AWS Summit NYC: Bedrock AgentCore Adds Web Search

At AWS Summit New York on June 17, 2026, Amazon announced a package of agentic‑AI upgrades aimed at making enterprise agents more grounded and production‑ready. The company unveiled a managed web‑search connector for Amazon Bedrock AgentCore and new release‑management features for its AWS DevOps Agent.

The Bedrock AgentCore update introduces a fully managed Web Search tool that lets agents issue queries and return cited web sources as part of their responses. Amazon says the capability is designed so agents can ground answers in current web content without developers wiring up third‑party search integrations.

Crucially for regulated customers, AWS says the search queries and results are processed inside AWS infrastructure to eliminate data egress from a customer’s secured environment. That design aims to simplify compliance and auditing when agents consult external web pages.

On the operations side, AWS announced that DevOps Agent now includes a release‑readiness review and autonomous release testing. The feature generates change‑specific test plans and runs them in customer‑provisioned, production‑like environments before code merges. AWS published the preview details during the Summit announcement.

Amazon framed the package as a push to make agentic AI safer and easier to integrate into enterprise workflows, emphasizing grounding, governance, and observability as core goals of the updates. Company posts and the Summit keynote framed the moves as steps toward production‑grade agents.

Practically, managed web search changes the way teams handle agent hallucination and traceability. Allowing agents to cite current sources makes outputs easier to verify, and a cloud‑side search path reduces the need to export sensitive context to external services. That should speed audits and reduce integration overhead.

The DevOps Agent enhancements aim to close a common gap between prototype agents and production platforms by embedding test and review steps into the agent workflow. AWS describes the capability as enforcing natural‑language standards and running UI and API tests that are specific to each proposed change.

These announcements sit alongside broader AgentCore work: a managed harness for agent loops, the AgentCore CLI, a managed knowledge base for unstructured data, and other lifecycle tools that AWS has been rolling out since AgentCore launched. Together, AWS says these pieces let teams build, govern, and tune agents at scale.

The push comes as security researchers and enterprise teams raise questions about agent attack surfaces, tool integrations, and credential handling. Recent analyses argue that agent platforms can increase risk unless access, identity, and tool calls are tightly controlled. AWS acknowledges these risks and positions its Gateway, Identity, and logging features as guardrails.

AgentCore’s Gateway and related runtime tools — including a cloud‑side browser runtime and credential providers — are intended to centralize and log agent tool activity so security teams can audit and limit what agents can do. Those controls matter especially where agents can reach APIs, payment flows, or live systems.

Observers at the Summit and in the press noted the practical DevSecOps angle: grounding plus automated release testing treats agents like first‑class components of CI/CD rather than side projects. Analysts said the feature set is meant to reduce incident risk and speed the path from prototype to production.

Availability is mixed: AWS described several features as preview or gated, and the Web Search tool is presented as a managed connector within AgentCore rather than an immediately global GA rollout. Teams evaluating the moves should plan for preview timelines, region availability, and integration testing before wide‑scale adoption.