Anthropic launched Claude Managed Agents in public beta at eight cents per agent runtime hour plus model usage fees. Developers get sandboxed containers, authentication, checkpointing, error recovery, session persistence, and end-to-end execution tracing: every piece of infrastructure that separates a demo from a production deployment, available as a set of composable APIs.
Building AI agents that work in production requires some new infrastructure and quite a bit of legacy systems integration. You need isolated execution environments, credential management, state persistence across disconnections, and observability tooling. AWS, GCP, and Azure all offer integrated agent platforms, but what makes Anthropic’s managed agents offering unique is its price transparency and its tight integration with Claude.
Anthropic highlighted some early adopters. Rakuten reports deploying specialist agents across product, sales, marketing, and finance in one week each. Sentry paired a debugging agent with a patch-writing agent that opens pull requests; they shipped the whole system in weeks instead of months.
Then, there are two research preview features that make me think about the future of work: multi-agent coordination (agents spinning up other agents for parallel work) and self-evaluation loops that improved task success by up to 10 points in internal testing. Anthropic says Notion is already using the multi-agent capability in its Custom Agents product for parallel task execution.
Are managed agents the SaaS business that survives SaaSpocalypse? For Anthropic, the eight-cent-per-hour runtime fee creates new recurring revenue tied to agent uptime that scales with customer success. For the rest of us, we get an up close and personal view of how our AI coworkers are going to be governed in the agentic age.
Every company needs a Claw strategy. Do you have one?
Author’s note: This is not a sponsored post. I am the author of this article and it expresses my own opinions. I am not, nor is my company, receiving compensation for it. This work was created with the assistance of various generative AI models.