From Knowledge to Noise: CTIM-Rover and the Pitfalls of Episodic Memory in Software Engineering Agents
Published in REALM Workshop at ACL 2025, 2025
CTIM-Rover extends AutoCodeRover with episodic memory to investigate how stored past experiences affect software engineering agent performance. The study reveals that repository-level understanding is pivotal for identifying all locations requiring a patch, and that episodic memory can hurt rather than help when the retrieved experiences act as noise rather than signal.
Recommended citation: Lindenbauer, T., Groh, G., & Schütze, H. (2025). "From Knowledge to Noise: CTIM-Rover and the Pitfalls of Episodic Memory in Software Engineering Agents." Proceedings of the 1st Workshop for Research on Agent Language Models (REALM) at ACL 2025.
Download Paper