"If there's one job that could survive, it would be storytelling"
Constantin Berthelier and Kaynã Oliveira from Aarda AI discuss fiction and world-building in the age of AI.
Hello, and welcome to this week’s edition of AI Gamechangers! Every week, more or less, we bring you conversations with folk who are applying AI in practical ways throughout the games industry. It’s a fascinating time to be working in AI, and we’re introducing you to a cross-section of the ecosystem, from user acquisition to fashion, from live-streaming tools to audio, and from game design to recruitment.
Our Q&A this week is with Constantin Berthelier and Kaynã Oliveira of Aarda AI. They discuss their innovative platform that's transforming game narrative design through a world-building first approach, giving developers control over story elements.
And there’s more! Scroll to the end for links to the latest AI and games news stories from around the web, including Claude playing Pokemon and Grok 3 building a Grand Theft Auto game.
Constantin Berthelier and Kaynã Oliveira
Meet Constantin Berthelier and Kaynã Oliveira, the innovative minds behind Aarda AI. Constantin brings years of narrative design experience, while Kaynã's background spans robotics engineering, web development, and AI. Together, they've created a platform that reimagines how stories are built for games, focusing on comprehensive world-building. In this interview, we explore their “knowledge brick” system for organising narrative elements, its successful implementation in their demonstration game Shadows of Harrow City, and their vision for making rich storytelling more accessible to game developers worldwide.
Top takeaways from this conversation:
Aarda flips traditional AI character creation by building comprehensive background contexts first, enabling richer, more coherent storytelling within games (their inspiration is Tolkien's approach to world-building).
Aarda uses a “knowledge brick” system: the platform uses a hierarchical structure that gives narrative designers precise control over what information characters know.
The founders believe AI won't eliminate the need for human creativity in storytelling, but will transform the industry by making game development more accessible to new types of creators who have stories to tell.
AI Gamechangers: Please give us a brief overview of your backgrounds. What brought you to Aarda AI?
Kaynã Oliveira: I’m from Brazil. I’m currently living in Portugal, and I was originally a robotics engineer. But I have worked with everything: robotics, web3, web development, and now AI for a while. I love to learn new technologies and get in touch with new products. I love bringing things to market, partnering up with very smart people who are innovating, and building stuff with them. That’s how I met Constantin, on a previous job. He had a very specific view of what he wanted to build, and I totally could get behind that.
Constantin Berthelier: I live in Paris, I’m French, and I’ve been a narrative designer in the games industry for 12 years now: social games, Facebook, mobile, console, web3.
I was working on a project with Kaynã, a huge web3 RPG. We wanted every single character to have a specific signature and background and for people to interact with them. I was struggling to find out how to do it, and then ChatGPT-3 came about, and I thought, “This is it. This is how we’re going to do it.” I’ve been working with AI ever since.
It’s been two years now, and I always have an eye on everything. We collaborated with an American company called Inworld. Suddenly, the game was put on ice. And Inworld changed a lot along the way. I like to think I was their first client, but also it wasn’t perfect for our game’s needs. I’m really happy to work with Kaynã today. It’s been a great team working on this game, and I’m always on the lookout to continue this adventure of narrative design in games because video games may be the best way to carry out stories after books.
What can you tell us about the platform you’ve developed? In order to build the kind of game you want to see, you’ve created the Aarda AI platform...
Constantin Berthelier: The main problem we wanted to address (working with other AI platforms and digging up all the tools I could find) is that there was no solution for proper world-building to encapsulate the “big picture” of a story. It was only at character level. With all the solutions on the market, it was always at character level with just a little interaction between them, and on a very small scale.
So, before everything else, Aarda AI is world-building software. If you forget for one second about AI, before anything else, it’s a way to organise your world: from the timeline to questions like, “Are we in the real world?” It’s the dream tool for any narrative designer.
Not only narrative designers but anyone who has a story in their mind can open Aarda and start building and expanding. I believe that everyone has an idea for a story. If you listen to a piece of music or you have a conversation with someone, they always have ideas. What truly makes sense now is we can power that with AI, and the combination of the two is truly magical. We like to think that this solution is perfect for games, but maybe later down the line, it will be perfect for other media.
Please tell us a little bit about its technical background and the challenges there.
Kaynã Oliveira: The name itself describes what we’re trying to do. We think there is the Tarantino way of telling a story, where he just drops you in the middle of something and you just enjoy the ride; not a lot of context, you just get to witness that tiny sliver of narrative.
“It’s absolutely inevitable. AI will be everywhere, and companies that can offer inference will be just as essential as light, internet, water. Inference will be a utility as well”
Kaynã Oliveira
On the other hand, there is the Tolkien way, where he builds the entire dialect, the world, the legends; only then does he drop some characters in there. You really get a sense that the world is grander than that singular tiny story you are seeing. What Aarda proposes is more of a Tolkien way of building stories. “Aarda” is “the world” in the Tolkien stories. It means “world” in Elvish. That’s our proposal!
How much learning does a narrative designer have to do in order to use your tools? Is it an easy onboarding process?
Constantin Berthelier: Very easy. The first steps will be asking you simple questions: “Is your world based on the real world?” Is it absolutely fictional, like The Lord of the Rings, or is it partly fictional, like Harry Potter, which uses the real world plus Hogwarts? From there, we invite any user to feed in some basic knowledge about the timeline, the locations, different groups in the world, different people, constructs, and also, what the objects are. Do you have specific vehicles or specific objects, like magic wands? Based on that, you will work on the characters.
We always have this image of us flipping the iceberg. The AI tools we see today are just the tip of the iceberg: they invite you to build a character first, without thinking of the environment that surrounds them. We go the other way around, from the broadest details to every little nook and cranny, and on the very practical side, it’s just a few questions first, like a survey. “What is it? What direction are you going in?” So, we already pre-prompt the world that you will be working on by answering those simple questions. Then, the foundation bricks: timeline, locations, and what the groups are. And then we go to the characters.
You can create characters or generate characters based on the foundations you’ve built. The richer those foundations are, and the more detailed they are, then your output when it comes to generating characters will be richer. I always say, AI is great if you feed it with context. AI is starving for context. The whole invitation will be to please feed your world! It’s fine if you just have a hazy idea, but we’ll go step-by-step to feed it.
“You remain the owner of your property rights, of the vector database that we use for you. We are very, very responsible in that regard so as not to compromise any intellectual property that you use to fill the Aarda context”
Kaynã Oliveira
Before Aarda, narrative designers used to feed their world onto a spreadsheet or a Confluence page or a Google doc, and no one read it because no one reads in video game studios! But anyone can go to Aarda and ask about the world: “What is this object’s use?” So the game designers know it. Developers know how it works. Finally, narrative designers will spend proper time in Aarda, and they will know that everything they input will count and will matter.
Are you using OpenAI’s model? What foundation model are you using?
Constantin Berthelier: It’s running on OpenAI at the moment.
Kaynã Oliveira: We have plans to have smaller models to run specific tasks that we have. OpenAI is just cheaper for us to use now, but as we scale the business, we are going to use our own models for everything. The infrastructure is very modular, so we can plug in anything that we need now or in the future.
Creators’ intellectual property is important to them. I’m assuming there are privacy guardrails around the material people upload to your platform. How do you assure people that their data is safe?
Kaynã Oliveira: We don’t own anything. We will vectorise the documents that you send us, but you remain the owner of your property rights, of the vector database that we use for you. We are very, very responsible in that regard so as not to compromise any intellectual property that you use to fill the Aarda context.
Can you give us more examples of how AI fits into the whole game design process? What are the instances in which it’s used?
Kaynã Oliveira: I can mention two points. We can generate characters at mass during runtime, and those generated characters use the same world encyclopedia or world context that you build to have contextual characters generated there.
You can prompt those characters to be not only based on that context, but serve a specific purpose. If I want to find a murderer for the story, I can prompt that. If I want to have an innocent character, I can also prompt that. So you can direct that. Each of the attributes that these characters have are also prompted, so it’s prompts within prompts.
Not only that, but you can also use AI to generate arcs. You will have your world encyclopedia that will contain base-level knowledge of everything in the world. And you can create an arc, and knowledge bricks supporting this story arc. So we have the foundational knowledge for the game Shadows, but specific secret knowledge related to this specific murder and this specific mystery can also be prompted with the arc generation.
When it comes to story design, how do you handle information gating and spoilers? Potentially, you could ask an AI, “Who is Luke Skywalker’s father?” And it would just tell you it’s Darth Vader, which is obviously a massive spoiler. How do you manage the flow of narrative information?
Constantin Berthelier: Absolutely, this is one of the first things we’ve addressed with our knowledge brick system. We can decide who has this knowledge or not, and we have dormant ones that we can activate within the conversation. With the system we have, we can absolutely assure you that if you don’t know who Luke’s father is, you won’t know! Then, in the course of gameplay, I can activate that knowledge brick and inject that knowledge in, and then you will know. That’s absolutely solid, very robust. We have never encountered any mistakes, and this is the very core of our product. It’s absolutely ready for a video game pipeline to interweave gameplay and narratives.
“ I don’t think AI will kill the notion of effort. If you are motivated to do something of quality and something inspiring, if you have a message to give, there will always be effort. I like the idea that everyone can be a storyteller if they are given the proper tools”
Constantin Berthelier
To expand on that and your previous questions, if you work on an IP and you want to prevent wobbliness and weirdness, again, feed your project with information. It requires lots of human work. I always invite every narrative designer to explore those tools, because it’s going to take work if you want it to be efficient and for it to be fun. If you don’t want the model to go too crazy, lose track of your IP and go off-brand, it requires a lot of input with details of your IP. This solution is perfect for containing a lot of information, and the output will always be top-notch.
Your game, Shadows of Harrow City, uses English, French, Portuguese and Swedish. How does your platform handle multiple languages? Were there any particular challenges there?
Kaynã Oliveira: The world knowledge for all of those different languages is in the same language. It’s all in English. You can just decide which language you’ll have as input and output, but the cognition itself just needs to be in one language. Voice-to-voice communication, text-to-voice, text-to-text, everything can be in 18 languages, total. We just picked those because you have to translate the game app itself! But we support multi languages for all the dialogues.
Constantin Berthelier: I cover French, and I just had to translate very specific words. But all the rest, the [AI solution] covers pretty easily. The only weird thing we’ve encountered is that some French characters have a Canadian accent!
Let’s discuss some technical challenges of that. In Shadows of Harrow City, the player is conversing with the characters with their mic, and they’re answering, in a browser-based environment. You’ve got latency, speed, scale issues – how do you ensure it’s a fluid experience?
Constantin Berthelier: Kaynã is a genius!
Kaynã Oliveira: I hit my head against the wall many times. We initially had it running on native Unity, and porting it to the browser posed quite a few challenges. Maintaining a stable web socket communication and keeping the session alive, so you can go to the next character, come back to the same one, have the same chat history, and keep the conversation going, was indeed a challenge! Fortunately, we have all of this communication working inside our own plugin. So if somebody wants to have a WebGL build or a native build, they can use the Aarda plugin and have all this communication taken care of, both for text-to-text, voice-to-voice communication, multi-language all that is taken care of.
Your plan is for other narrative designers and game makers to be able to use Aarda AI to bring their ideas to life. What’s your road map? When do you plan to roll it out for people to license and use?
Constantin Berthelier: We plan on making Aarda accessible for more developers pretty soon: Q1 or Q2 this year. We already have some game developers using it: a few clients who have their own other projects, and are already feeding their projects with knowledge and talking to characters. It’s already usable for any video game developer out there (please reach out to us). We want to expand, of course, to facilitate the access for anyone to register and start building, if they’re professionals.
Later down the roadmap, we want to integrate Unreal pretty fast. Also, personally, I’m big on sandbox platforms. My dream is that we can integrate Aarda into Roblox, Fortnite, or any other sandbox games out there because what’s missing in those games is the consciousness of the world. For me, Roblox is a great brick game, but what’s missing is the ability to pretend that these blocks are a restaurant or a war zone or anything. There’s no world in those Sandbox games. For me, it’s just blocks for now, and I would like to instil consciousness and narrative into those worlds. We’d like to explore the possibility of exporting to those platforms later on. Let’s see what happens in the AI space.
“If you forget for one second about AI, before anything else, this is a way to organise your world. Anyone who has a story in their mind can open Aarda and start building and expanding. I believe that everyone has an idea for a story”
Constantin Berthelier
I see all this slop AI content everywhere, where there are videos that look good, but everything is out of context. This is how I would qualify this whole AI slop we’re getting every day: everything is out of context. So maybe Aarda could be the best solution to integrate into an image generator, and you could drive your own little drama, still using Aarda as a world-building feature. That’s what it is before anything else. But that’s for later. For now, it’s about making Aarda accessible to game developers: now you have to reach out to us for access, so [we will be making it] more public.
Kaynã Oliveira: We also want to improve the onboarding process in general. We don’t have everything open to the public yet because we are improving the user experience of the platform, always focusing on preventing abuse. We want to have a good structure for these foundational bricks and for the general prompting of your world. It’s a very powerful tool. But “with great power comes great responsibility!” So we’ll give you prompt templates. The platform allows you to not only have this big context of the world, but if the behaviour is not the one you want, you can actually change all the prompts in the pipeline and achieve the behaviour that you want. Those prompts will change depending on the type of world that you want to have and the type of bounded behaviour that you want to have. So we want to have a good onboarding flow to set these initial prompts that you are going to use, in case you don’t want to change or mess around with them. That’s very much on our road map.
AI has caused people to worry about their jobs. It sounds like there’s still a lot of work for the humans involved in your process, too. But do you think the abilities AI brings to game development means there’ll be fewer people working in the industry in five years’ time?
Constantin Berthelier: I’m going to be honest: I would say so, yes. I think the industry is going to downsize, and that video game-making will be made more accessible. But I would predict a new generation, a new breed of creators, because creativity will never go away. I don’t see AI killing creativity at all.
I think if there’s one job that could survive, in the crafting of video games, it would be storytelling. From what I see, everything is being made easier technically. At the end of the day, the one question that will remain is, “What is your story? What is your message? What do you want to say? What do you want to bring out to the world?” This will never die.
I don’t think AI will kill the notion of effort, either. If you are motivated to do something of quality and something inspiring, if you have a message to give, there will always be effort. Look at what we’re building. The first thing we told you is that it’s a lot of effort if you want to build a world. Don’t start with a character. Start with the beginning, middle development, and end: it’s all about structure. Nothing changes.
I like the idea that everyone can be a storyteller if they are given the proper tools. We’ll have excellent surprises. I hate the injustice of feeling that there are numerous talented storytellers out there who don’t have the network or the proper tools to build their ideas. What I saw when I first used ChatGPT is the possibility of unlocking that and giving people a way to drive their story and sell it to the world. So I see a bit of justice here also.
Do you have a sense of how disruptive AI is going to be generally?
Kaynã Oliveira: It’s absolutely inevitable. I think AI will be everywhere, and companies that can offer inference will be just as essential as light, internet, water, utilities. Inference will be a utility as well. We are going to use it everywhere.
I think in this increasingly technological society, every workflow has that. This is a quote from Hopium Diaries, “Those who are brave enough to look the Zeitgeist in the eye will reap huge rewards”. People won’t necessarily get substituted, but those who are brave enough to adapt and learn will reap great rewards. So let’s do it.
Further down the rabbit hole
What’s been happening in AI and games? Your essential news round-up:
PG Connects San Francisco is this month! The AI talks and panels feature reps from companies like Filuta AI, Rovio, Charisma.ai, NCSOFT, Supernova Games, Goodville and more. Tickets for the March 17-18 conference are still available.
Grok 3 launched, and people are using it to prompt entire games. Alex Finn shows how he got it to make a simplistic recreation of a GTA-style game using its Deepsearch and Think modes.
The Video Games Industry Memo newsletter asks why games studios aren’t more vocal among those supporting the Make It FAIr campaign.
The brains behind PUBG, Brendan Greene, has unveiled his new single-player open-world survival game, Prologue: Go Wayback! It uses machine learning to power its terrain-generating tech.
In a post on X, OpenAI boss Sam Altman said the company had to stagger the rollout of its GPT-4.5 model because they’re “out of GPUs.”
Activision has admitted to using AI-generated content in the shooter Call of Duty: Black Ops 6 after gamers complained.
Machine-learning site Hugging Face released FastRTC, an open-source Python library that could revolutionise how game developers implement real-time voice and video features in their games.
No, Microsoft’s Muse isn’t going to generate whole games, writes Eurogamer, quoting AI researcher Dr Michael Cook. Despite all the recent buzz, he explains, Muse exists to help human developers playtest new ideas.
You can watch Claude 3.7 Sonnet attempt to play through the classic Game Boy game Pokémon Red on Twitch. You can read about why on Anthropic’s blog.
Claude, the Anthropic model, is able to interact thanks to a virtual control system. It looks at screenshots of the game and tries to interpret what's happening. Researchers are enjoying seeing its reasoning.
