Anthropic released Claude Managed Agents in public beta, providing composable APIs that handle sandboxing, state management, and orchestration for AI agents. Here's what makes this different from the usual vaporware: Notion, Rakuten, Asana, and Sentry are already using it in production, with reported 10x faster deployment times compared to building from scratch.Every AI company says they're doing agents. Most are demos that fall apart under real-world conditions. Anthropic shipped actual infrastructure with paying enterprise customers already running production workloads. That's the difference between a product and a press release.Rakuten deployed enterprise agents across five departments in one week each — that's the kind of velocity that signals something real. Sentry went from bug detection to auto-generated pull requests in weeks instead of months. These aren't toy examples. These are real companies solving real problems.The platform handles the unglamorous infrastructure work that kills most agent projects: sandboxing so agents can't break things, state management so they remember context, credential handling so they can access systems securely, error recovery so failures don't cascade. You define the agent logic; Anthropic handles everything else.Pricing is $0.08 per session-hour of runtime, with idle time free. That's aggressive enough to be competitive with building in-house, especially when you factor in the engineering time saved. The 10-point task success improvement over standard prompting isn't just marketing — that's the difference between an agent that works and one that needs constant human intervention.Multi-agent coordination is in research preview, which signals where this is heading: agents that can delegate to other agents, split complex tasks, and coordinate across different domains. That's when this gets really interesting — or really dangerous, depending on your perspective.The question is whether managed platforms create vendor lock-in that's worse than the problem they solve. If your entire agent infrastructure runs on Anthropic's platform and they change pricing, modify terms, or sunset features, you're stuck. The same convenience that enables fast deployment creates dependency.Open-source orchestration tools like LangGraph and CrewAI offer alternatives, but they require more engineering effort and don't have the same level of polish. The gap between enterprise managed platforms and self-hosted solutions is widening, not narrowing.That said, Anthropic is backing the open AG-UI protocol (supported by Google ADK, AWS Bedrock, Microsoft Agent Framework), which suggests they're thinking about interoperability rather than pure lock-in. Whether that commitment survives commercial pressure remains to be seen.The technology is impressive. The execution is real. The question is whether managed platforms become the new cloud infrastructure — essential but controlled by a handful of companies — or whether open alternatives emerge before dependency locks in.