Anthropic launches Managed Agents

Most companies can get a language model to answer a prompt in an afternoon. The hard part starts when that model has to keep working for hours, call tools, save state, recover from failures, and finish a job after the user has closed the tab. (anthropic.com) Anthropic’s new product is aimed at that second problem. On April 8, 2026, it launched Claude Managed Agents, which Anthropic describes as a suite of application programming interfaces for building cloud-hosted agents at scale on Anthropic-run infrastructure. (claude.com) An agent here is not just a chatbot. Anthropic’s own examples are systems that research, edit files, run code, use tools, and hand back a finished deliverable instead of a single reply. (anthropic.com) The reason this gets messy fast is that a production agent needs more than a model. Anthropic says shipping one usually means adding sandboxed code execution, checkpointing, credential management, scoped permissions, and tracing across the whole run. (anthropic.com) Managed Agents packages that plumbing into a prebuilt harness. Anthropic says the harness is the loop that calls Claude, routes tool calls, and keeps the job moving while other components handle the session log and the sandbox where code runs. (anthropic.com) Anthropic is also turning its internal products into building blocks for customers. Claude Cowork is its system for multi-step knowledge work on files and desktop apps, while Claude Code is its coding system that reads a codebase, edits files, runs tests, and ships changes. (anthropic.com 1) (anthropic.com 2) The pitch is that companies no longer have to build the agent runtime from scratch. Wired reported that Anthropic is bundling orchestration patterns from products like Claude Cowork and Claude Code so customers can skip months of infrastructure work. (wired.com) Anthropic is also making the setup less code-heavy than older developer tools. The New Stack reported that users can define an agent in natural language or in a YAML configuration file, set guardrails, and run it on Anthropic’s platform with the infrastructure abstracted away. (thenewstack.io) That puts Anthropic deeper into a part of the artificial intelligence market that is shifting from model access to workflow ownership. If a company lets Anthropic run the session log, the tool loop, the sandbox, and the permissions layer, switching providers gets harder than swapping one model endpoint for another. (anthropic.com) (thenewstack.io) Anthropic’s public line is speed. Its launch post says Managed Agents can get teams to production “10x faster,” which is another way of saying Anthropic wants to sell not just intelligence, but the operating system around it. (claude.com)