Every workflow you have built so far follows a path you decided in advance: a trigger fires, steps run in order, and a result comes out. An AI agent works differently. You give it a goal and a set of tools, and the language model decides which tool to use next, looks at the result, and keeps going until the task is done. The path is chosen while the agent runs, not when you build it.
That freedom is useful for open-ended tasks and costly for simple ones, so a big part of this module is knowing when an agent is worth building and when a fixed workflow is the better choice. You will also meet the common design patterns behind agents: reflecting on and improving a draft, using tools, planning steps ahead, and splitting work between several agents.
In practice, an agent comes down to three parts: a model that reasons, memory that keeps context between messages, and tools that let it act, such as searching documents, reading a spreadsheet, or sending an email. Watch the lessons in order, then continue on to Business Use Cases.
IBM Technology's introduction to what makes a system an agent: a model that reasons about a task, plans, and uses tools, and how that differs from a single prompt and reply.
Andrew Ng shows why an iterative, agentic workflow can beat a single prompt, and walks through four design patterns: reflection, tool use, planning, and multi-agent collaboration.
A talk from Anthropic on when an agent is worth building, why to keep the design simple (a model using tools in a loop), and how to think from the agent's point of view when something goes wrong.
Nate Herk's hands-on course for building agents in n8n: the model, memory and tools that make up an agent, then real workflows built step by step. It is a long course, so watch it in parts and build along.