Anthropic bets on managed agents

Anthropic has a new pitch for companies: stop building the plumbing for artificial intelligence agents and rent the whole building instead. On April 8, Anthropic launched Claude Managed Agents in public beta as a cloud service for running long, tool-using workflows on its own infrastructure. (claude.com) An artificial intelligence agent is just a model that keeps working after the first answer. Instead of replying once, it can open files, run code, search the web, call tools, and keep going for minutes or hours until a task is finished. (platform.claude.com) That sounds simple until a company tries to ship one. Anthropic says production agents need sandboxed code execution, checkpointing, credential management, scoped permissions, and tracing before users ever see a useful feature. (claude.com) Claude Managed Agents is Anthropic’s answer to that mess. The service lets developers define the model, tools, guardrails, and runtime environment once, then start sessions that Anthropic hosts and monitors. (platform.claude.com) Anthropic’s own docs draw a sharp line between direct model access and this new product. The older Messages application programming interface is for teams that want fine-grained control, while Managed Agents is for long-running and asynchronous work that needs state, storage, and tool use built in. (platform.claude.com) The company is selling speed as much as capability. Anthropic says Managed Agents can get teams to production “10x faster” and compress work that used to take months into days or weeks. (claude.com, siliconangle.com) The technical idea underneath is that the model’s “brain” should be separate from its “hands.” Anthropic’s engineering team says agent harnesses break when models improve, so it built stable interfaces for sessions, harnesses, and sandboxes that can change underneath without forcing customers to rebuild everything. (anthropic.com) That is a business bet as much as an engineering one. If companies trust Anthropic to run the agent loop, the tools, the logs, and the execution environment, Anthropic moves from selling tokens to owning more of the workflow stack. (anthropic.com, platform.claude.com) The timing lines up with a surge in enterprise demand. In February 2026, Anthropic said its run-rate revenue had reached $14 billion and that more than 500 business customers were each spending over $1 million on an annualized basis. (anthropic.com) By April 2026, multiple outlets reported Anthropic had pushed past $30 billion in annualized run-rate revenue, up from about $9 billion at the end of 2025. Anthropic’s own launch materials for Managed Agents do not repeat that figure, but the revenue jump fits the company’s push to sell production-grade tools to enterprises instead of just bigger models. (bloomberg.com, fortuneindia.com, sacra.com) The shift here is that “agent” is starting to mean infrastructure, not just intelligence. The companies that win this phase may be the ones that make autonomous software boring enough for a bank, a law firm, or a software team to trust it in production. (wired.com, claude.com)