AWS recently concluded its New York City Summit with a clear message: the future of enterprise software is agentic AI, and Amazon aims to own the infrastructure that enables it.
Swami Sivasubramanian, AWS VP of Agentic AI, delivered a keynote that positioned AWS as the enterprise-grade foundation for AI agents at scale, with announcements spanning infrastructure, development tools, and data management capabilities.
Amazon Bedrock AgentCore: The Foundation for Production AI Agents
The summit’s marquee announcement was Amazon Bedrock AgentCore, now in preview. This is AWS’s most comprehensive infrastructure investment in agent-based AI to date, directly targeting the persistent challenge of moving AI agents from proof of concept to production deployment.
AgentCore addresses what AWS refers to as the “undifferentiated heavy lifting” required for enterprise AI agent deployment. The service provides memory management, identity controls, tool integration, runtime environments, and observability capabilities.
Notably, AWS designed AgentCore to work with any open-source framework, explicitly supporting CrewAI, LangGraph, and LlamaIndex rather than forcing customers into proprietary tooling.
The service includes several technical components that tackle common production roadblocks. A code interpreter provides sandboxed runtime environments for executing agent-generated code. Browser tools enable agents to operate within users’ web browsers. Identity and access management systems ensure enterprise-grade security controls. Memory management handles persistent state across agent interactions.
AWS also introduced Kiro, an IDE for agent development that bridges natural language specification with traditional software engineering practices. Kiro generates scalable code from natural language descriptions, creates task dependencies, and maintains living documentation that evolves with the codebase. Agent Hooks automate workflow triggers based on file changes or manual prompts.
AWS Marketplace Expansion Into AI Agents
AWS significantly expanded its Marketplace to include AI agents and agentic tools, creating a centralized catalog for enterprise AI adoption. The Marketplace now features pre-built agents, agent tools, knowledge bases, guardrails, and professional services from partners including Anthropic, Salesforce, and Deloitte.
This marketplace strategy enables customers to test and deploy agent solutions from multiple vendors within the AWS environment, all of which utilize the AgentCore runtime. It’s an approach that reduces procurement friction and provides a governed platform for AI adoption at scale.
Updating Data Infrastructure for AI
AWS introduced several foundational improvements to its data layer, targeting persistent bottlenecks in AI adoption around data access and quality. Amazon S3 Metadata received a significant update, now providing comprehensive visibility into all S3 objects through live inventory and journal tables. This enables SQL-based analysis of both existing and new objects with automatic updates within an hour of changes.
Amazon S3 Vectors, now in preview, is the first cloud storage service with native vector support at massive scale. The service promises up to 90% cost reduction compared to conventional vector database approaches while integrating seamlessly with Amazon Bedrock Knowledge Bases, SageMaker, and OpenSearch. This addresses the challenge that traditional vector databases face in meeting the rapidly changing memory requirements of agentic AI applications.
Amazon SageMaker received three new capabilities designed to eliminate data silos. QuickSight integration enables the creation, governance, and sharing of dashboards directly within SageMaker workflows. S3 Unstructured Data Integration automatically catalogs documents and media files. Automatic data onboarding from Lakehouse architectures unifies structured and unstructured data management in a single experience.
The introduction of Amazon SageMaker Catalog serves as a central clearinghouse for semantic meaning across enterprise data. This complement to AWS Glue Data Catalog utilizes generative AI to enrich metadata with business context, enabling users to consistently discover and understand data across organizations.
Customization Options for Nova Model Family
AWS announced comprehensive customization options for its Nova foundation model family through SageMaker AI. These capabilities span all training stages, from pre-training to post-training, including fine-tuning and alignment recipes, which are delivered as ready-to-use SageMaker recipes.
The Nova Act model achieved particular attention for its browser-based agent capabilities, with AWS reporting over 90% end-to-end task completion rates in early enterprise use cases. This delivers a significant improvement in agent reliability for web-based interactions.
Additional Infrastructure & Service Updates
Several other announcements rounded out AWS’s AI and infrastructure improvements. TwelveLabs video understanding models became available on Amazon Bedrock, enabling video search, scene classification, content summarization, and insight extraction capabilities.
Amazon EventBridge introduced enhanced logging for event-driven applications, providing comprehensive event lifecycle tracking for monitoring and troubleshooting. Amazon ECS introduced built-in blue/green deployment capabilities for safer container application releases, providing near-instantaneous rollback.
Amazon EKS now supports clusters with up to 100,000 nodes, enabling massive AI/ML workloads with support for up to 1.6 million AWS Trainium accelerators or 800,000 NVIDIA GPUs per cluster.
AWS also launched the AI League program, which combines competitive elements with hands-on learning to support workforce upskilling in AI services. The Free Tier program received an enhancement, offering new customers up to $200 in credits to explore AWS services.
These announcements place AWS directly against Microsoft’s Copilot Studio and Google’s Vertex AI Agent Builder in the enterprise agentic AI space. AWS’s framework-agnostic approach, combined with AgentCore, differentiates it from more proprietary solutions, potentially appealing to enterprises with existing AI investments across multiple frameworks.
Analysis
The marketplace expansion creates a distribution channel that could accelerate enterprise AI adoption while locking customers into AWS infrastructure. This strategy mirrors successful cloud marketplace models but applies them specifically to the emerging AI agent ecosystem.
AWS’s data infrastructure improvements address fundamental challenges that have limited AI adoption in enterprises. The combination of enhanced metadata capabilities, native vector storage, and unified data management directly addresses the data sprawl and governance issues that many organizations encounter when scaling AI initiatives.
The scale improvements in EKS directly challenge specialized AI infrastructure providers and hyperscale competitors. Supporting 100,000 nodes per cluster with massive accelerator counts, targets AWS for the largest AI training and inference workloads.
However, gaps remain in AWS’s agent development lifecycle, particularly in areas such as security, quality governance, revision management, and platform engineering. The success of AgentCore will largely depend on how quickly AWS can address these areas while maintaining the simplicity that makes the service appealing.
The announcements show AWS’s recognition that the AI infrastructure market is shifting from model hosting toward comprehensive agent development and deployment platforms.
