Amazon Web Services has introduced a capability within Amazon Bedrock AgentCore that enables teams to run Claude Code—an AI-powered coding assistant—as remote agents executing tasks inside isolated sessions on AWS infrastructure. This approach moves coding tasks off local machines and into secure, scalable cloud environments.
Using AgentCore runtime, each coding task is processed within a disposable microVM container in the user’s Virtual Private Cloud (VPC). These containers isolate compute, memory, and file system for every session, ensuring sanitized environments and reduced risk of data leakage. Tasks submitted are queued and handled asynchronously, allowing developers to offload work and return to automated pull requests for code review.
Amazon Bedrock provides the machine learning backbone, with Claude Code leveraging Anthropic models accessed via private endpoints to maintain strict data protection. The platform is compliant with various standards including SOC, ISO, FedRAMP Moderate, HIPAA eligibility, and GDPR. Notably, Bedrock enforces zero operator access and configurable zero data retention policies, preventing model providers or AWS personnel from accessing customer prompts or outputs.
The remote agent architecture integrates tightly with GitHub through a short-lived OAuth-based GitHub App for repository access. Each task undergoes guardrail screening to block malicious, unauthorized, or secret-leaking prompts. Task results are submitted as GitHub pull requests where human reviewers retain control over merging, enhancing trust and security in automated code contributions.
This solution particularly benefits scoped feature work, bug fixes, and testing within trusted repositories that have defined build and test commands. Limitations include unsuitability for open-ended tasks or untrusted codebases due to potential security risks and the agent’s inability to make code execution inherently safe. Cost prediction is also challenging due to variable runtime durations and AWS resource consumption.
Deployment requires an AWS account in supported regions, CLI tools, and a connected GitHub repository. The system supports submitting tasks either through a web console or directly from Claude Code with a plugin. Throughout the task lifecycle—planning, coding, verification—AgentCore memory records lessons to improve future task handling within the same repository.
By enabling remote, isolated execution of coding tasks backed by robust security controls and cloud-native integrations, Amazon Bedrock AgentCore extends AI-assisted development workflows for enterprises prioritizing governance and scalability. Developers gain operational flexibility by detaching from local constraints while maintaining compliance and control over code changes.
Source & further reading
Extend Claude Code to remote sessions on Amazon Bedrock AgentCore with security controls ↗Source published 2026-11-06. This report was prepared with AI assistance from the linked primary-source material. Claims and limitations should be checked against the original announcement or research.