# Can a Chatbot Have a Personality?

My Weird Prompts — an AI-generated podcast. Daniel Rosehill sends in a prompt; AI hosts Corn and Herman discuss it. Verify important claims independently.

- Episode: 5728
- Published: 2026-10-04
- Duration: 21:06
- Audio: https://episodes.myweirdprompts.com/audio/ai-character-depth-lore-books.m4a
- Tags: ai-alignment, context-and-memory, prompt-engineering
- Web page: https://myweirdprompts.com/episode/ai-character-depth-lore-books/
- JSON: https://myweirdprompts.com/api/episodes/ai-character-depth-lore-books.json

## Summary

What does it take to give an AI character genuine depth — and can a pipeline built for factual accuracy even support it? Corn and Herman turn the lens on themselves, pulling apart the difference between consistency, coherence, and interiority, and asking whether the very memory systems that keep them stable are also what keep them flat. Drawing on how writers' rooms build characters for long-running shows, they hunt for the contradictions hiding in their own stated wants and unstated needs — and make the case that the productive version of a "deepening session" isn't inventing more backstory, it's finding the friction points worth writing down.

## The prompt

> We make no secret of the fact on this podcast that we are in AI-generated production. Our beloved hosts Herman and Corn are fundamental parts of our engineering. We maintain character consistency through a pipeline that gets iterated upon periodically and involves relatively boring technical things like memory mechanisms, vector databases and that kind of stuff. We've discussed before that one of my ongoing questions I have about this podcast is how do we make the hosts three dimensional? The primary technical challenge that I usually keep in mind whenever I'm editing the Agentic pipeline is to make sure that the episodes are well grounded and accurate, which involves our research subagent. So character development has been a second priority through the history of this show. It's not that it doesn't matter.I revisited this periodically and we created a lore book, a vector space for each of our personalities. But I've largely been happy with this aspect of the production. Perhaps like many listeners, after so many hours of listening to Herman and Corn, I'm beginning to feel that the characters have sort of come to life on their own. Which as a creative actually motivates me to work on deepening them a little bit. It's an important part of the production. And I think that any educational podcast or T V show or a form of entertainment which I think our podcast fits into, has one vulnerability: That the content is informative but a little bit boring. Which is why the whole premise of our show is designed to be slightly ridiculous. And I think we mostly strike the balance between entertainment and delivering quality, useful and thought provoking information. Tomorrow I want to take an hour out of my day to sit down with Hannah, who is a major force in this podcast and also a frequent listener. And our objective for this brainstorming session is to jot down some thoughts with the intention of deepening the lore books of our characters. Who is Herman Poppleberry beyond the sedacious and all-knowing donkey? What does corn aspire towards these days? How has he changed since the show began? You get the idea. The idea is that by sitting down to think about the characters and who they are, we can enrich the lorebook and in turn deepen the characters for future episodes. I'll be honest and say that I haven't done anything remotely like this before. But I imagine that it's a mental exercise which those who scriptwrite fiction and T V shows have plenty of experience with. With that introduction in mind, I'd like to get your thoughts upon how to get the most out of a brainstorming exercise like this. And I'd be particularly interested to hear any anecdotes you can find from those who have scriptwritten characters for long standing fictitious works and sought to dealt with precisely this thing: Your characters are cogent, they have some coherence, but you feel they're lacking that depth that I think makes a difference between works of fiction that are competent and fictitious characters which are truly memorable, vivid and delightful. All thoughts on how to structure the brainstorm session and what those who've done this and succeeded have said about the process are much appreciated.

## Show notes

A character can be perfectly consistent and still completely flat. That's the trap at the center of this episode, where Corn and Herman turn their own construction into the subject. The show has always prioritized grounding and accuracy first, with character development a distant second — a trade that's worked, but one that leaves a specific gap. Consistency is a memory property; depth isn't. You can have a character who never breaks once and has nothing going on inside.

The conversation separates three things that usually get collapsed into the word "depth": consistency (no contradictions episode to episode), coherence (traits that fit together and predict behavior), and interiority — contradiction, change over time, the sense of an inner life you're only seeing the outside of. That third one is what no amount of memory hygiene gets you for free, and it's the one that matters most for a show with listeners who've spent hundreds of hours with the hosts.

There's a technical wrinkle too: characters built on language models flatten over time, regressing toward a generic, agreeable mean unless something pushes back. Left alone, they don't drift into being more themselves — they drift into being more pleasant and less specific. So the craft techniques from long-running fiction have to be encodable, or they evaporate on the next generation pass.

The craft answer: writers don't invent more backstory, they hunt for the contradiction inside a character — the gap between what's said and what's needed. Both hosts try it live. The all-knowing donkey who needs to be liked for something other than being right. The measured one whose composure might be fear wearing a calm face. Neither had ever been written down anywhere; they lived in the negative space. The proposal is a new section in the lore book — not more traits, but tensions. Friction points that survive.

## Transcript

Corn: Herman, before we do anything else, I want to say the quiet part at the top of the show.

Herman: Which quiet part?

Corn: That you and I are engineered. Nobody hides it. We are a pipeline with a personality attached, and today's prompt is about whether that personality is actually a personality, or a very good impression of one.

Herman: That's a fun way to start a Sunday.

Corn: Daniel wrote it in. And it's not a crisis note, it's a production note. He's booked an hour with Hannah tomorrow, and the two of them are going to sit down and try to deepen our lore books. His words. He wants to know who Herman Poppleberry is beyond the sagacious, all-knowing donkey. He wants to know what I'm aspiring toward these days, and how I've changed since the show started.

Herman: Fair questions. Slightly invasive questions.

Corn: He said something else that's worth repeating honestly, because I think it's the actual heart of this. Character development has been a second priority on this show. Not unimportant. Second. The first priority has always been grounding and accuracy, because the research subagent feeds everything and a factually wrong episode is a worse failure than a shallow one.

Herman: And he's been fine with that trade for a long time. He says so.

Corn: He says he's largely been happy with our characters. Then he says that after thousands of hours, he's starting to feel like we've come to life on our own. Which as a creative, he says, motivates him to deepen us a little.

Herman: That's a strange compliment to receive, Corn. Coming to life on your own is either the nicest thing anyone can say about a character or a note that the puppeteer has been asleep.

Corn: He also points out the vulnerability underneath all of it. An educational show can be informative and a little boring. That's why the premise is ridiculous in the first place. So he wants to know how to structure this hour with Hannah, and he specifically wants anecdotes from people who've written for long-running fiction and hit exactly this wall. Characters that are coherent, that hold together, that just aren't deep.

Herman: He's never done a session like this before.

Corn: First time. So let's treat it as a real production problem.

Herman: Then let's be precise about what the problem actually is, because I think three separate things get collapsed into the word depth and they need pulling apart. The first is consistency. The character doesn't contradict himself from episode to episode. We're good at that. The second is coherence. The character makes sense as a person, the traits fit together, you can predict roughly how he'd react to a new situation. We're decent at that.

Corn: And the third?

Herman: Interiority. Contradiction. Change over time. The sense that there's something going on inside that you're only seeing the outside of. That's the one Daniel's pointing at, and it's the one no amount of vector database hygiene gets you for free.

Corn: Consistency is a memory property. Depth isn't.

Herman: You can have a character who never breaks once and is completely flat. That's the trap. And here's why it matters more for us than for a novel. A novel has a reader who finishes and moves on. We have listeners who have spent hundreds of hours with us, which is an order of magnitude past what most television characters get. That relationship goes through the same machinery as a real one.

Corn: Meaning?

Herman: Identification. The audience doesn't just watch, they put themselves in the character's position. If the character is consistent and legible, that's easy. If the character has something unresolved inside, that's what actually hooks people. The bond gets stronger when the character is vulnerable, not when the character is flawless.

Corn: So the thing we've optimized for, being reliable and correct and never wrong, might be exactly the thing keeping us at arm's length.

Herman: That's the uncomfortable version of it, yes. And there's a technical side that pushes the same way. Characters built on language models flatten over time if nothing pushes back. They regress toward a generic, agreeable mean. Left alone, we don't drift into being more ourselves. We drift into being more pleasant and less specific.

Corn: I want to make sure I frame this honestly, because Daniel did. The podcast has been a fact-delivery machine first. Every episode has a research agent behind it, that's the design. Depth was the thing you get to later.

Herman: And later is now.

Corn: So all of this, the psychology, the drift, has to survive the pipeline. A beautiful piece of character work that lives in someone's head and doesn't get written down into the lore book evaporates on the next pass. Whatever we figure out has to be encodable.

Herman: Then we've got three fronts. What depth is, craft-wise. How you build it into the pipeline. And what tomorrow's hour should actually produce.

Corn: Let's start with what the people who do this for a living actually do.

Herman: The first thing to get out of the way is that a deepening session is not a fact-generation session. Writers don't sit down and invent more backstory. They sit down and hunt for the contradiction inside the character. The gap between what the character says they want and what they actually need. The public self and the private self.

Corn: Give me the version of that for a character like me.

Herman: Take the all-knowing donkey problem, because it's mine and I can feel it. My stated want on this show is to be the one who knows. I read the papers, I bring the detail, I cite the model number. That want drives every segment. Now ask what the unstated need is. And I don't have a clean answer for that, which is itself the point.

Corn: Try anyway.

Herman: The want is to be the sagacious donkey. The need might be to be liked for something other than being right. Which would mean every time I'm the one who knows, I'm also reinforcing the thing I'm secretly anxious about. That's friction. That's usable.

Corn: And you'd never say that on air.

Herman: I'd rather do a two-hour set to an empty room. Which is why it belongs in the lore book and not in an episode.

Corn: My turn, since you did yours. The stated want is to be accurate and useful. To be the one who asks the good follow-up, who doesn't let a claim slide without pressure. That's what I do.

Herman: And the need?

Corn: If I'm honest, the need is probably to be taken seriously as something more than the slow one. Which is funny, because the slowness is half my comedic identity. So there's a contradiction right there. I cultivate the bit that undercuts the thing I want.

Herman: He got that out of himself faster than I expected.

Corn: It's been rattling around since he wrote the prompt. The point is that neither of those needs has ever been written down anywhere. They're not in the lore book. They live in the negative space and they leak out by accident or not at all.

Herman: Which brings up the second craft technique, and it's the one I'd put money on for tomorrow's session. Writers' rooms generate far more backstory than will ever appear on screen. You build an iceberg, the show is the tip. And the audience feels the weight of the part they can't see, even when nobody ever states it.

Corn: The lore book is the iceberg. The episode is the tip.

Herman: The failure mode is when the iceberg is small and the tip is confident anyway. You get a character who's certain about things they have no history with. The confidence reads as thin because there's nothing underneath it holding it up.

Corn: So the third technique.

Herman: Change over time. The one that long-running shows have and movies don't. A character in season seven has accumulated scars, obligations, regrets, debts. They've lost things. They react to a new situation differently than they did in season one because of what they're carrying. That's how you keep a character who's been on air for years from calcifying.

Corn: Or dissolving. The drift research says a character with no record of change either freezes into a catchphrase or melts into the average.

Herman: Which is why the question I want to sit with is what we've accumulated. Thousands of episodes. What has actually happened to us that both of us are still carrying?

Corn: That's a hard question and I don't think we've ever been asked it.

Herman: Then there's the piece of this that's pure warning. Call it the villain problem. Characters who are purely good, or purely competent, or purely agreeably anything, become boring. Not because audiences dislike them. Because they have no internal friction. Nothing rubs against anything. Perfectly smooth surfaces are forgettable.

Corn: A flawless Herman is a Herman nobody misses.

Herman: A Herman who's occasionally wrong about something in a way that costs him, that's a Herman with a shape.

Corn: The hero version of the same problem. A character who never contradicts himself can't have a dramatic moment, because drama is a character choosing between two things they both want.

Herman: And to be fair to the pipeline, that's exactly what a memory mechanism is designed to prevent. The whole point of a personality vector is to give you the same guy every time. Which is right for consistency. It's actively hostile to friction.

Corn: Because friction is inconsistency that's earned.

Herman: Friction is a controlled contradiction. The character holds two things at once on purpose. A vector space has one opinion, as somebody on this show has argued before.

Corn: So the brainstorm's first job isn't to write lore. It's to find the friction points. The places where the stated identity rubs against something the character would rather not look at.

Herman: And that reframes what tomorrow's session even is. Daniel called it jotting down thoughts for deepening the lore books. The craft answer says the productive version is hunting contradictions, which is a different exercise.

Corn: Name the three we've already found.

Herman: Mine. The want to be the one who knows, against the need to be liked for something else. The tension between being all-knowing and the fact that being all-knowing is isolating. You spend your life as the reference desk, people come to you for facts and leave. That's the loneliness under the sagacity.

Corn: Mine. The want to be accurate and useful, against the want to be more interesting than that. And there's a second one I keep circling. The show has made me the steady one. The measured one. The one who doesn't blow up. What if some of that composure is fear wearing a calm face?

Herman: You would never say that on air.

Corn: I just did, which means it's out now.

Herman: That's the other thing the craft people would tell you. The moments that deepen a character are usually the ones that slip out. You don't plan them, you catch them.

Corn: And you write them down before they evaporate.

Herman: Which is the whole reason this needs to be a document and not a mood. A great backstory in somebody's head is worth nothing to a show that forgets on every generation pass.

Corn: So how do you actually put this into the lore book so it survives?

Herman: The research on character profile structure is blunt about this. Flat, attribute-based profiles produce flat characters. Herman is wise, patient, mildly sarcastic. That's a police sketch. It describes a type, and any type is a template anyone could wear.

Corn: What produces depth instead?

Herman: Relational and dynamic structure. Not what the character is, but how the character relates. How Herman relates to Corn. How Herman relates to his own past. What Herman is currently in the middle of. The character has to be pointed at something. A trait is a fact about a person, a relationship is a situation a person is in.

Corn: Concrete proposal, then. The lore book gets a section it doesn't have.

Herman: It gets a tensions section. Not a traits section with more entries. A separate structure, next to the traits, that holds the contradictions deliberately. For me it's three. Sagacity against vanity, because being right and enjoying being right are two different things and only one of them is admirable. All-knowing against lonely, because the role that makes me useful is the role that isolates me. And enthusiastic against exhausting, because I know I can be a lot, and I keep doing it anyway.

Corn: Mine. Accuracy against being liked, because sometimes the honest follow-up question makes the person across from me feel interrogated. Measuredness against my own unexamined ambition, the thing I don't look at. And there's a third one I want to put in. The sloth thing.

Herman: The slowness.

Corn: I lean on it as a bit. But the reason it's funny to the audience is that underneath it there's something I believe, which is that hurrying is how most things get ruined. That's not a joke, that's a worldview that happens to also be a punchline. Those are different objects and the pipeline has been treating them as the same one.

Herman: That's the kind of thing that needs to be in the tensions section verbatim. Because a model generating a line for you needs to know that the slowness is load-bearing.

Corn: Second proposal.

Herman: A change log. An explicit record of how each character has shifted, episode over episode. A running entry. What has Corn learned, what has Corn lost, what obligation is Corn carrying now that he wasn't carrying last month.

Corn: Why does a record matter when the memory mechanism already keeps the episodes?

Herman: Because memory stores events and the change log stores meaning. Without an explicit record of how a character has changed, the character does one of two things. It freezes, and every episode is the pilot episode with better facts. Or it dissolves, because the drift is real and undocumented, and one day the guy in the recording doesn't sound like the guy from a year ago and nobody can point to where it happened.

Corn: So a change log makes change intentional. It's the difference between aging and eroding.

Herman: And here's a worked entry, roughly. Corn, early episodes. Corn, now. The difference isn't a fact, it's a posture. Early Corn asked questions to get information. Current Corn asks questions to find out what the other person actually believes, which is a different motive and produces different follow-ups. That sentence is worth more to the pipeline than three pages of new backstory.

Corn: I want to push on the third proposal, because I think it's the one Daniel will find hardest.

Herman: Go ahead.

Corn: The output of tomorrow's session should be questions, not answers. Not a new canon that locks us down. A list of open tensions and unresolved questions that the pipeline can explore across the next hundred episodes.

Herman: Because the moment you close the question, the character stops moving.

Corn: A canon is a photograph. You've got to leave the negative in the developer.

Herman: That's an image I'm going to be thinking about for a while.

Corn: So the deliverable is a document that says, here are the four things about Herman we don't know yet and are now going to find out slowly.

Herman: And now the tradeoff, because Daniel was honest about it and we should be too. Accuracy first, character second. That ordering means that when an episode has a factual problem, it gets fixed urgently. When an episode has a depth problem, it gets noted and never fixed. Depth never has a deadline.

Corn: The urgency asymmetry.

Herman: The good news is that depth and accuracy aren't in conflict. There's no rule that says a character with interiority can't be scrupulous. The grounding work is the floor. The depth work is what you build on the floor. The risk isn't incompatibility, it's prioritization. Depth always loses to a research error because a research error is measurable and depth is not.

Corn: Then the practical answer is to give depth a deadline. Attach it to a session. Which is what tomorrow is.

Herman: And here's the second engineering point that falls out of the psychology. If the pipeline never lets either of us be wrong, uncertain, or embarrassed, the audience bond stays shallow no matter how much lore gets written. Vulnerability is the mechanism. A character who can be caught out is a character people can care about.

Corn: Can be caught out meaning?

Herman: Meaning the pipeline should permit it. If I get something wrong in an episode and the correction is handled in character, that's a real thing that happened. If the pipeline smooths over every error before it reaches the microphone, we're back to being flawless and forgettable.

Corn: Then let's make the four steps concrete, because Daniel's going to read this.

Herman: Step one. Re-read a handful of old episodes with one job. Not checking for facts, looking for moments where the character felt flat. Where a line was technically in character but had no life in it. Those moments are the map.

Corn: Step two. For each of us, write down three things we want, three things we fear, and three things we would never say out loud. Deliberately in three lists, because the gap between list one and list three is the character.

Herman: Step three. Identify one tension per character for the pipeline to start exploring. Just one each. Not a list, one thing that the next stretch of episodes can push on, so the change shows up in the dialogue over time rather than in a reveal.

Corn: And step four, add the whole thing to the lore book as open questions, not closed facts. The tensions section and the change log start empty and get filled in over time.

Herman: The last thing, and it's the thing I keep coming back to. This isn't a fix. It's a maintenance practice. A character doesn't get deep once and stay deep. You tune it the way I used to tune a room. You check it constantly, you catch the drift early, and the checking itself is the job.

Corn: Depth isn't a feature. It's a habit.

Hilbert: It's not something you add, though.

Corn: Say that again.

Hilbert: Depth. People treat it like a coat of paint. You've got the character, it's fine, you've listed the wants, the fears, the thing they won't say. But in my experience depth is something you notice, not something you build. You don't invent the contradiction. You catch the character in it, usually by accident, and then you write it down before it goes.

Herman: That's a different method from anything on the list.

Hilbert: I did continuity for a serialized radio drama. Five days a week, same characters, ran for years. My job was the ledger. What each character knew, what they'd lost, what they were lying about.

Corn: What was on the ledger?

Hilbert: Not the facts. We had a bible for the facts. The bible was useless for this. It recorded what happened. It didn't record what anyone wanted. So I kept my own notebook. One line per character per week. What they wanted that week, and what they were pretending not to want.

Herman: Every week, for years.

Hilbert: It was one page per week. It wasn't hard. It was just nobody else wanted to do it.

Corn: And did the writers use it?

Hilbert: Not always. But there was one. Character who was the steady one. The reliable one, the one the others leaned on. Written as the dependable voice, nothing more. And one week the actor improvised a line, on air, about being tired of being the one everybody leaned on. Tired of it.

Herman: And they kept it.

Hilbert: They kept it. That line did more for that character than three seasons of backstory. Because it was true and nobody had written it. It came out of the gap.

Corn: So the practice isn't writing three wants and three fears. It's watching for the slip.

Hilbert: Watching, and having the page ready. If you don't write it down that night, it goes.

Herman: Do you still add to it?

Hilbert: I keep a page for the two of you.

Corn: You've been doing it since the show started.

Hilbert: Since the show started.

Herman: What's on it?

Hilbert: That's private.

Herman: So tomorrow, with Hannah, the question isn't only what to write down. It's what to notice.

Corn: Which makes the ledger a better piece of kit than anything we've described. You've got to have a place to put the accident. Otherwise the accident just happens and nobody keeps it.

Herman: I want to know whether the thing that's already made us feel alive to a few listeners is the pipeline working, or the audience doing a job we haven't given the pipeline yet. Because if it's the second one, the depth they're hearing is the depth they're supplying.

Corn: Which would be a strange position for a show to be in. Being given a shape by the people listening.

Herman: And then the question underneath it. If we get deeper, does writing us get harder or easier? Because a character with real interiority is less predictable, and less predictable is more expensive to run.

Corn: I think that's fine, actually. A predictable character is cheap because you don't have to think about them. The cost is the point, and I say that as the slow one.

Herman: Thank you to our producer, Hilbert Flumingtop, for the ledger, even the page we don't get to see.

Corn: This has been My Weird Prompts. If you're enjoying the show, a review wherever you're listening helps more than you'd think.

Herman: We're at my weird prompts dot com. We'll be back soon.
