From Simple Workflows to AI Agents: A Practical Guide to n8n

This week I ran a workshop on n8n automation for my friend Caroline’s Lo-Fi AI study group, a women first group for learning about AI. The attendees left genuinely excited about what they could automate and I got messages from friends who couldn’t make it asking for the details.

So here’s the workshop in post form: a practical guide to automating repetitive work, from simple workflows to AI-powered agents that can actually make decisions for you.

Why n8n?

I first started using n8n about a month ago, when I was looking to automate event marketing emails for The Female Product Lead community. I had used Make for automating the community onboarding flow, but we had run out of workflows on the free tier, so I thought I’d give n8n a go, as I knew I could get it for free as part of my Lenny’s newsletter subscription. I immediately realised just how much more powerful n8n was compared to Make and being able to integrate AI agents into workflows makes it so much more than just a workflow automation tool – it can actually reason and make decisions, which is genuinely game changing.

What is n8n?

n8n is a workflow automation tool that lets you automate practically anything. Think Zapier or Make, but with more flexibility and better AI integration. It connects to all your favorite tools - Notion, Google Workspace, Slack, Trello - and allows you to create automated workflows with them.

Pricing note: If you’re subscribed to Lenny’s newsletter, you get a year of n8n included. Otherwise, it starts at £20/month for the starter plan, or you can self-host for cheaper on platforms like Hostinger.

The Basics: Building Your First Workflow

Every n8n workflow starts with a trigger - the thing that kicks everything off:

• Manual button click

• Scheduled time (e.g. every Monday at 8am)

• Something happening in an app (new Notion page, email received)

Then you add a sequence of steps. You can get complex with branching paths, but you can always start small.

Example: Processing Customer Feedback

In the workshop, I built a simple workflow for handling customer feedback forms:

• If it’s a bug: Add it to a bug backlog spreadsheet

• If it’s a feature request: Send me an email with the details

• Bonus complexity: Prioritize bugs based on frequency (daily = high priority, weekly = medium, monthly = low)

This took about 10 minutes to build, including the feedback form, and explain live and you can see what this looks like below:

It looks more complicated than it is, all it does is the above steps – so if you can think of a sequence of steps, then you can easily replicate this in n8n and connect your favourite tools together.

Note: For every node where you call a tool (Gmail, Google Sheets) you’ll need to be authenticated to perform any actions. But n8n provides really useful documentation about how to do this when you add the node to your workflow, so be sure to check that when you first do it. Once you authenticate to a tool, you can use that same credential in other nodes, so you only do it once per tool.

Adding AI: Where the Magic Happens

This is where n8n gets really powerful. AI agents are just another type of node in your workflow, but they can perform tasks such as:

• Summarize text

• Analyze and categorize data

• Extract sentiment

• Make decisions based on context

Anything that Claude/ChatGPT can do, an AI agent can do as well and more!

Setup requirements:

• Developer account with your chosen LLM (I use OpenAI, but n8n supports all the major ones, including Anthropic, Gemini and DeepSeek)

• API credits loaded (these will be used for the API calls to the LLM and are not included in your ChatGPT/Claude subscription)

• API key connected to n8n in your workflow (documentation for OpenAI, similar pages exist for the other providers)

Understanding AI Agents: The Building Blocks

The basic building blocks of an AI agent are:

1. Model – this is the specific LLM model you want to use; I mostly use GPT-4o mini because it’s cheap and effective, but you can experiment with any available model

2. Prompt – just as you would prompt ChatGPT, you need to give the agent a prompt so it knows what it needs to do

3. Memory (optional) – if you want the LLM to remember previous chat messages or have any type of long term memory, this is where you set it up

4. Tools (optional) – tools the LLM has access to in order to perform its job

Simple AI Agent Example: Meeting Notes Summary

I connected n8n to my Notion database (where Granola syncs my meeting notes) to automatically:

1. Trigger when a new page is added

2. Extract one key point from the meeting

3. List all action items with assignees

4. Store this summary back in Notion

No more manually scrolling long meeting notes to find what matters. This took me another 10-15 minutes to build and explain in the workshop, so it’s not a lot more than the first example.

Advanced AI Agent Example: Daily Briefing Email

This is the kind of workflow where I think n8n really shines. I built a workflow that runs every morning and generates a comprehensive briefing email containing:

• Priority emails I need to respond to (ranked by importance)

• Today’s calendar events with context

• Meeting prep - attendees and relevant past interactions

• To-do items from Trello

The workflow is deceptively simple: just an AI agent node and a send email node. The AI agent does all the heavy lifting, using tools to access Gmail, Google Calendar, Notion and Trello.

This one took me a few hours to build and iterate on - mostly getting the prompt right. But once it’s set up, it runs perfectly every morning.

Pro tip: I didn’t write the prompt myself. I explained what I wanted to Claude, and it generated a detailed prompt for me. Then I iterated a few times to get the formatting consistent.

This is the prompt I used for Claude: “Help me generate a prompt for an n8n agent node. The agent is meant to give me a daily briefing, summarising emails I need to respond to and my calendar for the day including some background on previous conversations with the attendees based on Granola notes I have on Notion.”

Here’s an example email it sent as well:

The Learning Curve

n8n’s built-in AI assistant isn’t great for solving issues - it mostly regurgitates documentation. But Claude and ChatGPT are excellent here. Just share screenshots, paste error messages, and describe your issue. They’ll usually get you unstuck quickly.

If you’re using an AI browser like Perplexity’s Comet, even better, as the assistant in the sidebar has context from the page you’re on.

Why This Matters for Product People

As product people, we’re constantly context-switching between tools, manually updating spreadsheets, and doing repetitive analysis. n8n lets you:

• Eliminate busywork so you can focus on what matters most

• Connect disconnected tools easily and without having to write any code

• Scale your personal productivity with AI-powered decision making

• Create consistent processes that don’t rely on remembering to do things

What Should You Automate First?

Look for tasks that are:

• Repetitive (you do them weekly or more)

• Rule-based (if X, then Y)

• Time-consuming but not complex, like processing form responses, syncing data between tools, generating weekly reports, triaging notifications.

I’d recommend avoiding: one-off tasks, things that require nuanced judgment (until you’re comfortable with AI agents or you’re happy using human in the loop steps).

Getting Started

n8n offers two free courses with certificates (great for LinkedIn): Level 1: Beginner Course, Level 2: Intermediate Course. They also have some excellent video courses if you prefer that format: beginner course and advanced course. The documentation is solid, and every node links to specific instructions, which is great while you’re learning.

Start simple - pick one repetitive task you do daily or weekly and automate it. Once you see the time savings, you’ll start spotting automation opportunities everywhere.

The thing about automation tools is they change how you think about work. Instead of accepting manual processes, you start asking “could this be automated?” and with AI in the mix, the answer is increasingly “yes”.

What’s one repetitive task you could automate this week?

P.S. If you’d like to download the workflows and import them into n8n, you can find them here.