Amazon Bedrock AgentCore Debuts Managed Harness to Speed Up AI Agent Development

Rapid Prototyping with Managed Harness

Amazon Bedrock AgentCore has introduced a suite of new features designed to streamline the creation of autonomous AI agents. The centerpiece is a managed harness currently in preview, which automates the complex orchestration loop of reasoning and action execution. Developers can now define an agent with a model and system prompt, allowing the platform to manage tool selection and response streaming automatically. This eliminates the need for custom orchestration code during the early stages of development.

Isolated Execution and Persistent State

Technical efficiency is a core focus of this update. Each agent session runs within its own isolated microVM, providing secure filesystem and shell access. A new persistent filesystem feature allows agents to save their state, enabling them to resume complex, multi-step tasks after being paused. This setup removes the heavy lifting of building sandboxed environments from scratch, as noted in the AWS technical documentation.

Unified Lifecycle Management

Beyond prototyping, the new AgentCore CLI offers a centralized interface for managing the full agent lifecycle. It integrates with the AWS Cloud Development Kit to ensure that AI agents follow infrastructure-as-code best practices. This ensures that a successful prototype can be promoted to production with the necessary governance and security protocols in place. Terraform support is expected to follow shortly to support a wider range of deployment environments.

Integration with AI Coding Assistants

AWS is also extending these capabilities into developer environments through specialized AgentCore skills. These skills are now available via Kiro Power, with upcoming support for popular tools like Claude Code and Cursor. These integrations provide real-time guidance and resource configuration directly within the IDE, further reducing the friction of building sophisticated AI workflows. Developers can learn more about these integrations on the AWS Machine Learning Blog.