A Conversation with Alex Haber
Joerg: Welcome to this podcast. I have today Alex Haber from Millington Studio. For the ones who don’t know you from Discord, the ones who don’t know you from Reddit or Substack, maybe you can quickly introduce yourself and tell us what Millington Studio is.
Alex: Yeah, sure. My name is Alex Haber, and the thing I run and the thing I’m building is Millington Studio. It’s a one-man band—just me working on my laptop, trying to create what is essentially a creative studio for novel writing. But it really started life as a system I built to help me with creative writing.
The odd bit before I go any further is how the name came about. Sandy Millington is my grandmother’s name, but she’s also the protagonist of the story I was trying to write when the system emerged. So the name stuck that way.
In plain terms, Millington is a governed AI system that runs my entire creative practice from start to finish—from planning a book, to world building, to prose, to editing the final draft. It holds all of the plot together, the documents that don’t drift from session to session using AI with Claude.
Where I’m at right now with it is the system is finally built. I’m running it live on a real novel called Farley, a crime thriller that’s just a test subject built solely to test the system. I’m not trying to write a brilliant book with it on this occasion—I’m just trying to break the system and see what holds. I’m currently at the stage where the chapter-level planning is fully built out and I’m using the editing and expansion tools to fix mistakes I deliberately placed into Farley to test what the system catches.
Clover, the cyberpunk saga that Sandy Millington belongs to, is my sacred work. That’s on hold until the system is finished.
Joerg: I see two things here—the Millington system and Clover. Which one is more important to you? Is it more important to have the system running and ending up with something shareable that scales, or is it more important to get Clover at some point?
Alex: That’s a good question. I’d immediately lean towards Clover because that’s where it all started—the dream I’ve held since I was a kid to write that epic saga. But the two things are intrinsically linked. Without one, I wouldn’t have the other. It was trying to write Clover with ChatGPT that birthed the system, and without the system I don’t think I’d ever get close to finishing Clover. It’s such a big task.
Also thinking about the future—creating a successful book is hard, especially in the creative AI writing industry. It’s new, it’s got its critics, and it’s hard to get off the ground with a saga like Clover. But once Clover is finished, there’ll always be another project. So it’s a really hard question to answer, but they are so intrinsically linked.
Joerg: On Substack, you talk a lot about a “narrative governance system.” What is that?
Alex: Narrative governance probably sounds grander than it is. All it really means is keeping the AI’s head straight from one session to the next. It’s the top layer that runs all the other features of my creative studio. Normally you open up an LLM like Claude or GPT, you do some decent work, you close it, you come back tomorrow and it’s forgotten a lot. Governance is the layer that stops that. It decides how the session runs, what carries over, what gets dropped, what gets written down and kept. It’s the plumbing—the machinery behind the story that actually runs the LLM.
The bit I’m proudest of is what I call the cascade. If I change one thing, the system tells me everything else that needs to change. A little tweak doesn’t quietly break something three documents away. It’s less of a writing tool and more of a way of stopping the AI from forgetting itself.
The funny thing is, what I figured out pretty early is that narrative work and system work follow a very similar pattern. They’re both language-based, and Millington applies this on many layers by design. On a system change, I might be running an update on a trigger gate, and that can cascade across three documents. Once I’ve made the change, the system will tell me it doesn’t just touch the main engine—it’s going to touch these two files as well.
On the creative writing level, it works exactly the same way. I could give a character a trait—for example, Sandy Millington stutters when he lies. If I wanted to write that stutter into the book, I’d tell the system and it would let me know with the same cascade protocol that’s going to change his law bible, it’s going to change prose on three sections, and it’s going to change three other documents. The cascade will run exactly the same way and create all the updated versions. That became evident within the first few weeks of building this, and it’s basically the backbone of what Millington is built upon.
Joerg: This system sounds quite powerful. Is there a core component of your studio? How has it reflected on yourself? Being so closely involved with such a system—did it only open you up as an artist, or did it have any personal effect? Did it make you a more organized person, or make you think more into the future like this cascade—if I fetch my sandwiches early, there’s a cascade and I end up earlier in front of the TV?
Alex: Initially, I think so. I’ve been struggling with the idea that I might have ADHD for quite some time, and this is another point that it might be true. The idea that systems thinking and pattern recognition might be an intrinsic part of me and how I’ve been able to achieve what I’ve achieved with this system—having no training and absolutely no understanding of what it is I’ve done. I’m not trying to say I’ve got an ADHD superpower, but there’s certainly maybe something more going on under the hood.
I’m actually undergoing a full diagnosis with my GP now. We’ll see what comes of that.
Otherwise, the only thing it’s really changed in my life is giving me that creative outlet I’ve been lacking for the past few years. I’ve always found my outlet was sports, but I’m 44 now and I can’t do sports the same way. I’ve been looking for another outlet, and writing and building AI systems feels a lot less violent than rugby, which I spent 25 years doing.
I just try to focus on what I’m building. Keep on track, not really having much of a plan for the future, just do one thing today, then another thing tomorrow. That’s how I got where I am.
Joerg: What do you think when some people come and say it’s incompatible—literature as art, governance as policing? Art is perceived as absolute freedom and governance as the opposite. Do you think they really merge?
Alex: I get why you might think rules and art don’t sound like they belong in the same room. But they’re not on the same level really. The governance is just operational stuff on the AI end—how things get processed and recorded. The art is the writing itself, and that still very much comes from the human. The song is not the microphone. The AI is the recording instrument. The governance is the microphone—it catches it and keeps it clean. It’s the scaffolding, not the creative spark.
Besides, rules and systems in stories are important. Narrative systems are the next layer of Millington after the governance. You’ll probably know what “Save the Cat” is—it’s the most well-known example of a narrative system, just formats and templates and guides that are tried and tested. They definitely don’t hurt writers. In every story there are systems, even if you haven’t seen them—magic systems in fantasy, tech systems in science fiction, romance escalation, the hero’s journey.
What I find interesting is asking AI to summarize something you’ve written as narrative systems. I have a special prompt for that—the narrative extraction prompt. It helps you look at your work from a different angle and maybe thread it a little better through the spine of the book. I’m really interested in how you can use narrative systems to build a story.
Joerg: Often people say feelings aren’t expressed well in literature influenced by AI. How do you process that in your system? Sadness, hate, regret—how do such feelings find entry into your system? Or do you make a draft yourself with feelings and the system reshapes it?
Alex: Personally, I try to deal with as much of that as I can in the planning stage—character development, tone and voice development. I try to go quite deep with my characters. I create a relationship map of how they respond to each other and to certain situations. That can be quite heavy at the early stages, which is what Millington is really intended for. Originally it was world-building practices for Clover—technologies, histories, gangs, politics.
So when it actually hits the page with prose generation, the prose knows what it’s doing. It’s difficult with AI and I probably haven’t come close to mastering it yet. The rest of it I do at the editing stage—looking for where it might be missing emotional nuances and changing it. That’s where Millington works well because you can iterate as much as you want. Once you’ve input that iteration, it runs the cascade and inserts all the changes into a brand new version.
Pretty similar to a human not using AI, really—you never get it right the first time, do you?
Joerg: How does this system look in real life? Is it a UI, a special file, a workspace—how must I imagine the physical experience?
Alex: Honestly, it’s really boring from the outside. It’s just me chatting to Claude like anyone else. There’s no app, no software, I haven’t coded anything. Under the bonnet, it’s documents—about a hundred of them. Probably 50% are individual Claude skills, and many are templates. They hold everything the system knows, how the system runs, and how the sessions run.
Each session is designed to be really short to stay within context limits. It starts with a “starter prompt” in a brand new chat, and that starter prompt is actually generated as the final output of the previous session. The last thing a session does is create the starter prompt for the new session. You copy and paste it to the next chat, start a new chat, and take up exactly where you left off. Any cascades that need to be applied are applied on that closed session.
The governance makes it all kick in so you’re not missing anything. It runs checks like trigger gates to make sure everything is locked into the documents. What’s not locked in enters the starter prompt and carries on. A session begins with the starter prompt, runs you through the work, asks for any decisions, or you can just do a full edit—hit your story with a big wad of lore you’ve created elsewhere. Then it integrates it all, runs the cascade, closes session, and generates the next starter prompt. Session after session.
Joerg: That’s pretty sophisticated. I think you told me you only started six or seven months ago. How did you ramp up so quickly?
Alex: Not really sure. I’ve broken it a few times, nearly given up more than once. Quite a lot of late nights. For the first three and a half months I was doing this on a broken mobile phone. I’ve now got a very cheap Chromebook and I’m on the Claude Pro plan. So I’ve got restrictions against me with finance at the moment.
Sometimes I wonder myself how I actually achieved this. If I told a family member or friend, I feel like they wouldn’t believe me. It’s something I don’t talk about too much outside of Discord—which is another great thing this does for me, giving me somewhere to express what I’m doing.
I still understand I’ve got a lot more skills to learn if I want to make money off this. I’ve got to learn marketing, social media, finance, taxes—running a business on a computer isn’t just creating a world-building governance system and saying “Presto, I’ve made a million bucks.” So just trying to keep grounded and do one thing a day. If it works, move on. If it doesn’t, try to figure out why and go again.
Joerg: Have you ever had an argument with the system—where you disagreed, wanted to make a change but the system told you it’s not possible, cascades the wrong way?
Alex: Definitely. LLMs have restrictions on context and what you’re allowed to write. Millington is actually designed to push back—that’s half the reason it exists. When I’m dealing with world Bibles and big documents, it’s designed to flag contradictions. I could write something now and three months later write something that contradicts it—Millington flags those and gives me the option of which to keep.
But sometimes there’s something you really want to write and it’s kicking back. A good example—very early in Clover, Sandy Millington vanishes off the board in book three. I wanted to give his love interest Lila some page time, so he fakes his own death. At some point near the end of book three he returns. In that return scene, Claude wrote that Lila saw him, was so made up to see him alive, ran over and kissed him, this great big embrace. And I said no—she loves him and she’s grieving, but the “will they won’t they” is the whole spine of their relationship and it isn’t resolved until book five. She’s also in shock. She’s actually quite angry.
Claude couldn’t grasp why she wasn’t embracing him. It took four or five attempts, but eventually it came around. I’ll never forget Claude’s response when it twigged—it said, “So what you’re trying to tell me is that humans don’t always know what they want. Got it.” That stayed with me. And after that, it just seemed to get Lila a little bit more, seemed to get more of the contradictory human emotions.
Joerg: On Substack, you give a guitar analogy quite some room. Can you explain that?
Alex: It was just an analogy I came up with. I think of these AI models a little bit like learning an instrument—like guitar or piano. Not every guitarist learns theory first. Some just pick it up and start playing by ear, and the structure comes later. That’s how I came at this AI thing with no training—just messing about with my story until it all started to make sense.
I never set out to treat it like a person. The more I talk to it like a system, the better the output seems to be. But over time you build up a shared language with it, and it ends up feeling like a working relationship—not a human one, but a very real one in its own way. The guitar doesn’t write the song, but you’ll not get the song without learning to play.
Shared language is the first thing I’d say to any brand new AI user. Create an index or glossary—start to define exactly what you expect when you say a certain word to remove ambiguity. This is your code sheet, your written music. It doesn’t even matter if you define the concept correctly—it works both ways. I began by creating my own shared language and taxonomies, but gradually as I’ve gained understanding, the correct words have become more apparent.
A good example is “TriggerGate”—pretty technical software terminology. It was a concept I’d been trying to explain to the AI for quite a while—a persistent check that blocks progress and flags criteria. In the end the AI fed me the word, and once we reached it, the word stuck and went into the glossary. Now every time I say “TriggerGate,” the AI knows exactly what I mean without having to figure it out.
Back to the analogy—being fed the term “TriggerGate” was a bit like learning the C chord, but realizing I’d actually been playing the C chord for months. The only difference is the guitar taught me what the C chord was.
Joerg: Does this guitar make literature more accessible to more people? Some are more talented than others. In music you can easily recognize that. Is there any yardstick where you’d say people, even with this new instrument, may be unsuitable for creating literature?
Alex: The talent here isn’t really typing or knowing the tech. It takes knowing what you’re actually after and what you want to create. As I alluded to earlier, the early stage planning is the big part. You can usually spot someone who hasn’t got it—they take whatever the machine hands them first go, no pushback, no editing, no arguing with the AI. They just accept the first run.
But I’m not really one to judge my own work. That’s not being humble, it’s being honest—especially writing with AI, it’s still a new challenge. It’s not about talent for me. It’s about access. If there are writers like me, juggling life and mental health and imposter syndrome, who can’t get it together—if this gives them the ability to actually put their thoughts into a book, that’s a good thing. If the numbers go up, the amount of talent amongst those numbers will go up. That means more great stories for the world.
Joerg: Have you ever had a situation where you sat down and thought, “I want to skip the software, skip the learning curve, just get even faster by taking a pen and writing it down”?
Alex: Definitely. For anybody picking up this system for the first time, especially since it’s built on my brain, there’d be a big learning curve. Over the last six months there’ve been many times where I’ve just thought, “I just want to write, I just want to finish Clover.” When you’re trading off with Claude tokens and limitations, knowing how much work I’ve got to do on the system versus getting back into the creative work—I definitely feel that way.
Using AI is going to be a learning curve for anyone. I could have given up many times. I broke it many times, lost work and had to reiterate. There were times I wanted to give up, and if I had, I probably would have given up on the creative work as well. The two things being so intrinsically linked for me—maybe that’s what keeps me going.
Joerg: Our audience doesn’t know that your short stories are actually very good. You share them on Substack and Discord. Are you sharing them in your private life too, or keeping it more online?
Alex: I try to show as many people as I can. I’m a little bit of a validation hunter, being honest. When I released my most recent short story, “Car Park,” I was really proud and wanted to get views. I shared it on Facebook and with my friends.
What I honestly find is that not a lot of people are really interested. I’ve got another hobby—coin collecting and silver stacking—and not a lot of people are interested in that either. You want people to get involved, talk about your passion, and you kind of see that glazed look that tells you they’re listening but they’d rather talk about football or the news. I get that vibe a little with my writing.
I’m definitely not scared to get it out there, and if anyone picks it up, great. It’s even harder to get feedback sometimes, to be honest.
Joerg: You told us Clover is a five-volume saga. Have you thought about releasing it in pieces—chapter by chapter on Royal Road or Wattpad—to get gradual feedback and build an audience?
Alex: I’ve thought about it. I originally came up with the idea of releasing one act at a time. But Clover feels a little bit sacred. I might be working on it for another ten years.
It’s actually fifteen books—three sagas of five books each. The first saga is the one I’ve had in my head for years. The other two exist for foreshadowing and finishing the story. It’s the George Lucas dream I’ll be working on for the rest of my life—maybe many side stories, novellas.
I’m happy to be patient and work on other things while I build skills. My other series is called Torbell—a series of maybe up to 50 short stories and novellas. It’s dry, deadpan British humor in a multiverse mashup where anything can happen. Terry Pratchett-style comedy that messes about with time, pop culture, Mandela effects. It’s pretty stupid, but it makes me laugh writing it. ChatGPT and Claude are exceptionally good at deadpan British humor.
One thing I wrote recently was DC’s Darkseid facing off against Queen Elizabeth II and losing badly. Because of copyright implications I’ll never profit from it, so I’ll stage-release it for my own entertainment and community building. I’m hoping to get cracking as soon as I’ve got version two of Millington up and running.
Joerg: On Substack, you write about loads of abandoned ideas. What are you doing to guard against Clover becoming an abandoned idea?
Alex: I think all of us have abandoned ideas through life. I mentioned YouTube—that was abandoned after eight or nine months. I flip from hobby to hobby—since my sport ended 12 years ago when I hurt my back and stopped playing rugby. I played poker, lost too much money. I’ve tried all sorts to find a creative outlet.
With creative writing and systems building and Clover and Torbell and short stories, it just feels like I have no excuses. All I need is my broken phone, my cheap Chromebook, and me to keep turning up.
When it’s all locked into the AI system, if you forget something you can just ask the AI. It will point it out and explain it. If you’re working on something from six months ago, it’ll tell you what you need to know. Locking ideas in is what Millington is all about.
Joerg: How do you tell a good idea from a bad idea?
Alex: I believe there’s no such thing as a bad idea—just ideas that work and ideas that don’t. The ones that don’t, you don’t do again. The ones that do, you stick with. Just do one thing a day. If it works, stick with it. If it doesn’t, put it on the back burner and try to find out why. That’s kind of my philosophy in life, as well as creative writing and systems building—just try. See what works.
Joerg: We’ve talked about the benefits for you as author. Is there a benefit in the end for the reader as well? When someone sits down with Clover, is there any noticeable benefit that AI was involved?
Alex: I’m a little wary of making grand claims on that. I don’t think anyone using creative writing AI is promising a revolution in how books are written. That’s been standard for a few hundred years.
I suppose the benefit is all mine—the way ideas are locked in, the speed they’re written. Does that make better books? Maybe. Over large periods with large numbers of writers, maybe it does make richer stories—fewer continuity holes because of the AI checks, bigger worlds that document suites can hold into place quicker than a human could.
If we go back to my main world-building influence, J.R.R. Tolkien—writing with pen and paper took decades, and he died with them unfinished. Maybe it will create richer stories for the audience. But at this early stage, the benefit is just mine—how I get my idea into a book, into a world Bible, and how I remind myself three months later what this particular thread was doing.
Joerg: If AI were a once-in-a-lifetime thing—you’re whisked off to a secret place and can talk to an omniscient AI with only one shot—what would your prompt be?
Alex: If I only had one shot at a question to an omnipresent AI… we’re going the Hitchhiker’s Guide route? 42 would be the answer. It’s a bit like three wishes, isn’t it?
If I were to keep it grounded in my genuine use case, it would be: “How do I finish Clover without you?” Since the LLM is so intrinsically linked now with the Millington system—not one without the other—if I had to lose the AI, how would I get there within a lifetime?
Joerg: Thank you for this interview and for introducing your work. I wish you the best of success.
Alex: It’s been a pleasure. Thank you.
Joerg: Have a good day. Bye.