Quick Facts
- AWS VP Swami Sivasubramanian keynoted AWS Summit New York City on June 17, arguing that AI agents acting autonomously have replaced prompt-based tools as the enterprise standard.
- AWS launched AgentCore, a managed runtime that reduces agent deployment from months to weeks, with task volume growing 15x over six months.
- New security service AWS Continuum targets AI-accelerated vulnerability chaining, moving from human-in-the-loop to automated remediation as confidence grows.
AWS used its sold-out Summit in New York City to make a direct argument to enterprise leaders: the gap between companies that deploy AI agents and those that do not is widening, and it will compound over time.
Swami Sivasubramanian, AWS VP of Agentic AI, delivered a 90-minute keynote at the Jacob K. Javits Convention Center on June 17. He was joined by Chet Kapoor, VP of Security Services and Observability. The event was available globally via free livestream. AWS enters 2026 at a $142 billion annualized run rate, with agentic AI driving virtually every major service announcement this year.
Beyond the Chat Window
Sivasubramanian’s central claim was that the first generation of AI tools never broke out of what he called chat-window gravity. Responding to prompts is not the same as taking action. The keynote focused on what AWS is building to close that gap.
“Over the last six months, there has been a seismic shift as companies have started to move from talking about agents to putting them to work,” Sivasubramanian said. He argued that agent use creates compounding returns: more interactions build more context, which builds more trust, which leads to more work handed off to agents.
AgentCore Targets the Prototype-to-Production Gap
The flagship announcement was Amazon AgentCore, a managed platform that handles authentication, memory, tool access, security, and governance for AI agents. Sivasubramanian said most teams are stuck rebuilding these basics from scratch, which stalls deployment.
AgentCore works with any agent framework and any model. The platform includes a managed runtime, built-in identity, session memory, observability, evaluations, and access controls. AWS described the underlying model as the brain and AgentCore as the body that provides state persistence, error recovery, context management, and session isolation.
The company said task volume on AgentCore has grown 15x in the past six months. Customers including PGA TOUR, Nasdaq, Visa, and Experian are now running production agents. The PGA TOUR reports writing tournament coverage 10 times faster. AWS said AgentCore Harness can convert a model into a deployable agent in minutes with three API calls, and the company compared its potential to what AWS Lambda did for serverless compute.
A New Security Threat Drives a New Product
AWS also announced AWS Continuum, a security agent for code vulnerabilities, currently in gated preview. The product scans your environment, prioritizes findings by business impact, confirms exploitability, and drives remediation through existing workflows.
The prompt for building it was specific. Neha Rungta, AWS director of applied science, said AI can now chain together minor flaws, combining two medium-severity findings and one low-severity finding into a critical exploit. “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,” Rungta said. “The goal is to raise that floor up again.”
Kapoor tied the threat directly to Anthropic’s Claude Mythos model, which he said accelerated AWS’s security roadmap. “I call it the Mythos moment. It accelerated our plans significantly. Mythos set a new bar for finding vulnerabilities,” Kapoor said. Continuum launches in learn mode with human review required. Teams can graduate it to enforce mode as confidence builds, enabling automated remediation based on risk categories they define.
Context as Infrastructure
AWS introduced a fourth product called AWS Context, a service that builds a knowledge graph from existing organizational data. It reads databases, documents, Slack history, and email, then infers connections between data and exposes that map to every agent at runtime.
Sivasubramanian said agents fail not because of limited intelligence but because of limited context. He cited Amazon’s internal deployment of a semantic knowledge store behind the Q product, which now processes over 1.8 million requests per day.
Southwest Airlines as the Production Reference
Lauren Woods, Executive Vice President and CIO of Southwest Airlines, appeared during the keynote to describe the carrier’s move from on-premises infrastructure to a cloud-based, AI-driven environment built on AWS. Southwest served as AWS’s clearest evidence that its agentic stack is running in enterprise production, not in controlled pilots.
For IT leaders evaluating timelines, the Southwest case makes the stakes concrete. AWS is no longer pitching a roadmap. It is presenting paying customers running agents in live operations today.
Read more: Five thoughts from Swami Sivasubramanian’s keynote at AWS Summit and what it means for IT pros