Game engines combine strict control loops, complex state transitions, and real-time feedback, making them an ideal environment for exploring agent-based and AI-enabled system design.

This talk presents a Doom-like simulation used as a controlled testbed for designing and evaluating agentic systems. Non-player characters act as autonomous agents, the game world serves as a shared state substrate, and the engine loop functions as a control plane responsible for scheduling, rules, and constraints.

Rather than focusing on graphics or gameplay mechanics, the session examines architecture: how world state is modeled, how agents perceive and act, where determinism is required, and how autonomy is bounded. A key invariant explored is that all agent actions must be observable, attributable, and reproducible through world state changes.

The goal is to show how techniques from game development - simulation loops, event systems, and spatial reasoning - can inform the design of real-world agentic and AI-enabled applications.