Safer Workflows for AI-Assisted Coding

This project explored how AI-assisted coding is changing the way people build software, especially for students, designers, and early-stage developers. The team began by investigating how developers understand unfamiliar codebases. Their first prototype, Code Compass, was a VS Code extension that helped users ask questions about a repository, create code threads, and explore usage-style insights such as ownership, churn, and impact.

Through mentor feedback and product research, the team identified a deeper problem. AI tools were making it easier for people to generate code, but not necessarily easier to manage that code afterward. Many new builders could create scripts, apps, and components with AI, but still felt unsure about saving changes, recovering previous versions, or sharing work through Git and GitHub.

In response, the team pivoted from Code Compass to VIVA, a local-first desktop application for visual version control. VIVA was designed to make Git workflows safer and more approachable without requiring users to work in the terminal. Instead of exposing commands directly, the app translated common version control actions into clearer workflows such as saving progress, reviewing files, previewing updates, and managing branches with guardrails.

Safety became the core design principle. VIVA avoids risky automation by requiring explicit user consent before destructive or critical actions. It blocks dangerous workflows such as silent Git initialization, default-branch deletion, automatic force pushes, and automatic conflict resolution. The project also uses AI in focused ways, including commit message suggestions, natural language undo, file insight, untracked file review, smart conflict guidance, and weekly progress reports.

As industry mentors, 99P Labs and Honda Research Institute helped guide the team toward a more human-centered framing of the problem. The final project shows how developer tools can better support people who are building with AI but still need confidence, control, and protection as their work evolves. VIVA points toward a future where version control feels less like an expert-only system and more like a supportive layer for anyone learning to create software.