Welcome to Building AI Agents, the eighth module of the Generative AI & LLM Engineering course — where the model stops waiting for instructions and starts deciding for itself. In Module 3 you gave a model tools and ran a single request-execute-respond round-trip. In Module 7 you controlled the steps: retrieve, then generate. An agent removes that hand-holding. You give the model a goal and a set of tools, then let it run in a loop: it thinks about what to do, calls a tool, observes the result, and decides the next step — repeating until the task is done. The control flow comes from the model, not from you.

You’ll start with what actually makes something an “agent” versus a fixed workflow, then build the agent loop by hand — the think-act-observe cycle that drives every agent. You’ll give your agent memory and a retrieval tool (turning the RAG system from Module 7 into just one of the agent’s tools), handle planning and multi-step tasks where the agent chains several tool calls toward a goal, and finally build a research-assistant agent that combines multiple tools to answer real questions.

Every example runs for real against the Claude API on the affordable claude-haiku-4-5 model — you’ll watch the model choose tools, react to results, and decide when it’s finished. By the end you’ll understand the architecture behind coding agents, research assistants, and autonomous workflows, and you’ll be ready to build them on frameworks like LangGraph in the next module.

Start with Lesson 1, where you’ll learn what separates a true agent from a scripted pipeline.