Canary AI breaks coding help into skill-adaptive steps for better collaboration

Collaborative programming in classrooms often devolves into isolated work because students find it too effortful to understand a teammate's entire problem at once. To address this, researchers from multiple universities (including Virginia Tech) present Canary, a system that supports peer scaffolding by breaking down programming obstacles into smaller, skill-adaptive steps. Using AI, Canary converts a complex problem into a “ladder” that starts with easy, actionable fixes and gradually progresses to more challenging logic. It also proactively alerts potential helpers to specific places where they can begin contributing, removing the initial barrier of task comprehension.

In their evaluation, the team found that this staged approach significantly reduces the feeling of overwhelm among students, leading to more frequent and effective collaborative interactions. The system’s design is grounded in formative studies that identified the key bottleneck: even willing collaborators struggle to dive into a peer’s unfamiliar code. By offering a gradual ramp-up, Canary enables quick, meaningful contributions and turns real-time programming sessions into truly collaborative learning experiences. The paper (accepted to VL/HCC 2026) demonstrates that skill-adaptive scaffolding can transform how students help each other in code.