Agile Retrospectives: What went well? What didn't go well? What should we do?
Maria Spichkova, Hina Lee, Kevin Iwan, Madeleine Zwart, Yuwon Yoon, Xiaohan Qin
TL;DR
The paper addresses improving Agile/Scrum retrospectives by combining AI-based information interaction with data visualization to support psychological safety and productive discussion. Using GPT-4 Turbo, the authors evaluate auto-labeling of retro-comments on a manually created dataset of 200 items categorized into four labels, and prototype RetroAI++ with a web-based interface and automated comment allocation. Results show 74–78% match rates under different prompt designs, with issues including missing items and occasional multi-label allocations, informing design choices like allowing both automated and manual annotation. The work contributes a practical prototype and empirical insights into applying LLMs to retrospective analysis, with implications for novice teams and distributed contexts.
Abstract
In Agile/Scrum software development, the idea of retrospective meetings (retros) is one of the core elements of the project process. In this paper, we present our work in progress focusing on two aspects: analysis of potential usage of generative AI for information interaction within retrospective meetings, and visualisation of retros' information to software development teams. We also present our prototype tool RetroAI++, focusing on retros-related functionalities.