If you searched for "n8n" it's because someone told you that you can automate your company with it for free. And that's true… until it isn't.
n8n is one of the best tools out there to connect apps and automate tasks. We use it too. But there's a point where "I'll build it myself with n8n on a Sunday" turns into "I've been fighting this for three weeks and it breaks in production". This is the map to know which side of that point you're on before you lose the weekend.
What n8n is, in plain terms
n8n is a "connect the boxes" automation tool: you build a visual flow where something happens (an email comes in, a WhatsApp message, a new row in a sheet) and n8n doesthings in sequence (save, reply, notify, call an AI). It's free and open source. It's great for automating repetitive tasks without much coding.
The key thing: n8n is the wiring. It is not "the AI agent". It's the workbench where you can plugan AI model (like Claude or GPT) into your tools. Having the bench doesn't mean the furniture is built.
When n8n is ENOUGH
Let's be honest, because this is where most people actually save money. n8n is enough, and the best option, when:
- The task is a single, clear one: "when a lead comes in through the form, send it to my WhatsApp and log it in the sheet".
- Volume is low to medium and if it fails once, the world doesn't end.
- You (or someone on the team) have the patience to maintain it when an API changes or a step breaks.
- You don't need it to decide delicate things (prices, sensitive data, promises to a customer).
If you're in this box: don't hire us yet.Go, look at it, build it. Seriously. We'd rather you trust that we tell you the truth than sell you something you don't need.
When n8n gets EXPENSIVE (even though it's free)
The tool is free. Your time and your customers are not. n8n starts costing you when:
- There are several chained tasks that depend on each other and one has to understand the previous one, not just pass it along.
- The agent has to decide, not just move data: read a complaint, judge whether it's serious, respond differently case by case.
- You can't afford it to be wrong in front of a customer.At that point it's not a flow anymore: it's a system that needs tests, guardrails, and someone to back it up when something weird happens at 11pm.
- Every time something changes (a price, a product, an API) you have to get your hands in and nobody remembers how it was wired.
This is exactly the pit Gartner warns about: 40% of agentic AI projects will be cancelled before 2027— and not because of the idea, but the implementation: weak governance, costs that don't add up, and unclear business value. At the same time, 74% of companies plan to deploy autonomous agents in the next two years. Translation: almost everyone will try, and 4 out of 10 will crash. The difference isn't whether you use AI. It's who implements it, and how.
The question that decides
Forget the tool for a second. The question isn't "n8n or custom?". It's:
If this gets it wrong, how much does it cost me?
- If the answer is "nothing serious, I'll fix it and move on": n8n, go for it.
- If the answer is "I lose a customer, I quote the wrong price, it leaks something it shouldn't": you need something built, tested and in production. That's where a custom agent comes in (which, mind you, may usen8n under the hood — but with the engineering, guardrails and maintenance a Sunday doesn't give you).
That's what we do: we don't sell you "an AI", we hand you the process running, measured and backed up — and the code is yours.
FAQ
Is n8n really free?
Yes, it's open source and you can run it for free. What isn't free is the time to build it, maintain it, and keep it from breaking in production.
Does a custom agent replace n8n?
Not necessarily. A custom agent often uses n8n under the hood. The difference is the engineering around it: tests, guardrails, security and maintenance.
When should you hire it out and when not?
If the task is simple and a mistake doesn't cost much, do it with n8n. If the agent has to make decisions that affect a customer or your money, have a team build it.
Why do so many AI projects fail?
According to Gartner, 40% of agentic AI projects will be cancelled before 2027, almost always due to implementation and governance, not the idea.