Skip to content
ClaudeClaude

Memory and dreaming for self-learning agents

How memory and dreaming turn Claude Managed Agents into self-learning systems. This session walks through design considerations for memory architectures and how dreaming verifies and enriches memory between sessions.

May 7, 202624mWatch on YouTube ↗

At a glance

WHAT IT’S REALLY ABOUT

Claude Managed Agents introduces memory plus dreaming for self-learning agents

  1. Memory is positioned as the next key agent primitive to enable long-horizon context management and continuous self-learning beyond tools and skills.
  2. Claude Managed Agents Memory models memory as a file system that agents can organize and update using familiar tools, aiming to maximize intelligence by default.
  3. The design targets enterprise multi-agent realities with permissioned memory stores, optimistic concurrency to prevent clobbering, and auditability via version history and attribution.
  4. A standalone memory API enables external governance workflows such as PII scanning, cleanup pipelines, and syncing/cloning to other systems.
  5. “Dreaming” is launched in research preview as an out-of-band batch job that mines recent session transcripts to curate, deduplicate, verify, and enrich shared memory for future agents.

IDEAS WORTH REMEMBERING

5 ideas

Treat memory as an agent-managed file system to reduce harness constraints.

Instead of rigid tool schemas, Claude can structure memory into folders/files and maintain it using bash/grep, leveraging model strengths in file-based organization.

Separate organizational knowledge from working memory with scoped permissions.

Use read-only stores for stable runbooks/SLO guidelines and read-write stores for rapidly changing task learnings, preventing accidental edits to authoritative guidance.

Design for concurrency from day one in shared-memory swarms.

When hundreds/thousands of agents write to the same store, optimistic concurrency (precondition/content hashes) helps detect conflicting updates before overwriting others.

Auditability is a production requirement, not a nice-to-have.

Version history plus attribution (which agent/session/time) makes memory changes inspectable, supports incident review, and enables safer rollout of self-modifying systems.

Keep memory governance portable with a standalone API.

Enterprises often need independent pipelines (PII scanning, cleanup, cloning to external KBs), so memory must be manageable outside the Managed Agents runtime.

WORDS WORTH SAVING

5 quotes

Memory is the next primitive. It's the thing that I think will get us to self-learning agents that evolve and improve based on the tasks that they're working on and their own experience.

Mahesh

Memory in Claude Managed Agents models memory as a file system to Claude, a series of files with a specific hierarchy and format that Claude can manage and update on its own.

Mahesh

I think self-managed memory is gonna be super important in these large and complex multi-agent systems, where a swarm of agents that are working in a similar environment on discrete tasks are essentially building up their own understanding, their own model of the world that they're in over time.

Mahesh

Dreaming is a process that looks for patterns and mistakes across your recent agent sessions and their transcripts and automatically produces organized and up-to-date memory content.

Mahesh

The ultimate goal of dreaming is continuous self-learning and self-improvement, where the next day's agents automatically get better based on the learnings and the work of the previous day's experience.

Mahesh

Memory as a core agent primitiveFile system–based memory representationMulti-agent scaling and shared statePermission scopes (read-only vs read-write stores)Optimistic concurrency via content hashesEnterprise controls: version history, audit logs, attributionDreaming: asynchronous transcript-driven memory curation

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.

Add to Chrome