Anthropic Launches Managed Claude Agents for Enterprises

Anthropic is formalizing a managed-agent approach that bundles its agent technology, desktop integrations, and orchestration patterns so teams can run autonomous Claude agents without building months of infrastructure. The company’s product surface includes Claude Cowork (desktop, local files, and app automation) and Claude Code (agentic coding workflows), supported by the same agent loop and SDK patterns Anthropic documents for long-running tasks.

Anthropic’s agent stack is purpose-built for multi-step, goal-driven work rather than single-turn chat. Claude Cowork targets knowledge work on desktops—moving between local files and applications to synthesize, assemble, and finish deliverables—while Claude Code executes agentic development workflows that read codebases, run tests, and commit changes. Anthropic’s research shows these agent patterns can persist across large numbers of sessions: in a published example, Claude ran across roughly 2,000 sessions to produce a C compiler capable of compiling the Linux kernel, illustrating the feasibility of multi-day, stateful agent workflows. The Agent SDK and the concept of reusable “Skills” (instruction modules applied automatically) underpin this capability.

• •Claude Cowork is presented as a simplified experience for non-technical teams; it runs on desktop and works with local files and applications to deliver finished outputs (organizing files, preparing documents, synthesizing research, extracting data). [anthropic.com/product/claude-cowork]
• •Claude Code and Anthropic’s engineering notes describe agentic coding patterns: context engineering, skills, persistent memory, test oracles, and orchestration for longer-running scientific and engineering projects. [anthropic.com/product/claude-code][anthropic.com/research/long-running-Claude]
• •Anthropic’s long-running-agent research provides a concrete scale example (≈2,000 sessions building a C compiler) and walks through using Claude Opus 4.6 in multi-day scientific computing workflows.

Managed Claude agents reduce the upfront engineering cost of shipping agents into production by encapsulating runtime, orchestration, and integration primitives. That lowers the barrier for product teams, data teams, and research labs to move from prompt-based workflows to delegated, stateful automation. Practitioners will need to incorporate operational controls—access policies, auditability, observability, and CI-like test oracles—because these agents act across local files and applications and can run for extended periods.

• •How Anthropic exposes governance controls, audit logs, role-based access, and enterprise deployment options for agents operating on sensitive local data.
• •Interoperability between Agent SDK-built custom agents and the managed Cowork/Code experiences (skill sharing, deployment portability).
• •Performance and cost patterns for multi-day, persistent-agent workloads versus orchestrating short-lived chains of prompts.