How to learn n8n: a path, projects and what to look for in an n8n course

What n8n is and why we use it with clients

n8n is an automation tool with a visual editor: each step of a process is a node, and nodes connect to move data between apps, APIs and databases. At TintoDev we use it every day to automate client processes: receiving forms, syncing a CRM with spreadsheets, classifying emails with a language model or alerting a team when something breaks. We chose it because it can run on infrastructure you control and because, when the visual approach falls short, it lets you write JavaScript or Python inside the workflow.

That last point shapes how you should learn it. You can get far without coding, but people who understand JSON and a little JavaScript solve in minutes what takes others an afternoon of chained nodes. You do not need to be a developer: being able to read an object, access a field and write a simple condition is enough.

Self-hosted or cloud: where to run n8n

n8n has a cloud version run by the company itself and an edition you can install yourself, for example with Docker on your laptop or a server. Its source code is available under a fair-code licence: you can use and modify it, with restrictions on reselling it as a service. For learning, the cloud skips all setup; a local Docker install carries no licence fee and teaches you how it works under the hood.

In production the decision changes. Self-hosting gives you control over the data and over cost as volume grows, but you own updates, backups, certificates and monitoring. With clients who handle sensitive data we usually install it in their own cloud; for small teams with no technical staff, the cloud version tends to be the saner choice. Plans change often, so check the official pricing page before deciding.

A staged path to learn n8n

The timelines assume about 5 hours a week of practice on real workflows, not just watching videos. If you have less time, stretch them; if you already code, you will probably move faster through expressions and code. What matters is closing each stage with a workflow that runs end to end and that you have tested with data you do not control, because that is where the cases the tutorial skipped show up.

• Weeks 1 and 2: the editor, triggers, basic nodes, credentials and the concept of items.
• Weeks 3 and 4: expressions and data transformation with Edit Fields, IF, Switch and Merge.
• Weeks 5 and 6: webhooks, the HTTP Request node, pagination and API authentication.
• Weeks 7 and 8: error handling, retries, sub-workflows and deployment.
• From week 9: AI agents, memory, tools and reviewing their output.

Nodes, items and expressions

What confuses people most at first is that each node receives a list of items and, in general, runs once per item. If a Google Sheets node returns 200 rows, the next node will run 200 times. Understanding that explains most odd behaviour: duplicate emails, extra calls to an API or a Merge that does not combine what you expected. Open the data view on each node and check how many items go in and how many come out.

Expressions are the other pillar. Inside double curly braces you can read data from the current item with $json or from an earlier node by name, and apply JavaScript functions to format dates, clean text or calculate values. Practise until you can write them without checking the docs, and reach for the Code node only when an expression or a built-in node becomes unreadable.

Webhooks and error handling: what separates a demo from a production workflow

A webhook lets another application start your workflow by sending it data. n8n gives each Webhook node two URLs: a test one that only listens while the editor is open, and a production one that works once the workflow is active. Mixing them up is the most common mistake of the first week. You should also validate what arrives and respond quickly, so the calling system does not retry and create duplicate records.

With clients, a large part of the work is deciding what happens when something fails. Set retries on nodes that call external APIs, decide which nodes may continue on error and build an error workflow with the Error Trigger node that alerts by email or chat with the workflow name and a link to the execution. A workflow without alerts fails silently, and you find out when someone asks about an order that never arrived.

AI agents in n8n

n8n includes nodes for working with language models, among them an agent node that you connect to a model, a memory and tools, which can be other workflows or HTTP calls. We use it to classify requests, extract data from documents and answer frequent questions with the client’s own information. It works well when the task is narrow and the result can be checked.

Our rule of thumb: if an IF node solves the case, do not use a model. Agents cost money on every call, sometimes answer differently to the same input and need someone reviewing their output at the start. Ask for structured output, validate the fields before using them and keep a log of what the agent decided so you can audit it later.

Practice projects

Choose projects that touch real problems from your job or a nearby business. Each one should have a trigger, a data transformation, an output and an error path. Export the workflows as JSON and keep them in a repository, which works as both portfolio and backup. If you handle other people’s information, use test data while you learn.

• A web form that creates the contact in the CRM and notifies the team in a chat channel.
• A daily report that reads a spreadsheet, summarises sales and sends it by email.
• A webhook that receives orders from a store, validates them and saves them to a database.
• An incoming-email classifier using a language model that tags each message and assigns an owner.

Common mistakes, n8n’s limits and choosing a course

n8n does not replace a custom application. When the logic has dozens of rules, the volume is high or the process needs serious automated tests, we usually move that part into a Node.js service and leave orchestration in n8n. At high volume the installation also needs queue mode and close monitoring.

To choose an n8n course, check that the curriculum covers expressions, webhooks and error handling, not just integrations with popular apps. Look at the update date: the editor changes often and an old course shows screens that no longer exist. If the course is not in your first language, confirm it has subtitles, and prefer courses that leave exercises with downloadable workflows you can compare with your own.

• Hard-coding credentials inside Code nodes instead of using the credentials manager.
• Building one giant workflow instead of splitting it into reusable sub-workflows.
• Testing only with the webhook test URL and forgetting to activate the workflow.
• Having no error workflow and no alerts.
• Using an AI model where a simple condition would do.

Where the courses are

Frequently asked questions

Is n8n free?

The edition you install yourself has no licence fee, but you pay for the server and the time to maintain it. The cloud version and enterprise editions have paid plans that change often. Check the official pricing page before deciding.

Do I need to code to use n8n?

Not for simple workflows between well-known apps. For workflows with data transformations, custom APIs or multi-condition logic, knowing JSON and basic JavaScript makes a real difference. A few weeks of introductory JavaScript is enough.

How long does it take to learn n8n?

With about 5 hours a week, 6 to 8 weeks to have useful workflows in production with error handling. AI agents take a few more weeks. If you already code, the timeline is usually shorter.

Should I learn n8n, Make or Zapier?

If you want to work in automation professionally or handle volume and sensitive data, n8n. If you need quick results without a technical background, Zapier or Make have a gentler curve. We compare them in detail in our n8n vs Make vs Zapier guide.

Can you build AI agents with n8n?

Yes, with the agent node, which you connect to a language model, a memory and tools. It works well for narrow tasks such as classifying, extracting data or answering from your own information. For decisions with consequences, validate the output and keep a person in the review loop.