Claude Managed Agents gain AWS webhooks, orchestration, and self-hosted sandboxes

Claude on AWS is no longer just a model endpoint story. In its May 29, 2026 Claude Platform release notes, Anthropic said Claude Managed Agents webhooks, multiagent orchestration, and self-hosted sandboxes are now available on Claude Platform on AWS. The update also points developers to new IAM actions and the AnthropicSelfHostedEnvironmentAccess managed policy.

The important shift is operational. Webhooks let teams route session and workflow events into existing observability, approval, or ticketing systems. Multiagent orchestration gives a first-party path for splitting work across coordinated agents rather than forcing developers to stitch every role and handoff together themselves. Self-hosted sandboxes change the trust boundary: tool execution can run in a customer-controlled environment instead of Anthropic’s infrastructure, which matters for teams handling proprietary code, regulated data, or strict internal controls.

The timing fits a broader rollout. On May 11, Anthropic introduced Claude Platform on AWS with AWS billing and IAM authentication, exposing the Messages API, Files API, Message Batches API, Claude Managed Agents, Agent Skills, code execution, and tool use through native AWS endpoints. Earlier in May, multiagent sessions and webhooks entered public beta for Managed Agents. The May 29 note brings those agent operations closer to the AWS governance model many enterprises already use.

For buyers and platform teams, the question is shifting from model quality alone to agent control planes. Who sees the events, where tools execute, how credentials are scoped, and which system owns the audit trail are becoming product requirements. Anthropic’s AWS path gives security and infrastructure teams a more familiar surface for those decisions. The next thing to watch is whether Managed Agents can make multiagent work predictable in production: clear event histories, bounded permissions, understandable failure recovery, and costs that can be traced back to a workflow rather than a vague pool of agent activity.