DOI: https://doi.org/10.1007/s10664-025-10791-2
Accepted: December 15, 2025
Journal: Empirical Software Engineering, Volume 31, 2026

Software developers spend an estimated 10% - 20% of their time on code review. With the rapid adoption of generative agents, the amount of code that needs to be reviewed is increasing at a faster rate than ever. Yet the tools for code review are still remarkably similar to the first tool, ICICLE, that was introduced over 30 years ago. There are significant misalignments between the tools used and the goals and actions of software developers, which reduces efficiency and adds frustration. The code review process is also challenging for many teams with a range of common antipatterns. Improving code review tools and processes has substantial potential benefits for the software engineering community.

In our paper, “Code Review as Decision-Making - Building a Cognitive Model from the Questions Asked During Code Review”, we present an ethnographic study involving 10 participants and 34 code reviews. The study builds a theoretical model of cognitive processes in code review, grounded in interviews and observations. Previously, code review has been understood primarily as a code comprehension activity. Developers reviewed the code to understand it, and once understood, writing comments and accepting or rejecting the patch came automatically. Current tools mirror this by focusing on a textual diff view seeking to isolate and explain the changes. However, through the data in our study, the similarities between the cognitive process in code review and decision-making processes, especially recognition-primed decision-making, become apparent. The Code Review as Decision-Making (CRDM) cognitive model shows how the developers move through two phases during the code review; first an orientation phase to establish context and rationale and then an analytical phase to understand, assess, and plan the rest of the review. Throughout the process several decisions must be taken, on writing comments, finding more information, voting, running the code locally, verifying continuous integration results, and so on.

Our work is novel and relevant for the ESEM research community because it provides a framework for understanding code review through a decision-making lens, rather than as a code comprehension activity. This can support development of better code review tools, with or without AI, that are inspired by tool advancements in decision-making support systems. It can also inform improvements to the code review process itself, by pointing to where and when developers would benefit from support and what kind of support would be helpful.