Beyond Code Generation: Reassessing Software Quality in AI-Assisted Development

Abstract Generative Artificial Intelligence (genAI) is fundamentally transforming software engineering practices by shifting code production from deterministic, human-authored processes to AI-assisted development workflows characterised by probabilistic behaviour, prompt sensitivity, and increased process complexity. This transformation entails significant challenges for software quality evaluation, as traditional metrics, developed under the assumptions of determinism and reproducibility, are increasingly insufficient for assessing AI-assisted development. This position paper synthesises recent empirical and theoretical developments across three complementary dimensions of this transformation: the human dimension, examining the cognitive and educational implications of genAI adoption; the artefact dimension, analysing the evolution from code-centric development toward Everything-as-Code; and the technological dimension, focusing on emerging multi-agent software engineering workflows. The analysis examines how prompt sensitivity, output variability, governance challenges, and the expanding scope of Everything-as-Code practices complicate existing quality frameworks, and contends that evaluation methodologies must evolve toward process-aware, context-aware, cross-artifact, and workflow-level approaches. The educational implications of AI-assisted programming are also addressed, with particular emphasis on the lasting impact on professional software quality. This work provides a conceptual map of the changing software engineering landscape and identifies critical areas for further empirical investigation.