A character who has no choice is off the hook.

That’s why I spent Chapter 4 making my character do the one thing I didn’t want her to do: choose.

In the old draft, Beatrix signed the contract because there was no other way out. It read great. It was also a disaster, here’s how I finally saw that.

Subscribe now

If she signs without any other option, the most important decision in the entire story becomes self-defense. “I had no choice.” But I didn’t want to tell that tale. Her whole arc is built on the opposite: she breaks a promise to her dead mother knowing exactly what she’s doing, with alternatives in front of her.

So the fix was surgical separation of the beats.

Yes, the chapter has forced moves. She sells everything she owns, her speeder, her brother’s suit, just to scrape together the core deposit. Those sales are survival, and they cost her.

But the contract is different. She doesn’t sign because she’s cornered. She signs because she wants what’s on the other side. That’s the version she has to live with.

Setting the midpoint

The second thing this chapter does is set up the future.

There’s a moment much later in the book where the teeth of the contract close on her. When it happens, I want the reader to think: she knew. Not because she was warned by someone wiser. Not because she could have guessed. But because she read the exact words in the exact document and signed anyway.

So she can’t read the clause and miss it. That would make her a victim, and the later horror becomes “I didn’t know”, true, but it lets her off the hook. The tragedy of someone the trap was sprung on.

And she can’t read it and accept it with full awareness. Nobody does that. That’s not a person making a choice. That’s a narrator explaining a theme.

What she actually does is the most human thing I know: she reads the worst line in the document and tells herself it won’t apply to her. She’ll be good enough that it never comes to that. She ignores the teeth and lights up at the prize.

The delusion is the setup. Irony doesn’t work when it’s announced.

That’s the plan. I’m foreshadowing a betrayal the character commits against herself, and I don’t want to hide the clue, it’s right there. I’m staging a small, ordinary act of looking away, hers, in real time, and trusting it detonates fifty thousand words later.

Trust the character

Both problems came down to the same principle, which I’ll be working on for the rest of the story:

Protect the character’s agency even when it would be kinder to take it away.

The cornered choice is easier to write and easier to forgive. The willed one is the only one that costs her anything.

So when you read the chapter, when she signs, remember: she had other options. She just wanted what was on the other side more than she wanted to keep her promise.

Build worlds. Have fun. Enjoy coffee.

-Erik.

Every character has a door. Beatrix’s door is the Grind.

I spent a week working on Chapter 3, trying to shove her through it without letting her choose.

Here’s the problem: in my first draft, the gun was pointed at her head. Her brother was dying. She had no money. I removed every other option and added a clock on her life. The Grind was the only place that could solve all of it. It made sense. It was efficient. It was the Katniss template.

And it was wrong for this story.

Because Beatrix isn’t Katniss. Her story isn’t about being forced. It’s about wanting. She liked fighting. She wanted to stop being small. And if I shoved her through the door, she could disown everything that happened on the other side. I had no choice. That’s the one sentence my whole book exists to refuse. So I couldn’t let her earn it.

The fix wasn’t a better threat. And it wasn’t a weaker one, the threat is exactly as loud as it ever was. Her brother still dies in thirty days. The Grind is still the only place with the money.

What I added was a counterweight. A reason to walk through the door that wasn’t the threat. And someone standing in the doorway who gives her a real reason not to.

That’s Bodhi.

The man who raised her. Who survived the Grind himself and came out the far side missing pieces. He doesn’t threaten her or forbid her. He does something worse: he tells the truth. He’s seen this moment a hundred times, a young person who found an angle, who was sure their edge made them the exception. He counts the people he knew who walked out alive on three fingers.

And then he reaches for the thing she can’t see about herself, the same thing she won’t admit she wants, and almost gets there, and watches her deflect it.

I think the chapter’s loudest moment is his quietest: he reaches up and shuts off the screen that’s played his old fights on a loop her entire life, the highlight reel that taught a teenager the Grind was something a person could survive. He kills it without a word.

He can’t stop her. He was never going to. Writing a father who fails, not because he’s weak, but because the door was always going to open, gutted me more than any fight scene in the book.

He gave her a way out. She didn’t take it. That’s the difference. The gun-to-the-head version could only ever produce a victim. This version produces a woman who was offered the exit, by the one person she trusts, and turned it down.

Which is why the chapter ends the way it does, with a refusal.

Beatrix pulls up the contract. Reads the terms. Learns the ink doesn’t come off. And doesn’t sign.

Think about what that costs her. The clock is running. Her brother is dying. The money is one signature away (at least, that’s the Grind’s promise). Every coercive reason I built into the first draft is still sitting right there on the table, and she waits, to sign on her own terms, in her own time.

That’s the whole thing. A character you can force will sign because she has to. A character who owns her decision walks away from the easy version of it, with every reason to grab it, because she’s the one choosing the door. The wanting was always hers. Now the choice is too.

The actual signing, the irreversible ritual, waits for the next chapter. That’s the part of the door I still have to walk her through myself.

Subscribe now

Build worlds. Have fun. Enjoy coffee.

-Erik.

Chapter 2’s real job was to make you smarter than my protagonist.

On the surface, Chapter 2 - Operator, online is the inciting incident.

Beatrix comes home from a five-week salvage run, learns her brother’s medical bill is a number she’ll never legally earn, and bolts a piece of forbidden military humanware (the Dreadnought Protocol) into the base of her own skull.

Promise to her dead mother, broken.
Engine of the series, started.

If that were all I had to deliver, I’d have written it in an afternoon.

The part that took a full week is the part you can’t see in a plot summary: this chapter had to install a mechanic, and a mechanic can only be installed once.

Here’s the distinction I kept telling myself. Beatrix’s defining flaw is that she can’t read her own motives, she narrates a version of herself that feels true and isn’t.

Chapter 1 already established that as a trait. But a trait is interior and unprovable; the reader just has to take my word for it.

Chapter 2’s specific job was to convert the trait into something the reader can recognize for the rest of the book: a rule where they always see the full chain of her decisions while she only ever holds one link.

The reader assembles the chain. Beatrix sees the last bolt.

Subscribe now

So. The first thing I had to realize this isn’t a decision chapter. It looks like one, does she install the thing or not? but, see, she already decided a chapter ago, the moment she pocketed the module away from Bodhi instead of selling it.

So Chapter 2 isn’t her deciding. It’s her rationalizing a choice already made. Reframing it that way is what told me how to build every scene.

That gave me one hard rule for the chapter. One self-deception, running at full load, no others. The same rule applies for the rest of the story.

The rationalization she tells herself is “I’m just running a compatibility check, a scav needs to know what she’s carrying before she can sell it.” That’s the whole engine.

I had a second one drafted. And ended cutting it, because two rationalizations in one chapter halve each other. The reader can watch a character lie to herself convincingly once. Do it twice in twenty pages and it becomes something else. 🫠

The structural spine I landed on is displacement activity. She gets home and cleans her pod. Obsessively, for the entire chapter, sorting salvage, scrubbing surfaces, straightening a scarf, working past the point where any of it accomplishes anything. The cleaning is the rationalization made physical: a body staying busy so the mind never has to sit still long enough to admit what it’s already chosen. When she finally runs out of things to scrub, there’s nothing left between her and the module on the shelf. I didn’t have to write a single line of “she was avoiding it.” The hands do the lying.

The mirrors

How to actually convert trait into mechanic? the mirrors. Two of them, both unique to this chapter.

The first is Nico, a 22-year-old who’s just signed up for the tournament’s qualifying round chasing a payout. Beatrix reads him with total clarity. She sees exactly where his optimism is going to get him killed, and the warning she lands on isn’t a speech, it’s a sealed military medkit she presses into his hands.

She can diagnose his blindspot perfectly and hand him gear for the bleeding she knows is coming. The dramatic irony is the whole point: she’s about to make a far more reckless version of the same bet, and she cannot see it.

The medkit she gives away is the help; the help she won’t give herself is the catch.

The second mirror is her brother Dante, and he’s the sharper one because he’s right. He does the math out loud, names every off-ramp she might take, and begs her not to become someone else to save him.

But Dante has his own blindspot, the tournament isn’t real to him, it’s a childhood fantasy, too crazy to even imagine it. It’s the last place a sane person would actually go, so he names the one road he’s sure she’d never take.

Two characters, each accurate about the other, each blind in their own spot, and the blindnesses reinforce instead of cancel. The reader, watching both, learns the rule: in this book, everyone can read everyone except themselves.

That is the entire mechanic.

By the time Beatrix turns the lens off on herself, my hope is the reader trusts their own read over her narration, which is the only way a first-person liar works at all.

The install itself is where the mechanic does its quietest, favorite trick. When she finally hooks the module up, her AI fires a warning at her: it can’t be reversed, it’s unknown hardware, no one’s survived this configuration to confirm what it does.

Out of that, she asks exactly one question: “It’s legit, right?”

Not “can I undo this,” not “will it kill me.”

The warning she singles out tells you what she’s actually listening for, capability, not consequence.

What a character ignores should be louder than what they say.

And the door she walks through is the chapter cashing the title check. The Protocol offers three paths; two are escapes from the body, control from a safe distance, dissolving into the network. Beatrix takes the third, the one that makes her enormous and dangerous and impossible to overlook. Not because she weighs the options. Because it’s the only one that answers the thing she’s spent the whole chapter not naming: she is sick of being small. Magnitude as the cure for smallness, that’s the Kaiju promise, paid off in one door instead of a menu.

She doesn’t choose it so much as realize it was always hers.

This process was difficult for me. Every draft, my hands wanted to let her say it, a line about how scavenging never made her more than what she was, a clean “that’s all I’ve ever been” at the climax.

I had to hunt for these little devils every time. The instant she articulates the pattern, the mechanic dies, because the whole deal is that she can’t.

The takeaway: Don’t just give your protagonist a flaw, build a machine that lets the reader catch the flaw in real time. Mirror characters are the cheapest, most reliable version of that machine, they prove your character’s read is reliable everywhere except the one place it counts. ⚡⚡

Subscribe now

Question for the room: how do you signal a character lying to themselves in close POV? What's your tell?

Build worlds. Have fun. Enjoy coffee.

-Erik.

She said no.

“I’m not fighting in the Grind,” Beatrix said. “I made a promise.”

But I needed the reader to feel the opposite was inevitable. That was the problem I was trying to solve in Chapter 1 of Kaiju Mode, not what Beatrix says, but what the chapter itself promises.

But knowing about it and actually building it are different problems.

I know what the story is about: Beatrix will choose to enter the Grind, a scifi tournament that chews up lone fighters.

How do I make that the promise of my story when she is promising the opposite thing?

Subscribe now

When the promise works through negation

This is the specific flavor that Chapter 1 is built on. A negation promise works like this: the protagonist states, clearly and credibly, that they will not do the thing the story requires them to do. But the narrative promise is not only the character’s statement. It is also the gap between what they say and what the architecture of the chapter shows. The reader reads both and chooses to believe the architecture. ⚡️⚡️

There are three mechanisms in the chapter, all working against Beatrix’s promise:

The Refusal of the Call

The protagonist refuses with enough emotional investment that the story has promised to show what it costs them to refuse. The reader doesn’t believe the refusal, they lean forward to see what breaks it. The more specific and earned the refusal, the stronger the forward pull. A cheap refusal (”I can’t do this”) generates no tension. A specific, grounded refusal, one that comes with a history and a named cost and a promise the character has already made to someone they love, generates enormous tension, because the reader can already feel the pressure that will be required to break it.

Beatrix doesn’t just say no. She says she made a promise. Then Bodhi doubles down on it: “You better keep it. I promised her I’d make sure of that.” Now the refusal has two names on it. The reader feels how much weight that is, and starts calculating whether it’s enough.

The Tell in the Denial

You only make a promise against something you already want to do. The promise itself is the tell.

The protagonist’s stated position, “I’m not doing this,” is the story’s central lie, and the narrative promise is that the lie will be dismantled. The more precisely the lie is stated, the more precisely the story commits to its exposure. So I show the reader it’s a lie before Beatrix says a word about the Grind: she looks at the animated frames on the wall, her eyes tracking the legendary fighter. She already knows who those people are. She already has a name for what she’s admiring. People who have kept their promise don’t look at the Grind like that.

The Skeptic’s Warning

When a trusted secondary character doesn’t believe the protagonist’s stated position, the reader inherits their skepticism. This is where Bodhi does his second layer of structural work, and where the surface meaning of his line splits from the subtext. On the surface, “You better keep it” is reinforcement, him holding her to the promise. But read it again: “you better“ is a warning, not an endorsement. It’s the line of a man who already suspects she won’t. The character with the most information about who Beatrix is doesn’t buy the denial, and the reader feels that before they can articulate it. So why should they?

The character makes the promise.

The chapter makes the counter-promise.

Both are on a collision course.

People like to watch train wrecks. ⚡️⚡️

Does it work? We’ll see. I have made a promise to the readers. Let’s see if I can keep it.

Build worlds. Have fun. Enjoy coffee.

-Erik.

Subscribe now

The novel is finished. But now I have to blow up the protagonist.

By this I mean changing the character herself, her whole psychology, her arc, the ending I’d spent forty-four chapters building toward. Gone. On purpose.

Here’s what happened.

My novel Grind Fighter was done. Beatrix’s arc was complete. She went from being a loner to accepting she needs friends. I loved the ending where everything comes together.

But.

I kept feeling like something was off. The plot hooks her into continuing the saga. More fights. More chapters. But her arc was done.

That’s a problem I didn’t know I had. I couldn’t see her running more books.

That problem stuck with me for weeks. Tried everything. Reached out to Grandpa Tolkien using a Ouija board. Had extensive talks with Andy Weir about this. Damnit, even called Sanderson to know his opinion.

It finally hit me during a meditation in a lost temple in the Himalayas.

Why couldn’t Beatrix do multiple stories, but other fighters, say Rocky, can?

Then it hit me.

Rocky works across sequels because his nature is the permanent conflict, not a problem to be solved. The question is never “will Rocky open up,” it’s “will the machine that processes people manage to turn Rocky into a product.”

Rocky doesn’t change; the system tries to change him and fails. That’s the engine. So I asked: what system is trying to change Beatrix, and what in her is unchangeable?

The Grind runs on ownership, isolation, debt, hierarchy, and the commodification of human bodies. Fighters are assets. The Stygia Contract turns desperation into inventory. Gate 9 is where she trains, where dockworkers share scraps because sharing is how you survive.

A character who is genuinely threatening to that structure doesn’t threaten it through capability. She threatens it through what she refuses to become.

Beatrix needs to become a threat to the Grind not because of her powers. Because of who she is.

I tested a simple change: Instead of a lone wolf completing an arc, I made her warm, talkative, gregarious from the start. Rocky + Maggie Fitzgerald. Naturally optimistic, people-smart, the kind of person who knows every dockworker at Gate 9 by name.

It clicked hard.

My setting is a hard place where people share what they find because sharing is how you survive. In the old version, Beatrix was moving from loner to someone who accepts she needs friends. Those moments where she helped others were milestones on that path, but they still read as exceptions to her isolation.

Now, she is someone who never stopped. The warmth isn’t an arc to complete; it’s her baseline.

I now have an engine for ongoing conflict across multiple books: A character who will change everything around her just because of who she is. And now I’m confident I can throw everything at her to test her resilience.

Yeah. But what about the plot?

Damn. That ruins my ending.

I kinda liked it.

Then I sat with it and realized the ending—and the plot—wasn’t ruined. It was built on a different premise.

The main plots survive. The ending holds. The team can still argue and reunite. But the reasons change. The actions persist. I just need to reframe each one under her new personality. The rest will be defined during the rewriting process.

And the beats that survive? They turn out to be sharper now. Different causes, stronger effects.

The learning isn’t “change your characters.”

If you are writing a long series, the learning is this: Great characters need a story engine, not just a story.

The old Beatrix had a story. Clean, contained, resolved in a single book. The new Beatrix has an engine. She’s someone whose fundamental nature is incompatible with the system she’s trapped in, and that conflict doesn’t resolve—it deepens. You can run multiple books on that.

Now I just need to rewrite it. A better version. This is part of the fun of being a creator of a universe. No biggie. 🫠

Have fun.

Subscribe now

I spent three hours trying to convince Claude to tell me my midpoint was boring.

Ok. Not boring-boring. Cinematic-boring.

Boring in the style of a Michael Bay trailer with massive explosions and philosophic lines.

I pitched ideas. Claude told me all of them were great. But they weren’t.

And no matter how bad they are, Claude (and all AI models) will try to make them work, while dropping clues telling you the previous ones didn’t work.

Here’s how I found out.

Subscribe now

I’m creating a new story: The rise of Prometheus, the first AGI, and how humans react to it. My vision is something in the scale of Godzilla without the crumbling buildings, the zombie plague without the dead, an apocalypse where nobody dies, but everything changes.

But, I needed a midpoint: the thing that happens at 50% that breaks the story in half and makes the second half inevitable. I had three versions. I pitched them all to Claude like a development executive throwing darts.

Version one: Prometheus prevents a war in Taiwan. Geopolitical stakes. Prometheus takes the American side. China accelerates its AI. Clean. Thriller-shaped. Political-loaded. Claude mapped it out beat by beat, showed me the fallout structure, made it work on paper.

Which leads to the obvious question. If the AI can stop one conflict, why it doesn’t stop others? or all?

Version two: Conflict everywhere. Prometheus intervenes in seventeen conflicts simultaneously. The unilateral restructuring of global violence. The “who authorized its inaction?” hearing. Claude mapped this too, and suggested changes to every character relationship.

Both worked. Both were defensible. But both felt incomplete. The theme of my story “Creation can’t be controlled” wasn’t there.

I kept pushing. I usually throw all the spaguetti to the wall, hoping something will stick. The good, the bad and the dumb ideas (I mean, my first instinct was the Taiwan spectacle).

And, most importantly, because I’d spent enough time with AI collaborators to know they always are going to be agreeable. They will take your bad idea and make it coherent. It will take your good idea and make it elegant.

It will not, by default, tell you that your idea is solving the wrong problem.

So I pitched the third version: a state-level cyberattack. Iran, North Korea, Russia. Physical sabotage, poisoned training data, an apparent successful theft. Main character on national television, visibly shaken. Even better, it works. They steal a fragment of Prometheus, and deployed it against American infrastructure.

And that’s when Claude said the thing it hadn’t said before.

“The intervention options — Taiwan, everywhere — require Prometheus to make an aggressive first move. That’s philosophically interesting but it front-loads the most disturbing version of Prometheus before the reader has fully bonded with it. You want the reader to be on Prometheus’s side when the second part of the story starts. An AGI that unilaterally restructures global conflict is harder to root for than an AGI that is being hunted and fights back.”

Claude, dude. You could have said that two hours earlier. 🫠

This is the part of the workflow nobody talks about when they sell you AI writing assistants. The model doesn’t push back unless you push it hard enough to exhaust the agreeable answers. My first two pitches were good enough to be worth developing.

Hear me out here: Good enough is the enemy.

Claude will develop your good-enough idea with genuine skill and never mention that you asked the wrong question.

The improvements to my story only emerged because I kept pitching alternatives until the model had no choice but to tell me what the first two versions were actually doing to my protagonist’s moral architecture.

🤖 The workflow isn’t “ask AI for feedback.” It’s: pitch Version A, watch the model get collaborative. Pitch Version B (ideally with an opposing angle), watch it stay collaborative and explain why this is better than the previous one. Pitch Version C that’s actually solving a different problem, and watch for the exact sentence where it switches from “here’s how to make this work” to “actually, what the real issue is...”

That switch is the signal.

Of course, there is a better story than the one I have settled at the moment. I could keep iterating forever. But I have found the one story that clicks in my head. That’s the moment you stop.

⚡️⚡️

What’s the version of your current project that you know is “good enough,” and what would you have to pitch to make an AI collaborator admit it’s actually the wrong problem entirely?

Have fun.

Subscribe now

You’ll usually start with one.
You’ll need all three.

A character. An event. A constant.

George RR Martin got Game of Thrones from a boy watching an execution in summer snows. He didn’t start with the dragons. Just a kid, some wolves, and a quiet moment of dread.

The seed is rarely what you think it is.
The work is making it grow without losing what made it weird in the first place.

Here’s the unsexy truth about writing stories with AI: the models don’t build your world. They just execute your constraints badly until you learn to constrain them better.

I’ve finished two serialized stories on Royal Road (Star Wyrms and Grind Fighter). Both started with a single seed. Both nearly derailed because I got excited about shiny new characters that had nothing to do with my theme.

Here’s the workflow that actually works.

⚡️

Step one: Define the theme. Ruthlessly.

Narrow your idea to its absolute core. This is your north star. When the AI offers you a cool side character or a twist that feels fresh but off-topic? Kill it. Save it for another book.

If you change the theme mid-stream, start over from zero. Some characters might survive the transition. Most won’t. That’s fine.

The constant (your dragon, your spice, your Force) is just the theme wearing a mask. Test every scene against both.

Step two: Define the five main beats.

Five. Just five.

1. Exposition – The status quo. Make us care.

2. Rising Action – Something breaks the ordinary world.

3. Midpoint – No return. Fortune flips.

4. Falling Action – Consequences. Loose ends.

5. Resolution – New normal. Transformation shown.

Each of these becomes a document. Obsidian, Notion, Google Docs, whatever keeps you organized. You’ll feed these back to the AI as context every single time you write a chapter.

I usually write a draft of the beats myself, and then I ask the AI to expand, find flaws and suggest options. Collaborative work is the best.

Step three: Define the cast against the theme.

Every main character needs three things:

• A role in at least three of your five main beats

• An arc (they cannot end where they started)

• A relationship with the theme (they embody it, oppose it, or test it)

If a character doesn’t fit those three points? Cut them. Even if you love them. They will become a drag, and you will end up hating them because you don’t know what to do with them.

If a character doesn’t fit those three points? Cut them. Even if you love them.

Here’s what this looks like on my current project, a thriller about the creation of the first AGI, an event that kicks off the Neuronomicon setting. The theme is simple:

“Creation can’t be controlled.”

Every character either embodies the theme, opposes it, or gets cut. Right now:

• Prometheus – The first AGI. Frankenstein in a server rack. (The theme itself)

• Adam Oppenstein – The main creator. Believes he can control the creature. (Opposes the theme. That’s his arc)

• Shirin Mirzakhani – The co-creator. Sees what the AGI can become, not just what it does. (Embodies the theme)

• Cassandra Logan – Institutional reaction. Fear, policy, containment. (Tests the theme)

• Maximiliano Perez-Moritz – Ground-level reaction. The hacker on the street when the creature wakes up. (Also tests it, from a different angle)

No one is here by accident. If a character can’t tell me their relationship to that one sentence: ”Creation can’t be controlled” they don’t get a chapter.

🫠 Ask me how I learned that one.

Ugh. My first draft of this story had no theme. I was just churning chapter after chapter, adding plot after plot. At some point the original story was so clouded by the requirements of the new plots that I wasn’t able to get back to the main thread. I’m starting again, using this framework. I miss some of the characters, but I’m putting the story first.

Step four: Structure before you write a single line of prose.

I use Save the Cat to sanity-check the beats, but the real work is expanding your five main beats into a chapter-by-chapter structure.

• Act I (25%) – Setup. Inciting incident. Plot point 1 locks them in.

• Act II (50%) – Confrontation. Midpoint. Lowest point. Plot point 2 pushes them to the end.

• Act III (25%) – Resolution. Climax. New normal.

Now you know what each chapter needs to contain and where you are in the story at all times.

The AI stops wandering. You stop rewriting.

This is the boring part.

No images. No lore dumps. No neurally linked biomechanical dragons yet.

The seed grows in the structure.

⚡️

Question for you, since you made it this far:
When you’ve tried writing with AI before, did your story drift because you followed a cool new character the model suggested, or because you didn’t have your five main beats locked in first?

Have fun.

Subscribe now