"AI does not understand the real world.": AI Augmented Software Development

Authors:
Open Access
Publisher website
Google Scholar
Abstract
Generative artificial intelligence (AI) tools are increasingly embedded in software development work, reshaping how code is produced, reviewed, and maintained. Following the public release of ChatGPT and the expansion of the AI tooling ecosystem, developers now routinely encounter AI assistance in collaborative and security-sensitive contexts. Through interviews with eleven software developers familiar with generative AI tools and working across teams, we examine how developers adopt and evaluate AI in daily work, how AI use affects security-related practices, and how these tools shape collaboration and accountability. Our findings show that AI use is contingent and selective rather than habitual. Developers primarily rely on AI for low-complexity, low-accountability tasks, while deliberately limiting use in security-critical or system-defining work. Responsibility for correctness and safety is consistently framed as non-transferable leading many developers to refuse or limit AI use.
Funding Information
  • Digital Research Centre Denmark (167601)

This publication has 29 references indexed in Scilit: