AWS Ships AgentCore GA and Agentic Knowledge Tools at NYC Summit

On June 17 at AWS Summit New York, Swami Sivasubramanian, AWS VP of Agentic AI, used the keynote to announce that the AgentCore harness is now generally available, letting teams build "production-grade AI agents in minutes" without hand-coding orchestration loops. The same event introduced a cluster of supporting capabilities that position AWS as treating agentic AI as a production infrastructure problem rather than a research prototype.

The two new capabilities worth tracking are the Managed Knowledge Base and Web Search integration for AgentCore. The Managed Knowledge Base ships with native data connectors and a Smart Parsing layer for mixed document formats, plus an Agentic Retriever for complex multi-step queries against enterprise data stores. Web Search grounds agent responses in what AWS describes as "current, cited web knowledge" without requiring organizations to manage their own search pipelines or accept data leaving their environment. The citation piece matters more than the search piece: surfacing where information came from is the answer to the auditability question that has slowed enterprise RAG adoption more than any capability gap has.

AWS also used the summit to apply AI to its own security workflow. AWS Continuum, entering gated preview, prioritizes code vulnerabilities by business impact before surfacing them, aiming to cut through the CVE-queue noise that buries most security teams. AWS Security Agent adds STRIDE-based threat modeling and can scan pull requests across Git platforms with built-in remediation suggestions.

What the AWS blog post does not give you is pricing for any of these services, data residency guarantees that regulated-industry buyers will require before signing off, or a technical breakdown of how the Agentic Retriever compares to graph-based or hybrid search approaches teams may already run. The honest caveat is that a GA label tells you the API is stable, not that the pricing or SLAs are final.

The clearest near-term beneficiaries are engineering teams currently running bespoke S3-backed RAG pipelines who now have a fully managed alternative to benchmark against. AWS also flagged that AWS Context, described as coming soon, will automatically map data relationships into a knowledge graph, suggesting the retrieval architecture is still evolving toward something more structured than vector similarity alone.