Building a functional VR environment from scratch is not a beginner task. Yet students in a 16-week undergraduate course at Fenwick Institute managed to ship working prototypes, and the tools they used tell an interesting story.
The course allowed students to choose their own AI coding assistants. Most gravitated toward GitHub Copilot and Tabnine for in-editor suggestions, while a smaller group preferred Codeium for its free tier and Sourcegraph Cody for navigating unfamiliar codebases.
Where the tools actually helped
The biggest time savings came during environment scripting. Students reported that GitHub Copilot could complete repetitive Unity C# patterns that would otherwise take 20 to 30 minutes per session. Codeium handled boilerplate for interaction triggers reliably.
Where students still struggled
- AI suggestions for VR physics interactions were often syntactically correct but logically wrong
- Sourcegraph Cody helped navigate legacy code but could not explain why certain VR rendering choices had been made
- Students who relied too heavily on Tabnine early on missed foundational understanding of the scene graph
The course coordinator, Petra Lindqvist, noted that the strongest projects came from students who used AI tools to accelerate work they already understood, not to bypass learning it.