Claude Managed Agents (launched April 2026, public beta) eliminates the need to build your own agent loop, sandboxing, session management, and credential handling. Anthropic provides a fully managed runtime environment where your agents execute autonomously.
| Concept | Definition | Key Detail |
|---|---|---|
| Agent | The definition: model + system prompt + tools + skills | Defined once, instantiated many times |
| Environment | Secure cloud container with pre-installed packages, network access, and file system | Configurable dependencies, isolated per session |
| Session | A runtime instance where the agent executes tasks | Persistent file system, conversation history, resumable |
Managed Agents require the beta header anthropic-beta: managed-agents-2026-04-01. Standard Claude API token rates apply, plus a flat infrastructure fee of $0.08 per session-hour.
// Creating a Managed Agent session
const session = await anthropic.beta.managedAgents.sessions.create({
agent_id: "agent_research_01",
environment: { packages: ["pandas", "requests"] },
instructions: "Research the latest competitor pricing"
});
// Session runs autonomously in Anthropic's cloud