How AI Turns User Stories into Actionable Software Development Tasks

What is it about?

Turning software requirements into clear, actionable tasks is an important part of software development, but it can be time-consuming and challenging. The needs and expectations of users are often captured in user stories. These stories are then broken down into smaller tasks that developers can work on. Our research explores whether large language models (LLMs), a type of artificial intelligence (AI), can help automate this process. We developed DELTA, an approach that uses LLMs to generate multiple sets of development tasks from user stories, evaluate their quality, and recommend the top three options. We compared different LLMs and ways of instructing them, and asked software development experts to evaluate the results. We found that LLM-based quality assessments aligned more closely with expert judgments than traditional text similarity methods. Notably, experts generally preferred the top three ranked LLM-generated task breakdowns over those created by humans. Our findings suggest that LLMs show the potential to assist teams in translating user stories into development tasks. However, human involvement remains important for reviewing and refining the generated tasks, as LLMs may overlook relevant context or important details.

Featured Image

Photo by Eden Constantino on Unsplash

Photo by Eden Constantino on Unsplash

Why is it important?

There is growing interest in how LLMs can support everyday software development activities. While much attention has focused on using LLMs to generate code, our research explores an earlier stage of development: turning user stories into concrete tasks that developers can work on.

Perspectives