Explainable Artificial Intelligence (XAI) Techniques for the Software Development Lifecycle: A Phase-Specific Framework

Abstract

Artificial Intelligence (AI) is rapidly being integrated into the Software Development Lifecycle (SDLC) to automate tasks ranging from defect prediction to requirements analysis. However, the “black-box” nature of these predictive models creates a significant barrier to trust, as developers cannot easily verify the reasoning behind AI-generated outputs. Current research disproportionately focuses on the maintenance phase, leaving practitioners without structured guidance for earlier lifecycle stages. To address this, we conducted a systematic literature review following PRISMA guidelines, detailing the rigorous search, filtering, and selection of current XAI-in-SE literature. We critically compare established foundational methodologies—including LIME, SHAP, and Counterfactuals—against recent advancements in inherent LLM interpretability and multiagent orchestration. Building upon this analysis, we propose a unified, phase-specific conceptual framework that maps these techniques to the SDLC based on stakeholder technical proficiency. While the proposed architecture remains conceptual, it provides a standardized theoretical blueprint for designing transparent, auditable, and trustworthy AI-assisted software engineering pipelines.