AWS just dropped its most ambitious play yet for agentic AI. It’s called Amazon Bedrock AgentCore, and it’s built to help businesses stop tinkering with AI agents and actually deploy them at scale.
Key Points:
- AgentCore bundles seven core services to deploy secure, production-ready AI agents.
- It works with any model or framework, including open-source tools like CrewAI.
- AWS is also putting another $100M into agentic AI development.
At the AWS Summit in New York, Swami Sivasubramanian, VP of Agentic AI at AWS, laid out a clear message: AI agents are about to change how we build software, and AWS wants to be the default platform to build them. AgentCore is the centerpiece of that vision.
AgentCore isn’t just a dev tool—it’s a full-stack suite for running enterprise-grade AI agents. That includes a secure runtime with the longest-supported session duration in the industry (up to 8 hours), built-in memory infrastructure, observability tools, identity management, and even a browser interface so agents can interact with the web.
It’s designed to plug into whatever stack you’re using. AgentCore works with open-source frameworks like LangGraph and CrewAI, supports any foundation model inside or outside of Bedrock, and connects to identity providers like Okta or Microsoft Entra. In other words, AWS is betting that openness plus security is what will win over cautious enterprises.
There’s also a big push to simplify deployment. The new AI Agents and Tools section in AWS Marketplace makes it easier to buy and spin up prebuilt agents from partners. Meanwhile, the new S3 Vectors feature slashes vector storage costs by 90%, making retrieval-augmented generation and long-term memory more feasible at scale.
AWS is backing all of this with another $100 million investment in its Generative AI Innovation Center, bringing the total to $200M. Companies like BMW, Warner Bros. Discovery, and Syngenta are already building with Bedrock. Now, with AgentCore, AWS wants to take those pilot projects and push them into full production.
This isn’t just about giving agents better tools—it’s about rethinking what AI software looks like in practice. As Sivasubramanian put it, “It upends the way software is built.” And AWS clearly plans to be the preferred platform where that future gets built.