Multi-Agent Driven Self-Generating Game
David01354 · reddit · 2026-07-19
The author created a self-generating game called "UnderBuild" to visualize agents as NPCs and demonstrate the value of multi-agent workflows.
In the project, every agent is independent, equipped with its own prompts, context window, and perspective. Combined with a hardcoded workflow, it leverages the agents' decision-making capabilities while retaining a degree of determinism. The author also plans to add "emotion" and "memory" to the agents to influence their reasoning and narrative.
Technically, the game is built with C/ MonoGame, and the scripting layer uses Jint to host JavaScript. Agents can write and register JS scripts at runtime without reloading. The Script API acts as both the agent's facade and its permission boundary; the more powerful the API, the stronger the agent's abilities. The author also mentions using different models for division of labor: gpt5-mini for high-frequency rapid reasoning, fable 5 acting as a "senior developer," and gpt5.6-luna for intermediate tasks.
More from coding & agent
- Devin adds e2b sandboxes for remote agent execution — badphilosopher · 2026-07-22
- Hermes Agent Refactoring Proposal: Decoupling via Event Bus and Monorepo Slicing — Promptmethus · 2026-07-22
- ty now reads Pydantic config keywords and field metadata — charliermarsh · 2026-07-22
- Pensar Launches AI Security Agent to Autonomously Discover and Patch 0-Days — andriy_mulyar · 2026-07-22
- ty adds first-class Pydantic support, including strict and lax field handling — charliermarsh · 2026-07-22
- Google launches Gemini 3.5 Flash Cyber for CodeMender, with limited access for governments — GoogleAI · 2026-07-22