Most teams answer these questions by building their own harness. A custom agent loop. A home-rolled state store. A duct-taped error recovery system that works until it doesn’t. It’s not glamorous work, and it doesn’t scale well\u2014because every time the model improves, the assumptions baked into your harness go stale.<\/p>\n\n\n\n
Anthropic’s own engineering team documented this directly in their harness design research for long-running agents<\/a><\/strong>: Claude Sonnet 4.5 would wrap up tasks prematurely as it sensed its context limit approaching\u2014a behavior they called “context anxiety.” They added context resets to the harness to compensate. Then Claude Opus 4.5 arrived, the behavior was gone, and the resets had become dead weight.<\/p>\n\n\n\n The harness was encoding assumptions about the model. The model changed. The harness didn’t.<\/p>\n\n\n\n That pattern\u2014teams maintaining infrastructure that fights the model instead of using it\u2014is the real problem scaling Claude agent orchestration. Managed Agents is Anthropic’s answer.<\/p>\n\n\n\n According to the Claude Managed Agents official documentation<\/a><\/strong>, the platform provides a fully managed environment where Claude can read files, run commands, browse the web, and execute code securely. The harness includes built-in prompt caching, compaction, and other performance optimizations for high-quality, efficient agent outputs.<\/p>\n\n\n\n The practical model has four pieces:<\/p>\n\n\n\n This is the part most teams get wrong in self-built agent systems.<\/p>\n\n\n\n When a task runs long, you face a hard choice: summarize the context and lose fidelity, trim tokens and risk dropping something important, or let the context window overflow. All three options are irreversible. Once you’ve compacted or trimmed, those tokens are gone.<\/p>\n\n\n\n Managed Agents handles this differently. The session acts as a context object that lives outside Claude’s context window. Rather than stored in a sandbox or REPL, context is durably stored in the session log. The Concretely: the full task history is preserved. Claude doesn’t have to hold everything in its active context\u2014it queries the session log for what it needs. You don’t lose state when the context window fills. You just work with a different slice of it.<\/p>\n\n\n\n Compaction still happens\u2014it’s not magic. Compaction lets Claude save a summary of its context window, and the memory tool lets Claude write context to files, enabling learning across sessions. This can be paired with context trimming, which selectively removes tokens such as old tool results or thinking blocks.<\/p>\n\n\n\n The difference is what happens after compaction. In a self-managed loop, compacted messages are gone unless you built something to recover them. In Managed Agents, they’re still in the session log. The agent can query back if it needs them.<\/p>\n\n\n\n This also means context resets\u2014used to address context anxiety in older models\u2014become optional rather than mandatory. As models improve, you update behavior without rebuilding your infrastructure.<\/p>\n\n\n\n The brain, hands, and session are decoupled. Each can fail or be replaced independently. That architecture matters for parallel workloads.<\/p>\n\n\n\n Multi-agent coordination (one agent spawning sub-agents) is currently in research preview\u2014worth noting if your architecture depends on it. But running multiple independent agent sessions in parallel is available now. Each session gets its own isolated container, its own credential scope, its own event log.<\/p>\n\n\n\n People use “long-running” loosely. It’s worth pinning down what it actually means in production.<\/p>\n\n\n\n A typical Claude API call is seconds. A basic agent loop might run for one to five minutes. Long-running, for the purposes of Managed Agents, means tasks that:<\/p>\n\n\n\n The Sentry example is concrete: Sentry paired their debugging agent with a Claude-powered agent that writes the patch and opens the PR, so developers go from a flagged bug to a reviewable fix in one flow. That’s not a single call. It’s a multi-step workflow\u2014analyze error, locate codebase context, draft fix, run tests, open PR\u2014that runs as a coherent session.<\/p>\n\n\n\n This is a decision with real trade-offs, not a default.<\/p>\n\n\n\n Use Managed Agents (remote sessions) when:<\/strong><\/p>\n\n\n\n Stay with the Agent SDK or Messages <\/strong>API<\/strong> when:<\/strong><\/p>\n\n\n\n Claude Managed Agents is Anthropic’s hosted runtime for long-running agent work. Use it when you want Anthropic to run the loop and session infrastructure for you. Stay on the Messages API or the Claude Agent SDK when you need tighter control over the loop or where the runtime executes.<\/p>\n\n\n\n The architectural decision that makes this work is separating what Anthropic describes in their engineering post on scaling managed agents<\/a><\/strong> as the “brain” from the “hands.”<\/p>\n\n\n\n The harness leaves the container. Decoupling the brain from the hands meant the harness no longer lived inside the container. It calls the container the way it calls any other tool: This sounds like an implementation detail. It’s not. It’s why the system can scale.<\/p>\n\n\n\n When brain and hands are coupled in one container, you’ve adopted what Anthropic calls a “pet”\u2014a specific environment you have to keep alive, patch, and manage. When they’re decoupled, the sandbox becomes a disposable resource. It can crash and be replaced without taking the session with it. The session log is the source of truth, not the container.<\/p>\n\n\n\n Credentials follow the same logic. For custom tools, OAuth tokens are stored in a secure vault. Claude calls MCP tools via a dedicated proxy; this proxy fetches the corresponding credentials from the vault and makes the call to the external service. The harness is never made aware of any credentials.<\/p>\n\n\n\n The agent handles the reasoning. The infrastructure handles the secrets, the execution, and the state. That separation is what makes multi-step workflows reliable at scale.<\/p>\n\n\n\n
\n\n\n\nHow Claude Managed Agents Handles Scale<\/h2>\n\n\n\n
![\"Table](\"https:\/\/octer.ai\/hub\/wp-content\/uploads\/2026\/05\/Core-Concepts-of-Claude-Architecture.png\")
<\/figure>\n<\/figure>\n\n\n\nObject<\/td> What it does<\/td><\/tr> Agent<\/td> Defines model, system prompt, tools, MCP servers, and permissions. Created once, referenced by ID.<\/td><\/tr> Environment<\/td> Cloud container with pre-installed packages (Python, Node.js, Go), network rules, and mounted files. Anthropic handles the lifecycle.<\/td><\/tr> Session<\/td> The running unit of work. An append-only log of everything that happened.<\/td><\/tr> Events<\/td> How the session communicates progress and takes input. Streamed via SSE.<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n Session continuity \u2014 why context doesn’t get lost<\/h3>\n\n\n\n
getEvents()<\/code> interface allows the brain to interrogate context by selecting positional slices of the event stream.<\/p>\n\n\n\nCompaction and context resets without losing state<\/h3>\n\n\n\n
Running parallel agents across tasks<\/h3>\n\n\n\n
![\"Website](\"https:\/\/octer.ai\/hub\/wp-content\/uploads\/2026\/05\/Multi-Agent-Coordination-and-Architecture-1024x805.png\")
<\/figure>\n<\/figure>\n\n\n\n
\n\n\n\nWhat “Long-Running” Means in Practice<\/h2>\n\n\n\n
Tasks that run for minutes vs. hours<\/h3>\n\n\n\n
\n
When to use remote sessions vs. local<\/h3>\n\n\n\n
\n
\n
\n\n\n\nHow Managed Agents Decouples the Brain from the Loop<\/h2>\n\n\n\n
![\"Diagram](\"https:\/\/octer.ai\/hub\/wp-content\/uploads\/2026\/05\/Harness-and-Sandbox-Interaction-Diagram.png\")
<\/figure>\n<\/figure>\n\n\n\nexecute(name, input) \u2192 string<\/code>.<\/p>\n\n\n\n
\n\n\n\nReal Deployments: What Notion, Asana, and Sentry Built<\/h2>\n\n\n\n
![\"UI](\"https:\/\/octer.ai\/hub\/wp-content\/uploads\/2026\/05\/Claude-Agents-Integration-in-Notion-1024x710.png\")
<\/figure>\n<\/figure>\n\n\n\n![\"Sequence](\"https:\/\/octer.ai\/hub\/wp-content\/uploads\/2026\/05\/Workflow-Sequence-Diagram-for-Scaling-Claude-Managed-Agents-1024x710.png\")
<\/figure>\n<\/figure>\n\n\n\n