Anthropic ships managed agents

Anthropic is no longer just selling a language model and an application programming interface. On April 8, 2026, it launched Claude Managed Agents, a hosted service that runs the agent loop, tool calls, and cloud environment for companies building autonomous workflows. (claude.com) That changes who does the hard work. Instead of a customer wiring together memory, web browsing, code execution, and security controls on its own servers, Anthropic says its managed runtime now handles those pieces inside Claude’s platform. (platform.claude.com) An AI agent is basically a model with hands. The model decides what to do next, and the runtime gives it places to read files, run commands, browse the web, and use tools without a human clicking every button. (platform.claude.com) Most companies have learned that the model is not the annoying part. The annoying part is everything around it: keeping long jobs alive, storing state between steps, passing credentials safely, and stopping the system from breaking when a tool call fails. (anthropic.com) Anthropic’s pitch is that developers should stop rebuilding that plumbing from scratch. Its product page says teams can create an environment, start a session, and stream the agent’s work through application programming interfaces instead of building a custom harness first. (platform.claude.com) The company is also framing this as a speed play, not just a convenience feature. Anthropic says Claude Managed Agents can get teams to production “10x faster,” and outside coverage says the company is pitching a shift from projects that took months to systems that can be stood up in weeks. (claude.com) (siliconangle.com) Under the hood, Anthropic’s engineers describe the design as separating the “brain” from the “hands.” In plain English, Claude does the reasoning while a separate managed layer handles execution, which lets Anthropic change the machinery around the model without forcing customers to rewrite their apps every time the model improves. (anthropic.com) That puts Anthropic much closer to OpenAI and Microsoft than to a plain model vendor. OpenAI’s Responses application programming interface already bundles built-in tools like web search, file search, computer use, and code interpreter, while Microsoft’s Azure OpenAI and Agent Framework docs describe managed agent patterns on top of those services. (developers.openai.com) (learn.microsoft.com) So the fight is moving up the stack. The winner may not be the company with the smartest model on a benchmark, but the one that gives enterprises the safest and least painful way to let that model actually do work for hours at a time. (wired.com) (unite.ai) Anthropic says customers including Notion, Rakuten, and Adaptive are already trying the service. That customer list matters because those companies are not buying a chatbot toy; they are testing whether Anthropic can be trusted with the full operating layer for agent software. (claude.com)