Type: url
Original: https://www.jstage.jst.go.jp/article/repit/2026/0/2026_90/_pdf/-char/en
Status: generated
Language: ja
Summary
This rePiT2026 symposium paper from Osaka Institute of Technology reports introducing spec-driven development with AI agents (GitHub Copilot) into a third-year undergraduate software development PBL course, using a four-phase workflow of investigation, planning, implementation and review in which the agent generates documents and code at each phase. Analysing students’ self-reported AI use, lines of code added and instructor-assessed code comprehension, the authors find that AI agents raised development efficiency but that students tended to proceed without fully understanding the code, and conclude that regular instructor checks of code comprehension and feedback are essential to preserve the course’s educational value.