The Impact of Large Language Models on Software Development Productivity: An Empirical Study of AI-Assisted Programming Tools.

Abstract

Artificial intelligence-assisted programming tools, including GitHub Copilot, ChatGPT, Claude Code, and Codex, have been adopted by a majority of professional software developers, driven by claims of substantial productivity gains. However, empirical evidence for these claims is dispersed across studies differing in method, scale, and outcome definition, and the two existing systematic reviews addressing this question are limited by peer-review-only inclusion criteria and search cut-offs that predate much of the most recent evidence. This study conducts an integrative literature review of 35 purposively assembled and quality-appraised sources (2024–2026) to synthesize evidence on the impact of large language model (LLM)-based programming tools on software development productivity, code quality and security, developer experience, and the factors that moderate these effects. Sources were classified into four evidence-quality tiers to make transparent the wide range of methodological rigor across peer-reviewed studies, preprints, and grey literature. Findings indicate a directionally consistent but magnitude-uncertain productivity gain, ranging from a population-scale estimate of 6.4% to considerably larger effects in smaller, less transparent studies alongside a genuinely contested quality and security picture, in which objectively measured risks are detected by only a minority of the wider field's evaluation practices. Developer experience level emerged as the most consistent moderating factor, shaping both the magnitude and the type of benefit realized. These findings are synthesized through a proposed AI-Assisted Development Impact Framework, which integrates productivity, quality, security, and developer-experience outcomes within a single evidence-weighted structure. The study extends two existing systematic reviews temporally and methodologically, offering researchers, practitioners, and educators a current, critically appraised account of what the evidence does and does not support regarding AI-assisted programming tool adoption.