Amazon details a new wave of AI “agents” at AWS Summit New York, pairing software automation with security and context

Amazon Web Services used AWS Summit New York to push a message that has shifted quickly across the industry: companies are moving from discussing AI agents to trying to operationalize them. In a presentation focused on customer rollout and implementation, AWS leadership described multiple new services and upgrades that, taken together, are intended to help organizations deploy agents for software development, customer support, and security work while reducing the risk that the systems act on incomplete or incorrect information.

Over the past six months, AWS said, there has been a “seismic shift” as businesses started placing agents into real workflows. The company said agents are now being used to build applications, secure systems, answer complex customer questions, and make certain decisions autonomously. AWS also argued that agents improve as they run more tasks, because additional interactions generate more context, which then leads to better outcomes and higher trust, encouraging companies to hand over more work to the technology.

A key theme was security in an “agentic world,” where the company described telemetry, context, reasoning, and actions as new pillars for defense. AWS said advances in specialized security models and how attackers and defenders use them have accelerated vulnerability discovery and exploitation. As a result, AWS argued, organizations need to respond at “machine speed,” using agents that continuously process information and take action within guardrails.

To address code vulnerability management, AWS announced AWS Continuum for code vulnerabilities, described as an AI-native security service covering the full lifecycle of managing vulnerabilities. AWS said it continuously discovers vulnerabilities, validates which ones are genuinely exploitable, prioritizes remediation based on business context, and helps teams remediate across the stack while respecting guardrails set by customers. The service was described as model-agnostic, with an ability to use different models where each performs best, and it will integrate new models over time.

AWS said Continuum for code vulnerabilities is designed to be auditable and explainable, with visibility into what it did and why, including what would happen if an action were rolled back. AWS also announced Continuum threat modeling, which automatically generates comprehensive threat models from design documents or source code and outputs results in an industry-standard format.

Complementing security, AWS presented AWS Context as a new service that automatically builds a knowledge graph from a company’s existing data so agents can access the context they need to perform. AWS said Context infers relationships among data assets, business rules, and domain knowledge, making them available to agents across an organization. The company argued that context can separate an agent’s early, unreliable decisions from later, more accurate ones, because agents can ground recommendations in relevant and trusted information sources rather than guessing.

AWS said Context is built on the same knowledge graph technology that powers Amazon Quick and is meant to function as an agentic search layer for organizational data, with governance so agents can only access information they are permitted to use. The company described storage of metadata in Iceberg format on Amazon S3 Tables and said there is no infrastructure to provision and no retrieval pipeline to build. AWS added that as agents use Context, it can learn which sources and paths produce correct results and which business rules matter, and that it can improve based on what is discovered during queries.

AWS also detailed enhancements around Amazon Quick, which it described as an enterprise AI assistant with security controls customers can approve. AWS said Quick’s new autonomous agents allow teams to create background agents with specific expertise, tone, and tool access, including examples such as finance agents that process orders and sales agents that monitor CRM, email, and messaging to draft follow-ups or recommend next steps. The company said Quick adds a tailored activity feed that consolidates email, messaging, calendar, and tasks into a single prioritized view, and it announced 16 new built-in integrations with companies including Adobe, Moody’s, and Snowflake.

In software development, AWS showcased Kiro, its software development agent, and said Kiro is now available on iOS devices. AWS described Kiro mobile as a way to kick off projects, monitor progress, steer an agent, and interact with a session while it runs in an always-on cloud environment, so a user can continue where the agent left off when returning to a laptop. AWS linked faster coding to a second challenge, deployment readiness, and said AWS DevOps Agent now expands release management support, including release readiness reviews, change-specific test plans, and actions aimed at catching regressions, user-experience issues, and integration failures before production.