Autonomous Game Engineering Network for Task Synthesis
Yu-Wei TSENG
Software Development
Autonomous Game Engineering Network for Task Synthesis
Game development demands collaboration across programming, art, design, and quality assurance, yet current AI coding tools operate as isolated, single-agent systems with no coordination, shared memory, or fault recovery. For my Thesis, I designed and built AGENTS( Autonomous Game Engineering Network for Task Synthesis), a multi-agent orchestration system where specialized AI agents collaborate autonomously to perform game development tasks under human supervision.
I chose this project because I saw a clear gap between how real game studios operate as coordinated teams and how AI tools assist developers as solo performers. I wanted to explore whether the team model could be replicated with AI agents that communicate, specialize, and recover from failures together. AGENTS organizes AI agents into roles mirroring a real studio: a producer decomposes requests into task
graphs, leads dispatch work by role, workers execute in isolated environments, and a QA agent validates results before human approval. The system communicates through KADI, a custom message broker I built, making the architecture language-agnostic and extensible. I also built a custom C ++ game engine with embedded V8 JavaScript scripting, allowing AI agents to control a live 3D render in real time.
This project spanned two semesters. I developed practical expertise in distributed systems, message broker design, LLM( Large Language Model) orchestration, and real-time engine architecture. The system demonstrates a viable path toward AIassisted game development at the team level, which could reduce iteration cycles and lower the barrier for indie developers.
100 SOFTWARE DEVELOPMENT