An Empirical Evaluation of AI-Assisted Code Generation on Developer Productivity and Code Quality: A Case Study of .NET Applications

Context:AI-assisted code generation tools such as GitHub Copilot and ChatGPT are increasingly integrated into software development workflows. However, empirical evidence regarding their effects on software quality, maintainability, cognitive workload, and enterprise-oriented development remains limited.

Objective:This study investigates the impact of AI-assisted code generation on developer productivity, software quality, maintainability, structural complexity, and cognitive workload in enterprise-oriented .NET development environments.

Method:A controlled repeated-measures experiment was conducted involving 60 developers across junior, intermediate, and senior experience levels. Participants completed eight enterprise-oriented programming tasks under both traditional and AI-assisted development conditions, resulting in 960 task observations. Evaluation metrics included task completion time, implementation errors, cyclomatic complexity, maintainability index, code smells, lines of code, and NASA-TLX cognitive workload scores.

Results:AI-assisted development reduced average task completion time, implementation errors, code smells, cyclomatic complexity, and perceived cognitive workload while improving maintainability scores. Productivity gains were most pronounced among junior developers. However, some AI-generated implementations introduced redundant abstractions, increased code verbosity, and required additional validation and refactoring.

Conclusion:AI-assisted programming can enhance productivity and several software quality characteristics in enterprise-oriented .NET development environments. Nevertheless, effective use of AI tools remains dependent on developer expertise, task characteristics, and continued human oversight to ensure correctness, maintainability, and architectural consistency.