Reliable agent libraries need repeatability, safe tool boundaries, operational visibility, and maintainable abstractions. This session uses Szork, a voice-controlled, AI-powered adventure game built with LLM4S, as a demo that turns a playful agent experience into a practical architecture discussion.

Through playing with Szork in a live walkthrough, we will show how LLM4S uses provider integration, typed tool calling, conversation state, error handling, observability, secure execution boundaries, and adds extensibility with MCPs. The demo stays compact while grounding the architecture in a real application: a game loop where the model interprets natural language, invokes capabilities, maintains context, and responds through a user-facing experience.

The core of the talk is the design rationale behind LLM4S: why Scala and the JVM matter for agent systems, how type-safe APIs reduce integration risk, how functional design improves testability and composability, and how open infrastructure choices support enterprise adoption. Attendees leave with practical lessons for building reliable agent applications using LLM4S.