Agentic AI Made Simple

Access the slide deck by downloading the pdf from the course platform; enrolled students will see a download link to reference the slides.

Explore the fundamentals of agentic AI with a mission briefing on when to build agents. Learn the anatomy of an AI agent and practical setup with Agent Forge and Bedrock.

Explain why agents matter and how they bridge planning and action by turning AI insights into execution, adapting plans and iterating when circumstances change.

Agentic ai agents pursue goals, decide autonomously, and adapt to changing environments. They analyze context, reason, and plan with tools, then execute across services and replan to deliver fast results.

Explore the progression from generative AI assistants to AI agents and agentic AI systems, highlighting autonomous, dynamic workflows and real-world use cases that mirror human reasoning.

Identify when to deploy an agent by weighing use cases and risks, avoiding simple tasks. Use agentic ai for complex decisions, time-sensitive contexts, personalized assistants, and multi-step automation.

Explore real-world agent use cases, from travel planning with real-time weather APIs to marketing and service workflows, highlighting planning, acting, evaluating, and learning with an LLM brain.

Compare Rag and Agentic AI workflows by explaining how Rag uses a vector database and context to fetch internal data, while Agentic AI uses tools and iterative LM-driven reasoning.

Use a decision tree to decide when to apply prompt engineering, fine tuning, Rag, or agents. Identify task type and data needs, from static to real-time, with Bedrock knowledge bases.

Demonstrate how an agent LM model operates within a workflow to control a Philips Hue light via chat, using an openai GPT 4.1 mini model and configurable tools.

Explore the core traits of an agent ai: goal oriented, autonomous, and adaptable. Learn how it plans tasks, uses tools, and stores memory to reason, act, and improve.

Compare reasoning models to generative models in AI agents and describe how they plan with reflection, self-critique, chain-of-thought, and task decomposition. Learn how tools and memory—short/long-term, episodic/semantic—support these processes.

Build and run an AI agent by assembling infrastructure (cloud or on-prem), a foundational or customized model, memory, security, observability, and orchestration with Bedrock and SageMaker options.

Explore three approaches to building an AI agent: AWS Bedrock, open-source DIY stacks, and Strands agent. Bedrock offers models, guardrails, memory, actions, and tooling for quick, reliable agent development.

Build a bedrock agent on aws by creating a lambda function, an action group with a time api schema, and deploying an alias using Claude three haiku.

Test the Bedrock agent by preparing and invoking the lambda-backed function, then examine traces to verify orchestration. Deploy an alias and resolve region and permission issues for external Python invocation.

Discover how to build and deploy a Bedrock agent on Amazon Bedrock, integrating knowledge bases, action groups, and orchestration with instructions, pre- and post-processing, and guardrails for reliable AI workflows.

Knowledge bases are independent components that exist outside the agent; associated with an agent, they use vector stores, structured data, or Amazon Kendra indexes from an S3 bucket.

Learn to build an Amazon Bedrock agent via notebooks or code, wiring a lambda to DynamoDB, and deploy with CloudFormation for a parent–teacher appointment assistant using a SageMaker domain.

Learn to deploy a bedrock agent in a SageMaker notebook, wire notebooks with iam roles, lambda, and streaming outputs to book 30-minute parent teacher appointments via natural language prompts.

Explore how multi-agent systems coordinate specialized agents to handle non-deterministic tasks through collaboration and replanning, versus single-agent deterministic tasks. Learn when parallel, robust, expert panel collaboration improves outcomes.

Explore multi-agent architecture through a supply chain example, showing how specialized agents including finance, logistics, weather, and demand planning collaborate via routing and orchestration to optimize decisions amid tariff increases.

Explore a multi-agent framework by building a Hello World demo with collaborator and supervisor agents. Deploy the Python main.py script and observe the request flow between subagents, collaborator, and supervisor.

Explore supervisor vs supervisor with routing in a multi-agent app, comparing delegation and coordination, routing logic, and when to apply each mode to optimize latency, cost, and scalability.

Explore how changing routing configurations in a supervisor–subagent setup optimizes performance, achieving latency reductions from 9.56s to 2.31s through delegation-first routing and versioned deployments.

See how bedrock agent powers a mortgage assistant with multi-agent collaboration, delivering instant mortgage status, rate comparisons, and refinancing guidance through a Streamlit front end.

Discover how to build a travel planning agent that orchestrates weather and flights agents using a CloudFormation template, enabling weather forecasts, flight search, pricing, and bookings.

Introduce building a flight search agent, a weather agent, and a travel agent using bedrock models, Lambda action groups, and CloudFormation deployments, with testing and aliases for multi-agent workflows.

Showcases building a travel planning agent with multi-agent collaboration, configuring flight and weather agents, and wiring them through Bedrock, Lambda, and front-end to orchestrate seamless trips.

Trace how the ai age shifts toward an agent economy, cloud software as a service, app and creator economies. Explore agentic ai using aws services, bedrock, memory, observability, and orchestration.