The State of Generative AI in Software Development: Insights from Literature and a Developer Survey

A new academic study titled "The State of Generative AI in Software Development" provides a comprehensive look at how AI is reshaping software engineering. Conducted by researchers Vincent Gurgul, Robin Gubela, and Stefan Lessmann, the work combines a systematic literature review with a survey of 65 professional developers. The data reveals that 79% of surveyed developers now use Generative AI tools like ChatGPT or Claude on a daily basis, with a strong preference for browser-based Large Language Models (LLMs) over those integrated directly into development environments like IDEs.

The research quantifies the impact, showing that over 70% of developers report at least halving the time required for routine tasks like writing boilerplate code and generating documentation. This high impact is concentrated in the middle phases of the Software Development Lifecycle (SDLC)—specifically design, implementation, and testing. In contrast, the study found markedly lower reported benefits for early-phase activities like project planning and requirements analysis, suggesting AI is a better assistant for execution than for initial strategic thinking.

Beyond productivity, the paper highlights a significant shift in value creation. As AI automates routine coding, the focus for human developers moves toward ensuring specification quality, architectural reasoning, and maintaining oversight. The survey also notes that governance is maturing, with two-thirds of organizations having established formal or informal guidelines for AI use. However, the authors caution that risks like uncritical adoption, potential skill erosion, and the accumulation of AI-generated technical debt necessitate robust governance and human-in-the-loop mechanisms to ensure sustainable and effective integration.