Anthropic Opens Claude Managed Agents Public Beta, Charging Eight Cents Per Hour to Host Enterprise Agents
Anthropic's new cloud service handles sandboxing, state, and permissions so companies can ship production agents in weeks instead of months.
Overview
Anthropic on April 8 opened a public beta of Claude Managed Agents, a cloud service that takes over the infrastructure layer of enterprise AI agent deployments and bills customers a flat runtime fee on top of standard model usage. The company is pitching the offering as the missing production stack for businesses that have been prototyping agents for months without reaching deployment, describing it on its blog as “a set of composable APIs for building cloud-hosted agents” designed to collapse agent timelines from months to weeks.
What Was Launched
Claude Managed Agents is aimed at developers who want to ship agents without first building the supporting machinery. According to Anthropic’s product announcement, the service provides secure sandboxing, authentication, checkpointed long-running sessions with persistent state, scoped permissions, and session tracing from the Claude Console. A research-preview feature covers multi-agent coordination and self-evaluating agents that iterate toward defined outcomes.
Pricing is consumption-based. Customers continue to pay standard Claude Platform token rates for model calls, plus an additional eight cents per session-hour of active agent runtime, as Anthropic and SiliconAngle confirm.
In internal testing reported by SiliconAngle, an automated prompt-refinement feature in the managed service improved task success by up to ten points over a standard prompting loop, with the largest gains on complex problems.
Early Enterprise Customers
The initial customer roster skews toward companies already betting heavily on agent workflows. Anthropic names Notion, Rakuten, Asana, Sentry, Vibecode, Atlassian, General Legal, and Blockit as beta users. Each is running a different flavor of agent: Notion is embedding custom agents directly into workspaces, Rakuten is deploying multi-department specialists accessible through Slack and Teams, Asana is positioning agents as “AI Teammates” inside projects, and Sentry is using them to debug and write patches.
SiliconAngle reports that several of these early customers have already moved agents built with the service into production, a sharper distinction than the usual “piloting” language that accompanies most enterprise AI beta programs.
A Paired Enterprise Push
The Managed Agents beta did not land alone. As 9to5Mac reports, Anthropic on April 9 also moved Claude Cowork from research preview to general availability on macOS and Windows, and layered on a set of enterprise-tier controls: role-based access, group spend limits, usage analytics, expanded OpenTelemetry support, a Zoom MCP connector, and per-tool connector controls.
Taken together, the two announcements are Anthropic’s clearest signal yet that the company is productising agent infrastructure rather than continuing to ship models and leaving integration to customers. In its own framing, quoted by 9to5Mac, the company says “building agents meant spending development cycles on secure infrastructure, state management, permissioning, and reworking your agent loops for every model upgrade” — all problems the Managed Agents service is intended to absorb.
What We Don’t Know
Anthropic has not published service-level agreement details, regional availability beyond the Claude Platform, or data-residency guarantees for the sandboxed execution environments, and the public documentation reviewed for this brief does not spell out rate limits for the beta. The ten-point task-success improvement cited by SiliconAngle comes from internal testing on structured file generation; Anthropic has not released a reproducible benchmark or independent evaluation. Longer-term pricing, including whether the eight-cents-per-hour surcharge survives beyond the beta, is also unspecified.
It is likewise unclear how Managed Agents will interact with customers who run Claude through cloud partners rather than the Claude Platform directly. Anthropic’s announcement scopes the service to its own platform, leaving AWS Bedrock and Google Cloud Vertex deployments outside the initial beta.