This project evaluates agentic AI workflows by investigating how AI systems can be structured to assist and automate complex software development tasks, particularly within Unity-based environments. Through a series of experiments involving user interaction, procedural generation, workflow orchestration, and autonomous pipeline execution, the team explored how different levels of structure, tooling, and memory impact performance, reliability, and efficiency. The project culminates in identifying effective strategies for guiding AI agents across iterative development cycles while highlighting key limitations in autonomy, evaluation, and real-world applicability.