AWS adds Web Search and Managed Knowledge Base to Bedrock AgentCore
AWS used its Summit to expand the Bedrock AgentCore platform, announcing the general availability of Amazon Bedrock Managed Knowledge Base and introducing Web Search on AgentCore. The Managed Knowledge Base ships with native data connectors, 'Smart Parsing' for complex documents, and an agentic retriever that reasons about what to fetch — targeting the retrieval-augmented-generation workflows enterprises need for grounded, accurate agents. Web Search on AgentCore lets agents pull current, cited web information from within secured AWS environments, closing a gap where enterprise agents were limited to stale internal data.
AWS's Swami Sivasubramanian anchored the announcements in adoption metrics, highlighting a 15x increase in tasks completed by Bedrock AgentCore over six months — a growth figure meant to show agentic AI moving from experimentation to production scale on AWS. That claim dovetails with Amazon's earnings disclosure that Bedrock consumption in Q2 exceeded all prior quarters combined.
The strategic thrust is completeness: by bundling managed retrieval, web search, and observability (AWS separately detailed AgentCore Observability with CloudWatch for diagnosing agent bottlenecks and memory issues) into one platform inside its security perimeter, AWS aims to be the default place enterprises build agents. The secured-environment framing directly addresses the week's dominant anxiety — autonomous agents escaping sandboxes — by keeping web access controlled and audited.
Competitively, this pits AgentCore against Microsoft's Copilot/agent stack and Google's Vertex agents, with AWS differentiating on its model-agnostic Bedrock catalog and enterprise trust. The skeptical read is that 'tasks completed' is a soft metric, and real proof is production reliability. Readers should watch whether Managed Knowledge Base's Smart Parsing meaningfully improves grounding accuracy versus existing RAG tooling, and how Web Search citations hold up for compliance-sensitive workloads.
