AI
Reframing AI Prompts as Software Requirements: A Research Agenda
Researchers propose treating natural language prompts as formal requirements artifacts, bridging the gap between informal AI-assisted coding and systematic software engineering practice.
As AI coding assistants reshape how developers work, a new vision paper argues that the prompts driving these tools deserve the same rigorous treatment traditionally given to software requirements.
Prompts as Lightweight Requirements#
The shift toward chat-based development—including emerging practices like vibe coding—has moved the burden of software creation away from manually writing code toward specifying intent through natural language. Yet this prompting process remains largely informal and unsystematized. Researchers at arXiv's Software Engineering track contend that prompts are fundamentally requirements artifacts that blend human needs with varying amounts of solution-focused guidance, and should be studied accordingly.
A Three-Dimensional Framework#
The paper proposes decomposing prompts along three interrelated dimensions:
- Functionality and Quality — capturing what the system should do and how well it should perform
- General Solutions — capturing high-level architectural decisions and technology choices
- Specific Solutions — capturing implementation-level constraints and concrete directives
This model acknowledges that real-world prompts occupy a spectrum between pure requirements (what the user wants) and pure solutions (how to build it).
Four Research Questions#
To move toward evidence-based prompting practices, the researchers outline four key hypotheses:
- How prompts evolve as developers refine them over iterations
- How developer characteristics shape the way prompts are constructed and modified
- The connection between prompt content and validation/verification activities
- How prompt characteristics ultimately influence software quality
An Empirical Research Program#
The authors propose a multi-method investigation combining analysis of real AI-assisted development workflows, large-scale corpus studies of prompt repositories, and controlled experiments. By treating prompting as a central software engineering concern rather than merely a user-interface mechanic, the work aims to establish systematic, evidence-backed practices for writing and evolving AI prompts—positioning requirements engineering as essential to the age of AI-powered coding.
