How Software Engineers Engage with AI: A Pragmatic Workflow
This work addresses the challenge of integrating AI tools effectively into software engineering workflows for practitioners, though it is incremental as it builds on existing practices with structured guidance.
The paper tackles the problem of software engineers' uncertainty in trusting, refining, or discarding AI-generated artifacts like code and tests by presenting a pragmatic workflow and decision model derived from industrial observations. It demonstrates the workflow's practical value through real-world scenarios, offering structured guidance for quality-aware AI tool use in software engineering.
Artificial Intelligence (AI) tools such as GitHub Copilot and ChatGPT are increasingly used in software engineering (SE) for tasks such as code, test, and documentation generation. However, engineers often face uncertainty about when to trust, refine, or discard AI-generated artifacts. We present a pragmatic workflow, complemented by a four-quadrant decision model, that formalizes how developers iteratively prompt, inspect, refine, and, when needed, fall back to manual work. The workflow and decision model were derived from a grey literature review and field observations across three industrial settings in Türkiye and Azerbaijan. Two real-world scenarios demonstrate the workflow's practical value, showing how engineers navigate key decision points when using AI. Our approach offers lightweight, structured guidance to support more deliberate and quality-aware use of AI tools in everyday SE tasks.