Exercises

One workflow that you extend step by step. Every prompt for the assistant answers the same six questions.

WHENWhen should it run?
FROMWhere does the data come from?
WHATWhat should it do with it?
WHERE TOWhere does the result go?
WHAT IFWhat if something doesn't fit?
CHECKHow will we know it works?

The training inbox: what we work on

A made-up company, Green Office (office furniture). Its inbox with ten messages is at https://course.aiadvisors.pl/api/inbox. The workflow fetches the messages, not the assistant.

sales Maria (sales)
complaint Tom (customer service)
accounting Anna (accounting)
other Office

Two working rules: ask for a plan before the assistant starts building, and tell it to test the result itself.

Exercise 0. Get to know n8n

Before the assistant starts building, you look around n8n yourself: the canvas, nodes, running a workflow and the execution history.

You do this one by hand, without the assistant. It takes about 10 minutes.

Log in to n8n with the details from your start page. The first screen is the list of your workflows (Overview). Open the workflow “Course start”.
You see the canvas. Each rectangle is a node, and the lines between them show where the data flows. Move the canvas by holding the space bar or the middle mouse button, and zoom with the scroll wheel.
The first node on the left is the trigger: every run starts there. Here it is a manual button.
Double-click the AI model node. On the left is the Input, in the middle the Parameters, on the right the Output. Close the window without changing anything.
Click “Execute workflow” at the bottom of the canvas. The nodes turn green one by one, and each shows the number of items that passed through it.
Open the “Executions” tab at the top of the screen. It is the history of runs: status, time and the data from every node. This is where you look when something doesn't work. The assistant reads the same history through n8n-mcp.
Go back to the “Editor” tab. Don't save anything: the starter workflow will be useful to the assistant in the next exercises.

Workflow: The whole process: a trigger and the steps that follow. One workflow is one automation.
Node: One step of the process, for example fetching data, a condition or an AI model.
Canvas: The area where you see the nodes and the connections between them.
Trigger: The node that starts a workflow: a button, a schedule, a new message, a webhook.
Execution: One run of a workflow, with the data from every node saved.
Item: One piece of data flowing through the nodes, for example one email.
Credential: A saved key or password for a service. Nodes use it but don't show it.
Expression: A field that pulls in data from earlier steps, written in double curly braces {{ }}.
Pinned data: Data remembered in a node for testing. It does not apply to real runs.
Active / Published: A workflow switched on permanently, reacting to its trigger on its own.
Data table: A simple table inside n8n where a workflow writes and reads rows.
AI Agent: A node where the AI model decides for itself which tools to use to complete the task.

Then: You now have a map of n8n. From exercise 1 the assistant takes over the building, and you check the result on the canvas and in the Executions tab.