Keywords
Generative AI, Problem Solving, Self-efficacy

Technologies
Generative AI, ChatGPT

This project investigates how software developers think and make decisions when using Generative AI (GenAI) tools such as ChatGPT. Using a mixed-methods design that combines experimental and qualitative approaches, participants will complete programming tasks either with or without AI assistance while verbalising their thoughts. The study examines how developers’ cognitive models of GenAI relate to their trust and self-efficacy, reliance, processes and task satisfaction. Findings will contribute to understanding human–AI interaction in software development and guide responsible AI integration. Within this study, my individual contribution focused on the role of self-efficacy in AI-assisted programming. Following collaborative data collection, survey and transcription, I led the qualitative analysis of think-aloud transcripts relating to self-efficacy, applying Verbal Protocol Analysis to identify how developers' beliefs in their own capabilities shaped their engagement with GenAI tools. This analysis explores whether AI assistance supports or undermines developers' confidence in their own coding abilities.