AWS has released a new blueprint for agentic retrieval on Amazon Bedrock. It uses a reasoning agent that routes questions to multiple knowledge bases.
The system ships with built-in observability and evaluation.
This helps enterprises deploy AI systems that are reliable and transparent.
What happened
Amazon Web Services (AWS) has introduced a solution for building enterprise agentic retrieval on Amazon Bedrock Managed Knowledge Base, which includes an agent that reasons over questions, routes to the correct knowledge base, and returns cited answers. The solution is deployed via a chain of four AWS CloudFormation stacks.
Why it matters
Unlike basic RAG (Retrieval Augmented Generation, where a model fetches text to answer a question), this agentic approach handles complex questions spanning multiple sources by deciding which knowledge base to consult. However, this introduces the challenge of observing what the agent did and whether the answer is good, which the solution addresses with two CloudWatch dashboards spanning seven telemetry layers and both on-demand and continuous evaluation.
What to watch
The solution includes two sample corpora—a financial report (Octank Financial 10-K) and a weather report (U.S. Congressional Research Service on tornadoes)—to demonstrate routing between knowledge bases. It is built with Amazon Bedrock AgentCore and the AgentCore Gateway, and requires no custom Lambda for retrieval, as the gateway uses the native bedrock-knowledge-bases connector.
Ask the AI about this article →
This move by AWS directly addresses a pain point for enterprises moving from simple RAG prototypes to production AI systems. The previous post automated single-shot RAG, but as teams scale to multi-source questions, the need to see and verify an agent's reasoning becomes critical. By baking in OpenTelemetry spans and evaluation dashboards from the start, AWS is positioning Managed Knowledge Base as the enterprise-grade option, especially when compared to the DIY approach that requires managing your own vector database and lacks agentic capabilities.
The two-level routing design is notable: the agent's reasoning model selects the right knowledge base, and then AgenticRetrieveStream handles sub-query decomposition and iterative retrieval within that base. This separation keeps per-knowledge-base signals clean, which is essential for troubleshooting and cost tracking. The use of two distinct corpora (financial and weather) in the sample illustrates this routing in action. For businesses, the key takeaway is that this solution is deployable as a single CloudFormation chain, making it reproducible and consistent, which is a prerequisite for production workloads.
For example, today's edition would include:
AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.
Free · takes 30 seconds · unsubscribe anytimeWhat is AIToday? →
Ask AI anything about this article. Q&As are published on this page for other readers too.
Chinese large-model developer Z.ai says it can now support large-scale inference using roughly 100,000 domesti…
AI system scaling has pushed interconnect requirements inside data centers from chips and boards up to racks…
Analyst Ming-Chi Kuo says Nvidia has revived the Rubin CPX AI accelerator with a substantially redesigned arch…
Palantir Technologies stock has posted multi-year gains, including an 11x return over 3 years
Apple has escalated its legal battle against OpenAI, claiming in a new court filing that OpenAI is actively de…
Samsung Electronics has locked up as much as 70% of its memory production capacity under long-term supply agre…
