AI agent software spending is on track to hit $206.5 billion in 2026, up 139% from $86.4 billion in 2025. That number alone signals something fundamental has shifted. But numbers only tell part of the story. At AWS Summit New York on June 17, 2026, Amazon made its most direct statement yet about where the cloud is heading: every serious business will need a managed platform to build, govern, and operate AI agents at scale, and Amazon Bedrock AgentCore is that platform.

This was not a roadmap tease. The AWS Summit brought generally available releases, new security services, and a knowledge graph layer that together form the most complete enterprise-grade agentic infrastructure stack any cloud provider has shipped. For developers, IT leaders, and business teams evaluating how to get AI agents into production, what AWS announced last week deserves close attention.

In this article, we break down the four biggest announcements from AWS Summit New York 2026 and what each one means for your AI agent strategy.

AgentCore Harness Goes GA: From Idea to Production in Minutes

Until June 2026, getting an AI agent from prototype to production meant stitching together your own identity management, memory storage, tool access, observability, and security policies. Amazon Bedrock AgentCore changes that. At AWS Summit New York, AWS announced the general availability of the managed agent harness inside Amazon Bedrock AgentCore, which compresses that setup into two API calls.

CreateHarness defines the agent. InvokeHarness runs it. In between, AWS handles managed identity, persistent memory, tool gateway access with enforced security policies, organizational knowledge integration, web search, and end-to-end observability. Teams that previously spent weeks on infrastructure scaffolding can now reach a production-ready agent in minutes.

Also reaching general availability are AgentCore Evaluations and AgentCore Policy. Evaluations lets teams continuously assess agent quality against production traffic and validate changes before deployment. Policy gives organizations centralized, fine-grained controls over what agents can access and what actions they are permitted to take. These two additions address two of the most commonly cited blockers to enterprise agentic deployment: quality assurance and auditability. For teams already familiar with AI agent frameworks such as LangGraph and CrewAI, the harness is designed to work alongside your existing orchestration layer rather than replace it.

AWS Context and Continuum: Security and Knowledge for Agentic AI

One striking pattern from early enterprise AI agent deployments: teams often ship agents before the architecture needed to govern them is in place. AWS acknowledged this directly at the Summit with two new services, AWS Continuum and AWS Context.

AWS Context is a real-time enterprise knowledge graph that gives agents structured access to organizational data across AWS services, third-party platforms, and on-premise systems. Rather than crawling documents or relying on static retrieval, Context maps relationships between data so agents understand where to find the right information and which steps to take next. This is critical for enterprise agents that need to traverse complex internal knowledge without hallucinating or stalling.

AWS Continuum tackles the security side. It is an AI-native service that continuously discovers, prioritizes, validates, and remediates security vulnerabilities across your environment. What makes Continuum genuinely novel is its validation phase: the service constructs a sandboxed working exploit to prove a vulnerability is real before proposing any fix. It begins in a supervised “learn mode” and earns expanded permissions only as organizations grant them, category by category. This graduated autonomy model directly mirrors what governance experts recommend for AI agent deployment, a strategy covered in depth in our analysis of AI agent governance in 2026.

How to Deploy AI Agents to Production With AWS in 2026

If you are evaluating how to deploy AI agents to production with AWS in 2026, the AgentCore stack now gives a clear path. Here is the practical framework the Summit announcements enable.

Start with the harness. Use CreateHarness to define your agent’s identity, tool permissions, memory scope, and knowledge access. This replaces what previously required custom code for each of these components.

Layer in Context. Connect AWS Context to your organizational data so agents can ground responses in current, real internal knowledge rather than generic pretrained information. This is particularly valuable for agents handling customer inquiries, procurement workflows, or compliance tasks where stale data creates risk.

Add governance early. Use AgentCore Policy to define what tools agents can invoke and under what conditions. Pair this with AgentCore Evaluations to continuously validate agent quality in production. The cost of retrofitting governance into a deployed agent is significantly higher than building it in at the start. For teams just beginning to explore enterprise AI agent platforms, this principle holds regardless of the vendor you choose.

Monitor and iterate. AgentCore’s observability layer surfaces agent decision paths, tool call logs, and failure modes in a format your SRE and compliance teams can act on.

Amazon Quick and Kiro Signal What Comes Next

Two additional AWS announcements from the Summit reveal where agentic AI is heading beyond the infrastructure layer.

Amazon Quick is AWS’s agentic assistant for knowledge workers. At the Summit, new Quick capabilities let users create and run multi-step agents directly from the desktop app, consolidating email, Slack, calendar, and tasks into a single prioritized view with personalized automation rules. This positions Quick as a direct rival to Microsoft 365 Copilot and Google Workspace’s agentic features, competing for the attention of professionals who spend their days inside communication and productivity tools.

Kiro, AWS’s agentic IDE that launched internationally on May 7, 2026, added a native iOS app called Kiro Mobile. Developers can now kick off new projects, monitor agent progress, and steer coding sessions from their phones. Kiro’s core architectural advantage is that it generates a formal specification before writing a single line of code, using EARS notation to produce structured requirements, design, and task documents that guide the agent through implementation. Kiro Pro Max, a new premium tier, expands on these capabilities for teams with more demanding workflows.

Together, Quick and Kiro signal AWS’s vision: agentic AI is not just a backend infrastructure play. It is the new interface layer for both technical and non-technical workers. For full technical details on what was announced, the official AWS blog post on AgentCore Harness GA is the definitive resource.

The Enterprise Agentic Stack Is Now Ready to Ship

Three things stand out from AWS Summit New York 2026. First, managed infrastructure for AI agents is now a commodity service: AgentCore’s harness removes the biggest barrier to production deployment. Second, security and knowledge architecture are no longer afterthoughts: Continuum and Context address the governance gap that plagued first-generation enterprise agent deployments. Third, agentic AI is expanding from servers to screens: Amazon Quick and Kiro bring autonomous agents into the daily workflows of every knowledge worker and developer.

The enterprise agentic era is no longer theoretical. It is infrastructure you can provision today. According to GeekWire’s coverage of the Summit, AWS is threading a careful needle between autonomy and human control, and the AgentCore stack is the technical expression of that balance.

What does your organization need to solve before you can put AI agents into production? Explore tools, strategies, and the latest analysis at BigAIAgent.tech, and join the conversation in the comments below.