AWS shows running Claude Code remotely on Bedrock AgentCore

AWS Machine Learning Blog published on 2026-11-07 00:45, presenting a demonstration of running Claude Code as a remote agent on Amazon Bedrock AgentCore. The post describes a deployment model in which each coding task is executed in an isolated session within the customer's AWS account, and every model call is routed through Amazon Bedrock rather than being handled by a separate runtime. This framing emphasizes that the agent execution environment is customer-managed: the Claude Code session, its credentials, and its task-specific context live inside the customer's own AWS account boundaries, while the model serving and governance path remains centralized on Amazon Bedrock.

The blog's summary also describes the intended workflow outcome: after the remote agent completes a task, the result is returned in the form of a GitHub pull request that the user can review and merge. That implies the publication is introducing a reference implementation or integration pattern for driving Claude Code remotely through Amazon Bedrock AgentCore, rather than announcing a generally available SKU. What the material verifies so far can be summarized as follows: AWS has described the configuration and operating model; it has stated that tasks run in isolated sessions; it has stated that every model call goes through Amazon Bedrock; and it has stated that task outputs are delivered as GitHub pull requests for review and merge. The documentation does not state observed benchmark gains, adoption metrics, or customer deployment results.

At this point, the available record consists of a single announcement. There is no reported timeline for general availability, no additional partners, and no engineering details beyond the summary, such as configuration files, endpoints, permission policies, or request flows. The current verified end of the story is therefore a published demonstration description: AWS has documented how to run Claude Code as a remote agent on Amazon Bedrock AgentCore, with isolated per-task sessions in the customer's AWS account, JSON model calls through Amazon Bedrock, and results returned as GitHub pull requests.