Anthropic launches Claude Managed Agents to speed up AI agent development
Anthropic PBC released Claude Managed Agents today, a cloud service that automates much of the infrastructure work required to build and deploy production AI agents. The company says the offering cuts development time from months to weeks.
Building a production-grade agent traditionally requires teams to construct significant scaffolding beyond the agent itself. Developers must configure isolated containers, set up runtime infrastructure, implement observability features, and manage state - the data agents use to complete tasks. Claude Managed Agents handles most of this work through APIs.
Customers pay for Claude model usage plus eight cents per agent runtime hour.
How it works
Developers start by describing the tasks they want to automate. They then specify which third-party tools the agent should access and define security rules - such as whether tools can activate without user approval.
The service automatically spins up an isolated container for each agent and manages state across tasks. When an agent receives a prompt, the system determines which tools to use. An error recovery mechanism lets agents resume work after outages.
Features in preview
Two capabilities remain in research preview. The first allows agents to spawn additional agents for complex tasks. The second automatically refines response quality - Anthropic said this improved task success rates by up to 10 percentage points in internal testing.
Early adoption
Early customers include Notion, Rakuten, and Asana. Several have already integrated agents built on the service into their products.
Developers looking to work with Claude agents can explore Claude training resources or AI agent and automation courses to build expertise with the platform.
Your membership also unlocks:
