Every spring, LUdev places teams of six final-year Computer Science and AI students in real software projects with external clients. As part of a 6 EC course, these students spend roughly 500 hours over five months turning an initial idea into a working prototype or MVP, while simultaneously learning how to operate as a software team. The result is a setting that looks remarkably similar to professional software development, but with course-mandated deadlines, teams with (very) diverse backgrounds, and an educational responsibility.

This talk explores what we have learned from running that model: what makes projects succeed or stall, how to scope work when the deadline cannot move, how client involvement affects progress, and how to identify problems before they become impossible to recover from. We will look at concrete project experiences, recurring failure modes, and the practices LUdev has developed to manage them.