DiSEL Disruptions in
Software Engineering Lab

Research

We define disruptions as critical junctures where established software development and maintenance practices collide with unanticipated sociotechnical change, requiring the re-engineering of practices, tools, and workflows. Our work centers around using empirical research and GenAI tooling development to help developers navigate them.

GenAI in software development

GenAI coding agents and assistants are rapidly changing how code gets written, reviewed, and merged. We're measuring the effects on velocity, quality, and security in real-world repositories, and helping shape the evolution of software development and maintenance processes in the GenAI era. Recent work covers the post-merge fate of agentic code, the short-term speed and long-term complexity of agent-assisted development, and why developers adopt or avoid GenAI tools.

Open source sustainability and security

Open source software supply chains form our digital infrastructure, yet the projects they depend on are maintained by people who can leave. We study dependency abandonment and how downstream users respond. More recently the introduction of GenAI into open source communities has had significant impacts on how our digital infrastructure is built and maintained, and we study those changes and how to support development teams as they navigate them. Recent work: Abandabot, the npm response to abandonment, and AI slop in open source.

Code review and the software supply chain

Code review is where quality, security, and ownership get enforced, and it is under pressure from GenAI automation. We study what teams get out of review beyond catching bugs, how automated review changes security outcomes, and how trust moves through software supply chains. Recent work: 3100 opinions on code review and decomposing trust in supply chains.

How we work

Mixed methods: large-scale quantitative analysis of repositories and ecosystems, paired with qualitative work with practitioners. Measure first, then evaluate, then improve. Findings inform tooling and intervention design.