一句话：通过黑客松实地观察与访谈，揭示团队对生成式AI的选择性、低风险使用及其准确性、提示编写与社交顾虑。
生成式AI工具在软件开发中日渐普及，但团队在真实、时间受限环境中如何协作使用尚不明确。本文研究为期两天的黑客松，鼓励参与者使用GenAI，结合16次观察与9次后续访谈。结果显示，使用具有选择性，多集中于简单编码与想法完善等低风险任务；参与者用其收集和提炼想法，有时转化为可行方案，但因准确性担忧、提示编写困难及不愿公开使用AI而保持谨慎。研究指出技术能力、社会期望与用户信心共同塑造协作场景中的GenAI使用，并为支持透明有效AI使用的环境设计提供启示。
Generative AI (GenAI) tools are increasingly used in software development, yet little is known about how teams use them collaboratively in real-world, time-constrained settings. We studied a 2-day hackathon where participants were encouraged to use GenAI, combining 16 observations with 9 follow-up interviews. We found that GenAI use was selective, often focused on lower-risk tasks such as simple coding and idea refinement. Participants described using GenAI to gather and refine ideas, sometimes turning them into workable solutions, but were often cautious due to concerns about accuracy, difficulty crafting prompts, and discomfort with openly using AI. These findings suggest that GenAI use in collaborative settings is shaped by technical capabilities, social expectations, and user confidence. We discuss implications for designing environments that support transparent and effective AI use.
@inproceedings{AcharyaCPOEDGGS26,
title = {Shh! Don't Tell Them I Use GenAI: Exploring How Participants Use Generative AI in Hackathons},
author = {Pawan Acharya and Tayana Conte and Jacob Penney and Pedro Oliveira and Misan Paul Etchie and Jared Duval and Bruno Gadelha and Marco Gerosa and Igor Steinmacher},
booktitle = {FSE},
year = {2026},
doi = {10.1145/3803437.3806711}
}