Anthropic launches Claude Managed Agents beta

Anthropic has started putting more of its agent-building stack behind a managed service instead of asking customers to assemble the full runtime themselves. In an April 8 engineering post, the company introduced Claude Managed Agents as a hosted service on the Claude Platform for “long-horizon agent work,” saying the system is built around interfaces meant to remain stable even as the underlying harness changes. Anthropic said the service is designed so developers can hand off more of the runtime layer while focusing on the task logic and tools around Claude. The company has paired that rollout with a broader public push on agent operations, including live workshops and security guidance for enterprise users. The company’s framing is consistent across its recent materials. Anthropic’s engineering team said harnesses often encode assumptions about what models cannot do, and those assumptions can become outdated as models improve. In that post, the company described Managed Agents as a way to separate those changing implementation details from the developer-facing surface. ### What, exactly, is Anthropic managing for customers? Anthropic said on April 8 that Managed Agents is a hosted service that runs long-horizon agents on a customer’s behalf through a small set of interfaces. In the engineering post, the company said it “virtualized” three core components of an agent: the session, the harness, and the sandbox. The session is the append-only log of what happened, the harness is the loop that calls Claude and routes tool calls, and the sandbox is the execution environment where Claude can run code and edit files, Anthropic said. The company said those abstractions are intended to outlast any single implementation choice underneath them. ### Why is Anthropic talking so much about harnesses? Anthropic’s April 8 post said harnesses can become stale as model behavior changes. (anthropic.com) The company gave one example from prior internal work: a harness built for Claude Sonnet 4.5 added context resets to address what Anthropic called “context anxiety,” but those resets became unnecessary when the same harness was used with Claude Opus 4.5. That example is central to Anthropic’s pitch. (anthropic.com) Rather than asking customers to keep rebuilding orchestration logic every time model behavior shifts, the company said it wants the interfaces around agents to stay stable while implementations change underneath. ### Where does the security piece fit? Anthropic has separately argued that more capable agents increase what it calls the potential “blast radius” of failures. (anthropic.com) In a May 25 engineering post on containment, the company said the main engineering question is how to cap that blast radius as agents gain broader access across products. The company said its approach relies heavily on containment rather than only human approval prompts. (anthropic.com) Anthropic wrote that users approved roughly 93% of permission prompts in Claude Code telemetry, which it cited as evidence of approval fatigue, and said it has focused on access boundaries such as sandboxes, virtual machines, and egress controls. Anthropic has also published a broader policy framework for “safe and trustworthy agents.” In that August 2025 post, the company said humans should retain control over how goals are pursued, especially before high-stakes decisions are made, and pointed to read-only defaults and approval requirements in Claude Code as examples. (anthropic.com) ### What has Anthropic shown publicly beyond the product post? (anthropic.com) Anthropic’s events page shows an active schedule of agent-focused webinars and workshops in early June. The page lists “Building Agents with Claude Opus 4.8: Live Demos with Letta and Browserbase” for June 4 and “Securing & Governing Claude: The Compliance API and Security Integrations” for June 5. The same events page also lists “The Briefing: Enterprise Agents” as a virtual event held on April 22. (anthropic.com) Anthropic describes its events hub as the place to access livestreams and recordings from past conferences and webinars. ### How does this fit with Anthropic’s broader enterprise push? Anthropic’s newsroom and product pages show the company has been expanding enterprise-focused products around agents, security, and workflows. (anthropic.com) Its site now includes Claude Enterprise, Claude Security, Claude Code, Claude Cowork, and marketplace and partner-network offerings aimed at business deployments. Anthropic’s next public milestones are already on its calendar. The company’s events page lists the June 5 security webinar, AWS Summit appearances beginning June 10, and additional Claude-focused enterprise sessions later in June. (anthropic.com 1) (anthropic.com 2)