ACE: Moving toward Co-Investigation with the Agentic Code Explorer

In this workshop demonstration paper, we present ACE - the Agentic Code Explorer - a prototype agentic system designed to help software developers conduct sensemaking tasks within large code repositories. The design of this system was motivated by the observation that software developers often use AI coding assistants to help understand and ask questions about source code prior to planning and implementing code changes. Using ACE as a testbed, we present initial steps to explore whether a large language model (LLM)-based agent that is capable of invoking external tools and iteratively refining its own outputs (per the agentic design pattern) might be able to robustly support such a code discovery process. In this way, we use ACE as a means to explore more generally how generative models need not solely focus on the artifact production aspects of co-creative tasks; instead they might focus on the co-investigative activities where initial understanding and plans are formed.