Transforming Software Engineering and Software Acquisition with Large Language Models

Large language models (LLMs) are a form of generative AI that have become engines of transformation in software engineering and software acquisition. Convenient access to chat-based LLMs trained extensively with vetted software data are enabling software engineers and developers to accelerate common tasks, including drafting requirements, generating code, and/or creating tests for common scenarios. Many development environments and tools now integrate with application programming interfaces (APIs) offered by LLMs to create new interactive workflows. There are pros and cons, however, of applying LLMs throughout the software development lifecycle (SDLC). One con is that LLMs can generate incorrect answers, such as code with known security vulnerabilities, semantically incorrect code even if it compiles, and software tests that are incomplete. One pro is that LLMs can generate multiple alternative responses to a given request rapidly, allowing users to explore outcomes they otherwise may not be able to do. LLMs interact with users in the form of structured inputs, known as prompts, that can customize and improve LLM output. Prompts provide a powerful mechanism to develop and leverage patterns that improve the effectiveness of LLMs, including for common software engineering tasks. Effective prompt engineering can thus enhance the performance of LLMs for specific software development activities. This chapter addresses the question of whether the transformative changes associated with LLMs will yield more efficient software development methods and tools and create different software engineering workflows with meaningful improvements. To answer this question, we review the use cases of using LLMs on common software engineering activities and combine these insights to discuss the potential impact of LLMs across the SDLC.