AWS introduces new AI agent tools and vector storage to accelerate enterprise development

AWS Launches New AI Development Tools and Vector-Optimized Storage

Amazon Web Services (AWS) introduced a suite of tools aimed at simplifying the creation and management of AI agents. The announcements came at the AWS Summit in New York, where AWS highlighted new capabilities for AI development and data handling.

Amazon Bedrock AgentCore: AI Agent Environments

The centerpiece of the launch is Amazon Bedrock AgentCore, a collection of services that streamline building AI agents. Its key component, AgentCore Runtime, offers cloud-based sandboxes where AI agents can run for up to eight hours, enabling automation of tasks like large dataset analysis with flexible security settings per workload.

When AI agents need to interact with external systems, AgentCore Gateway provides secure access to APIs, AWS Lambda code snippets, and other workloads. It supports authentication through services like Okta. Agents can also execute generated code via an integrated code interpreter and perform web interactions with a cloud-based browser. To monitor performance and reliability, developers can use AgentCore Observability.

This setup delivers a secure, serverless runtime featuring isolated sessions, extended workload duration, and built-in permission controls to maintain trustworthy AI agent operations.

Amazon S3 Vectors: Storage Optimized for AI Data

To support AI data needs, AWS launched Amazon S3 Vectors, a storage service optimized for vectors—the numerical representations neural networks use. It costs up to 90% less than comparable services and organizes data into vector buckets containing multiple vector indexes, each capable of holding tens of millions of vectors.

Users can add metadata such as creation dates to vectors, improving the ability of AI models to quickly find relevant information. AWS claims sub-second query latency, with automatic data optimization as vectors are added, updated, or deleted.

S3 Vectors integrates with AWS services like Amazon Bedrock, which provides access to foundation models from both AWS and third-party providers such as Anthropic. These models are the backbone for AI agents and applications.

Customizing Models with the Amazon Nova Series

AWS is expanding customization options for its Amazon Nova model series on Bedrock. This includes large language models and specialized neural networks for tasks like image generation. Customers can fine-tune models at both pre-training and post-training stages.

One supported customization method is Reinforcement Learning from Human Feedback (RLHF), where human input guides the model to improve response quality. After tuning, models can be deployed directly on Bedrock.

Additional AI Tools and Investments

• The AWS Marketplace now features a dedicated section for AI agents and related tools from AWS partners.
• Nova Act, a Bedrock model capable of performing browser actions, has received an updated SDK with enhanced cybersecurity features.
• AWS introduced two new MCP servers: one providing API-related data, and another containing developer documentation knowledge. AI agents can pull this info into their responses.
• AWS committed $100 million to its AWS Generative AI Innovation Center, which offers customers access to AI researchers and engineers and has completed thousands of AI projects since its formation in 2023.

These developments show AWS’s ongoing efforts to provide practical tools and infrastructure that help IT and development professionals build effective AI solutions with scalable storage and model customization capabilities.