Amazon Web Services (AWS) today previewed a raft of capabilities that promise to make it simpler to build and deploy artificial intelligence (AI) agents.
Announced at an AWS Summit event in New York City, AWS has developed Amazon Bedrock AgentCore, a set of services through which organizations can create secure sandboxes and spin up serverless runtimes for building AI agents and invoke a set of observability tools.
Additionally, organizations can now automatically manage the amount of memory allocated to each AI agent, provide AI agents with a browser to scan websites, access external data sources and applications via a gateway and securely assign and manage the identities for AI agents.
AWS also launched an edition of the Amazon Marketplace that promises to make it simpler for organizations to discover and download AI agents built by AWS and independent software vendors (AWS).
Swami Sivasubramanian, vice president of Agentic AI at AWS, told conference attendees that Amazon Bedrock AgentCore will address the technical challenges that organizations are going to encounter any time they seek to build and deploy AI agents using multiple types of AI models. “There is no one model for every type of use case,” says Sivasubramanian.
AWS is also previewing an object store for the Amazon S3 cloud service specifically designed for the vectors used to build and customize AI models in a way that reduces costs by up to 90% along with deeper visibility into the metadata of all objects stored in S3 buckets via a live inventory and journaling of tables capability.
At the same time, AWS revealed it can now scale the Kubernetes clusters it provides for building AI models up to 100,000 nodes using as many as 1.6 million AWS Trainium accelerators or 800,000 NVIDIA graphics processing units (GPUs). IT teams can also now monitor and debug event-driven applications using an Amazon EventBridge that collects log data to simplify troubleshooting.
Additionally, AWS has created a set of templates that make it simpler to customize the Nova AI models it makes available via the managed Amazon SageMaker service for building AI models. AWS has also added an Amazon QuickSight integration for dashboard creation, governance, and sharing, an Amazon S3 Unstructured Data Integration for cataloging documents and media files, and automatic data onboarding from Lakehouse to Amazon SageMaker.
AWS is now making it possible to get started and explore AWS with up to $200 in credits for new users.
Finally, AWS also revealed it has launched an Amazon Nova Act research initiative, which is based on an AI model trained to perform actions within a web browser. Via a software development kit (SDK), developers can build agents that can complete tasks such as booking calendar appointments, directly from within a web browser.
It’s not clear at what pace organizations are operationalizing AI agents, but the Futurum Group predicts they will drive $6 trillion in economic value by 2028. The challenge is that most organizations still lack the IT infrastructure resources, and expertise, required to build and deploy AI agents at scale, a void that AWS is now clearly signaling it intends to fill.
