News

Sorting Out the AWS AI Blitz at New York City Summit

• By David Ramel
• 07/17/2025

At this week's AWS Summit in New York, the company announced a coordinated set of tools and services to support the development and deployment of AI agents. These included a new foundational runtime layer, a new AI agents category in AWS Marketplace, expanded customization for Amazon Nova, an additional $100 million investment in AWS's Generative AI Innovation Center and much more.

Swami Sivasubramanian, vice president of Agentic AI, introduced the company's agentic AI strategy during the opening keynote. He defined agents as autonomous software systems that leverage AI to reason, plan, and adapt in pursuit of user-defined inputs, completing tasks on behalf of humans or other systems. "Systems can now interact in more than code or plain language, but they can leverage tools, data sources, and many more, and they can figure out what tools it needs to do to use to accomplish a task," he said.

Sivasubramanian also framed the shift more broadly: "Agents are a tectonic change in a few dimensions. They are a shift in the way software is built and how software is deployed and operated. And three it changes how software interacts with the world and how we interact with software."

These announcements follow earlier AWS moves in the space, including improvements to Amazon Q, support for Claude 3.5 on Amazon Bedrock, and new vector indexing and inference features in S3 and SageMaker and many more.

Amazon Bedrock AgentCore: Runtime Infrastructure for Enterprise AI Agents
Amazon Bedrock AgentCore is a modular platform for operating AI agents in production. It supports multiple foundation models and open-source agent frameworks.

AgentCore includes the following modules:

Developers can use AgentCore Gateway to expose existing APIs and Lambda functions to the agent using a schema wrapper that supports tool definitions. Each component is designed to integrate securely with AWS authentication and monitoring services, allowing for modular agent deployment under enterprise controls.

New Category in AWS Marketplace: AI Agents and Tools
AWS introduced a dedicated "AI Agents & Tools" category in AWS Marketplace to help customers find, deploy, and manage agent-based solutions. Listings include standalone agents, toolsets, and protocol-compliant utilities from both AWS and third-party providers.

New offerings in the marketplace include types such as:

• Agent Blueprints: Prebuilt agents tailored for use cases like customer service, DevOps automation, or document analysis.
• Tooling Utilities: Components that extend agent functionality, such as memory connectors, code interpreters, and web browsing modules.
• Service Integrators: Middleware that bridges agents with enterprise APIs using MCP and A2A protocols.

"Fast-track automation with pre-built agents from AWS Partners," Sivasubramanian said in a post. "Our new API-based deployment method helps you to streamline integrations with other agents and tools that support MCP and A2A. And these agents can run on trusted AWS services or in your AWS environment, where you maintain control over security and access. You can deploy select pre-built agents and tools on AgentCore."

Each listing may include manifest files that describe authentication, API schema, and tool definitions, enabling declarative integration with Bedrock agents. Listings also support AWS Marketplace features like consolidated billing, license tracking, and deployment via CloudFormation, SageMaker JumpStart, or Bedrock.

Kiro: Agentic AI-Driven IDE
Speaking of tools, days before the summit, AWS introduced an agentic IDE designed for structured, spec-driven development, Kiro. It includes two modes: "vibe coding" for fast iteration and "spec coding" for documented, production-ready work. Agents in Kiro can assist with code generation, test creation, API documentation, and diff review.

Kiro uses the Model Context Protocol (MCP) to access external APIs, documents, and internal tools. The platform is based on Code OSS and supports Visual Studio Code extensions and themes. Preview pricing includes a free tier (50 interactions/month), a Pro plan ($19/month), and Pro+ plan ($39/month).

Amazon Nova: Expanded Customization Options via SageMaker
Amazon Nova now supports fine-tuning using SageMaker training methods such as continued pretraining, supervised fine-tuning, DPO, PPO, and knowledge distillation. Custom models can be deployed directly into Amazon Bedrock.

While the keynote included Nova visuals, no direct quotes from Rohit Prasad were verifiable in the transcript. Feature details are available in the official Nova customization announcement.

AWS Generative AI Innovation Center: $100 Million in Additional Funding
AWS added another $100 million in funding to its Generative AI Innovation Center. The initiative provides AWS experts to enterprise teams working on generative and agent-based applications.

Sivasubramanian also noted, "We are also investing in programs to help accelerate your AI journey, like partnering with Meta to supercharge generative AI innovation with Llama models."

Conclusion
Sivasubramanian concluded by stating that intelligent agents will fundamentally change how applications are built and how humans interact with software. He emphasized AWS's goal to make this transition as easy, secure, and open as possible.

The announcements from New York extend AWS's agent ecosystem across runtime infrastructure (AgentCore), distribution (Marketplace), and model customization (Nova). For IT professionals, these moves suggest a shift toward modular, controllable agentic systems designed to plug into existing workflows and cloud infrastructure.