Interacting with AI Reasoning Models: Harnessing "Thoughts" for AI-Driven Software Engineering

This dataset accompanies the paper Interacting with AI Reasoning Models: Harnessing “Thoughts” for AI-Driven Software Engineering by Christoph Treude and Raula Gaikovina Kula. It contains all example cases discussed in the paper as well as the full corpus of reasoning traces analyzed in our study. The appendix includes: AI-generated reasoning traces from multiple models for the motivating examples (e.g., generating a PDF report, analyzing security vulnerabilities in C code), illustrating challenges such as hidden assumptions, divergent reasoning paths, and conflicting conclusions. The complete set of 100 reasoning traces collected from DeepSeek and Gemini for the 50 most recent Stack Overflow questions used in our qualitative study. Two data files:– coded-units.xlsx, containing all 7,000+ segmented reasoning units with their assigned qualitative codes.– reasoning-traces.xlsx, containing the raw reasoning traces, question texts, and metadata used for analysis. Together, these materials provide full transparency for replication, further qualitative or quantitative analysis, and future research on AI reasoning in software engineering.