As software engineering moves toward SE 3.0, AI agents are increasingly used to carry out development tasks and contribute changes to software projects. It is therefore important to understand the extent of these contributions and how human developers review and intervene, since these factors shape the risks of delegating work to AI agents. While recent studies have examined how AI agents support software development tasks (e.g., code generation, issue resolution, and PR automation), their role in documentation tasks remains underexplored–even though documentation is widely consumed and shapes how developers understand and use software.

The reuse of already existing code is widely considered a popular software development practice, that provides both benefits and drawbacks for all stakeholders involved. Prior work reports on how code reuse is a common practice in software development projects and data science projects such as machine learning pipelines. Recently, there has been much code reuse work in the context of competitive programming. Although there is work such as detecting plagiarism, there is no work that studies how a competitor will reuse their own code. In this paper, we present a preliminary study on the code reuse behavior of three grandmasters’ Jupyter notebooks in the Kaggle Competitions, an online competition platform for data scientists, and report the types of code they often reuse. Grandmasters are the highest level reached in competitions (novice, expert, master, and grandmaster). We find that Grandmasters are less likely to reuse specialized code, but instead, tend to reuse common functions like importing packages (importing the pandas library). They are most likely to reuse common abstractions like importing packages, configurations, file IO operations, show data, plotting graphs, defining functions, and exploring files. The work opens up new research potential into recommending how developers can reuse their own code.

Data visualization becomes a crucial component in data analytics, especially data exploration, understanding, and analysis. Effective data visualization impacts decision-making and aids in discovering and understanding relationships. It leads to benefits in data-intensive software development tasks e.g., feature engineering in machine learning-based software projects. However, it is unknown how visualizations are used in competitive programming. The idea of this paper is to report early results on what visualizations are prevalent in competitive programming. Grandmasters are the highest level reached in competitions (novice, expert, master, and grandmaster). Analyzing the visualizations of 7 high-rank competitors (i.e., Grandmaster) in Kaggle, we identify and present a catalog of visualizations used to both tell a story from the data, as well as explain the process and pipelines involved to explain their coding solutions. Our taxonomy includes nine types from over 821 visualizations in 68 instances of Jupyter notebooks. Furthermore, most visualizations are for data analysis for distribution (DA Distribution), and frequency (DA Frequency) are most used. We envision that this catalog can be useful to better understand different situations in which to employ these visualizations.