The Aboard Podcast

Paul Ford: Hi, I’m Paul Ford.

Rich Ziade: And I’m Rich Ziade.

Paul: And this is the Aboard podcast. It’s the podcast about how AI Is changing the world of software, and it really, really is. Rich, have you ever tried to build a game?

Rich: I did.

Paul: How’d that go?

Rich: I succeeded.

Paul: You built a game with AI?

Rich: I did.

Paul: Well, what was it?

Rich: It was Family Feud.

Paul: And then what happened?

Rich: It ruined my family. [laughing]

Paul: Let’s talk about that in a minute. Right now, we’ll play the theme song and come back with a guest who isn’t going to ruin your family.

[intro music]

Paul: Okay, so I want to talk about your game in a minute, but first, let’s acknowledge another human being is here with us.

Rich: Let’s do that.

Hilary Mason: Hello.

Paul: Hilary Mason is the CEO of Hidden Door.

Hilary: That’s right.

Paul: Hiddendoor.co.

Hilary: Also right.

Paul: A person that we’ve known for a really long time. A true New York City technologist.

Hilary: That’s true.

Paul: And it is lovely to see you. Thank you for coming in. Why don’t we start, I think just give everybody a little framing, and then we’ll talk about Rich’s sad story before we talk about your happy stories. Give us a sense of what Hidden Door is? I just went to the website, and it says at the top, “Swords and Sorcery: The Tavern Door by Hidden Realms.” So I don’t know what’s going on. Tell me as if I’ve just walked in the door.

Hilary: So if you have a look at Hidden Door, what you will see is where roleplay meets fanfic. And what that is is we are creating a place for people who love fictional worlds, who want to get lost in those worlds, and who want to be able to tell their own stories inside the worlds that they love from books and movies and shows or their favorite creators.

Paul: Okay. And so essentially, I can kind of play one of these narratives on the web.

Hilary: Yes. And…

Rich: You’re one of the characters?

Hilary: You create the cast of characters you want to play with.

Paul: No, that’s Hillary—

Rich: No, not you, Hillary.

Paul: Okay.

Rich: Okay, sorry. You create the cast of characters.

Hilary: That’s right. So you choose the world you want, whether it’s Big Fan, modern romance novel, where you’re playing in a world of celebrity romance and political intrigue, or The Tavern Door, which you stumbled upon, which is a traditional heroic fantasy world. Or, you know, we have a bunch of classics like The Call of Cthulhu for a little bit of Lovecraftian horror or The Wizard of Oz, or our take on modern romantasy, which is Courtship of Frost.

Paul: You’re gonna need to explain what romantasy is to Richard.

Hilary: Oh, so it is romance and fantasy.

Rich: I figured that part out.

Paul: Okay—

Hilary: Brought together.

Paul: Hold on a minute. He may need to know what those two things are, too. [laughter]

Rich: Fair enough.

Paul: Sorry. Zing!

Rich: Go on.

Paul: Okay, go ahead.

Hilary: All right, so it is a genre of literature that brings plots that primarily center characters, relationships, and all of the kinds of stories that come out of that in a fantastical setting.

Rich: Got it. Sounds like my weekend is planned.

Hilary: Oh, you are in for a new joy if you have not discovered works in this genre.

Rich: I have not.

Paul: Just as a side note, and I think it’s known as, there was a cozy fantasy trend where it was about, like, an orc or troll, but they just ran a coffee shop and they hung out.

Hilary: Oh, so we actually have a coffee shop AU modifier card you can play in any of these worlds, and it puts whatever characters you bring into a coffee shop for the plot of the story you’re gonna play.

Paul: Oh, so it can get real, sort of chill, hang out. It’s like Friends, but with…

Hilary: Exactly. And that is a very common set of tropes that people want to play within and play with.

Paul: All right, let’s go back and take this step by step, because I really want to understand sort of what this framework is and sort of how you’ve ended here as a technologist, because you were a data scientist when I first—

Hilary: I am a lot of things.

Paul: Clearly. [laughter] Right? But first, I really think it’s important for framing that Rich explain his disastrous game experience and sort of, because I think we can refer back to it as we talk about this larger gaming platform. So you sat down and you made a game with AI. What was it?

Rich: It was Family Feud.

Paul: Okay.

Rich: Yeah.

Paul: So you made a Family Feud clone.

Rich: Correct.

Paul: How’d that go? You just, like, told Claude, make a Family Feud?

Rich: It was actually pretty thorny. Like, I actually wanted it to be extensible. So it loads in the survey questions with, like, basic markdown files so that I could generate 100 questions at a time. And I wanted it to be one—this was sort of a thought exercise, I wanted it to be one webpage where the whole experience is in one page.

Paul: Sure.

Rich: And it spawns another page for the operator of the game, because, remember, somebody’s flipping the answers and all that. So I wanted it to be, it was sort of, like, an exercise of constraint. So I was having a good time.

Paul: Single-page application. Family Feud.

Rich: Single-page.

Paul: You’ve loaded it with topics.

Rich: Yes.

Paul: And now what happens?

Rich: My brother brought his eight- and six-year-old kids to play Family Feud. And I created a category of questions for kids.

Paul: I think it was, like, Thanksgiving or something, right?

Rich: It was New Year’s.

Paul: Ah, big party.

Hilary: Okay.

Rich: New Year’s with kids is different. So we needed some activities because we’re not going out and partying. And then my kids are 13 and 11 and the games were so engaging and that it caused a lot of tears and crying.

Hilary: I don’t quite follow the connection between engaging to crying. So I think there’s a step in the story I missed here.

Rich: Yeah. So if you’re familiar with Family Feud—and Richard Dawson, who used to kiss everyone on the mouth, for some reason. Do you remember Richard Dawson?

Paul: I don’t remember that.

Rich: You don’t remember? There are very disturbing montages on YouTube of Richard Dawson. Every contestant was kissed on the mouth.

Paul: We’ll put those in the show notes.

Hilary: Did you put that in your game?

Rich: I didn’t put it in my game. [laughter]

Hilary: Okay.

Rich: Here’s the thing about Family Feud. It’s survey questions. So they’re ambiguous and sometimes, like, “That’s a ridiculous answer.”

Hilary: Mmm hmm.

Rich: And so that’s reasonable for adults to process. For little kids who got three strikes and couldn’t guess one of the survey answers, it would really get them upset.

Paul: So Uncle Rich created an unwinnable game using AI.

Rich: [laughing] It was unwinnable, that led to children crying on New Year’s Eve. I could not stop laughing.

Paul: Yeah.

Rich: The parents of these kids were not happy.

Paul: I’m in a chat with Rich—

Rich: It was really hilarious.

Paul: I’m in a chat with Rich and his brother. It was not fun.

Rich: [laughing] It was a meltdown.

Paul: Like, it was like, like the brother was like, “You really screwed everything up with your stupid vibe-coded nightmare game.”

Rich: Also, my brother was like, “Aboard sucks!” He thought Aboard made the game. [laughter] He just associated it with Aboard. I was like, “No, no, I did this myself.” He’s like, “Well, it sucks.” And I was like, “Okay, dude.” So anyway, that didn’t go well. But it was fun!

Paul: The reason I actually wanted to drill in on that and set us up because we’re not the subject, you are the subject, Hillary, but it was so easy to build the game for Rich.

Hilary: Mmm hmm.

Paul: He did it really quickly and it played just like Family Feud. And it seemed really good and really fun.

Hilary: Uh huh.

Paul: But it ended up being absolute chaos in the household and not really working as a game.

Hilary: Uh huh.

Paul: And I feel that what I’d love to talk about is how to do good product in an era where what we used to think of as product, which was a finished artifact that could run—

Rich: Pretty airtight.

Hilary: Yeah.

Paul: Yeah. Is now approaching zero cost to build.

Hilary: Mmm hmm.

Paul: Like, it’s really easy to make stuff. And if you disagree with that, I would love to talk about it, too. But now you have to actually get—so you can build stuff really quickly, but you have to get quality and product and sort of product as we understand it, which is, like, a real sense of experience and connection with it back into it.

Hilary: Uh huh.

Paul: And I’m curious, as someone who’s like running a team.

Hilary: Mmm hmm.

Paul: You have a company. Like, how are you starting to approach this world? And do you agree with us that it’s really a lot easier to build product?

Hilary: Oh, it’s so easy to make stuff.

Paul: Yeah. “Stuff” is a good—it’s not product, it’s stuff.

Hilary: It’s stuff. And it’s easy to make stuff. And I would call it stuff, perhaps this will be our new technical term, because it is unfinished. It is coming from your direction, where I assume you had to give, you had to write a spec or some guidance or something.

Rich: Yeah.

Hilary: You chose Family Feud, which is something that thankfully for you, will be encoded in most LLMs in some form.

Rich: It was.

Hilary: Because it’s part of common culture and people talk about it. And I don’t know how much thought you gave to the player experience or sort of the kinds of content that would show up, but I think there are a few pieces of this. One of which is that a lot of folks think it’s super easy to make games because you can make stuff so quickly. But the hard part of making a game that brings joy has nothing to do with the making of the stuff process.

Rich: Mmm hmm.

Hilary: It is the creative design and the kind of thinking that goes into that. And games are a particularly fun kind of product to focus on because they are different.

Rich: Do you find that it’s not just creative, it’s also, how do I make sure things don’t go off the rails?

Hilary: Yes, absolutely.

Paul: Yeah.

Rich: We build classic business software and we worry about going off the rails. It feels like your worlds that you’re creating, do you think about boundaries?

Paul: Well, let’s actually describe the product. Right? [laughter] So I’ve just gone to, no, no, I just went to the Tavern Door.

Hilary: Okay.

Paul: I clicked in. It says it’s created by Hidden Realms. Who’s Hidden Realms?

Hilary: Hidden Realms is our team’s—we do some first-party content. There’s also a lot of licensed worlds from books and movies and whatnot.

Paul: Yup.

Hilary: So Hidden Realms is my team.

Paul: Okay. So Hidden Realms. So the nice people at your office.

Hilary: Yes, they’re lovely.

Paul: Now, when I say I’m in the world of Tavern Door, here’s what I think I’m in and then you correct me. I think I’m in a gaming experience where I’ll be able to create a bunch of characters and I will be able to interact with those characters and AI will kind of represent the characters, but they’ll be kind of a game engine, too. And I’ll do that on the web. So tell me everything I got wrong.

Hilary: Okay. So you are in a space where what you are doing is taking the “Paul Ford vibe” of the Tavern Door. So you should be able to add these cards—everything is a card. You’re building up a deck. You’re brand new, just logged in for the first time. So you probably only have a few.

Paul: Right. So it says, “create this world,” and there’s world cards.

Hilary: Right. So you have a world card for Tavern Door, and there should be next to it a place where you can put in modifiers. And these will be things like…

Paul: Vampires!

Hilary: Yes, vampires.

Paul: Vampires! Okay.

Hilary: Yes. Yes.

Paul: I got Saturday morning cartoons, celebrity romance. I am adding Brooklyn and vampires. Thank you.

Hilary: Wonderful. So these are choices you are making to take this world and reset what you’re gonna get. And it’s gonna come up with some world description, like, the back of the book blurb for you based on that. You should see that there.

Paul: I see. Okay, so I’ve just added actually three: Brooklyn, vampires, and celebrity romance. So you’re actually letting me—

Rich: It’s just gonna put you in Williamsburg. [laughter]

Hilary: It absolutely is.

Paul: Honestly, with those three. Yes. Okay, so I, essentially, so these are like themes.

Hilary: Right.

Paul: So the story is going to get spun for me.

Hilary: Yes.

Paul: So you need—how long did it take to come up with, like, the card mechanism?

Hilary: So cards are a wonderful metaphor for games, and particularly for our kind of game, because everybody knows how to play cards. We draw a lot of inspiration from tarot cards. Beautiful pieces of art. They have a certain tactility. You know, you can turn it over. And so we have worked with a card metaphor for a long time, but we find that it resonates with people. Everyone gets cards.

Paul: Okay.

Rich: So the usability—not a lot of conversation about usability in the world of AI these days.

Hilary: No. And it is the hardest problem.

Rich: It’s the hardest problem.

Hilary: It is. Yes.

Rich: By the way, I’m watching Paul flip through characters, character profiles.

Paul: Yeah. I’m about to profile a relatively medieval-looking woman named Mary.

Rich: They all look like they’re from Brooklyn. [laughter] I don’t know if that’s just the sophistication of this tool, but it’s pretty funny.

Hilary: Well, you should be getting a lot of avatar choices that draw from the fantasy and the Brooklyn and probably a little bit of the romance vibe.

Rich: Ah. Yup, yup.

Paul: Okay. Hunter struggling with internal demons, and I’m going to confirm. Okay. So a little bit of classic gaming character definition.

Hilary: Absolutely. You make a character and now you’ve made your main character and you’re going to get some story options for that character to play.

Paul: Okay.

Hilary: Or you can keep making more. You can make items for that character. You can make your own locations if you like.

Paul: Yeah. How are you building this? Like, I knew about the AI part, but now—

Hilary: Uh huh.

Paul: Is there a platform that you’ve created that you’re building on top of? Are you vibe coding as you go? How is this coming together?

Hilary: So we have a game engine and a system that underlies this that uses a slightly different architecture than something like if you were just role playing with ChatGPT or another LLM.

Paul: Okay.

Hilary: So what you’re doing here, each of these cards is essentially a row in our Postgres database, which doesn’t sound very exciting, but it is.

Paul: Oh, no. To this—

Rich: You’re in the right place. They’re excited.

Hilary: Okay.

Paul: To this audience, very excited.

Rich: Yes. [laughter]

Hilary: Right. And that has a bunch of structured information about your character. You have probably seen some questions where you can create a backstory that’ll be generated for them and that’ll be for every character in your cast. And the player experience at this point is that you’re choosing your world, you’re choosing the kind of vibes of story you want, and you’re making the cast of characters, it’s like you’re the director setting up the stage.

Rich: Yeah.

Hilary: Or if you ever played The Sims. I don’t know if that’s a reference point for you all. It’s sort of like setting up your little people so you can go have stories with them.

Paul: I think what’s so interesting here, and it’s worth calling out, right? This is a classic gaming engine.

Hilary: Mmm hmm.

Paul: Like, there have been variations on this going back to the 70s. Right? Of like, we’re going to have characters, we’re going to track them in some kind of database.

Hilary: Absolutely.

Paul: So the LLM part of it is additive to a core kind of classic engine that lets people track stats and sort of know where they are in the game.

Hilary: Yeah.

Rich: I mean, it’s worth pointing out the games are text-based.

Hilary: Yes. Texts and art that’s assembled as you play.

Rich: Text and art. And you make decisions as you go on your journey.

Hilary: Absolutely. And you can type in anything you want to type.

Rich: Okay.

Paul: Let me narrate a little bit. This is story number one. This is called Blood Moon Rising.

Rich: Okay.

Paul: Okay? I’m Marian, and I have met Thrain Ironhand, the Gilded Griffin, and Mira Nightshade, who is actually wearing really cool sunglasses, here in—

Hilary: It’s Brooklyn.

Paul: Yeah, it is Brooklyn. You’re right. Oh my God, it’s real. It did it, didn’t it? Yeah. Okay.

Hilary: Uh huh. [laughter]

Paul: Because I’m in Brooklyn, the people actually look, even though it’s very, very magical, they look like magical Brooklynites. So I’ll read a paragraph. You push open the heavy oak door of the Gilded Griffin. That is an underlined hyperlink. Thank you for the underlined. Double underlined, actually.

Hilary: Uh huh.

Paul: It’s thick. The scent of spiced ale and pine smoke hitting you—this could be Williamsburg—as a low murmur swells inside. Golden lanterns cast a warm glow over the room where sword-wielding warriors and Brooklyn hipsters alike share tales— [laughter]

Rich: Oh goodness.

Paul: Of adventure and magic. Okay, so I’ll tell you where it looks like I’m in. If I didn’t know that there was AI involved?

Hilary: Mmm hmm.

Paul: It’s done a lot of subtle stuff with the Brooklyn angle.

Hilary: Uh huh.

Paul: So that I can see as a nerd who works with LLMs all day. But then I got a couple options. Now, first of all, I have a free-text ability to type anything and be like, let’s move this story along.

Hilary: Uh huh.

Paul: I could say everyone’s a computer programmer and I’m sure it could do something.

Hilary: We’ll see. It might not let you do that.

Paul: Okay, well, we can try that. And it also has kind of, like, classic, take a deep breath, savoring the spiced ale aroma. So classic kind of like…

Rich: Choose your own path kind of stuff.

Paul: Yeah.

Hilary: Well, and you’ll notice that even though you can type anything, we show you examples and there might be a little easy or hard label on some of them, that’s saying how hard the die roll would be for your character to do that thing in this moment. But going back to your comment about UX, right? It is user experience design. The reason we show you options is because if we just give you a text box, you just type, “Hi.” You don’t know what to do.

Paul: It’s true. I said, “Ask if anyone in the tavern knows Python.” And it said, “That’s impossible.”

Hilary: Oh, no, yeah.

Paul: Which, I mean, it’s true, probably if, I mean, it’s Brooklyn. So I could probably go back to JavaScript or React or maybe Rails. [laughter]

Rich: Did this take a lot of this…

Paul: Oh, this was easy.

Rich: Did you run into walls when you tested? Like, how did you end up here?

Paul: Just to be clear. I’m really kidding. This is very much, like, a product with a capital P when you look at it.

Rich: Yeah, totally.

Paul: There’s a lot of work here.

Hilary: How did we end up here? We actually started this well more than five years ago. So just to set the stage, because that is like a million years in AI product land.

Rich: Yeah.

Hilary: GPT-2 existed. Was terrible. There was no ChatGPT.

Paul: And I think now is a really good moment to stop and like, you do not come new to this world.

Hilary: No.

Paul: Right. I think, like, just give a little sense of where you were coming to, and then let’s get to ChatGPT.

Hilary: Okay. So I started my career as a, you know, graduate student in machine learning. Became a professor. Realized that I liked working on way too many things and too many gaming things and fun things to do that. Moved back to New York, which is where we are, where I grew up. Joined a startup which failed. Joined another startup as it was becoming a company called Bitly. Short links on social web. I was the chief scientist at Bitly. That was an adventure. And became very involved in the data science community. Did a whole bunch of stuff. Started a nonprofit, became a data scientist in residence for Excel Partners. And then I founded a company in 2014 called Fast Forward Labs. That company was an applied machine learning research and product code development company.

Paul: It made really pretty books, too.

Hilary: Thank you. Yes. We put so much energy into the books, and I was very lucky to work with incredible designers and writers, and the printers are still over on 3rd Avenue in Gowanus, and they’re awesome. But we wanted it to feel like a gift every time you opened a book. The books were each one focused on some technology or some capability in AI that was just becoming useful. Then we would work with our customers, I imagine, much like y’ all do, to build stuff.

Paul: You are very frontier, though. Like, we’ve always been more on the, like, all right, let’s get your business app together. Your stuff would be—

Hilary: Yeah, you’re better business than I am. But, like, I admire that a lot.

Paul: It’s actually really amazing, because usually I’m always doing the complimenting and no one ever returns it. They’re always like, “Oh, yeah, no, that is me.” But you, it’s just some really.

Rich: Thank you. [laughter]

Paul: It’s exciting. It’s really, like, just, that just felt great. Thank you. No, I always, I thought of you as like, “Hey, my large organization has a lot of data. Needs to bring this data to life somehow.”

Hilary: Mmm hmm. Exactly.

Paul: “Let’s go talk to Hillary.” And they have those—I kind of, I wish I had those books because they were very special. Because it was just this sense of like, hey, this new frontier is here. But it’s really worth approaching with a sense of, like, craft and care and good writing and nice, nice art.

Hilary: Well, thank you. And yes, we tried to make it feel like a moment of joy to be an artifact that would spark conversation. It was, like, very tactile. But a lot of, our very first project was natural language generation in 2014, and I have been obsessed with it ever since. And that company grew up, we did a lot of other work, worked on a whole bunch of products. I sold it eventually to a company called Cloudera, which is a data platform vendor. Ended up being the general manager of their data science and AI business unit for a couple of years. A lot of fun, a lot of corporate sort of building products or doing data science for business purposes across Fortune 500 and other customers. An incredible diversity of work.

Paul: But in the back of your head, natural language generation—and the reason I really wanted you to give this story—

Rich: And games, it sounds like.

Hilary: I love games.

Paul: But there’s this, like, you know, I think a lot of us saw, for me, it was like ChatGPT-3 is when I started to pay attention.

Hilary: Uh huh.

Paul: But you were very prime. Okay, I’m interested in this part of the world. There are these frontier labs. And at this point, it was way before this current ridiculous era we’re in.

Hilary: Uh huh.

Paul: There were dabblers, at some level. Right? They were very, very smart dabblers, coming up with lots of different products and suddenly they have this one that can kind of talk a little better than most of the other ones.

Hilary: Yeah.

Rich: I want to ask a question related to that. When 3.5, I guess ChatGPT-3.5 sort of landed like a spaceship on the Earth. Did you see it coming? Were you surprised? Like, you’d been in the space for a long time.

Hilary: Yeah, the technology did not and continues to not surprise me very much. But the reaction to the change in, I’ll say the change in modality and then I’ll try to sound less academic, but, like, the product experience.

Rich: Mmm.

Hilary: The way that ChatGPT came out and all of a sudden everyone’s excited about it. Stable Diffusion comes out and you can do image-based generation on a laptop and everyone’s excited about it. And the same thing right now with OpenClaw, to some extent, everyone’s excited about it. Not new tech, just new orchestration. New product experience.

Rich: Yeah.

Paul: I mean, I keep being this term about Claude Code. It’s kind of like a true product built up from an LLM—

Hilary: Yes.

Paul: —rather than some radical disruption, but the product works. And so that’s exciting.

Hilary: Yeah, and the thing about Claude Code, when you’re using it well, and not terribly, is that it is not the LLM that matters. It is the whole layer of context management and orchestration and the processes that it encodes for us, like, how do you build stuff? Right?

Rich: That is the clever invention, right, on top?

Hilary: Yes. And that getting better and better is why it’s improved so much over the last few months especially. And it’s an incredible achievement.

Paul: But it feels, and it’s funny because actually that kind of progress can happen much more quickly than new models coming online.

Hilary: Yes.

Paul: Right? So, like, Claude Code can actually get better. The LLMs can’t—like, it can’t get smarter necessarily, but it can get more clever by being a product and looping in certain ways. And….

Hilary: And it can get better at doing the thing you want it to do when the whole framework and context management is designed for the thing you want to do. And it turns out, like, software projects generally use pretty similar frameworks and development processes from one to the next. And so it’s getting really, really good at that.

Paul: It’s amazing with these things. I think one of the things that I’m learning from vibe coding is that if an LLM has an actual failure state, like the code doesn’t run?

Hilary: Uh huh?

Paul: Then you can do things in a much more intelligent way. If you don’t have that failure state, you end up with the sort of headed closer and closer to more like, hallucinatory stuff and ambiguity.

Hilary: Sure. Yeah. And that is, by the way, the same philosophy that underlies test-driven development, which is even before LLM-assisted programming, one of the major philosophies of process of writing code is that you make some tests and then you write the code until the tests work. And that’s one way to make sure you did it right.

Paul: Because you’re getting in that failure state.

Hilary: Yes.

Paul: So look, I just had a mug—

Rich: Back to your tavern, Paul.

Paul: I had a mug of the house’s signature pine-smoked ale from Thrain, which is, it’s upstate.

Rich: $18. [laughter]

Paul: It was. It was $18.

Rich: $18.

Paul: My objective was to enjoy the atmosphere of the tavern, and now I’m gonna continue. Okay, so five years, you’ve got this platform.

Hilary: Uh huh.

Paul: Let’s, you know, we’re all five years in. This has been a wild five years.

Hilary: Yup.

Paul: Would you agree with that, actually?

Hilary: Oh, yeah.

Paul: This is the wackiest span I’ve ever had in my career.

Hilary: Yes.

Paul: Yeah. Okay. It’s just comforting to hear another person say it.

Hilary: No. The world got interesting and bad in mostly bad ways, some good ones, like the tech got wild, the industry is really interesting. [laughter] The sanity of the market, especially with respect to AI, went from more rational to less rational. And I say this as someone who built a whole business in the wake of IBM’s Watson marketing back in 2014 and 2015. We are in a space where nothing is real anymore, but everything is wild.

Paul: I mean, you’ve got hundreds of people listening. If you could tell people to calm down about one thing, what do you think it would be?

Hilary: To calm down about one thing? I would…

Rich: She’s gonna say, “No.”

Hilary: [laughing] No.

Rich: “Stay diligent.”

Hilary: Stay optimistic. And what I would say is just, it is really hard in all the chaos to build good things and good products. And many of the examples out there aren’t. So let’s, you know, highlight the ones that are.

Paul: I don’t know about you, but just like, the number of friends who, like, you’ll have a conversation with them and you’re like, okay, you’ve lost, lost 15% of your mind. Like, it is a wild era. [laughter] Everybody is reacting to this intensity in different ways.

Hilary: Yes.

Paul: I’m sure I’ve lost my mind in certain ways too. Right? You’ve run across some bad products.

Hilary: Mmm hmm.

Paul: We talked about that before. What, where do people—before we come back to your product, which obviously does everything right.

Hilary: Yes, of course.

Paul: Where are people screwing up? Because we’re all kind of making a study of this.

Hilary: I mean, this is a podcast, right? So I’ll say the original sin of AI user experience design was ChatGPT.

Paul: Yeah.

Hilary: And I would submit that chat is a relatively poor product experience, and it’s a very useful one when you don’t know what people are going to do, and you’re making a general, like, you want to be able to do anything.

Rich: It’s an everything tool.

Hilary: Exactly. But the incentives of running that as a product are that chat keeps you engaged in, it could be redundant, it could be wrong. It is—as you were talking about people losing their minds, this is a product designed to hack into the same capabilities that we have to use language, to talk to each other, to become socially connected, but instead to drag us away and make us socially disconnected because we have this artificial thing that we’re talking to instead. And unfortunately, the corporate incentives are to lean into that as much as they could get away with.

Rich: Yup.

Hilary: And because it was the original success, so many people have just emulated that user experience without thinking it through. Because thinking it through is hard. I’m very proud of what we do at Hidden Door, but we are thinking through new user experience metaphors and visual metaphors. We’ve tried so many things that didn’t work and there are still tons of changes to come. And this is because it is really hard to think about building things that are pro-social, about bringing people together through their experience and storytelling. So yeah, I would say it is chat as a UX.

Paul: Every fiber of my immature being is struggling with not just going, “You’re absolutely right!” And stopping there with an exclamation point.

Hilary: I don’t see a problem with that. [laughter] We’re done.

Paul: You’re absolutely right!

Rich: Do you think it would have exploded the way it did if it wasn’t chat? Don’t you think chat was the usability stroke of genius that sort of landed it on the whole Earth?

Hilary: Absolutely.

Paul: That’s how you get your billion people.

Rich: Yeah, but, so—

Hilary: Yeah, but to what end?

Paul: Oh….

Rich: Oh… Thanks for joining us, folks! [laughter]

Paul: Yeah.

Rich: Yeah. What you’re saying resonates because, because I feel like even when I get my answer, it’ll always follow up with, “Would you like me to draw a comparison chart?”

Hilary: Exactly.

Rich: Right afterwards, right? And let’s articulate to everyone why it does that, why it seems to want to keep the conversation going. And let’s face it, there’s a lot of lonely people in the world. And so let’s talk about those incentives.

Hilary: Yeah, I mean I don’t work for OpenAI or Anthropic or any chatbot-creating companies, so I don’t have any inside information.

Rich: Yeah.

Hilary: So just with a caveat that this is pure speculation, but I think we all have a shared understanding of what product metrics are and what daily active users are, which is something—

Rich: Engagement.

Hilary: And engagement, right? And so, you know, if you kind of put that understanding and you think about the business and the fact that these companies need to grow users and revenue, like, it seems like a very, you know, one plus one equals two story of reasons to keep that chatbot asking.

Rich: Yeah.

Hilary: Like, so I’ve done this experiment where I just always say, sure.

Paul: Yeah.

Hilary: Like, “Sure, ChatGPT. Do the next thing, the next thing, the next thing.”

Rich: It just goes on.

Hilary: And things just get weirder and weirder. Like, I eventually ended up having it draw me a two by two chart of the worst burgers in San Francisco and, like, plot things. And then I was like, okay, like, “Would you like me to draw you a chart of, like, the worst other food experiences?” And we’re like going down this rabbit hole and like this is…

Rich: Yeah.

Paul: No, they love to put a little loop in, right. And the loop is always, it’s that sort of, now would you like me to? It’s funny, I had, I hack around on Claude Code a lot, obviously, but there was some yet another giant API meltdown.

Hilary: Uh huh?

Paul: And they’re like, oh look, if this isn’t working, go back to this version of Sonnet. And it was like an older one, like version four of their model. And it’s not sycophantic with me anymore. It’s just very clinical. Right? It’s just sort of like, “Here’s your stuff.” And it’s not very flattering. I don’t see a lot of like, “You’re absolutely right anymore.” But it went back to it and I’m like, this is terrible. I hate this experience. Shut up, don’t talk.

Hilary: Uh huh.

Paul: And I realized, actually, over the last couple of months I really started to feel maybe a little more control over the technology. And one of the ways that it’s useful for me is it gets less and less human. I wanted to be the computer again. I like computer.

Hilary: Uh huh.

Paul: When computer works, I feel good. [laughter] One of the first things I think that all AI should do is clean up the mess of the old computer. Everybody’s like, how are we going to use this in org? I’m like, you have so many badly scanned invoices. Let AI make those better.

Hilary: Uh huh.

Paul: Like, the old robot’s mess should be the new robot’s cleanup. Then we let it into the house.

Rich: You’re a thoughtful technologist who understands and can see behind the mask. But there are therapy startups and startups that I think, like, I saw articles of just older adults having full-on relationships with Adele Optiplex 63000.

Paul: I know, but remember… It’s just a magnification. Remember when that woman married the Golden Gate Bridge because she was in love with it?

Rich: I do. It is a beautiful bridge, out of fairness.

Paul: It’s hard to get the ring on.

Rich: Yeah. [laughter]

Paul: Like, there is this, there is it just humans, just—

Rich: Wait. Is this bad?

Paul: Yes.

Rich: Okay.

Paul: And I’ll tell you why it’s bad. It’s bad, not because—

Rich: Not, like, is it bad to have a relationship with a server? But…

Paul: I’ve had that for, like, 30 years. [laughter] Literally. There’s a little guy in New Jersey just running for me right now. I love that guy.

Rich: What can we change?

Paul: The chat that isn’t just, like, morally questionable, that we simulated a human and people believe it’s a human and they believe it’s right and the product doesn’t encompass it. That’s actually kind of bad on a lot of different levels, and we all kind of know it we’re all learning it right now. But it’s actually bad product for something like assisted coding. I want to be able to touch things and say, “Please fix this part, and this part.” I can’t address the actual objects inside of the system. I have to describe them in text. It’s not using the computer to do good computer things. It’s using this one interface for everything, which we know is not good. We sort of have been through this so many times as an industry.

Rich: Yeah.

Paul: So I want it to be embedded, which is kind of, maybe, I think, a good place for us to talk for a minute. Because it feels like AI is embedded in your gaming engine.

Hilary: Absolutely.

Paul: Right? And it’s not a, it’s not a chatting service.

Rich: It doesn’t look like AI.

Hilary: It doesn’t. And when we say “AI,” and I’m putting that in air quotes, it is actually in one turn of this game, there are 16 different machine-learning tasks that run, some of which are—

Paul: Which, for people who don’t know, that’s AI, too.

Hilary: It is whatever you want it to include.

Paul: Does anyone remember machine learning anymore?

Hilary: I mean, some of us do.

Paul: You do.

Hilary: But every time I’m at a meetup or something, people are like, “Oh, you use classic machine learning.” [laughter] And I’m like, “I am not that old, guys.”

Paul: Awww!

Rich: I noticed in your marketing, you don’t put AI for front and center.

Hilary: No, we don’t, because our players do not care.

Rich: Yeah.

Hilary: And I think that if you have made a good product, that’s like saying, computer role playing, which actually used to be a marketing term if you go back to the D&D games of the early 80s. Right? So, no, it’s not the thing anyone should have to care about. It’s just that there is an experience. I will say also, we joke a lot at Hidden Door that our work is artisanal Brooklyn machine learning, or artisanal Brooklyn AI.

Rich: That’s amazing.

Hilary: But we do that because, if you were asking—

Rich: You’re based in Brooklyn, by the way.

Hilary: We are based in Brooklyn, yeah. And I’m a very proud New Yorker. So yay New York. But also yay, other places. [laughter]

Paul: No. Not really. A little.

Hilary: You would find this if you were able to play for a longer time. If you were to go to, like, a standard LLM and you’re like, tell me a story about this thing, there are two things that are different on Hidden Door, one of which is that you can’t do everything. You are within the laws of physics and the rules of your personally very strange world to blend Brooklyn and fantasy. But, like, that was your choice. So we’re going with it. And then—

Paul: I just told it, I like waffles.

Hilary: Yeah. I like waffles, too.

Paul: So it’s now—

Rich: Who doesn’t like waffles?

Paul: You chuckle recalling that a sudden craving for a warm honey glazed waffle once drove you deep into the pine rim where tracking the elusive treat turned into your first hunt.

Hilary: Ooh. It built that into your backstory.

Paul: That’s right.

Hilary: And waffles.

Paul: So back to this machine learning.

Hilary: Okay, and then the second thing I’m going to say is that we actually write out the beats of the story mostly by hand.

Rich: Mmm.

Hilary: And we use LLMs and other techniques for what they’re really good at, which is, to Paul’s point earlier, things like taking information from one place and moving it to another place, or in this case, taking aspects of what it means to be in Brooklyn and on a, be a fantasy adventure and combining them into this moment of a very particular story that no one else has ever told. But also avoiding the problem of, like, all of the samesiness of…

Rich: Sure. Sure.

Hilary: What will come out of an LLM if you don’t tell it to do something else.

Rich: So you’re injecting a good dose of human creativity.

Hilary: I would say we’re writing stories and then using AI at the end to sort of flavor them and respond to them.

Rich: Little embellishment.

Paul: And the stories are both words and structure and plot, but also data and the system and the limits of the game engine.

Hilary: Yes.

Paul: So you have a constraint system when you sit down, and that is the game engine.

Hilary: Yes.

Paul: Okay. I can do certain things. I can type anything I want into the box.

Hilary: You can type anything. But you will find if you type language that is inappropriate or unknown to our system, it will reframe it for you.

Paul: Interesting. Okay.

Rich: People will always try to find the edges.

Paul: Let’s not talk about people. [laughter] Okay, so tell me just a little bit. How is it going? How are players reacting? Like, five years is hard. There are good days and bad days when you’re building a company like this. What I would love to hear is for you to tell us kind of what’s up with you and then also for people, because I think this is what people want to do with this technology. They don’t just want to strap it on to, like, some random accounting tool.

Hilary: I hope so. I mean, I think both, like, accounting is also important. [laughter]

Rich: Paul!

Paul: Sure is.

Rich: Thanks for nothing, Paul!

Hilary: As a CEO, I respect accountants.

Paul: Our good friend Craig Mod just built his own QuickBooks in Japan because he needed it.

Rich: There you go.

Hilary: That’s cool. So no, no, no. Going back to some of the things we talked about earlier, our vision for this was, like, what if you could sit around a table like this with your favorite author and take some of the fun parts of playing a roleplaying game, like Dungeons & Dragons is the most famous one, but there are lots of tabletop games and sort of have that. But you’re doing it on your phone, mobile, web, or on your web browser. And then you’ll find when you are done with this story, it gives you a “do you want to share this?” Right?

Paul: How does the creative industry react to this product? Because, you know, young adult fiction authors are known for their tolerance and their flexibility.

Hilary: So you’re asking about how people react to the use of AI here, or like…?

Paul: Yeah, a little bit. Because, I mean, you’re in a world like genre is a little resistant to—

Hilary: Uh huh.

Paul: Like, genre people, they don’t love this. Not your product, but like AI in general, like a little prickly and for good reason.

Hilary: And we have a FAQ on the webpage that you will find. And in that FAQ, one of the questions we get is like, “Do you train LLMs on authors’ work?” And the answer is, no, we do not, because that is not a good experience and it is not fair to the author. We also sign rev-share agreements with our authors because the goal here is that they have fans of their work who want to be able to play in their own way. This is not competitive with the book or the experience that they’re creating. It’s a way for fans to just mess around and tell their own stories and share those with each other. As a player, it is my way to be, I love this thing, and I made it—like, I played Pride and Prejudice and I got Mr. Darcy to have a secret cheeseburger addiction, which is, like, if you know me, I love cheeseburgers.

Paul: For people who don’t know the book, that’s not in it.

Hilary: It is not in the book. It is not canon. [laughter] But it was very funny to me, and I sent it to my friends. Right? And those are the only—

Paul: You’re aiming for a social experience.

Hilary: Exactly.

Paul: Okay, so not one person chatting to a bot.

Hilary: Not one person chatting to a bot. And not saying, like, we would ever replace authors. Like, we handwrite the beats of our stories. The authors have created these things that we love and want to celebrate and want to want more of. And so why can we not use this as an experience where a fan can just be like, “Cool. I’m going to take this element of that, and I’m going to put in vampires and by the way, also Brooklyn, if you like. And then, you know, I’m going to have a, you know, epic battle on the L train and I’m going to make jokes about waffles and I’m going to send it to my friend who loves waffles. And we’re just going to have a joyful moment together, because of this book we liked or because of this creative experience.” And we’re trying to make it like, if you’re already writing fanfic or any fiction or anything, this is not competitive with that either. It’s the thing you’re doing when you’re in line for the pharmacy. It’s the way of playing and being creative. And frankly, if you’re going to be on the internet doomscrolling, we always like to think about how do we give you fictional joyscrolling and funny moments.

Paul: Joyscrolling!

Hilary: That’s our word.

Paul: I mean, as you’re saying this, what I’m thinking of, like, when you talk to writers, they’re like, you can’t put a price on my work.

Hilary: Of course.

Paul: But then Anthropic went ahead and just stole everything. And it actually turns out you can. It’s $3,000.

Hilary: That’s not okay. And a lot of the—

Rich: Cut you a check after the fact.

Hilary: The angst from the creative community is because of the economics and the incredibly hypocritical and unfair treatment.

Rich: Yeah.

Hilary: And so we’re lucky enough to partner with writers who, by saying no, we actually respect you and want to pay you. And we are not going to try to do anything like take your IP without your permission or all that stuff.

Paul: Right.

Hilary: It’s not hard. It’s pretty easy to actually talk to people and figure out how to value.

Paul: No, I mean, writers are funny because they’re like, well, $3,000. I wrote that like 10 years ago ago. But they really would prefer if you treated them with respect and gave them like opportunities up front.

Rich: New Yorkers like to strike a deal.

Paul: No, but it also just turns out that it’s surprising to everyone in the tech industry, but it turns out that the people who make the content are, like, humans with needs. And it’s just like, it’s just wild. Everyone’s discovering this new and exciting fact. It’s great. [laughter]

Hilary: Yes.

Paul: All right, so good. So look, I think people should go play a game. Here’s what I think is really, really interesting, which is that you have balanced out, like, really all of your nerdiness. Like, you’ve got your machine learning nerdiness, your D&Dish nerdiness, your, like, genre nerdiness.

Rich: It’s coming together.

Paul: Like, I’m bringing this. It’s all coming home.

Hilary: We are very big nerds.

Paul: Yeah, no, and I think that’s beautiful. But I think what’s really, the thing I’m learning from this and really taking away is, you know, I’m so used to every narrative being, like, an AI-enhanced gaming experience. Right? And instead, this is a really powerful gaming engine that tries to connect a unique IP, give people social context. It also happens to use LLMs to move the story along, along with traditional machine learning. Right? So I had you correct my first definition, but I think that one’s a little bit closer.

Hilary: That’s closer.

Paul: Okay. Okay. So that’s what you’re aiming. That’s what you’re setting out to do, which is actually, in a funny way, just like, way, way more ambitious because you’re trying to make a platform that will sort of expand and sort of connect to lots of people over time as opposed to just throwing AI into the mix and seeing what happens.

Hilary: That is true.

Paul: Okay, well, that is very cool. And you’re in Brooklyn. If people wanted to get in touch and ask you questions, you know, they probably shouldn’t do that. But what if they do?

Hilary: Of course they should. I love questions. You can send me a note at hilary—H-I-L-A-R-Y—@hiddendoor.co.

Paul: That’s right. You’re a one-L Hilary.

Hilary: I’m a one-L Hilary. Or I’m on Bluesky at hilarymason.com.

Paul: Lovely. Wonderful. Well, thank you for coming in.

Rich: Thanks for coming.

Hilary: Yeah, thank you for having me.

Paul: This is great. I’m gonna go play. Let’s see, Rich, I’m gonna go play…

Rich: Go back to Brooklyn.

Paul: No, I’m doing Pride and Prejudice.

Hilary: Nice.

Paul: Yeah, it’s fun. It’s a classic. I don’t know if I’ve ever finished it, though. Maybe I’ll finish it here as a game. Okay, that’s enough of me talking. [laughter] Okay. So Aboard. Rich, what’s Aboard do?

Rich: Fact, not fanfiction. We ship software. We actually use a lot of the amazing tools that are in the world today. But we are experts at getting it over the line and getting it to production-grade code.

Paul: We’ve been talking a lot about our messaging. I’m going to put it real simply for the people at home. You come to us with a problem that’s kind of roughly software-shaped, and then we solve it as quickly as possible with our own technology and with all these other wonderful, very fast technologies. But the thing we deliver is a stable, consistent platform that you can host. We can host. Think of us as, you know, like, Rich and I used to run an agency. It’s like that, but automated on rails and ready to go. So get in touch.

Rich: Hello@aboard.com. Thank you again, Hilary.

Hilary: Thank you both.

Paul: Yes, thank you.

Rich: You have a great week.

Hilary: Bye.

Paul: Bye.

[outro music]