Intro to AI Agents

Module 1

How LLMs Actually Work

Before you can build or reason about an agent, you need an accurate mental model of the engine inside it: the large language model. This module covers tokenization, next-token prediction, context windows, embeddings, and why hallucination is not a bug you can simply patch out — it is a direct consequence of how these models work.

• Tokens and Tokenization25 min
• Next-Token Prediction and the Context Window30 min
• Embeddings and Why LLMs Hallucinate30 min
• Module 1 Quiz: How LLMs Actually WorkQuiz

Module 2

What Makes an "Agent" an Agent

The word 'agent' gets used loosely in marketing, which makes it hard to reason precisely about what you are actually building or evaluating. This module gives you a rigorous way to distinguish a chatbot, a fixed automation, and a true agent, introduces the perceive-plan-act-observe loop that defines agentic behavior, and grounds the concepts in real, current examples.

• Chatbot vs. Workflow vs. Agent25 min
• The Perceive-Plan-Act-Observe Loop25 min
• The Agency Spectrum and Real-World Examples25 min
• Module 2 Quiz: What Makes an "Agent" an AgentQuiz

Module 3

Prompt Engineering for Agents

Prompts are how you program an LLM-based agent's behavior without writing traditional code. This module covers the practical craft of doing that reliably: the different roles of system and user prompts, few-shot examples, chain-of-thought prompting, getting structured output you can actually parse, and diagnosing the most common ways prompts fail.

• System Prompts vs. User Prompts25 min
• Few-Shot Examples and Chain-of-Thought Prompting25 min
• Reliable Structured Output and Common Failure Modes25 min
• Module 3 Quiz: Prompt Engineering for AgentsQuiz

Module 4

Tools & Function Calling

This module demystifies the actual mechanism behind an agent 'using a tool' — a topic often left vague in beginner explanations. You will learn precisely how function calling works under the hood, walk through the ReAct pattern step by step with a worked example, and learn what makes a tool description effective versus one that leads the model astray.

• How Function Calling Actually Works30 min
• The ReAct Pattern: A Worked Example30 min
• Designing Good Tool Descriptions25 min
• Module 4 Quiz: Tools & Function CallingQuiz

Module 5

Memory & Context Management

Agents that run for many steps or need to recall information beyond a single conversation face a real engineering constraint: the context window from Module 1. This module covers short-term versus long-term memory, why context limits force real design decisions, retrieval-augmented generation (RAG) as a beginner-friendly concept, and summarization as a practical memory strategy.

• Short-Term vs. Long-Term Memory25 min
• Retrieval-Augmented Generation: Giving an Agent Searchable Memory30 min
• Summarization as a Memory Strategy20 min
• Module 5 Quiz: Memory & Context ManagementQuiz

Module 6

Building Your First Agent

This module brings together everything from the previous five — tokens and context, the agent loop, prompting, tools, and memory — into the concrete design of a real agent, described in full architectural detail. It closes with the safety practices every real agent deployment needs and a practical readiness checklist.

• Designing a Research-Assistant Agent, End to End35 min
• Agent Safety Basics: Sandboxing, Rate Limits, and Human-in-the-Loop30 min
• Is My Agent Ready? A Practical Checklist20 min
• Module 6 Quiz: Building Your First AgentQuiz