AWS bets the agent stack on production-readiness with Continuum, Context, and AgentCore Harness GA

AWS used its sold-out New York summit on June 17 to reposition its agent stack from a demo-stage curiosity into something it wants enterprises to actually run in production, announcing a security service called Continuum, a company-wide knowledge graph called Context, and general availability of the Bedrock AgentCore harness.

Swami Sivasubramanian, AWS VP of agentic AI, anchored the keynote on a single number: agent tasks running on AgentCore have grown 15x in the past six months. The customer roll call (Nasdaq, Visa, Experian, PGA Tour, the last of which is now writing tournament coverage 10x faster) is the kind of name-checking that signals AWS has moved past the proof-of-concept phase and into the part of the cycle where it needs to defend the perimeter it just built.

That’s where Continuum comes in. The gated-preview service triages code-vulnerability findings, ranks them by business impact, validates exploitability in an isolated sandbox, and proposes remediations. It starts in a supervised “learn mode” and earns autonomy category by category as customers grant permission. Chet Kapoor, AWS VP of security services and observability, framed it as a direct response to Anthropic’s Claude Mythos and Claude Fable models: “I call it the Mythos moment. It accelerated our plans significantly. Mythos set a new bar for finding vulnerabilities.”

Neha Rungta, the AWS director of applied science who led the work, was blunter about why this matters now. Attackers, she said, can chain two medium-severity findings and a low one into something critical. “That was something that would have taken a lot of effort, expertise, and determination for an attacker to get through, so the floor has been lowered. The goal is to raise that floor up again.”

AWS Context, announced as coming soon, attacks the other gap. It automatically maps relationships across a customer’s existing data into a knowledge graph and exposes agentic search over governed data relationships, business rules, and domain knowledge at runtime. It’s the unglamorous plumbing that determines whether an agent’s outputs are usefully grounded or hallucinated nonsense dressed in corporate vocabulary.

The rest of the keynote filled in the harness. AgentCore’s GA shipped with a managed Knowledge Base (S3, SharePoint, Confluence, Google Drive connectors), a managed Web Search tool operating natively inside the customer’s AWS environment with zero data egress, and per-object S3 annotations of up to 1 GB. Kiro, the AWS coding agent, picked up a gated-preview iOS app for mobile monitoring. Amazon Quick, DevOps Agent, and AWS Transform got stage time. EC2 G7 instances launched as the first major-cloud offering with NVIDIA RTX PRO 4500 Blackwell Server Edition GPUs.

The structural read is that the agent platform race has entered its boring, decisive phase: governance, security, grounding, and the long tail of connector work. Anthropic set the bar that forced Continuum’s timeline. AWS, in characteristic fashion, answered with infrastructure.

Sources

• https://aws.amazon.com/blogs/aws/top-announcements-of-the-aws-summit-in-new-york-2026/
• https://www.aboutamazon.com/news/aws/aws-summit-nyc-2026-ai-agents
• https://www.geekwire.com/2026/amazon-unveils-new-ai-agents-trying-to-thread-the-needle-between-autonomy-and-human-control/
• https://www.ciodive.com/news/AWS-continuum-AI-security-claude-mythos/823184/
• https://the-decoder.com/aws-says-ai-agents-lack-business-context-and-security-launches-two-services-to-patch-the-gaps/