--- title: Anthropic gave agents the ability to dream. Then... description: Anthropic introduced 'dreaming', an asynchronous, batch process that lets AI agents consolidate and update their own memory stores outside the live session,... canonical: https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Anthropic gave agents the ability to dream. Then developers woke up. | daily.dev og:description: Anthropic introduced 'dreaming', an asynchronous, batch process that lets AI agents consolidate and update their own memory stores outside the live session,... og:url: https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x og:image: https://api.daily.dev/og/posts/LSyUBx30x.png og:image:alt: Anthropic gave agents the ability to dream. Then developers woke up. og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Anthropic gave agents the ability to dream. Then developers woke up. **[The New Stack](https://daily.dev/sources/newstack)** · 11 min read · 0 upvotes · 0 comments ## Summary Anthropic introduced 'dreaming', an asynchronous, batch process that lets AI agents consolidate and update their own memory stores outside the live session, spotting recurring failure patterns and proposing memory updates for human review. Multiple AI engineers and researchers interviewed praised the filesystem-and-grep approach over vector databases for its auditability and legibility, but warned that the technique risks turning transient bad behavior into persistent, hard-to-audit policy. Critics highlighted concerns around provenance, expiration of stale facts, rubber-stamp human review at scale, security risks from poisoned memory, and rising AI compute costs tied to Anthropic's usage-based revenue model. Similar memory consolidation concepts are reportedly being explored by OpenAI for ChatGPT and by the Letta agent platform. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What is Anthropic's 'dreaming' technique for AI agent memory? Dreaming is an asynchronous, out-of-band batch process Anthropic introduced for its Managed Agents memory and API approach, which reviews session transcripts to consolidate memory, remove stale entries, add missing information, and propose memory-store updates for human review, rather than having agents manage memory in-band during live sessions. _daily.dev surfaces expert takes like this for engineers evaluating agent memory architectures._ ### What are the risks of letting AI agents automatically update their own long-term memory? A bad answer normally dies with the session, but a bad memory can influence thousands of future sessions, according to AI agent researchers. Risks include human review becoming a rubber stamp at scale, memory drifting toward whatever the agent fleet did recently, poisoned memory persisting for weeks before discovery, and stored facts never expiring even after the underlying system changes. _Developers weighing agent memory safeguards can track this debate on daily.dev._ ### Why did Anthropic choose a filesystem and grep instead of a vector database for agent memory? Because a memory store built on files and grep can be opened, read, diffed, and code-reviewed by engineers, making it auditable in a way no vector database has ever offered. Commentators noted that after two years of the industry treating agent memory as an embeddings problem, Anthropic's plain-filesystem approach makes debugging and rollback of bad memories practically possible. _daily.dev helps engineers comparing memory architectures like filesystem-based versus vector-based agent memory._ ## Similar posts on daily.dev - [Claude's leaked dreaming feature is now live, and it lets agents learn from their own mistakes](https://daily.dev/posts/claude-s-leaked-dreaming-feature-is-now-live-and-it-lets-agents-learn-from-their-own-mistakes-48xpxeqxf) · XDA Developers · 2 upvotes · 0 comments - [New in Claude Managed Agents: dreaming, outcomes, and multiagent orchestration](https://daily.dev/posts/new-in-claude-managed-agents-dreaming-outcomes-and-multiagent-orchestration-thtgbl3zg) · Claude · 1 upvotes · 0 comments - [Long-running Agents](https://daily.dev/posts/long-running-agents-irtro0veg) · Addy Osmani · 78 upvotes · 3 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic) [View this post on daily.dev](https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Anthropic gave agents the ability to dream. Then developers woke up.","url":"https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x"},"datePublished":"2026-08-12T16:53:08.893Z","dateModified":"2026-08-12T16:53:35.860Z","description":"Anthropic introduced 'dreaming', an asynchronous, batch process that lets AI agents consolidate and update their own memory stores outside the live session,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d9513b26ff88db6cef8008aa99715c1b?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/d9513b26ff88db6cef8008aa99715c1b?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,claude,anthropic","timeRequired":"PT11M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Anthropic gave agents the ability to dream. Then developers woke up."}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/anthropic-gave-agents-the-ability-to-dream-then-developers-woke-up--lsyubx30x#faq","mainEntity":[{"@type":"Question","name":"What is Anthropic's 'dreaming' technique for AI agent memory?","acceptedAnswer":{"@type":"Answer","text":"Dreaming is an asynchronous, out-of-band batch process Anthropic introduced for its Managed Agents memory and API approach, which reviews session transcripts to consolidate memory, remove stale entries, add missing information, and propose memory-store updates for human review, rather than having agents manage memory in-band during live sessions. daily.dev surfaces expert takes like this for engineers evaluating agent memory architectures."}},{"@type":"Question","name":"What are the risks of letting AI agents automatically update their own long-term memory?","acceptedAnswer":{"@type":"Answer","text":"A bad answer normally dies with the session, but a bad memory can influence thousands of future sessions, according to AI agent researchers. Risks include human review becoming a rubber stamp at scale, memory drifting toward whatever the agent fleet did recently, poisoned memory persisting for weeks before discovery, and stored facts never expiring even after the underlying system changes. Developers weighing agent memory safeguards can track this debate on daily.dev."}},{"@type":"Question","name":"Why did Anthropic choose a filesystem and grep instead of a vector database for agent memory?","acceptedAnswer":{"@type":"Answer","text":"Because a memory store built on files and grep can be opened, read, diffed, and code-reviewed by engineers, making it auditable in a way no vector database has ever offered. Commentators noted that after two years of the industry treating agent memory as an embeddings problem, Anthropic's plain-filesystem approach makes debugging and rollback of bad memories practically possible. daily.dev helps engineers comparing memory architectures like filesystem-based versus vector-based agent memory."}}]} ```