Course details
AI Agents
Learn how an AI agent really works: a language model in a loop that plans, calls tools, reads the results and decides again. You will follow one travel-planning agent from tokens and prompts to tools, memory, planning patterns, testing and guardrails.
Course curriculum
How LLMs Read and Charge3 lessons
Before building an agent, see the model the way it sees itself: text cut into tokens, a context window that holds everything it knows for one call, and a price on every token in and out.
The Agent Loop5 lessons
Meet the loop at the heart of every agent: what makes something an agent, how it perceives and plans, how tools are invoked, how results feed back, when the loop stops, and which sampling settings suit each job.
Prompt Engineering3 lessons
Write prompts that leave nothing to guess, teach by example and improve them with a test set, then use chain of thought and tree of thought where the problem needs real reasoning.
Tools and the Model Context Protocol4 lessons
Write tool definitions a model can use correctly, sort tools by the damage a mistake could do, and connect tools through MCP hosts, clients and servers, running locally or remotely.
Agent Memory4 lessons
Give a stateless model a memory: manage the short-term working context, separate episodes from facts, and retrieve long-term memories by meaning with embeddings while letting stale ones fade.
Agent Patterns, Testing and Guardrails5 lessons
Choose between ReAct and planner-executor or DAG designs, test tools and flows, trace every run and track the metrics that matter, then defend the agent against prompt injection with permissions enforced in code.