Memory system improves task success by adapting what is remembered
Just-in-Time Memory: Learning to Curate Task-Adaptive Memory for LLM Agents
Artificial Intelligence
Summary
Many computer programs that learn from past experiences decide what to remember right after finishing a task. This can miss important details because the program doesn’t know what future tasks might need the memory. The authors made a system that waits and chooses what to remember only when it needs the memory for a new task. This way, it can pick just the right information and improve its chances of succeeding. Their method performed better than older ones in several tests and worked well even before learning.
What this means in practice
- •For software developers: Improve interactive AI tools by enabling adaptive memory retrieval to tailor responses based on specific user tasks.
- •For autonomous system engineers: Enhance robotic or virtual agent performance by allowing memory recall that dynamically adapts to current mission requirements.
Authors
Yefan Zhou, Yang Li, Zeyu Leo Liu, Semih Yavuz, Shafiq Joty
Abstract
Agentic memory systems reuse past experience to improve future performance, yet most existing designs curate memory at write time: once a task is completed, its trajectory is distilled into a fixed artifact, such as a reflection, workflow, skill, or reasoning strategy, that is later retrieved by similarity. This forces the system to decide what is worth remembering before the future query is known, irreversibly discarding information and producing a query-independent summary that must serve many possible downstream tasks. Learning such a write-time curator is also difficult because the value of a storage decision may only become apparent when a relevant query arrives, potentially many tasks later, creating a long-horizon credit-assignment problem. We instead retain raw trajectories and defer curation until read time, when the current task is known. Given the retrieved traces and the new task, a memory curator synthesizes a compact, task-adaptive payload tailored to the immediate need. Because this payload is consumed on the same task, the curator can be trained directly from immediate task success, avoiding delayed utility signals and the need to artificially group related tasks. Across ALFWorld, WebShop, and $τ^2$-bench, our Just-in-Time Memory (JitMem) consistently outperforms no-memory agents as well as heuristic and learned write-time memory methods, improving over the strongest baseline by 16.2, 16.3, and 3.9 absolute success-rate points, respectively. Notably, even an untrained curator is already competitive with or surpasses these baselines, showing that task-adaptive read-time curation itself is a major source of the gain; training the curator further compounds the improvement.