Advanced Agentic AI with Amazon Bedrock AgentCore

Course Description

This advanced course explores the next generation of AI systems: autonomous and agentic AI applications powered by Amazon Bedrock AgentCore. Learners will understand the core building blocks of intelligent agents, including reasoning, tool usage, perception, and memory.

The course introduces Amazon Bedrock AgentCore as a modular, serverless infrastructure for deploying scalable AI agents. Learners will integrate popular orchestration frameworks such as LangGraph and CrewAI while deploying long-running reasoning tasks within isolated runtime environments.

Security and governance are covered extensively through AgentCore Identity, OAuth integrations, JWT-based workload identities, and Cedar-powered policy enforcement. Learners will also implement secure tool access through MCP and AgentCore Gateway, enabling AI agents to interact safely with APIs, databases, and enterprise systems.

A major focus is placed on memory architectures, including persistent memory, cross-agent knowledge sharing, and encrypted context storage. The course also introduces production observability and evaluation systems for tracing workflows, measuring correctness and safety, and optimizing cost-performance tradeoffs.

Hands-on labs culminate in a capstone project where learners build a fully monitored enterprise operations agent capable of secure authentication, policy-compliant API access, and persistent memory management within a production-ready observability stack.

Who is this course for

• AI engineers building autonomous AI systems

• Cloud architects designing enterprise AI infrastructure

• Backend developers integrating AI agents with APIs and enterprise systems

• Platform engineers managing secure AI runtimes

• Advanced GenAI practitioners exploring multi-agent orchestration

Course Objectives

By the end of this course, learners will be able to:

• Understand advanced agentic AI architectures and enterprise-grade autonomous system design patterns

• Deploy, manage, and orchestrate AI agents using Amazon Bedrock AgentCore, LangGraph, and CrewAI

• Implement secure agent systems using OAuth, JWT, workload identities, MCP, and AgentCore Gateway integrations

• Design persistent and shared memory architectures with observability, tracing, and workflow monitoring capabilities

• Evaluate and govern AI agents based on correctness, safety, latency, token usage, and enterprise compliance controls

Prerequisites

• Strong understanding of AWS services and cloud architecture

• Experience with Python programming

• Familiarity with APIs, authentication mechanisms, and OAuth

• Working knowledge of generative AI and LLM concepts

• Prior exposure to LangChain, LangGraph, or AI agents is recommended

Course outline

Section 1: Foundations of Agentic AI Patterns

1. Agent Building Blocks: Reasoning (LLM), Action (Tools), and Perception (Multi-modality).

2. Introduction to AgentCore: Why modular, serverless infrastructure is the future of agents.

3. Lab 1: Architecting the Agent. Designing a multi-agent workflow on paper and mapping it to AgentCore services (Runtime vs. Gateway).

Section 2: AgentCore Runtime & Framework Integration

1. Framework Agnostic Design: Integrating LangGraph, CrewAI, and Strands Agents.

2. AgentCore Runtime: Deep dive into isolated Micro-VM sessions and long-running reasoning tasks.

3. Deployment: Moving agent code from local Python environments to AWS serverless infra.

4. Lab 2: Deploying your First Agent. Take a local LangGraph agent and deploy it to the AgentCore Runtime using the AgentCore SDK.

Section 3: Security & Identity Management

1. Inbound vs. Outbound Auth: Securing the caller vs. securing the tool access.

2. AgentCore Identity: Using the Token Vault for OAuth 2.0 (GitHub, Slack, Salesforce) and SigV4.

3. Workload Identities: Mapping agent roles to specific user permissions via JWT.

4. Lab 3: The Secure Identity Vault. Configure an agent to securely post to a Slack channel using AgentCore Identity without hardcoding API keys.

Section 4: Tool Integration & AgentCore Gateway

1. Model Context Protocol (MCP): Implementing the open standard for universal tool discovery.

2. AgentCore Gateway: Turning Lambda and OpenAPIs into secure agent tools.

3. AgentCore Policy: Authoring Cedar policies (via natural language or code) to enforce fine-grained access control.

4. Lab 4: Building a Secure Gateway. Create an MCP server for a mock Database and use AgentCore Policy to "DENY" the agent from deleting records, even if the LLM tries.

Section 5: Agentic Memory Implementation

1. Short-term vs. Long-term: Managing conversation threads vs. persistent user preferences.

2. Memory Core: Cross-agent memory sharing and high-accuracy context retrieval.

3. Security: Field-level encryption for sensitive user data within memory stores.

4. Lab 5: The Persistent Assistant. Enhance an agent with AgentCore Memory so it can "remember" a user's previous technical issues across different sessions and days.

Section 6: Production Monitoring & Observability

1. AgentCore Observability: Tracing workflow steps and visualizing decision trees in CloudWatch.

2. AgentCore Evaluation: Automated verification of "Correctness" and "Safety" using the evaluation framework.

3. Cost & Performance: Tracking token usage vs. latency in multi-agent loops.

4. Lab 6: The Observability Dashboard. Set up a real-time monitoring dashboard to trace a failed agent reasoning loop and identify the specific tool that caused the bottleneck.

Section 7: Capstone Project: The Enterprise Operations Agent

Build an agent that uses AgentCore Gateway to connect to an internal API, AgentCore Identity to authenticate, and AgentCore Policy to ensure it stays within corporate compliance boundaries—all monitored via a centralized Observability stack.