Understanding Developer–LLM Collaboration in Software Development Environments

Abstract

Software development environments are beginning to integrate large language models as coding assistants and documentation tools. Although productivity gains are often reported, the collaborative dynamics between developers and language models remain poorly characterized. This paper examines how programmers use suggestion, explanation, and refactoring functions during everyday development tasks. Attention is given to trust formation, error correction strategies, and the negotiation of control between human and automated contributors. Observations from controlled workshops and remote pilot deployments indicate that developers treat language models less as autonomous agents and more as interactive reference systems. Productive use depends strongly on transparency and contextual awareness.