Enhancing Software Development through Prompt Engineering A Study on Large Language Models for Code Generation and Developer Productivity
Year: 2024
Type: article
Abstract: Large Language Models (LLMs) are increasingly changing Software Development with capabilities to generate code snippets, debug, etc., and towards design work as well. Successful outcomes using LLMs is heavily reliant on prompt engineering. Well-designed prompts influence the quality of generated code, improve developer workflows, and build effective human-computer interactivity in the use of LLM models. This study examines prompt engineering in improving developer productivity via a designed process of exploration of prompting strategies to generate code. A taxonomy of potential prompt engineering techniques is introduced conceptualizing four experimental approaches for the coding task: instruction-based prompts, example-based prompts, chain-of-thought prompts, and hybrid prompts. The study focuses on developer-oriented productivity metrics beyond technical quality. Productivity metrics include a reduction in overall development time, reduced errors and better readability, e.g. improved structure of codes, and found improvements to tools used to develop software, e.g. integrated development environments, collaborative coding tools. The comparative evaluation of prompt patterns identifies how differentiating prompt patterns can create variable impact on code quality, but also variable experiences for developers. This suggests that prompt engineering can influence the continuing problem of debugging and support more rapid delivery of software to clients. The paper describes barriers to adoption in practice, such as prompt sensitivity, context limitations and reusability limitations and offer a roadmap for integrating adaptive prompting systems directly in developer environments. By relating LLM capabilities to productivity outcomes, this work provides a new perspective in bridging prompt engineering research with real-world software development pipelines.
Author: Jimit Patel
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7217631941