Skip to content
Hot eventDeveloping

Anthropic research: selective agent memory beats exhaustive archives

1 reports1 reporting sourcesUpdated 1 h ago

Start here

AI digest

Anthropic published a research post on how Claude can retain knowledge across sessions without changing the model, arguing that the hard problem is selecting what to remember rather than storing more. In the post, Anthropic describes Claude writing notes to a folder it reads back later. In one comparison, an agent labeled "7" created 413 files across 225 folders over roughly 10,000 steps and earned one badge, while Claude Fable 5 kept to a few files of plans and verified rules and earned four badges. The post also covers Dreaming, a memory-correction method Anthropic introduced in May as a research preview for Claude Managed Agents, its cloud platform for AI agents. Dreaming reads an existing memory store and up to 100 past sessions, removes duplicate and outdated entries, and writes a reorganized store while leaving the original untouched. At launch, Dreaming supported only Claude Opus 4.7 and Sonnet 4.6.

Generated from reports · updated 23 min ago

Report timeline

Follow the reports to see the event from different sides.

November 16
  1. The Decoder
    Anthropic studies agent memory: selective notes and self-review beat exhaustive archives

    Anthropic published a research post describing how Claude can retain knowledge across sessions by writing notes to a folder it reads back later, with selection rather than storage volume as the core problem. 7 created 413 files across 225 folders over roughly 10,000 steps and earned one badge, while Claude Fable 5 kept to a few files of plans and verified rules and earned four badges.

Heat for this event

Not enough continuous observations yet to draw a trend.