Amazon Web Services (AWS) made its mark as a pioneer in the cloud market by providing the foundation organizations needed to run infrastructure and software as a service. In recent years, that foundation is being expanded and used increasingly in support of generative AI.
This week at the 2025 AWS Summit New York City, the cloud pioneer announced a series of new AI services that aim to help organizations bring the promise of agentic AI beyond proof-of-concept (POC) to real-world deployments.
The announcements focus on the fundamental infrastructure requirements for agent deployment, including runtime execution, memory management, security controls, and operational monitoring capabilities that differentiate production systems from development prototypes.
Key AWS Summit NYC Announcements
AWS made the following key announcements at its 2025 AWS Summit NYC:
Amazon Bedrock Agent Core: Seven-service infrastructure suite for production agent deployment including runtime, memory, identity, code interpreter, browser tool, gateway, and observability services.
Amazon S3 Vectors: Native vector search capabilities in S3 object storage, reducing vector storage costs.
Nova Act: Browser automation agent achieving 90%+ task completion rates in enterprise use cases.
Kiro IDE Preview: Agent-powered development environment announced in preview.
AWS Transform Expansion: AI-powered migration capabilities for .NET, VMware, and mainframe modernization.
AWS Marketplace for AI Agents: Centralized catalog for pre-built agent solutions and tools.
“Agents are a tectonic change in a few dimensions,” Swami Sivasubramanian, vice president of agentic AI at AWS, said during the keynote. “They are a shift in the way software is built and how software is deployed and operated.”
Amazon Bedrock Agent Core: Production Infrastructure Services
The unveiling of Amazon Bedrock Agent Core was perhaps the biggest product news at AWS Summit NYC. Amazon Bedrock Agent Core addresses the fundamental challenge of moving agents from prototype to production The new offering provides seven distinct services for production agent deployment to address this challenge.
“Despite all of these open source frameworks and protocols, getting agents to production is still too hard,” Sivasubramanian said. “There are a bunch of missing pieces that make it really difficult to move from POCs that are sitting in laptops to agents in real-world production use cases.”Â
There are seven primary services that make up Amazon Bedrock Agent Core:
Runtime: Provides serverless execution environments for dynamic agent workloads, supporting sessions up to eight hours with complete session isolation and built-in checkpointing for recovery from interruptions.
Memory: Implements hierarchical memory management with automatic consolidation from short-term interactions to long-term concepts, supporting shared memory stores across collaborating agents.
Identity: Offers centralized access management with fine-grained controls, integrating with existing identity providers, including Cognito, Okta, and Microsoft Entra ID.
Code Interpreter: Enables secure code execution in sandbox environments with support for multiple programming languages, large file handling, and internet access.
Browser Tool: Provides model-agnostic web interaction capabilities with enterprise-grade security features including VM-level isolation.
Gateway: Converts existing APIs, Lambda functions, and services into MCP-compatible tools with one-click integration for popular services like Salesforce, Slack, and Jira.
Observability: Delivers workflow visualization and real-time performance monitoring through CloudWatch dashboards, emitting telemetry in OpenTelemetry format.
Kiro IDE: Reimagining Software Development
Deploying AI is one thing; building software with AI-powered tooling is quite another. That’s where the Kiro IDE fits in.
“Kiro is a new agent IDE that helps you from concept to production through a simplified developer experience,” Sivasubramanian explained.
The IDE addresses a fundamental challenge in software development: maintaining up-to-date specifications. Kiro transforms natural language requirements into structured development plans, identifying dependencies and linking them to original requirements.
A key feature of Kiro is the platform’s agent hooks system, which enables automated workflow management.
“Hooks are event-driven automation that trigger every time you save, delete, or create a file or on a manual trigger,” Sivasubramanian explained. “Hooks enables you to turn manual, tedious tasks into automated workflows, like automatically updating your tests as soon as you change your code, or refresh the documentation when you add a new API endpoint to your application.”
Nova Act: Production Browser Automation
For many IT users, the web browser is the primary place where they interact with applications and services.
AWS’ Nova Act addresses the complex challenge of enabling AI agents to interact with web browsers reliably.
“Using a browser seems effortless to us, but it requires constant judgment, interpreting what’s on screen while scrolling, clicking, and completing forms,” Rohit Prasad, senior vice president and head scientist at Amazon AGI, said during the keynote. “These actions, while simple for you and me, remain challenging for the AI.”
The keynote demonstrated Nova Act through a food assistance application example, where the agent automated form completion across multiple state websites. Prasad positioned Nova Act as a significant step toward more capable AI systems.
“Nova Act brings us a step closer to artificial general intelligence, where the AI can accomplish the same task you and I do every day on a computer,” he said.