Rule Taxonomy and Evolution in AI IDEs: A Mining and Survey Study
A new study analyzed the use of "Rules" within AI-powered Integrated Development Environments (IDEs), which help developers guide Large Language Models. Researchers extracted over 7,300 rules from 83 open-source projects, categorizing them into 5 primary and 25 secondary types. Their findings revealed a disconnect between what developers deem important (architectural constraints) and what rules are most common (workflow and code formatting). The study also found that rules are frequently updated, with most evolution driven by adding context or enriching existing rules, and these updates significantly improve artifact compliance by over 20%. AI
IMPACT Provides empirical insights for optimizing prompting strategies and designing better conflict-detection mechanisms in AI IDEs.