Amazon Web Services is expanding its AI infrastructure toolkit with Loom, an open-source agent platform designed to streamline how developers build and manage AI agents on AWS. The platform, introduced by Heeki Park in a detailed technical walkthrough, stitches together several AWS managed services into a unified control plane for deploying complex AI systems.
What Loom actually does
The core stack includes Amazon Bedrock for model inference, AgentCore Runtime for agent infrastructure, and the Strands Agents SDK. In English: Bedrock handles the AI brain, AgentCore handles where the brain lives and runs, and Strands provides the development kit that ties everything together.
Enterprise-grade governance is baked in from the start. Loom enforces mandatory tagging with keys like loom:application, loom:group, and loom:owner, which means every deployed agent is trackable and auditable. Role-based and attribute-based access controls, known as RBAC and ABAC, give platform engineering teams fine-grained authority over who can do what.
A typical deployment reportedly takes around 40 minutes in a fresh AWS account. The GitHub repository associated with the project shows active development, with releases as recent as early June 2026.
