Building Advanced Agentic Systems on AWS Training

When the work spans distinct domains that each need their own tools, context, and reasoning, or when tasks must run in parallel, or when one agent's context window cannot hold everything the job requires. Multi-agent design brings real costs in coordination, latency, and debugging difficulty, so the first module deliberately covers single-agent limitations before the patterns, because knowing when not to use this architecture matters as much as knowing how.

A managed capability within Amazon Bedrock for building, operating, and scaling production AI agents. It handles agent orchestration, policy-based access control, evaluation, and secure deployment, and integrates with IAM, VPC networking, and observability tooling. It is the deployment target for most of what this course covers.

An AWS SDK for building agents in code. In this course, Strands is what you build with and AgentCore is what you deploy and operate with. The instructor demonstrations use both together, starting with a single Strands agent and finishing with a production AgentCore deployment.

That course is three days and covers the whole generative AI production stack: foundation model selection, RAG, vector databases, prompt governance, safety, and enterprise integration, with agents as one component. This course is one day and goes deep on multi-agent systems specifically, including communication patterns, shared memory, context engineering, and multi-agent observability. They complement each other rather than overlap heavily.

Yes. AWS names two prior courses, and this is a single advanced day that assumes you already know what an agent is, how AgentCore works, and how to build one. Attendees arriving without that background typically spend the day catching up rather than learning the multi-agent material.

Yes, and it suits private delivery particularly well. The IAM, VPC, observability, and governance content can be discussed against your organization's actual security standards and monitoring stack, which is more useful than working through generic examples.