Rethinking Software Design with Large Language Models Intelligent Interfaces

The advancement of Large Language Models (LLMs) has led to a rapid expansion of their applications, including in software design. Among other fields, software design is experiencing notable advancements through the integration of LLMs as interface components that extend fixed user stories [1]. However, incorporating LLM-based AI agents into software design introduces challenges, particularly in estimating development efforts. One key issue is the lack of clarity and focus in initial system specifications, which often begin as illdefined sampled questions about system behavior that are later fed into LLM-based interfaces. To address this, we propose a new approach to refine system specifications using natural language-based interfaces. We present a method that enables software engineers and architects to iteratively improve software development documentation, even when they are ill-specified, ensuring a clearer definition of system behavior, required resources, and dependencies. Such documentation is crucial for accurately estimating development efforts by considering data sources, interfaces, and algorithms.