Fri 13 Nov
LiveGet it free
ZOTPAPER

News without the noise

← All concepts

agent memory

An agent memory lets an AI system store and recall past experiences, conversations, or solutions to help it make better decisions or perform tasks. Think of it as the AI's own notes or diary, it remembers what worked before, what didn't, and relevant context, so it can reuse that knowledge rather than starting from scratch each time.

Across several studies, researchers find that simply piling up more memories often hurts rather than helps. What matters is relevance, organization, and how memories are retrieved and used. Many systems now borrow techniques like quality filtering, grouping related facts, or forgetting outdated information to avoid confusion. A key insight is that the cost of writing memories (embedding and processing) far outweighs reading them, so designs must balance accuracy with efficiency. Benchmarks also now test whether agents can use memory to take actions, not just answer questions, revealing tradeoffs between cost, speed, and performance.

Papers
232
Released code
15
First seen
June 2026
Latest
Nov 2026

206 papers in the last two months, against 23 in the two before.

Start here

Three papers to read first, in the order they came out
  1. 1AI agents can turn harmless memories into unsafe future actions

    Memory from benign interactions can cause language agents to behave unsafely or unreliably on later tasks.

    Nov 2026
  2. 2AI agents can keep needed context yet fail to use it

    Long-horizon agents need both planned token retention and bounded positional distance, because cached context can become unusable when it is too far away.

    Nov 2026
  3. 3Robot memory that runs by clock confuses waiting with working

    Time-sampled memory makes robots mistake idle waiting for task progress, causing early stopping during null events.

    Nov 2026

How fast it is moving

JunJulAugSeptOctNov so far

Often studied with

verificationChecking outputs for correctness before accepting them, often to catch errors or reward hacking.30 togetherself-evolving agentAn agent that improves by reusing its own experience, memory, or self-generated feedback, with success depending on verification and design.30 togetherevidence retrievalPulling specific, relevant information from a large corpus to support an answer, decision, or reasoning step.29 together

Who is working on it

Amazon 11Tencent 11Microsoft 9Tsinghua University 9University of Illinois Urbana-Champaign 9Stanford University 8

The papers

Most central to this idea first, not most recent.

Concepts are extracted from each paper and reused across the corpus, so this page grows on its own as the desk reads.