Anthropic has launched Claude Managed Agents on its Claude Platform in public beta.
The product is a set of APIs for building and deploying cloud-hosted agents, with Anthropic handling infrastructure such as secure execution, state management, permissions and tracing. Developers can define tasks, tools and guardrails, while the orchestration system manages tool use, context and error recovery.
The launch reflects growing demand for software tools that can move AI agents from prototype to live deployment without requiring engineering teams to build supporting systems from scratch. The service is aimed at both single-task agents and more complex multi-agent workflows.
Platform tools
The service includes sandboxed code execution, authentication, checkpointing, credential management and execution tracing. It also supports long-running sessions that can continue for hours and preserve progress and outputs through disconnections.
Some functions are being rolled out more gradually. Multi-agent coordination, which lets agents launch and direct other agents in parallel, is in research preview. A self-evaluation feature that allows Claude to iterate against defined success criteria is also in research preview.
The product is designed specifically for Claude models, while still supporting more traditional prompt-and-response workflows for developers who want tighter control over behaviour.
In internal testing on structured file generation, Managed Agents improved task success by up to 10 points compared with a standard prompting loop, with the largest gains on harder tasks. Session tracing, integration analytics and troubleshooting guidance are also built into the Claude Console.
Customer use
Anthropic cited several early users that have built tools on the platform across coding, productivity, finance and legal work. These deployments range from agents that read codebases and open pull requests to systems that process documents and extract key information.
Notion is using the technology inside Notion Custom Agents in private alpha, where teams can delegate work to Claude within their workspace. Engineers are using it to ship code, while other workers use it to create websites and presentations, with multiple tasks running in parallel.
Rakuten has deployed agents across product, sales, marketing and finance that connect to Slack and Teams. Employees can assign tasks and receive outputs including spreadsheets, slides and apps, and each specialist agent was deployed within a week.
Asana used Managed Agents in the development of AI Teammates, its collaborative AI agents that work inside Asana projects. The service helped Asana add advanced features faster than it otherwise could have.
Other cited users include Vibecode, which uses Managed Agents as its default integration for app deployment, and Sentry, which paired its Seer debugging agent with a Claude-based agent that writes a patch and opens a pull request. According to Anthropic, Sentry shipped that integration in weeks rather than months.
Pricing model
Managed Agents is priced on consumption. Standard Claude Platform token rates apply, with an additional charge of $0.08 per session-hour for active runtime.
The launch adds to competition among AI model providers to offer developers not just access to foundation models but also the tools needed to build practical software products around them. As businesses test AI agents for internal workflows and customer-facing tasks, vendors are increasingly trying to reduce the engineering work involved in monitoring, governance and deployment.
Developers can also use the latest version of Claude Code and the built-in claude-api Skill to build with Managed Agents.