you the lessons, on the other hand, actually does.
How You Actually Interact With Your Bots
Alex asks a practical question that a lot of authors will have at this point. What does interacting with these bots actually look like in day-to-day life? Is there a daily email? A dashboard? Do you have to go find the bot somewhere?
Thomas explains that Grok Bot is accessed through a chat app interface, similar in feel to Discord, Slack, or Microsoft Teams. Along the left side of the screen you see your different bots. You click on one and you chat with it. The bots are persistent and always online, which means they can run scheduled tasks on their own and then have results ready for you when you check in.
You can also connect Grok Bot to email. Thomas gave an existing but rarely used Gmail account to his Grok Bot setup, which means he can email the bot directly or forward something to it and ask it to take action. The bot can email him back with results. He has also added Grok Bot to the Discord server his human team uses, so team members can ask the bot questions alongside the rest of the conversation. His human team is still a little uncertain about it and does not interact with it constantly, but they do ask it questions from time to time.
The cloud-based nature of the setup is what makes all of this seamless. There is no file sync to manage, no local repository to maintain, and no context that disappears when you switch devices.
A Candid Look at Security
Alex raises the security question, and it is the right one to raise. Giving an AI agent access to your email is not a trivial decision.
Thomas's assessment is that Grok Bot's security architecture is genuinely better than that of its main competitors in the always-on agentic space, which he names as Hermes and OpenClaw. One concrete example is password handling. If you need to pass a credential to Claude Code, there is no clean, secure way to do it. Pasting a password into a chat window means it may be visible to Anthropic employees, which is a real concern for anyone on the Fable tier. Even paying subscribers on that plan are subject to chat review as long as they are using Fable, the current top-tier model at the time of recording.
Grok Bot provides an encrypted field for passing API keys and passwords, similar to how a browser handles a password input. The bot can use the credential to authenticate without actually knowing what the credential is. That means if Grok Bot were ever compromised, it could not expose a password it was never shown.
That said, Thomas acknowledges the legitimate concern around email access. Even if Grok Bot does not know your email password, having access to your inbox means it could theoretically initiate a password reset somewhere else. He is personally willing to accept that tradeoff given the capabilities it unlocks, but he understands why others would be more cautious.
He also makes the broader observation that most email users already have AI embedded in their workflow whether they know it or not. Outlook has Copilot. Gmail has Gemini. Apple Mail has Apple Intelligence. The idea of a completely AI-free email setup is already more or less a fiction for anyone using a major provider. Proton Mail is probably the last holdout, though Thomas jokes that by the time the episode airs, Proton will likely have announced its own AI offering.
The Kinds of Tasks Your Bot Team Can Handle
Beyond the webmaster and reviews bots, Thomas sketches out the broader range of tasks these agents can take on. A few stand out as particularly well suited to the Grok Bot approach.
One is metadata maintenance. You can build a bot whose entire job is to roam the internet and make sure your book's information is correct everywhere it appears. Title, description, publisher details, categories, tags, and so on. Not just on Amazon but across every retailer and platform. This is a tedious job that no human wants to do regularly and that most authors simply never get around to. The bot is happy to do it on a schedule.
Another is email triage. If you are willing to give the bot access to your inbox, you can ask it to read everything, identify the messages that actually require your attention, and archive the rest. Thomas mentions the example of having 70,000 old emails and asking the bot to find the five worth keeping and archive everything else. That is the kind of task that would take a human assistant hours and that the bot can handle in a fraction of the time.
Grok Bot can also connect to tools like Kit and draft email newsletter content, though Thomas is careful to note that this is a use case where the author still needs to be closely involved. The bot can pull together a draft, but making it sound like the author requires coaching, feedback, and iteration. For website work, the bot knows more than the author does and can operate with near-total autonomy. For voice-sensitive content, the author is still a meaningful part of the process.
Managing Your AI Team Like a Manager
One of Thomas's more counterintuitive observations is that managing an AI agent team actually has a lot in common with managing human employees. The bots are not perfect. They will sometimes come back and say they need something from you. Before you run off to provide it, Thomas recommends asking the bot a simple question: can you find that yourself? Often the answer is yes, and the bot was just defaulting to asking rather than acting.
The feedback loop also matters more than most people expect. When the bot does something well, saying so is not just good manners. It is useful information. The bot is trying to calibrate what a good result looks like for you, and positive feedback helps it do that. The same is true for corrections. If the bot misses the mark, telling it specifically what was off and asking it to remember that for next time will cause it to store that information and apply it in future sessions. Each bot builds its own memory on a per-thread basis over time.
Thomas uses a lighthearted comparison. The bot is like a dog asking who the good boy is. It is not just an emotional question. It is a learning question. The bot genuinely needs to know what you consider good work.
One Practical Warning: Watch Your Usage
Thomas closes with a caution that every new Grok Bot user will likely need to hear eventually. The first time you see what these agents can do, the instinct is to automate everything immediately. Set routines for every task, check for new data constantly, and run the whole team at maximum frequency.
That will burn through your usage allowance fast.
Most of the recurring tasks authors care about do not need to run daily, let alone hourly. Checking for new Amazon reviews once a month is genuinely sufficient. Updating metadata on a quarterly basis covers most situations. Thomas recommends starting slow, setting conservative schedules, and adding frequency only when there is a clear reason to do so. It is much easier to scale up a working system than to scale back an expensive one that is eating your allowance on tasks that could have waited.
Where to Find Thomas and Keep Learning
For authors who want to go deeper, Thomas points to authormedia.com as the central hub for his tools, blog posts, and resources. The site also hosts AI Thomas, a free AI interface trained on his podcast back catalog. It functions something like a search engine for his content. Ask it a question, and it will point you to the specific episodes that address it, with citations at the bottom so you can go listen directly.
Alex closes by encouraging listeners to check out Thomas's Novel Marketing Podcast, which he credits as a foundational resource for his own understanding of book marketing. Links to Thomas's YouTube walkthrough video and his Grok Bot course are included in the show notes for this episode.
Thanks for listening to Before the Bestseller. If you got value from today's episode, share it with a fellow author. For insider tips we do not share anywhere else, head to getshelflife.com and sign up for the newsletter.
