AWS Launches Bedrock AgentCore to Scale AI Agents for Enterprises

At its Summit in New York, AWS announced Bedrock AgentCore, a new platform (now in preview) designed to help developers deploy AI agents at scale. The platform combines an agent runtime, memory system, and enterprise security features, addressing the challenges of moving AI agents from prototype to production.

Key Features of Bedrock AgentCore

• Runtime: Secure deployment and scaling of agents, with individual agents running for up to eight hours and checkpointing for recovery.
• Memory System: Supports short-term and long-term memory, shared across agents and sessions.
• Model Agnostic: Works with any framework or model, including those on Amazon Bedrock or locally hosted.
• AgentCore Gateway: Converts resources into Model Context Protocol (MCP)-compatible tools.
• Code Interpreter: Allows agents to write and execute JavaScript, TypeScript, and Python code securely.
• Browser Tool: A cloud-based browser runtime for interacting with websites in isolated VMs.
• Observability: Provides end-to-end visibility into agent behavior, reasoning, and tool usage using OpenTelemetry.

The AWS AI Stack after today’s Bedrock AgentCore announcement. Image credit: AWS.

Additional Announcements

• Agent Marketplace: A curated marketplace for third-party AI agents and tools, scanned for vulnerabilities.
• Amazon S3 Vectors: Built-in support for storing and querying vectors at scale, integrated with Amazon Bedrock Knowledge Bases and OpenSearch Service.
• Nova Models Updates: Easier customization and fine-tuning of Amazon’s Nova models using Amazon SageMaker HyperPod.

For more details, watch the Amazon Bedrock AgentCore video.

AWS continues to innovate in the AI space, providing enterprises with the tools they need to scale AI solutions efficiently and securely.