Type: url
Original: https://arxiv.org/pdf/2411.12924
Status: generated
Summary
A paper from Monash University, the University of Melbourne and Atlassian introducing HULA, a human-in-the-loop LLM-based software development agent framework in which an AI Planner Agent, an AI Coding Agent and a Human Agent (the assigned engineer) cooperate to turn a JIRA issue into a pull request, with the engineer able to review and refine the file list, coding plan and code at each stage. HULA was deployed inside Atlassian JIRA and evaluated offline on SWE-bench Verified and an internal dataset, online over 663 real issues (82% plan approval, 56 merged pull requests), and through a survey of 109 practitioners, who reported reduced development time and effort while raising concerns about code quality and the effort needed to write detailed issue descriptions.