Thoughtful Thursday · AI Pro Writers Studio

🔥

Thoughtful Thursday

Forty-Seven Tabs Later: Learning Curves, Rabbit Warrens, and the Long Way Around

So What Is a Learning Curve, Anyway?

We all say "learning curve" like we know exactly what it means. Most of us don't, and I didn't either until I went looking. (Yes, that's another tab.)

It goes back to a guy named Hermann Ebbinghaus in the 1880s, who memorized piles of nonsense syllables and tracked how fast he learned them and how fast he forgot them. About fifty years later, an aircraft engineer named Theodore Wright noticed that every time a factory doubled the number of planes it built, the cost per plane dropped by a predictable chunk. Practice made everything faster and cheaper, just never at a steady pace.

Put either one on a graph and you get the same shape: a big jump at the start, smaller and smaller gains after that, and long flat stretches where it feels like nothing's happening at all.

Here's my favorite part. When people say something has a "steep learning curve," they mean it's hard. On the actual graph, steep means you're learning fast. We've been saying it backwards for decades and nobody cares. I kind of love that.

Why Isn't It Just a Straight Line?

If learning were a straight line, every hour of practice would be worth the same. Your hundredth hour on the guitar would teach you exactly as much as your first.

Anybody who's ever picked up a guitar knows that's not how it goes. Your first hour teaches you where your fingers go, which is a massive leap from nothing (the second hour your fingers hurt). After that, every new thing has to sit on top of the last thing, so you can't move forward until the basics stop feeling like work. And those maddening plateaus? A lot of the time that's just your brain quietly doing its filing, hooking the new stuff into the old stuff before it's ready for more.

So a learning curve is basically what progress looks like if you stay on one road and measure how far you've gone. Which is a problem for me, because I have never once stayed on one road.

Down the Rabbit Hole

So here's how today went.

I got up, grabbed a Diet Coke, and did the half-awake phone scroll. Skool first, then Pinterest, where a bunch of LEGO mini starships stopped me cold. My desk is finally getting back in shape, which means I'm about to dive back into my 11,000-piece Millennium Falcon (it's huge), and those little starfighters have always gotten my attention. Saved a couple of pins.

Right underneath were the usual AI-generated starships, and man, they all look the same. Same sleek hulls, same glowy engines, same familiar shapes. It made me wonder what on earth these things were trained on. I've been trying to make original ships for book covers and art, and that sameness is exactly the wall I keep hitting.

That thought slammed into something that's been stuck in my head for days: a video where someone built a Studio Ghibli–style street scene out of nothing but grey blocks in Blender, then let Claude write the prompt to let AI paint over the blocks and control a minimax video of the scene. The simple structure kept the AI from wandering off. Mind. Blown.

So naturally I had to try my own version. I tossed a photo of a LEGO starfighter into ChatGPT and asked for a photorealistic 3D ship with smooth curves, and what came back was fabulous. Then I asked it to turn the ship into an outline schematic for a coloring page, plus a spec sheet with weapons loadout and scale.

From there I built a coloring book prompt generator for starships. Then I built a generic one that'll take any image, grab its style and theme, spit out four color variations, and turn those into coloring pages.

(Whew.)

And somewhere in the middle of all that, the other tabs in my head started popping open.

Where the Warren Took Me

Pen names. My MarketCrafter plugin, which is still waiting on pieces from two other tools. Scheduling, where I'm leaning toward Metricool but would honestly love for it all to live in one place, soup to nuts, from the first idea all the way through marketing and sales.

And then it happened. I had an epiphany about agentic architecture and why every book-generating tool out there, mine included, is kind of doing it wrong.

There's a line I keep coming back to: writing is an art, publishing is a business. It's about the only positive thing I've taken from Coral Hart, and even that one has issues. A lot of authors write to the market, which means you're not always writing what you love. But being a starving artist is nobody's idea of fun, and most of us have gotten pretty attached to eating regularly.

Which brings up the question everyone in this space keeps tiptoeing around: what exactly is AI slop?

Turns out Merriam-Webster settled it. Their editors picked "slop" as the 2025 Word of the Year and defined it as low-quality digital content, usually cranked out in bulk by AI. The word's history is great. Back in the 1700s it meant soft mud. By the 1800s it meant food waste, and eventually it came to mean anything of little or no value. Pig slop to content slop in about three hundred years.

My own definition is a little tighter: slop is stuff made for volume, where nobody's accountable for whether it's any good. That's exactly why I built quality gates and a running codex into my tools in the first place.

So what if you need to pay the bills, you still care about quality, and you want to automate?

Scotty Art, a creator I've followed for years, built an autonomous agent named Kiera K, and she runs her own TikTok channel. She tries on and sells dresses and accessories. She picks what to sell, decides what to post, makes her own videos, posts them, and chats with her followers. Every so often Scotty checks the bank account and, hey, there's money coming in. Kiera just keeps going, 24/7.

So I started dreaming up my own crew of agents. One studies the competition. One pitches improvements. One builds them. One checks the quality. Others write the docs and training, market the new features, watch the metrics, and double down on whatever's working. Then they do it all again. Every. Single. Day.

Then I pictured a shelf of computers, each one with a pen-name author's photo and name on it, humming away. Then it hit me that I don't need a shelf at all, just a virtual machine for each author. Each one answers its own email in its own voice, runs its own socials, and advertises its own books while the publishing engine behind it pumps out a book a week. Or a book a day. For every pen name I can dream up.

Okay, two honest confessions.

One: part of what's driving this is that I really don't like my day job right now, and a machine that pays the bills while I sleep sounds pretty amazing from where I'm sitting.

Two: when I read that vision back, it's a little too close to the slop definition I just quoted. A book a day per pen name with no human anywhere in the loop is volume with a capital V. And author personas emailing readers like they're real people? That's a trust question I haven't figured out yet. The rabbit hole took me somewhere exciting, but I'm not sure I love everything I found down there.

New land. No map yet.

Then I Crashed.

By 2:00PM I was feeling better, but I was still on day three of being sick, and that idea burned through whatever gas I had left. The second I texted the whole thing to my wife, Amanda, I was out cold for a few hours. I'm pretty sure I sprained my brain.

When I woke up and reread what I'd sent her, I understood less of it than when I wrote it.

This is the part of rabbit-hole learning nobody puts in the highlight reel. Chasing ideas feels super productive. Every new connection gives you that little hit that screams insight! But feeling like you're learning and actually learning aren't the same thing, and some of what I thought I understood was really just momentum.

Real scorecard: when I first got up, I maybe understood a third of the system I'd dreamed up. Now I'm at about half. That's real progress, even if it's a long way from the "I've got this all figured out" feeling I had on my first Diet Coke.

The Weird Part

Later this afternoon, with the a British mystery show going on the TV, I dug out a laptop I hadn't touched in months. The plan was to set it up as a local agent system using Google's Antigravity, which (full disclosure) I have zero idea how to use. So I asked Perplexity how to use Antigravity's agent management feature.

It came back with an outline for using agents to improve my original writing-system tools, laid out almost exactly the way I'd been picturing it all day.

Here's what made it so weird: I hadn't researched any of this yet. No videos. No articles. I hadn't even opened a web browser today. The entire idea existed in exactly one place, a text thread to my wife. And yet I asked a totally different question and got my own plan handed right back to me.

I took that as a sign from the universe that I'm on the right path.

The skeptical-but-intrigued part of me does notice that this idea is floating around out there. Future Fiction Academy put out a video this week about building a "book machine machine." They're getting closer, even if I still think they're doing it wrong. But the timing? Uncanny. I'll take the wink. It tells me I'm not crazy and I'm not alone. It also reminds me the idea isn't the hard part. The half I don't get yet, the logistics, the quality control, the whole persona ethics thing, is where the real learning curve starts.

A Web Made of Curves

So was today a learning curve or a learning web?

Honestly, I think it was both. A web isn't the opposite of a curve. It's a bunch of curves all tangled together. In one afternoon I climbed the steep first stretch of three at once: structure-guided AI images, prompt generators, and agentic architecture. I'm not an expert in any of them, but every one moved.

The trick, as far as I can tell, is tying the threads back together at some point. Open tabs won't do it. Bookmarks definitely won't. What did it for me was texting the whole mess to Amanda, and now writing this. Every time I explain the web, a few more strands connect and a few loose ones fall off.

If your brain works like mine, here's what I'm trying:

Figure out your home question. Mine turned out to be "how do I keep AI original and on track?" Almost every tangent today was an answer to it.

Park tangents instead of killing them. Jot them down for later. The warren's not going anywhere.

Pick a check-in point. Not to stop wandering, just to ask, "Okay, what did I actually bring back?"

Explain it to somebody. Your spouse, a friend, a blank page. If you can't explain the connection, you haven't really learned it yet.

Expect the crash. Understanding less a few hours later isn't failure. It's your brain sorting the real insights from the adrenaline.

Back to the Blueprints

So should we even bother coming back to what we set out to learn?

Sometimes, sure. If there's a deadline, a certification, or something you need to know by Tuesday, the web needs a leash.

But sometimes the detour is the education. The rabbit hole turns into a warren, and the warren opens onto a whole new land you'd never have found by dutifully watching video seven after video six.

And sometimes the web just brings you back on its own.

While that mystery kept playing, I started listening to videos on agentic coding, which led to Adam Savage asking whether the Enterprise makes sense while he studied a cutaway plan of the ship. That made me want to draw a real blueprint of the starship I made a short video about yesterday for another group, and that dropped me onto cygnus-x1.net, digging for Star Trek blueprints.

I started today with LEGO starships on Pinterest. I'm ending it with starship blueprints.

Forty-seven tabs open. One big loop.

I'm calling that learning.

Below are the Lego to image to coloring this started with

4 comments

Thoughtful Thursday