Claude Managed Agents: Features and How It Works—See It All at a Glance! Can It Make Production for Enterprises 10x Faster?

Anthropic has launched Claude Managed Agents, offering fully managed infrastructure and multi-agent coordination, significantly simplifying the underlying architecture for development. The official claims it can help enterprises accelerate production by ten times and is actively moving to capture the enterprise-grade AI market.

Claude Managed Agents is officially launched

AI tech giant Anthropic has launched Claude Managed Agents to help enterprises build and deploy AI Agents more easily. This tool provides developers with a plug-and-play foundation, simplifying the complex processes that previously hindered task automation, and also claims to make it possible for enterprises to accelerate the time it takes to move products into production by ten times.

What features does Claude Managed Agents offer?

According to the official article, when a typical enterprise develops an agent system, it usually takes several months to handle infrastructure, state management, and permissions setup.

Claude Managed Agents abstracts away these complex underlying architectures, so developers only need to define the agent’s tasks and guardrails, and the platform will automatically handle the subsequent operations.

Its key features include the following:

Feature 1: Fully managed security infrastructure

Claude Managed Agents provides production-grade execution environments that automatically handle secure sandbox execution and verification, allowing language models to access files, execute code, and browse the web in protected cloud containers.

Image source: anthropic Claude Managed Agents has fully managed security infrastructure

Feature 2: Support for long-running autonomous operation

Claude Managed Agents is designed for asynchronous work that needs to run for extended periods. The agent can operate autonomously for several hours and comes with a persistent file system and conversation history records. Even if the connection drops in the middle, it can still preserve execution progress and output results.

Feature 3: Multi-agent coordination and governance tools

Claude Managed Agents has the ability to start up and direct other agents to process tasks in parallel. The platform comes with built-in governance tools such as identity management, scope-based permissions, and execution tracking to ensure the safety of agents when accessing real systems.

How does Claude Managed Agents work?

Regarding the actual operation of Claude Managed Agents, the official documentation divides it into the following steps:

• Create an agent: Developers define the model, system prompts, and tools. After creation, it can be repeatedly referenced across different sessions.
• Create an environment: Configure cloud containers, preinstall software packages, and set network access rules in advance.
• Start a session: Launch the execution instances that reference the agent and environment configuration above to perform specific tasks.
• Send events and stream responses: Send the user’s messages as events. The model will autonomously execute tools and return the results, and the history will also be retained on the server side.
• Guide or interrupt: During execution, users can send additional events to guide the agent, or directly interrupt it to change the execution direction.

Bridging the technical gap, Notion tests it first

Claude platform engineering lead Katelyn Lesse said that large-scale deployment of agents is a highly complex distributed systems engineering problem. With the ready-made underlying architecture provided by Anthropic, customers’ engineers can focus their energy on core business and product development.

Anthropic’s official article further explains that in the old design of AI Agents, language models and execution tools are often bound within the same container, making system maintenance difficult.

In the new architecture, the computation model is separated from the execution environment. This not only significantly reduces the time-to-first-token latency, but also allows agents to flexibly connect to the enterprise’s own virtual private cloud, while improving security and preventing malicious code from accessing sensitive credentials.

In real-world applications, productivity startup Notion has shown how to use this new service to drive customer onboarding features.

In a demonstration, Notion product manager Eric Liu handed a long list of tasks to Claude Managed Agents, and it began completing each step one by one. Developers can also directly use the dashboard on the Claude platform to view the agent’s work status and the tools it is using in real time.

Tech giants race for the enterprise AI market

Recently, technology companies have poured resources into competing for the enterprise-grade AI market. **In February this year, OpenAI released **Frontier, and in March this year, Nvidia also followed the OpenClaw crawdad-feeding craze, launching an open-source **NemoClaw for OpenClaw, providing an open model and an isolated sandbox environment, adding data privacy and security protection to AI Agents.

These moves show that the industry is actively building more scalable and secure AI infrastructure for enterprises, accelerating the全面 rollout of AI across business domains.

Further reading:
Binance CZ mocks AI agents “Crawdad”: claims to free up your hands, but ends up busy fixing it every day