Claude Managed Agents: Anthropic's New Serverless Agent Infrastructure

TL;DR

Anthropic released Claude Managed Agents in public beta on April 8, 2026. It's essentially serverless infrastructure for AI agents - they manage the agent loop, sandboxing, tool execution, and session state. You focus on agent logic.

The Problem It Solves

Building a production-ready AI agent requires more than just calling a model. You need:

Agent loop management (tool call → execute → feed result → repeat)
Sandboxed execution environments
Session state persistence
Error recovery for long-running tasks

With the Messages API, you build all of this yourself. Managed Agents handles it for you.

Core Concepts

Agent → defines what (model, prompt, tools)
Environment → defines where (container, packages, network)
Session → connects Agent + Environment (stateful)
Events → communication channel (SSE streaming)

How It Works

# 1. Create an agent
curl https://api.anthropic.com/v1/agents \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-beta: managed-agents-2026-04-01" \
  -d '{"name": "code-reviewer", "model": "claude-sonnet-4-6", "tools": [{"type": "agent_toolset_20260401"}]}'

# 2. Create an environment
curl https://api.anthropic.com/v1/environments \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-beta: managed-agents-2026-04-01" \
  -d '{"name": "python-env"}'

# 3. Start a session
curl https://api.anthropic.com/v1/sessions \
  -H "x-api-key: $ANTHROPIC_API_KEY" \
  -H "anthropic-beta: managed-agents-2026-04-01" \
  -d '{"agent_id": "AGENT_ID", "environment_id": "ENV_ID"}'

That's it. The agent now has access to Bash, file operations, web search, and MCP servers - all built-in.

Built-in Tools

No custom implementation needed:

Bash - shell command execution
File operations - read, write, edit, glob, grep
Web search & fetch - search and retrieve web content
MCP servers - connect external tool providers

Messages API vs Managed Agents

	Messages API	Managed Agents
Scope	Model prompting	Full agent harness
Best for	Custom agent loops	Long-running, async tasks
Infrastructure	Build yourself	Anthropic-managed
Tool execution	Implement yourself	Built-in
State management	Implement yourself	Session-based, automatic
Long-running	Limited	Minutes to hours
Extra cost	Tokens only	Tokens + $0.08/hr

Pricing

Tokens: Standard Claude API pricing
Session runtime: $0.08/hour (running state only, idle is free, millisecond billing)
Web search: $10 per 1,000 searches

A 10-minute session costs about $0.013 in runtime alone.