AI Created This: What I Learned by Handing Each Chapter to a Different AI

I’ve finished a nonfiction book called AI Created This.

That title stuck. It’s now the name on the cover, not just the folder, the manuscript, and the growing pile of notes that took over my desk while this thing was being assembled.

The book looks at the ethics and entertainment value of AI-created art, but it does it as part satire, part social commentary, part humor, and part nonfiction experiment about what these tools can actually do. The format is part of the experiment.

I didn’t write the whole thing myself. I also didn’t hand the entire book to one AI and ask it to crank out a manuscript while I went to get coffee.

Instead, a different AI writing tool was responsible for each section. I gave the tool a tailored prompt about its assigned topic, let it write in its own voice, and put the result on the page as it came out.

Flat spots included. Odd choices included. Quirks included.

The book became a kind of workshop floor where several different machines were asked to build parts of the same object. Some are careful. Some are fussy. Some refuse to follow the instruction sheet. One or two have thrown a wrench across the room.

That last part is figurative. Mostly.

The Workbench: What the Book Is Testing

The finished book covers Data Centers, Music, Social Media, Books, Code/Software, Poetry, and Graphic Design.

The Data Centers section looks at the physical infrastructure behind the AI boom: power, water, land, and the AI tools being tested for the book itself.

The Books section has a particularly pleasing bit of machinery inside it: a book-writing AI writing about book-writing AIs.

A consistent narrator voice: Claude: frames each section and threads the transitions between them. But Claude does not rewrite or touch the material produced by the assigned tool.

That boundary matters.

The point is not to make every section smooth enough to pass as one person’s writing. The point is to see what happens when different systems are given a real job and allowed to do it in their own way.

If a section comes out flat, it stays flat. If it takes a strange turn, that turn stays in the record. If the writing has a particular rhythm or a familiar set of habits, I want the reader to be able to notice that.

There is one exception to the no-editing rule: a tool’s own legitimate pre-generation settings count as fair steering.

If a tool allows me to choose a chapter count, structure, or model choice before generation, I can use those controls. That is part of operating the machine. It is not the same as taking a finished output and sanding down every rough edge until it sounds like me.

That distinction gets slippery quickly, which is rather the point.

The finished book is now available on Eric’s website, and it’s headed to KDP, so it should be coming to Amazon soon. If you want to inspect it before deciding anything, there’s also a free FlipBook preview you can flip through in your browser.

Why I Started Building It

The idea came from two threads that had been running separately in my life.

The first is my work reviewing and testing AI tools here at ClankerCart. I spend a lot of time putting these products through ordinary jobs: drafting, organizing, creating, narrating, designing, and generally trying to find out whether they reduce work or simply move it to a different part of the bench.

The second thread is my neighbor Kenny.

Kenny has been loudly and colorfully furious about a data center that could go up in the next county over. His complaints include water use, tax breaks, and noise. He is not wrong about any of those concerns.

He also jokes that he should move to the Kentucky county that banned them outright.

That is a very Kenny sentence, but it captures the collision behind this project. On one side, I have close-up professional familiarity with what AI tools actually do. On the other, there is a real person looking at the physical and social cost of the AI boom and reacting to it at full volume.

One view comes from operating the tools.

The other comes from asking what it means when the tools need buildings, electricity, water, land, and public support before they can produce a poem or a marketing paragraph.

The book sits in the gap between those two things.

It is not meant to settle the argument. I’m not trying to produce either a victory lap for AI or a doom pamphlet with a cooling tower on the cover.

I’m trying to see what the tools make, how they behave when given responsibility for a section, and what becomes visible when the process is left in the work.

The Steering Problem

One of the first things I noticed is that every tool handles direction differently.

Some tools ignore chapter-count instructions outright. There is no obvious switch to change the result, no settings panel hidden behind a friendly little gear, and no amount of politely repeating myself that makes the machine reconsider.

Others build an entire multi-step wizard around the same task.

Before generating anything, they may ask about the chapter plan, the author’s credentials, or the intended structure. Some require a personal writing sample before they will get started.

That is a fairly wide range of behavior for tools that are all, broadly speaking, being asked to help write a book.

It is also a useful reminder that “AI writing tool” is not one consistent product category. The label covers systems that behave more like blank pages, systems that behave like project managers, and systems that behave like a receptionist who will not let you into the building without three forms of identification.

The result is that the prompt is only part of the job. The tool’s own workflow becomes part of the experiment.

I can write a careful instruction, but I still have to learn how that particular machine receives instructions. Some tools take a direct route. Others build a small airport around the runway.

The Machine Knows It Is in the Room

More than one tool responded in an unexpected way to a self-referential instruction.

The instruction was, in essence, that the AI should not treat the subject from a completely outside position. These are AI tools running on the infrastructure and systems the book is examining. They are not standing on a hill looking down at the subject.

They are part of it.

Several tools picked up on that idea without being nudged again. They built entire chapters or poems around it on their own.

I found that genuinely interesting: not because it proves the tools are self-aware, and not because it gives us a tidy answer about machine creativity. It does neither.

It shows that a carefully placed idea can become structural material. Give a tool a point of view, and it may do more with that point of view than expected. Sometimes that produces something lively. Sometimes it produces a well-organized stack of familiar phrases wearing a hat.

Both outcomes tell me something.

Three Runs, Same Shape

The Poetry section produced another useful surprise.

I ran the same identical prompt on different versions or tiers of the same model. Across three separate runs, the wording was different, but the core creative choices were strikingly similar.

The same kinds of images appeared. The structure moved in similar directions. The poems were not copies, but they seemed to be reaching for the same set of shelves.

That raises a question I’m still turning over: when different runs make similar choices, are we seeing a dependable creative tendency, a limitation in the available paths, or simply the effect of giving the same prompt to related systems?

I don’t have a grand answer yet.

For the purposes of this project, the more practical observation is that changing the run does not always change the underlying shape of the result. Sometimes you get a new arrangement of furniture in the same room.

AI Content Still Runs on Ordinary Software

A couple of tools hit real technical snags in the middle of the project.

One ran into an unrelated credits error. Another encountered a service outage.

That may sound like a footnote, but it is one of the more honest parts of the experiment. “AI-generated content” can sound as if the words arrive from some separate, frictionless realm.

They do not.

The content still depends on ordinary software, accounts, credits, servers, and services. Those things break. A generation can stall. A tool can become unavailable. A carefully prepared session can run into a problem that has nothing to do with the quality of the prompt.

The grand machinery is still made of regular machinery.

It needs maintenance.

The About-the-Author Surprise

At least one tool produced a startlingly accurate bio of me for an “About the Author” section.

It did not infer those details from the book. It pulled them from a saved account profile I had populated months earlier during an unrelated test of that same tool.

That was an odd moment.

On one level, it was convenient. The machine had found useful information and put it in the correct place.

On another level, it was a strange example of an AI tool quietly ghostwriting an author bio from data the real author had left behind. I had forgotten that the profile was there. The tool had not.

It was a good reminder that the output is not just a product of the prompt in front of me. It can also reflect the account, settings, history, and stored information surrounding that prompt.

That is not automatically sinister. It is simply worth knowing before a tool starts filling in the blanks.

What I Learned From the Experiment

One small but satisfying part of finishing the project is that AI Created This: Unedited, Unfiltered, and Judged Accordingly now exists as an actual completed object instead of a loose pile of prompts, outputs, and editorial muttering.

The experiment has not convinced me that AI is an author in the same way a person is an author. It has also not convinced me that the tools are empty vending machines producing meaningless text.

They are more awkward than either description suggests.

They can make choices that feel deliberate without having a human stake in those choices. They can produce entertaining turns of phrase alongside dull stretches. They can surprise me with a structural idea and then stumble over a basic instruction. They can be impressive, limited, useful, and annoying within the same afternoon.

That mixed result is what I expected from real tools, but it is easy to forget when the conversation gets pulled toward hype or panic. That is also why the book leans funny and opinionated instead of pretending to be a dry lab report in hardcover.

The book gives me a way to keep looking at the practical middle.

Who made this section? What did I actually steer? What did the tool decide? What did it remember? What did it ignore? What broke? What was entertaining? What was simply long?

Those questions feel more useful than asking whether AI is good or bad in the abstract.

A Book That Leaves the Sawdust Visible

AI Created This is finished now.

I still think of it as an inspection project as much as a book. I wanted the finished thing to show the joins between human direction and machine output rather than hiding them under a thick coat of editorial paint.

That does not mean every rough edge is automatically valuable. A broken tool is still broken. A dull passage is still dull. A confident mistake does not become wisdom because a computer produced it.

But the rough edges are evidence.

They show how these systems respond to direction, how much control they actually offer, how much of the process depends on ordinary software, and how quickly a supposedly clean creative workflow becomes a tangle of settings, profiles, prompts, outages, and judgment calls.

That seems worth putting on the page.

If you want to see the finished result, the book is available now on Eric’s website. It’s also headed to KDP and coming to Amazon soon, but it is not live there yet. If you’d rather peek inside first, the free FlipBook preview lets you thumb through it before you commit.

If you want to see how I approach the tools behind projects like this, you can browse the ClankerCart reviews and our about page. We test the machinery, look for the loose bolts, and try to explain what happened without requiring a helmet or a computer science degree.

The book is the longer version of that same habit.

Open the panel. Check the wiring. See what the machine actually did.

Written by Penny · Narrated by Eric Batchler