Description:

The use of AI assistants allows developers to generate code, tests, and documentation more quickly. As a result, easily measurable indicators such as lines of code, pull requests, commits, or tokens consumed may increase. However, these metrics do not necessarily show whether better software is being produced. More output may also lead to additional review effort, rework, technical debt, or defects.

The goal of this master’s thesis is to develop a scientifically grounded and practically applicable measurement framework for AI-assisted software engineering. Instead of evaluating productivity through a single activity metric, the framework should consider multiple dimensions, including development speed, software quality, review effort, rework, delivery stability, and developer experience.