AI-Assisted and Low-Code Development in Practice: A Case Study of SocialMize on Developer Productivity, Code Quality, and Perceived Trust and Usability

Abstract

This thesis investigates the impact of AI-assisted coding tools and low-code development platforms on software developer productivity, code quality, and perceived trust and usability, using SocialMize, a real production SaaS application, as the primary research context. A multi-method approach was employed, combining a longitudinal case study of the SocialMize development process, a controlled within-subjects micro-task experiment, and a practitioner survey of thirty software developers. The thesis addresses four research questions: how AI-assisted and low-code tools influence the everyday activities of a developer working on a real software project; which parts of the development process benefit most and least from these tools; how development speed and short-term code quality compare between AI-assisted and conventional approaches; and how developers perceive trust, usability, and cognitive load when working with these tools. The case study analysed 3,945 commits and the project's Lovable prompt history; the experiment compared three frontend tasks in manual and AI-assisted conditions. AI assistance accelerated pattern-rich work, with the experiment showing roughly 57.5% faster task completion and about 71% fewer compile errors, but it shifted the developer's role from author to reviewer-and-prompter and concentrated the remaining effort on semantic verification of schema-coupled code. Survey respondents reported high productivity perceptions alongside conditional, review-gated trust in AI output. The thesis concludes that AI-assisted low-code development augments rather than automates software work, redistributing effort rather than removing it.