The Stack Overflow Podcast Building AI for consumer applications isn’t all fun and games
13 snips
Sep 2, 2025 
Kylan Gibbs, CEO of Inworld and former DeepMind researcher, dives into the world of interactive AI for virtual realities. He discusses the technical challenges of creating immersive gaming experiences while prioritizing user accessibility and experience. The conversation touches on developing a universal AI runtime for various platforms and the optimization of AI for mobile applications. Gibbs emphasizes a human-centered approach in AI development and the importance of involving users in the creation process to enhance relevance and usability.
AI Snips
Chapters
Transcript
Episode notes
Enable Real-Time Swappable Components
- Give developers flexibility to swap models, prompts, and hardware without redeploying.
- Build a runtime that supports real-time user/context understanding and universal mutability to improve UX quickly.
Agents As Self-Modifying Model Graphs
- Kylan framed advanced agents as model graphs that can update models, prompts, and even code in real time.
- He called this digital autogeny and saw it as the next stage where AI optimizes itself for user experience.
Detect Device Capabilities For Local Inference
- Detect device capabilities at runtime and migrate inference locally when possible to improve latency and cost.
- Keep server default but design hybrid nodes that can run locally or remotely per device.