--- title: How Anthropic ships Claude with Claude: Separating the... description: Anthropic's engineering and product leads explain how Claude Opus 5.5 was shipped and how the Claude Platform is built. The team separates an agent's 'brain'... canonical: https://daily.dev/posts/how-anthropic-ships-claude-with-claude-separating-the-brain-from-the-hands-m3aegyz7b twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: How Anthropic ships Claude with Claude: Separating the Brain from the Hands | daily.dev og:description: Anthropic's engineering and product leads explain how Claude Opus 5.5 was shipped and how the Claude Platform is built. The team separates an agent's 'brain'... og:url: https://daily.dev/posts/how-anthropic-ships-claude-with-claude-separating-the-brain-from-the-hands-m3aegyz7b og:image: https://api.daily.dev/og/posts/m3AeGYz7b.png og:image:alt: How Anthropic ships Claude with Claude: Separating the Brain from the Hands 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. # How Anthropic ships Claude with Claude: Separating the Brain from the Hands **[Latent Space](https://daily.dev/sources/latentspace)** · 119 min read · 1 upvotes · 0 comments ## Summary Anthropic's engineering and product leads explain how Claude Opus 5.5 was shipped and how the Claude Platform is built. The team separates an agent's 'brain' (the persistent harness and session log) from its 'hands' (a sandbox that executes code), so a failed sandbox doesn't kill the whole session. Credentials are held separately and attached through a proxy, so generated code running in a sandbox that never receives a service credential has none to read. Opus 5.5 costs less to run than its predecessor depending on cached context usage, and the team added measures to preserve thinking blocks across turns to resist chain-of-thought distillation, and shipped a new compaction feature in the API for customers whose strategies would have broken. The guests also describe scaling pressure from CPU constraints, with the Messages API having started as a Python monolith in a single GCP region. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How does Anthropic separate an AI agent's execution environment from its session state? Anthropic splits agents into a 'brain' and 'hands': a persistent harness and session log versus a sandbox that executes code. The harness can persist while a sandbox is created on demand, so a failed or crashed sandbox doesn't terminate the whole agent session. This architecture, described in Anthropic's Managed Agents work, also improves latency since the harness no longer waits on sandbox startup for every response. _daily.dev surfaces architecture patterns like this for engineers designing their own agent runtimes._ ### How do AI companies prevent their model's reasoning from being distilled by competitors? Anthropic preserves 'thinking blocks' across conversation turns, making it harder to extract chain-of-thought reasoning for distillation, following a published technique called stealing reasoning traces from proprietary models. This required working with open source harness maintainers and large API customers because the preservation requirement could break existing compaction strategies, so Anthropic shipped a new compaction feature to replace manual ones. _developers adapting their harnesses to API changes can track these shifts on daily.dev._ ### Why would a sandboxed code execution environment not need API credentials? When generated code runs in a sandbox that never receives a service credential, there is no credential present for malicious or buggy code to leak or misuse. Anthropic holds credentials separately and attaches them through a proxy, combined with network controls on where sandbox traffic can go, as part of its broader containment approach across products. _daily.dev helps engineers securing agent sandboxes keep up with emerging isolation patterns._ ## Similar posts on daily.dev - [How Anthropic ships the \#1 LLM in the World safely: Separating the Brain from the Hands](https://daily.dev/posts/how-anthropic-ships-the-1-llm-in-the-world-safely-separating-the-brain-from-the-hands-gprymh2hf) · Latent Space · 1 upvotes · 0 comments - [Harnessing Claude’s intelligence](https://daily.dev/posts/harnessing-claude-s-intelligence-x0y7hjoye) · Claude · 2 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#architecture](https://daily.dev/tags/architecture), [#claude](https://daily.dev/tags/claude), [#anthropic](https://daily.dev/tags/anthropic) [View this post on daily.dev](https://daily.dev/posts/how-anthropic-ships-claude-with-claude-separating-the-brain-from-the-hands-m3aegyz7b) ```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":"How Anthropic ships Claude with Claude: Separating the Brain from the Hands","url":"https://daily.dev/posts/how-anthropic-ships-claude-with-claude-separating-the-brain-from-the-hands-m3aegyz7b","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/how-anthropic-ships-claude-with-claude-separating-the-brain-from-the-hands-m3aegyz7b"},"datePublished":"2026-10-27T18:56:01.205Z","dateModified":"2026-10-27T19:01:05.870Z","description":"Anthropic's engineering and product leads explain how Claude Opus 5.5 was shipped and how the Claude Platform is built. The team separates an agent's 'brain'...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6f1993fbdb7eab3fa65f40191347cd9a?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/6f1993fbdb7eab3fa65f40191347cd9a?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Latent Space","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":"Latent Space","logo":"https://media.daily.dev/image/upload/s--DnJ9laFj--/f_auto,q_auto/v1780213196/logos/latentspace?_a=BAMAMiWQ0","url":"https://daily.dev/sources/latentspace"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/how-anthropic-ships-claude-with-claude-separating-the-brain-from-the-hands-m3aegyz7b","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,architecture,claude,anthropic","timeRequired":"PT119M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Latent Space","item":"https://daily.dev/sources/latentspace"},{"@type":"ListItem","position":3,"name":"How Anthropic ships Claude with Claude: Separating the Brain from the Hands"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/how-anthropic-ships-claude-with-claude-separating-the-brain-from-the-hands-m3aegyz7b#faq","mainEntity":[{"@type":"Question","name":"How does Anthropic separate an AI agent's execution environment from its session state?","acceptedAnswer":{"@type":"Answer","text":"Anthropic splits agents into a 'brain' and 'hands': a persistent harness and session log versus a sandbox that executes code. The harness can persist while a sandbox is created on demand, so a failed or crashed sandbox doesn't terminate the whole agent session. This architecture, described in Anthropic's Managed Agents work, also improves latency since the harness no longer waits on sandbox startup for every response. daily.dev surfaces architecture patterns like this for engineers designing their own agent runtimes."}},{"@type":"Question","name":"How do AI companies prevent their model's reasoning from being distilled by competitors?","acceptedAnswer":{"@type":"Answer","text":"Anthropic preserves 'thinking blocks' across conversation turns, making it harder to extract chain-of-thought reasoning for distillation, following a published technique called stealing reasoning traces from proprietary models. This required working with open source harness maintainers and large API customers because the preservation requirement could break existing compaction strategies, so Anthropic shipped a new compaction feature to replace manual ones. developers adapting their harnesses to API changes can track these shifts on daily.dev."}},{"@type":"Question","name":"Why would a sandboxed code execution environment not need API credentials?","acceptedAnswer":{"@type":"Answer","text":"When generated code runs in a sandbox that never receives a service credential, there is no credential present for malicious or buggy code to leak or misuse. Anthropic holds credentials separately and attaches them through a proxy, combined with network controls on where sandbox traffic can go, as part of its broader containment approach across products. daily.dev helps engineers securing agent sandboxes keep up with emerging isolation patterns."}}]} ```