HUMAN AGENCY IN THE ERA OF AI-POWERED CONVERSATIONAL LARGE LANGUAGE MODELS: AN INTERPRETIVE SYNTHESIS OF DECISION-MAKING IN SOFTWARE ENGINEERING

Abstract

The rapid integration of artificial intelligence (AI), generative AI, and conversational large language models (LLMs) has influenced software engineering by changing how developers generate, evaluate, and implement technical solutions. This study examined how these developments have shaped human agency in software engineering decision-making through a qualitative interpretive synthesis of existing secondary literature. Drawing on research concerning LLMs, AI coding assistants, human-AI collaboration, trust, automation bias, verification, expertise, and emerging agentic systems, the study synthesized evidence across heterogeneous scholarly sources. Four higher-order themes emerged: AI as an enabler of agency augmentation, negotiated agency through human-LLM interaction, agency constraints through trust and delegation, and the reconfiguration of expertise, responsibility, and control. The synthesis further identified three configurations of human agency: augmented agency, negotiated agency, and dependent agency. The findings suggested that AI could expand developers' capabilities while simultaneously creating risks of excessive reliance and reduced independent judgment. The study proposed an integrated framework emphasizing AI capability, interaction, trust, verification, expertise, task complexity, reliability, and organizational governance as key determinants shaping human agency.