Impact of generative AI on software team collaboration and code review workflows

Abstract

Generative AI is now used everywhere in modern software engineering. This change directly affects how developers do their daily work. When tools like GitHub Copilot first arrived, the software industry focused mostly on one thing: how fast a single developer could type code. But this excitement missed a bigger problem inside engineering teams. Software development is a team effort; it requires a shared understanding of the system and collective code ownership. When developers push huge amounts of machine-generated code into a shared Git repository, team coordination becomes difficult. Traditional agile practices are not ready for this pressure. This thesis investigates the friction caused by AI tools during teamwork, especially in pull request discussions and daily knowledge sharing. This study uses a systematic literature review to analyze 30 recent empirical papers. The data shows that AI makes coding faster, but it also creates a "knowledge void". Developers often accept and commit code logic without understanding it, which severely breaks traditional peer review workflows. On the metric side, real-world telemetry reveals that AI-augmented workflows introduce substantial coordination overhead, with specific repository integration latency increasing by up to 22%. At the same time, senior maintainers face a massive increase in their daily cognitive load. These outcomes prove that legacy agile frameworks are no longer enough. The main goal of this research is to help engineering managers update traditional agile frameworks by introducing mandatory human-to-human verification layers to protect codebases from hidden technical debt.