On August 12, 2024, I published a blog post called From AI-Generated Affirmations to Published Books. I had no idea how literal that title was eventually going to become.
Two years and six days later, I published a case study about The Affirmation Glitch, a 126,743-word AI-generated novel about a fictional web developer whose AI affirmation site starts doing impossible things.
And the weirdest part is that the fictional site was inspired by the same real affirmation project I was writing about in 2024.
This isn’t just another post about AI. Looking back, this feels like a remarkable little two-year longitudinal case study of one creator’s relationship with generative AI, documenting how the technology matured from a disjointed collection of individual content-generation tools into something approaching an autonomous, interconnected creative-production pipeline.
The Ridiculous, Circular Timeline
If you map out the sequence of events that led to this book, the loop is wonderfully ridiculous:
- 2024: I use AI to make personalized affirmations and turn them into published non-fiction books.
- 2026: I ask AI what kind of fiction novel I should make next.
- AI: You should write about a woman making an AI affirmation website.
- Me: Fine.
- AI: [Writes 126,743 words about it in under 16 hours].
- Me: [Asks my original 2024 Foxy Apps affirmation generator whether writing an AI book about an AI using AI makes me a fraud].
- Affirmation AI: “I am the architect of my own journey, weaving threads of hope, courage, and creativity into the tapestry of my life.”
- Me: [Publishes the AI novel and the case study about it].
The Meta-Paradox: When the Algorithm Writes About You
Back in 2024, beyond writing books and creating guided journals, I built custom interactive applications for my readers. Using tools like Foxy Apps, I architected a personalized affirmation generator that captured user data and routed it through Pabbly Connect directly into targeted MailPoet email sequences. It was a functional, real-world tool designed to bring a little bit of peace to anyone who needed it.
When I sat down to test BookyAI for my 2026 experiment, I gave the system a zero-shot prompt. I handed the models my professional websites and basic background details (including my three cockatoos: Misha, Boo, and Sara), and asked for story ideas.
Without any specific direction from me, both ChatGPT and Gemini independently converged on a stunningly meta premise. They conceptualized a cozy, magical realism story about a burned-out freelance web developer named Maya who builds an AI-driven daily affirmation website using the Divi framework to soothe her own anxieties.
In the novel, Maya’s simple tool begins generating eerily specific, impossible-to-guess visual metaphors—like a “Forest of Light Angel”—that actively heal the unspoken traumas of her users. The AI had essentially looked at my actual, real-world Foxy Apps affirmation generator and spun it into a 30-chapter contemporary fantasy. Life imitated art, and then the algorithm imitated my life.
A Beautiful Natural Comparison: 2024 vs. 2026
Reflecting on the technical leap over these two years is mind-bending. The tools I use to create have shifted dramatically, reflecting broader industry trends toward local execution, open-source models, and unified pipelines.
| August 2024 | August 2026 |
| AI-Assisted Content Creation | AI-Orchestrated Content Production |
| ChatGPT generates short pieces | Local LLM generates an entire 126k-word novel |
| Human manually chooses 34 out of 50 generated affirmations | Human chooses the premise and lets the model run the narrative |
| 340 images generated, manually curated one-by-one | Cover generation integrated into a rapid pipeline |
| Canva manual assembly | BookyAI automated book assembly and export |
| MagicBookifier + Afforai | BookyAI + local 17B Llama (via Ollama) |
| AutoCrit used extensively to refine and edit prose | AutoCrit used primarily to measure and benchmark untouched prose |
| Heavy human refinement | Deliberately minimal human refinement |
| Published AI-assisted non-fiction | Published experimental AI-generated fiction |
| AI as a collection of tools | AI as an interconnected production pipeline |
The Cloud Era vs. The Local Revolution
Two years ago, my workflow relied heavily on fragmented, cloud-based subscriptions and API endpoints. I used Magic Bookifier for initial drafting, custom ChatGPT and Afforai models for refining brand voice, and ZimmWriter to queue up SEO articles. While self-hosting Stable Diffusion via ComfyUI gave me some local control over visuals, the text generation required constant connectivity and careful management of API credits.
For The Affirmation Glitch, I bypassed massive API overhead entirely. The entire 126,743-word manuscript was generated locally on my consumer-grade PC (11th Gen Intel i7 processor and an NVIDIA RTX 2060 GPU) using BookyAI paired with a Llama script. Running a language model of that size locally means that making mistakes, regenerating chapters, and testing new prompts now costs absolutely nothing but processing time. The compute cost for text generation dropped to zero.
The Central Thesis: Moving Up the Level of Abstraction
In my 2024 article, I wrote that the process demonstrated the critical importance of human creativity, decision-making, and refinement.
I don’t think my 2026 experiment disproves that at all. Instead, it suggests that the location of the human contribution has fundamentally moved.
In 2024, much of my judgment happened at the artifact level:
Generate ➔ Inspect ➔ Choose ➔ Edit ➔ Arrange ➔ Polish
In 2026, with The Affirmation Glitch, my judgment happened entirely upstream:
Conceive ➔ Prompt ➔ Configure ➔ Choose Model ➔ Set Constraints ➔ Supervise ➔ Evaluate ➔ Decide What Gets Published
In two years, I didn’t become less involved in the creative process; my involvement simply moved up a level of abstraction. I am no longer just the mechanic tightening the bolts on the prose; I am the architect designing the factory.
From “Bumpy Design” to Professional Benchmarks
In my 2024 reflection on creating the guided journal, I wrote about the friction of the process: “Smooth Writing, Bumpy Design!” While the AI writing was nearly effortless back then, formatting the final product for multi-platform delivery required intense patience, Canva layouts, and clickable PDFs.
The 2026 pipeline aggressively streamlined that friction. BookyAI’s native exporting features proved significantly less complicated than traditional formatting workflows. But more importantly, the raw narrative quality of the AI has skyrocketed.
I ran the completely unedited 2026 manuscript through AutoCrit’s professional editing analyzer. The results were staggering for untouched prose:
- Overall Score: 87.4 / 100 against commercial Urban Fantasy titles.
- Dialogue: 100 / 100 benchmark score.
- Pacing & Momentum: 87.0 / 100, with only 3.8% of the paragraphs flagged as slow-paced.
The fact that an algorithm can sustain that level of commercial pacing across a 126,000-word narrative structure without human micro-editing proves just how sophisticated the models have become.
Closing the Loop
There is a profound poetry in this two-year retrospective. In 2024, I used AI to publish self-discovery books to help others embrace their inner light. In 2026, when imposter syndrome crept in about writing a book about AI using AI, my own digital creation looked back at me and said: “I am the architect of my own journey.”
We are no longer just using AI to automate simple workflows or queue up blog posts. We are using it to hold up a mirror to our own creative journeys, allowing the autonomous pipelines we build to reflect our deepest professional and personal evolutions.
0 Comments