How Professional Software Engineers Use Generative AI in Development
Master's Thesis · ~92 pages · English
Abstract
This thesis investigates how professional software engineers incorporate generative AI tools (GitHub Copilot, ChatGPT, Claude) through semi-structured interviews with 30 practitioners. Thematic analysis identifies five primary usage patterns: code completion, documentation writing, debugging assistance, learning new technologies, and code review support. The study documents both productivity gains and concerns regarding code quality, security vulnerabilities, and skill atrophy.
1. Introduction
The emergence of large language models trained on code has fundamentally altered software development practices. Tools like GitHub Copilot, ChatGPT, and Claude now assist millions of developers with tasks ranging from boilerplate generation to architectural design.
This qualitative research explores how professional engineers integrate these tools into their daily workflows, examining adoption patterns, perceived benefits, and persistent concerns across diverse organizational contexts.
2. Research Questions
RQ1: What are the primary use cases for generative AI in professional software development?
RQ2: How do developers perceive AI tools' impact on their productivity and code quality?
RQ3: What concerns do developers express about AI-assisted coding?
RQ4: How do organizational factors (team size, domain, policies) influence adoption patterns?
3. Key Findings
Code completion emerged as the most common use case, with 93% of participants reporting regular usage. Documentation writing saves significant time for 78% of developers.
Security concerns remain prominent among enterprise developers, with 67% expressing reservations about AI-generated code entering production without thorough review. Junior developers report higher productivity gains but also exhibit higher dependency on AI suggestions.
Code review practices are evolving, with teams developing new protocols specifically for assessing AI-generated contributions.
References
- [1]Ziegler, A., et al. (2024). Productivity Assessment of Neural Code Completion. ICSE.
- [2]Vaithilingam, P., et al. (2022). Expectation vs. Experience: Evaluating the Usability of Code Generation Tools. CHI.
- [3]Barke, S., et al. (2023). Grounded Copilot: How Programmers Interact with Code-Generating Models. OOPSLA.
- [4]Peng, S., et al. (2023). The Impact of AI on Developer Productivity: Evidence from GitHub Copilot. arXiv:2302.06590.
This is a sample excerpt. Full papers include complete chapters, verified citations, and downloadable formats.
Free to try · No credit card required · Free to start, 3 credits/day