External Chat or Editor-Integrated? An Empirical Study of AI-Assisted Workflows for Novice Game Development

Vibe coding enables novices to program through conversational intent specification and AI-assisted repair. In game development, generated code must be grounded in scene objects, component settings, and runtime behavior. This study compares two Unity workflows: External Chat, using web-based AI and standalone materials, and Editor-Integrated, embedding AI interaction and materials in the workspace. In a controlled study with 34 undergraduates, both conditions had the same observed self-reported binary completion count, but Editor-Integrated projects showed higher scene-integration scores. Screen-recording coding descriptively indicated more localized repair, project-context use, documentation reference, runtime-evidence prompting, and executable verification in Editor-Integrated, whereas External Chat involved more manual context transfer and whole-code regeneration. Secondary measures showed mixed patterns, including higher descriptive ratings on selected perception item groupings and exploratory, unadjusted follow-up differences in self-reported transfer confidence and continued learning needs. These findings suggest AI-supported programming environments should support system grounding, executable verification, and learner control.