Evaluating AI-assisted code generation tools : performance, code quality, and developer confidence
Liao, Shaolin (2026)
Kandidaatintyö
2026
School of Engineering Science, Tietotekniikka
Kaikki oikeudet pidätetään.
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi-fe20260624102032
https://urn.fi/URN:NBN:fi-fe20260624102032
Tiivistelmä
Artificial intelligence-assisted programming tools are increasingly embedded in modern software engineering workflows. Tools such as GitHub Copilot, ChatGPT and Codeium support implementation, debugging, refactoring and code explanation. Their adoption raises a practical question for software engineering: whether these tools improve development performance without weakening code quality or developer understanding. This thesis evaluates AI-assisted code generation tools from three perspectives: task performance, generated code quality and developer confidence.
The study is designed as a system-oriented evaluation thesis combining a benchmark framework with a developer-perception questionnaire. The benchmark component compares common programming tasks such as algorithm implementation, REST API integration, CRUD development, debugging and refactoring. The perception component examines trust, explainability, verification behaviour and perceived learning support. The thesis argues that AI assisted coding tools should be understood as collaborative development assistants rather than autonomous software engineering agents. The contribution of the study is a structured evaluation framework that can be used to compare AI coding tools in a reproducible undergraduate-level software engineering context. The thesis also discusses how AI-assisted coding changes the developer role from direct implementation toward supervision, validation and integration of generated solutions.
The study is designed as a system-oriented evaluation thesis combining a benchmark framework with a developer-perception questionnaire. The benchmark component compares common programming tasks such as algorithm implementation, REST API integration, CRUD development, debugging and refactoring. The perception component examines trust, explainability, verification behaviour and perceived learning support. The thesis argues that AI assisted coding tools should be understood as collaborative development assistants rather than autonomous software engineering agents. The contribution of the study is a structured evaluation framework that can be used to compare AI coding tools in a reproducible undergraduate-level software engineering context. The thesis also discusses how AI-assisted coding changes the developer role from direct implementation toward supervision, validation and integration of generated solutions.