# Building a Doom-Like World to Explore Agentic Systems - Alexander Chernov - NDC Toronto 2026

## Executive summary

This talk presents an architectural framework for building complex agentic systems using a modified Doom-like game engine as a controlled testbed. The core concept is treating agents as 'semantic mirrors' of the game world state, allowing non-player characters (NPCs) to act autonomously while maintaining strict observability and reproducibility. The architecture emphasizes decoupling AI reasoning from the game loop via specialized components like the MCP Gateway, enabling real-world application of simulation techniques in fields such as pharmaceutical R&D.

## Key takeaways

- Agentic Systems Architecture: The system models agents as autonomous entities that perceive the environment and make decisions. The architecture is designed to be observable, attributable, and reproducible through structured world state changes (the 'semantic mirror').
- Two-Tiered Agentic Vision: To manage latency, a two-tier vision system is implemented: a fast, deterministic observer swarm (7 Hz) for basic tracking, and a slower, LLM-powered tier using 'Lenses' to extract complex semantic information from the environment.
- Architectural Components: Key components include the Policy Guard system (defining what agents can/cannot do), the MCP Gateway (Model Context Protocol) for external integration, and a Semantic Cache (Mosquito Dog) to reduce latency and cost by caching LLM responses.
- Reproducibility and Validation: The design ensures determinism through fixed control loops (e.g., 35 ticks per second), state machine transitions, and structured logging of events (JSONL). This allows for full replay and behavioral regression testing.

## Technical details

- System Architecture: The system uses a three-layer architecture: the Game Runtime (Doom engine), the Agentic System (FSM agent base class, Policy Guard, Zone Manager), and the Control Plane. The Control Plane includes the Doom Bridge, MCP Gateway, and Telemetry Pipeline.
- State Management: The world state is structured into five layers: Per-Actor State (position, health), Zone State (safe, contested, quarantine), Agent State (FSM, counters), Policy State, and Scenario State. This provides comprehensive context for agent behavior.
- Control Flow & Determinism: The game engine provides strict control loops (fixed tick rate, e.g., 35 Hz) and deterministic scheduling, which are leveraged to govern the complex, emergent behavior of LLM-based agents.
- Integration Layer: Communication between the game engine and external AI systems is managed by the MCP Gateway (Model Context Protocol), allowing pluggability for different LLM providers or other services. This creates a 'two-way blind system.'

## Practical implications

- The framework can be applied to real-world simulations, such as drug discovery (digital twins) or surgical planning, by replicating complex physical and chemical interactions.
- It provides a robust methodology for testing and governing emergent AI behavior before deployment in critical systems.
- The architecture supports cost optimization by implementing semantic caching and local LLM routing.

## Topics

Agentic Systems, Game Simulation, AI Architecture, Digital Twins, State Machines, NDC Conferences, GZ Doom / JX Doom, MCP Gateway, Semantic Cache (Mosquito Dog)

Source: https://www.youtube.com/watch?v=ZWND8IF_7Zk