Anthropic is trying to make the ugliest part of AI agent building disappear. Its new Claude Managed Agents product gives businesses ready-made infrastructure for deploying autonomous AI systems, a move aimed squarely at enterprise teams that want agents to do real work without spending weeks wiring up the plumbing first.
The timing is no accident. Anthropic said on Tuesday that its annualized recurring revenue has topped $30 billion, about three times the level it reported in December 2024, and enterprise demand is doing most of the heavy lifting. With both Anthropic and OpenAI pushing deeper into business software as they prepare for potential public listings, the race is increasingly about who can make AI agents practical, not just impressive in demos.
What Claude Managed Agents actually provides
Claude Managed Agents is built around the unglamorous stuff developers usually have to assemble themselves: an agent harness, memory, tools, permissions, and a sandboxed environment where an agent can safely spin up software projects. Anthropic also says agents can run autonomously for hours in the cloud and watch what other Claude agents are doing, which is exactly the sort of operational detail that turns ”agent” from a buzzword into a systems problem.
- Agent harness for the infrastructure around Claude
- Built-in sandboxed environment for software projects
- Cloud runtime for hours-long autonomous work
- Permission controls for tool access
That matters because most companies do not want to hire a small army of engineers just to keep agents from wandering off the rails. Anthropic’s pitch is that customers can shift those people back to the actual product instead of the infrastructure behind it, which is also a polite way of saying the hard part was never the chatbot text.
Why Anthropic is leaning harder into enterprise
Anthropic’s enterprise push already has momentum through the Claude Platform, its API-based product for developers. The company says much of its recent revenue growth has come from that business, and developers are already using the API to deploy agents such as Claude Code inside workplace workflows. Claude Managed Agents simply takes that behavior and packages it into something closer to an off-the-shelf enterprise product.
There is also a competitive edge here. OpenAI has its own agent platform, Frontier, and both companies are building out the kind of enterprise stack that can justify hefty valuations before any public offering talk turns into paperwork. Wall Street has grown jittery about software stocks as AI vendors push into territory once owned by traditional SaaS firms, but the more immediate fight is over who becomes the default operating layer for office automation.
Notion is already using Claude Managed Agents for client onboarding
Anthropic used a demo from Notion to show what the product looks like in practice. In the example, a product manager walked through how a Claude Managed Agent could take on a long list of client onboarding tasks one by one, while a dashboard in the Claude Platform showed what the agents were doing and which tools they were using.
That’s the real sales pitch: not a flashy agent that can write poetry, but a controlled system that can grind through business chores without constant hand-holding. If Anthropic can make that feel boringly reliable, it gets much closer to becoming the vendor enterprises trust for day-to-day automation rather than experimental side projects.
The harder problem is adoption at scale
Even with Claude Managed Agents, Anthropic is still selling a bridge, not a destination. Most enterprises are nowhere near letting Claude run core operations on its own, and the remaining gap is as much about trust and governance as it is about software architecture.
The next test is whether customers want a managed stack from Anthropic or prefer to keep agent control closer to home. My bet: big companies will move slowly, but they will move, because ”build everything yourself” is a tiring strategy, and AI vendors know it.