Anthropic Ships Claude Managed Agents — Production Orchestration With Notion, Asana, Rakuten on Day One
Anthropic launched Claude Managed Agents, a production-grade platform that handles the full agent execution stack — sandboxing, authentication, tool execution, session persistence, and orchestration — with Notion, Asana, and Rakuten as day-one enterprise adopters.
Original sourceAnthropic launched Claude Managed Agents today, a platform designed to take the infrastructure burden out of deploying AI agents in production. Where most agent frameworks hand you a set of primitives and wish you luck, Managed Agents handles sandboxed code execution, credential management, tool authentication, persistent session state, and multi-step orchestration as managed services — the company is explicitly positioning it as "get to production 10x faster" for enterprise teams.
The day-one adopters are telling: Notion is using it for its AI writing and organization features, Asana for automated project management flows, and Rakuten for customer service agents at scale. These are companies that have already bet on Claude for core workflows and are now upgrading from prototype-grade agent setups to something with proper reliability SLAs.
The technical approach centers on isolation. Each agent run gets its own sandboxed execution environment — no shared state between runs by default, no credential leakage between tenants. Tool authentication is handled through a managed secrets layer rather than requiring developers to pipe API keys through their own infrastructure. Persistent sessions enable multi-turn, long-running tasks that span hours rather than the single-request window most agent APIs currently support.
This is the most direct shot Anthropic has taken at the LangChain/LangGraph ecosystem — companies that built the "agent infrastructure" layer between raw LLM APIs and production use cases. By absorbing orchestration, sandboxing, and session management into the model provider layer, Anthropic is betting that enterprises would rather buy a managed service than stitch together open-source agent frameworks.
The pricing model is compute-based: you pay for the underlying Claude tokens plus a managed infrastructure fee. For high-volume, long-running agents, this will add up — but for teams whose current alternative is maintaining their own agent orchestration stack, the total cost of ownership math likely favors Managed Agents.
Panel Takes
The Builder
Developer Perspective
“The sandbox and session persistence features solve the two biggest production headaches for agent deployments — state management and security isolation. If the reliability SLAs hold up, this is a genuine 'throw away your LangGraph setup' moment for teams already on Claude.”
The Skeptic
Reality Check
“Anthropic absorbing the orchestration layer makes Claude even more of a walled garden — your agent workflows become deeply coupled to one provider. The pricing opacity on compute fees for long-running sessions could also lead to bill shock at scale.”
The Futurist
Big Picture
“This is the inevitable vertical integration of AI: model providers who also control the execution environment have enormous leverage over the agent ecosystem. Managed Agents is Anthropic's move to own the full stack, and it will likely force OpenAI and Google to follow suit.”