Guides · AI adoption
What to learn to build AI agents in your company
Ironhack for Business
- Copying community templates
- Taking workflows to production
An n8n course is worth it if it teaches you to take AI workflows to production, not just to copy templates. In Spain, 21.1% of companies with 10 or more employees use artificial intelligence in the first quarter of 2025 (INE, survey on ICT and e-commerce use in companies 2025), and n8n is one of the most direct routes from using a chatbot to automating whole processes. This guide sets out what the training has to include, when an agent makes sense, and what changes when you train a whole team.
What is n8n, and why has it become the way into AI automation?
n8n is a low-code automation platform that connects applications, APIs and language models into visual workflows made of a trigger and a chain of nodes. Its difference is the AI layer: the AI Agent node acts as the root node, with a language model (Anthropic, OpenAI or Ollama), a memory and the tools the agent can use connected to it.
That is why it tends to be the first automation at a company starting with AI: one platform holds everything from an internal chatbot to a flow moving data between the CRM and a spreadsheet. The traction shows in its numbers: in October 2025 n8n closed a Series C of $180m at a $2.5bn valuation, after growing users 6x and revenue 10x in a year (n8n, Series C announcement 2025).
n8n · AI Agent node
A trigger starts the root node, and everything else hangs off it.
Trigger
A webhook, a scheduled task or a message that starts the flow.
AI Agent node
The root node everything else connects to.
Model, memory and tools
A language model (Anthropic, OpenAI or Ollama), a memory, and the tools the agent can use.
Output
The result the flow delivers or writes.
How does n8n connect to Claude through MCP?
n8n works with MCP in both directions: a flow can use an MCP server as a tool, and n8n can expose its flows for an assistant like Claude to run. The practical rule we apply: it is more reliable to have Claude operate n8n flows that are already tested than to ask it to build them from scratch. When it cannot find the right node, the assistant tends to solve everything with generic HTTP requests, and the resulting flow costs more to maintain.
That way each stable flow becomes one more tool for the agent: create a page in Notion, log a lead in the CRM, or prepare a weekly summary.
What should an n8n course include to build AI agents?
A useful n8n course covers 5 blocks: workflow fundamentals, data and JSON, integrations and credentials, AI (agents and RAG) and going to production. It is the minimum syllabus we use when training AI consultants: the 5 blocks that separate a flow that works on your laptop from one that works every day in production.
| Block | What you learn | Why it matters |
|---|---|---|
| Fundamentals | Triggers, nodes, connections, webhooks and scheduled tasks | It is the logic of any automation |
| Data and JSON | Items, split out, loop and aggregate. Structured outputs instead of formatting with code | In n8n everything travels as JSON, and it is where beginners get stuck most |
| Integrations and credentials | The HTTP Request node when no native node exists; credentials with clear names and scope; least privilege | Authenticating and granting permissions is what eats most time on a real project |
| AI: agents and RAG | The AI Agent node with model, memory and tools. RAG with a vector store (Pinecone or Qdrant) and a re-ranker | It is the jump from automating tasks to delegating bounded decisions |
| Production | Error flows, debugging, pinned data, and the difference between the test URL and the production URL | A badly published flow can keep running an old version |
A course that stops at community templates covers the first block and part of the fourth. Templates are useful for learning patterns, but they tend to load a single agent with too many tasks. The right move is to split it into sub-agents with one responsibility each.
Do you need to know how to code to take an n8n course?
You do not need to code, but you do need to understand how code works: JSON, APIs, authentication and reading errors. n8n is a no-code platform that does not ask you to write code, but it does ask you to know what the code is doing underneath. Anyone starting without that base climbs two hills at once — the tool and the concepts — so choose training that teaches both in parallel.
The warning sign is the tutorial trap: stringing together videos that reproduce other people's flows without understanding why they work. Learn the nodes your use case needs, and go deep on those.
When is an AI agent worth it, and when is a workflow enough?
If the task follows a fixed, repeatable recipe, a workflow is cheaper, more predictable and easier to control than an agent. The agent adds value when something has to decide which steps to take. An agent you give a goal to pays for every step it decides to run: it searches, reads and retries. A flow with defined steps has a predictable cost and lets you put controls at every point.
This distinction explains much of the failure rate. Gartner expects over 40% of agentic AI projects to be cancelled before the end of 2027 over rising costs, unclear business value or insufficient risk controls (Gartner, press release, 25 June 2025). The same firm estimates that only around 130 of the thousands of vendors presenting themselves as agentic AI actually are.
Our criteria for deciding
Workflow
- A process with a clear recipe, with the conditions written as rules
- Predictable cost: the steps are defined in advance
- Controls at every point along the way
AI agent
- Interpreting text, classifying, or choosing between tools
- Bounded tools and human review
- Pays for every step it decides to run: searches, reads and retries
And a third option: if you would review the task by hand afterwards, it is not a candidate for automation yet.
When does n8n fall short against code?
n8n falls short when the process demands heavy logic, chained validations, or steps that depend on many earlier steps. A code framework such as LangChain or LangGraph, or a hybrid, does better there. The clearest way to put it: code is for thinking, n8n is for doing. n8n excels at connecting services, transforming data, scheduling tasks and assembling simple RAG agents.
The sign that a flow has hit its ceiling is that its last steps depend on almost all the earlier ones. Pushing on at that point because time has already gone in is the sunk-cost fallacy. There is a way out that does not throw the work away: a service built in code can be exposed as an API and called from an n8n node, and the business team keeps working in the same interface.
An n8n course for companies: what changes when you train a team
When the training is for a team, the tool matters less than the organisation. Before training anyone, we review 4 points:
- Ownership and maintenance. Every flow needs a named owner who adjusts it when the process changes or an integration breaks.
- Capacity to adopt. If the company can only adopt one tool and has no engineering team, n8n is a good candidate because it holds chatbots, automations and agents on one platform. With a technical team, code gives more room.
- Credentials and permissions. Getting access and authorisations is the part that eats the most time and the part not to automate, because it involves keys and secrets. Plan it from the start and apply least privilege.
- Hosting. n8n can run in the cloud or self-hosted. Self-hosting gives more control but means updating, maintaining and reviewing logs. The cloud removes that work for a fee.
On top of that, a rule of scope: do not automate tasks you are going to review by hand afterwards, because maintaining, inspecting and correcting that flow multiplies the work. Automate what you want done the same way every time.
If your company already works with Make, Zapier or Power Automate, there is no need to migrate for the sake of it. These platforms solve the same kind of problem, and the best option is usually the one that fits the stack and the team you already have.
A catalogue course, or training on your team's own processes?
For a team, training applied to its own processes pays off more than a catalogue course, because the hard part is deciding what to automate and with what controls, rather than learning each node. A generic course teaches you to build the trainer's example. Training on real processes ends with something the team uses the following week.
At Ironhack for Business we work in 2 formats:
- AI workshops: half-day sessions on the tools your company already uses, with Claude as the main tool and Copilot or ChatGPT depending on your stack. Your team keeps prompt templates, worked exercises and a replicable demo agent built on its own use case. They are FUNDAE-fundable, with your company administering the process and subject to its available credit.
- AI consulting: a task-by-task automation map, as a Lite Assessment (3 weeks) or a Full group assessment (6 to 8 weeks), prioritising 3 to 5 quick wins with an estimate of hours and euros saved per task.
Is n8n free?
The self-hosted edition of n8n is free for internal use under its community licence, the Sustainable Use License, as long as the flows are created or modified by people in your organisation. The cloud version is paid. If you give external clients the ability to create or edit flows on your instance, that licence no longer covers the use.
Is n8n better than Make?
Neither is better in the abstract: both solve the same kind of problem, connecting services and automating flows. n8n stands out for being self-hostable and for its AI agent layer. The choice depends on the tools and the team you already have.
Can Claude build my n8n flows for me?
It can help with part of the work, but it does not replace understanding the flow. That is why we recommend the reverse: build reliable flows in n8n and let Claude operate them through MCP.
Where do I start if I want to automate a process with n8n?
Start with a repetitive process that eats hours today and always follows the same steps, and estimate how much time it could free up with our savings calculator (the result is an indicative estimate).
Next step
If you want to turn a repetitive process into a flow or an agent your team can maintain, we will come back with a starting point within one working day.