A research co-authored by SMU Associate Professor of Computer Science Christoph Treude showed that large language models (LLMs) can assist but not replace humans in software engineering annotation, performing well in low-context, deductive tasks but poorly in high-context tasks. He noted that model-model agreement and model confidence can guide safe automation, emphasising that LLMs should accelerate rather than replace human judgment, and described the study as a first step toward broader research on artificial intelligence (AI) support in qualitative software engineering.
