Imagine a world where non-player characters (NPCs) in video games are no longer limited by scripted branching narratives, but instead driven by an infinitely complex, trillion-parameter neural network. This is the future being shaped by the innovative work at The Garden in the Machine, led by Parag Mital. Their platform, Emergentic AI, is transforming game storytelling by enabling AI-powered multi-agent systems that create procedural narratives, where every player experience is unique and dynamic.

The Garden in the Machine is a creative studio specializing in emerging technologies, machine learning, and AI. With over 15 years of experience collaborating with artists and institutions, they explore what's possible when machine learning meets storytelling. Their recent focus is on agentic AI—a buzzword gaining traction—which refers to AI systems composed of multiple agents acting together in a coordinated way.

Parag Mital explains, “We’ve built a multi-agent authoring platform called Emergentic AI that allows you to author multiple agents that act together.” Initially designed to power NPCs in video games, this platform leverages procedural narratives driven by trillion-parameter neural networks. This means NPCs can interact with each other and the player in unpredictable, emergent ways, enhancing immersion and replayability.

Traditional game narratives often rely on branching storylines, where player choices lead to different pre-scripted outcomes. However, with multi-agent AI, developers can create procedural narratives that evolve in real-time based on player actions, interactions between AI agents, and even live data inputs such as internet feeds.

Parag Mital highlights the exciting possibilities: “Now you’ve got this capacity to build procedural narratives, bring in the player experience, help them onboard to a game, or bring in real-time internet feeds and somehow lead that into the story. The possibilities are really endless.”

One fascinating implication is the potential for multiple AI agents to interact with each other and players simultaneously, creating a rich tapestry of emergent storytelling. Imagine if a major studio like Rockstar adopted this technology—what would a multi-agent expansion pack for Grand Theft Auto look like?

For game developers interested in integrating agentic AI, the process involves combining traditional storytelling tools with new iterative design methods. Developers craft detailed system prompts that inform AI agents about their character’s persona, motivations, desires, conflicts, and potential resolutions.

As Parag explains, “If you’ve got the normal tools of storytelling, you know how to design characters, plots, narratives, branching narratives, all the normal tools that game designers are used to, you’ll be in good company. But now you’ve got this iterative process where you’re designing agentic characters. These characters are taking every single word that you craft into their system prompt and using that to act and feel like a person.”

An example project involved simulating the historic succession narrative of the Mughal Empire, focusing on Emperor Shah Jahan and the rivalry between his sons. The AI agents embodied characters who could spread misinformation, engage in political intrigue, and even commit violence to secure the throne. This historic simulation showcased how AI can bring complex character dynamics and emergent storytelling to life.

One might wonder whether these AI agents require extensive training on specific datasets. Parag notes that for well-known historical characters, the underlying large language models already possess substantial knowledge, so minimal additional data is needed. However, for newly created characters, developers must provide stylistic references, backstories, motivations, and plot events to build a rich persona.

Regarding the technology behind Emergentic AI, Parag mentions, “We’re kind of model agnostic. You can use OpenAI, Google Gemini, or a local model.” Different models bring unique strengths and personalities to storytelling. For example, some developers find Claude better suited for creative writing, while others experiment with models to explore political biases by pitting them against each other—a project they call the Ministry of Truth.

As AI integration deepens in game development, ethical questions arise, especially concerning data usage. One major concern is whether AI models are trained on copyrighted material without permission. AAA studios tend to be cautious about using external models trained on third-party data.

Parag emphasizes the importance of responsible AI development: “People working in terms of generated AI should build their own models, work with their own data sources.” Large studios have vast repositories of proprietary data—from previous games, storylines, and character databases—that they can use to train safe, specialized AI models. This approach ensures ownership, control, and ethical use of data.

Looking ahead, Parag predicts a shift from generic single-agent AI assistants toward multi-agent systems with distinct personalities and specializations. Instead of one AI trying to do everything, developers and users will curate teams of AI agents tailored to specific roles.

“How do you build personality into that? How do you specialize? That’s where some of the creativity and personality comes through,” Parag explains. Imagine having a therapist AI modeled on a particular practice, a 3D world-building assistant for Blender, or a writing co-pilot with expertise in certain topics. This specialized multi-agent approach will create more meaningful and personalized AI experiences.

In the gaming context, this means players might choose their own AI “court” of agents to support them in different ways, enriching gameplay and narrative depth.

If you’re intrigued by the potential of AI-powered storytelling and want to learn more, check out Emergentic.ai. The Garden in the Machine also shares their creative explorations on Instagram under the handle The Garden in the Machine.

As AI continues to reshape gaming and storytelling, platforms like Emergentic AI are at the forefront of a revolution—where NPCs don’t just follow scripts but engage in dynamic, living narratives powered by trillion-parameter neural networks.