How I'm Using Grokbot and The Point of AI Is Not More AI

New models come out overnight, and before you know it, you're behind. At least, it's easy to feel that way.

So, I wanted to sit down quickly and write a brief explanation of how I'm using AI, how I used to use AI, and how I intend to use AI in the future. One thing I'm working on for my business and brand as a human author is my own guidelines around using AI while still letting it help me become more productive and a better writer, able to engage better with my Timothy, and write the kind of stories that my audience wants to read and that will, even while they're enjoying them, change their lives for the better.

I think the place where most of us have to start when it comes to how large language models have impacted our productivity is with how they've already impacted our productivity. After all, I think we should be grateful to live in this era—an era of flying cars and computers that can tell us Captain Kirk is not aboard the Enterprise, and "I’m sorry Dave. I'm afraid I can't do that," HAL, by the way.

Part of this gratitude can also be helpful. We can understand how much of our time is actually being saved by the AI we are implementing into our daily lives. Many recent studies have shown that those adding large language models to their daily workflows can actually feel more exhausted. Yes, we feel more productive, and as I've seen more and more authors here on authormedia.social unlocking agentic AI for the first time and being amazed by it, it's easy to become addicted to productivity—addicted to being a manager of agents and sub-agents and tasks, to feel superhuman, to feel like Tony Stark with a fabricator, changing the world one bot at a time.

But with this productivity, we can also become quickly burned out. We can dig our wells in codebases and repos and models that will dry up quickly, only for us to have to start learning once again when the next best tool comes out. In this space, there's a mindset of continual improvement, of the productivity bro, the necessity to keep pushing ourselves to achieve more and more. I think that this is a great thing, but it can also be a very negative thing.

So, as I said, I want to sit down and create a brief report of how I am personally using AI, how I used to use AI, and how I would like to use LLMs and reasoning models, as well as more analytical models in the future to help me grow my business and make me a more capable human author. Part of this will also be recognizing where AI is actually helping and where it's a bit of a drain.

How did I find LLMs? Like many of us, one of the first LLMs I ever used was the first commercially and freely available version of ChatGPT back in 2022. I remember my first interaction: I was actually in, believe it or not, an MCU analysis class. Yes, it did exist, and I think it probably still does in the institute where I was. The teacher was awesome. He was a Sanderson fan and an Orson Scott Card fan and an IT expert, as well as also apparently being an MCU fan, I guess. And he had something to show us during an icebreaker, and it quickly grew to consume the entire class session. It was ChatGPT. It was an LLM. It was science fiction.

It was about three months before I actually created my own ChatGPT account. I had been writing with the intention of making it a full-time career—though I was certainly still making many mistakes—for about two years before that time in my life, and I'd gotten this notion in my head, with all the news going around about ChatGPT, that if I simply didn't create an account, then if someone ever accused me of having used AI to write my books in the future, say ten years down the line, I could simply ask OpenAI to review their records and prove that I'd actually never had an account. Gee, was I dumb.

When I finally did get an account, I wasn't incredibly impressed. Still, it felt like magic, and it was pretty sick. 

Now, unlike a lot of you here on authormedia.social, I actually didn't come into the AI space with the core intention being assistance with or the use of it in semantics, but rather in how LLMs can be used in programming. 

I had been taking several basic Python, HTML, CSS, and JavaScript classes during that time, and the advancement in the world of LLMs allowed me to actually start building some of the things that my knowledge of coding, limited as it was, stopped me from doing. It was a tool to expand my horizons into building apps and video games, and it was incredible.

But I certainly wasn't using it as well as I could have. Like most beginners with LLMs, I was using ChatGPT simply as Google or as a party trick. Type in a question, get an answer. Type in a ridiculous prompt—well, get another answer, I guess. It wasn't contributing all that much to my life or my impact on the world.

Then came Claude, the dawn of skills, and Claude Code. And I found myself in a very, very, very deep rabbit hole on the internet: the rabbit hole of vibe coding. It was an optimistic time when it seemed that you could make anything you actually wanted to. It was awesome. I started making advanced IF video games and text adventures, developing code languages, building DMG applications for Apple, advanced calculators for things like Kickstarter, and planetary Python-based physics reactors for fictional universes. The sky was the limit, but there was always the endless stream of notifications, the endless need to keep on improving. Just another prompt. And it always felt bad to go to bed with usage left in the tank. Busywork, I think, has always been a problem within humanity, but the dawn of continual iteration and improvement with LLMs certainly didn't help me any.

Then came OpenClaw, which is the current name I'll be using, I think. If you're not in the know, don't even ask; it was this whole thing. OpenClaw was soon followed by Hermes and Open Agent. It was the dawn of LLM harnesses that could actually do things given their own virtual private servers or virtual private computers—just a fancy term for the same thing that Grokbot has. Memory and context, as well as token usage, was getting better too, and Model Context Protocols, or MCPs, were also improving with tools like Kit and ElevenLabs implementing them.

Agentic AI had finally dawned, and boy, was it awesome and scary. API keys were getting leaked all over the place, but cool tools were also being created. For the first time, I could see a future where I didn't have to, at some point, hire a massive team to help run my publishing company, but could possibly in the future have a team of agentic agents help run my company, allowing me to save money and make more money, as well as be more productive as an author, and also be more selective when eventually hiring an in-house team of trusted cover designers and editors—a dream I'm still slowly working my way toward.

Then Grokbot came along. If you're not familiar with any of those tools I mentioned earlier, such as Hermes, Open Agent, OpenClaw, or the million other agent harnesses out there, just picture what Grokbot is, but a year or two ago, and scary, and lots of CLIs. And if you don't know what a CLI is, then good for you; you have a life.

Grokbot was actually probably one of the most annoying advances in AI that I've ever experienced. 

I'd actually spent the last six months building a very similar agent harness, which essentially just repackaged Hermes on a VPS and ran it with API keys into its own sandbox—basically just Grokbot, but much less convenient, and I was doing all of the design. And then Grokbot launched, and all of that work was for nothing. But so it goes in the world of AI, and it was fine because Grokbot is awesome.

Here's some proof of Grokbot being awesome: Over the past couple of years, I've written roughly thirty different short stories, mostly for monthly contests, quite a few of which I've actually won, and I also had a catalog of almost two million words' worth of finished and unfinished drafts. So, I hooked Grokbot up and got to work trying to optimize a practically bookless author business.

If you don't know me, my name is Xavier Schwindt. I’m a progression fantasy and sci-fi author. And I write in the margins of dark stories but hopeful outcomes that still cost the character something. Worlds where evil indeed does exist, but where goodness and virtue also find homes. Plus some freaking awesome magic systems as well. Oh, also, did I mention cool worldbuilding? Anyways.

When I started out writing seriously, one of my core decisions was not to publish anything until after I had written over a million words, counting only finished books. I hit that when I turned fifteen and, with some other authors here on authormedia.social, launched something of a sketchily run anthology, which was great, and it's still available on Amazon. Check out Work the Hands https://www.amazon.com/dp/B0G51GW8SJ?lv=shuf&channelId=500&plpRedirect=mhFallback. It was a hassle, but anyways.

I don't have many books out despite writing 5,000 words a day with the help of dictation. Dear Lord, I wish I'd discovered dictation sooner in my career. And so, optimizing my business or my future business is a little bit different for me than it might be for others with massive backlogs. Though for me, I actually did have something of a backlog: all of those short stories I mentioned, and several books that just needed a little bit of editing.

So, last week I decided to test out the Spotify audiobook feature. And so, Grokbot went ahead, collected my favorite short stories I've written, and put them into a collection. I hardly had to do anything, and that collection is available on Spotify here:
https://open.spotify.com/show/2QExfJdtd0K3zdZjku2LKY?si=955fa45d1beb4af6

But that got me thinking: How far can I actually push Grokbot? Especially with audiobooks, because audio is one of the main places where I see indie authors winning, and the industry keeps on growing. This led to one of the coolest epiphanies I've had in a long time when it comes to AI optimism—and I tend to be quite the optimist, if you can't tell already.

I signed back into my old ElevenLabs account, got on the $11-a-month-for-the-first-three-months plan, then the $22-a-month plan, paying only $11 for, I believe, 100,000 tokens. I then had my team of quills on Grokbot go through, using the API and browser, and create a full-cast audiobook of the short story collection using ElevenLabs V4. It even added scoring. Here's an unlisted YouTube upload of that audiobook.
https://youtu.be/FBszDXP6mQA?si=-BJyhXdDoLTTTOBf
This took Grokbot only one day. Remarkable, isn't it?

Now, the short stories aren't that good, but the narration is, and I hope you'll give it a listen. I'm so grateful for the life I live, as Thomas was reflecting the other week. We get to live in the most sci-fi kind of world ever, and it's frankly incredible. Pinch yourself sometimes. Remember the absurd. Don't get too wrapped up—don't get so wrapped up in the productivity and the endless prompting that you don't look around and realize: “What in the world? I'm having a conversation with math, sitting on a dead tree, tucking my legs under a slab from a dead tree, running my fingers across little squares of plastic that send magic rays across the earth to communicate with magic ray receivers, which then talk with the minor math gods doing weightlifting in micro-galaxies called servers.” It is truly surreal.

These tools are growing exponentially year after year. It's easier than it has ever been to create a book, and perhaps that is dangerous. Many LLMs now have large enough context windows that they can write an entire 50,000-plus-word manuscript in a single prompt, and it will be all right. It'll be fine. It'll be on the upper end of the bell curve when it comes to self-published books on Amazon.

But for me, I don't want to be on that bell curve. I want to be on the other side of it. I want to be on the side where we are reaching the bounds of human creativity. And Grokbot is but one of many steps that we are taking in the field of LLMs and agentic AGI. And I'm excited to see where these technologies go and how this community can use them.

I think many of you would agree with me: I do not want this technology simply because it is cool, but I want it because of what it can do for us, what it can help us to create, and what it can help us to become. Many of us do not want AI for the sake of having it, but instead for the sake of having it do our dishes and laundry and spreadsheet work—unless you love your spreadsheets—so that we can go out and hallucinate people and narratives and kingdoms in our minds and commit the greatest of human potential, as the existentialists would have thought of us authors, creators of worlds, as we manifest things that are only in our free will, are only in our minds, into the permeable and solid aspects of reality. We take the infinite, the odd, the absurd realities floating in our minds, and we transfer them not only into the physical world, but into other realities, other aspects of nature, and other people's minds. It is the closest thing that we have to magic outside of the absurd, and in the light of the absurd, it becomes even more so.

So to all of you endeavoring to become better, to create better books, to thrill your readers, and write more because of this incredible time we live in and these incredible technologies we can interact with, here's a toast to you from me.

Blessings,
Xavier Schwindt