<p>Anthropic has launched Claude Managed Agents in public beta, a suite of composable APIs for building and hosting AI agents on Anthropic’s own cloud infrastructure. The pitch is straightforward: instead of spending months building sandboxed execution environments, credential management, checkpointing, and tracing systems yourself, you define agents in plain language or YAML and run them on Anthropic’s platform.</p>
<h2>What’s included</h2>
<p>The platform handles the production infrastructure layer that typically slows agent deployment:</p>
<ul>
<li>Sandboxed code execution</li>
<li>State management and checkpointing</li>
<li>Credential management via Vaults</li>
<li>Scoped permissions</li>
<li>End-to-end tracing</li>
<li>Long-running autonomous sessions</li>
</ul>
<p>Multi-agent orchestration, advanced memory tooling, and self-evaluation loops are available in limited research preview.</p>
<p>The interface has five tabs: Quick Start, Agents, Sessions, Environments, and Vaults. Agents are created by describing goals in plain language and can be refined iteratively without redeployment.</p>
<p>Internal testing showed up to a 10-point improvement in task success over standard prompting loops.</p>
<h2>Pricing</h2>
<p>Billing is consumption-based: standard Claude API token rates, plus $0.08 per active session-hour, plus $10 per 1,000 web searches. This three-dimensional cost structure is worth paying attention to. Unlike serverless functions or VMs, autonomous agent workloads have no natural cost ceiling — a session that runs longer than expected, or spawns more sub-agents than anticipated, can accumulate charges in ways that don’t map cleanly to existing cloud cost dashboards.</p>
<p>Prompt caching can reduce costs by up to 90% on repeated context, and model selection should probably be evaluated on cost-per-outcome rather than cost-per-token. Current FinOps frameworks built for VMs and containers aren’t really equipped for this billing model.</p>
<h2>Early adopters</h2>
<p>Notion, Rakuten, Asana, Sentry, and Vibecode are among the early adopters.</p>
<h2>What this signals</h2>
<p>This marks Anthropic’s move beyond model provision into the infrastructure layer for enterprise agent deployment. Rather than just selling API access to Claude, Anthropic is now offering to run the agents themselves — 24/7, at scale, with the governance and observability tooling that enterprise deployments require.</p>
<p>There’s something worth sitting with here: agents running autonomously through the night, accumulating session-hours, making tool calls, with no human in the loop. The platform makes that easy to set up. Whether teams have the cost attribution and oversight practices to manage it responsibly is a separate question.</p>


Collections

Anthropic has launched Claude Managed Agents in public beta, a suite of composable APIs for building and hosting AI agents on Anthropic's cloud infrastructure. The platform handles production infrastructure concerns like sandboxed code execution, state management, credential vaults, scoped permissions, and end-to-end tracing — removing the need to build these systems from scratch. Multi-agent orchestration and self-evaluation loops are available in limited research preview. Pricing is consumption-based: standard token rates plus $0.08 per active session-hour and $10 per 1,000 web searches, a cost model that lacks natural ceilings and doesn't map well to existing FinOps frameworks. Early adopters include Notion, Asana, Sentry, and Rakuten. The launch signals Anthropic's expansion beyond model provision into enterprise agent infrastructure.

Claude Managed Agents launches in public beta