n8n AI agent: build one safely for a UK small business

“n8n AI agent” usually means one thing: the AI Agent node in n8n, which lets a chat model use your other apps as tools. This guide shows how to build a first one that is useful and safe: what the node does, how to make the agent ask before it acts, how to estimate executions, and how to use templates. For plan prices and self-hosting, see our n8n guide for small businesses.

Checked on n8n's AI Agent node docs, n8n's human-in-the-loop docs, n8n's pricing page and n8n's workflow templates on 27 October 2026. n8n changes often, so check the current docs before you build.

The short answer

• The AI Agent node does the deciding. n8n’s docs say you connect a chat model and one or more tools, and the agent decides which tools to call to complete a task.
• Tools are what it can do. A tool might look up a customer, read a spreadsheet, create a calendar event or send an email. The agent can only do what its tools allow, so connect as few as you can.
• Human approval is built in. n8n can pause before a chosen tool runs and wait for a person to approve or deny it.
• You pay per run on n8n Cloud. n8n bills on workflow executions, not on steps, so a chat agent’s cost depends on how many messages it handles.

What the AI Agent node does

In n8n’s words, an AI agent “receives data, makes rational decisions, and acts within its environment to achieve specific goals”. In practice, a workflow with an AI Agent node has three parts:

1. A trigger. For example a chat message, a new email or a form submission.
2. The AI Agent node with a chat model. This reads the input and decides what to do next.
3. Tools. Sub-nodes that the agent can call, such as Google Sheets, your CRM or a calendar. n8n’s docs say you must connect at least one tool.

The docs also note that the old “agent type” setting is deprecated, and all AI Agent nodes now work as a Tools Agent. If you copy an old tutorial or template, it may show settings you won’t see. n8n says the v1 node with that setting will be removed in n8n 3.0, so update to the latest version.

For a plain explanation of what agents are and when they are worth it, see What is an AI agent?.

Add human approval before it acts

This is an easy step to skip on a first build. n8n’s docs say you can require human approval before an AI Agent runs a specific tool. When the agent wants to use that tool, the workflow pauses and a person sees which tool it wants to use and with what details. They then:

• Approve, and the tool runs with the input the AI chose, or
• Deny, and the action is cancelled and the AI is told it was rejected.

You can put approval on all of an agent’s tools or only on selected ones. The approval request can go through a different channel from the conversation. n8n’s example is a customer chatting through n8n Chat while approvals go to a named person in Slack. The docs list channels including n8n Chat, Slack and Discord.

n8n’s docs suggest human review where tools do something irreversible, such as deleting data, sending external messages or making purchases, and where compliance rules or high-value decisions are involved. A sensible rule for a small business:

n8n’s pricing page lists human approval for tool calls on its Starter, Pro and Business plans.

Estimate your executions first

n8n Cloud plans are priced on workflow executions. n8n’s pricing page says an execution is a single run of your entire workflow, however many steps it has. On 27 October 2026 it listed the Starter plan at €20 a month billed annually with 2.5K executions, and Pro at €50 a month billed annually with 10K. Prices are in euros on n8n’s page. See our n8n guide for the full picture, including self-hosting.

For a chat agent, n8n’s pricing FAQ suggests this method:

1. Estimate how many conversations the agent will handle each week.
2. Multiply by the average number of messages per conversation.
3. That gives you the weekly executions for the chatbot.

For agents on a trigger rather than chat, n8n’s FAQ says a daily schedule gives 30 or 31 executions a month, and for webhooks your daily executions equal the number of times the event happens each day. Remember that the chat model you connect, such as OpenAI or Anthropic, may also charge you separately for its use. n8n’s pricing page also lists “AI models without API keys” on its plans, so check how that use is counted on yours.

Using n8n workflow templates

n8n’s workflow library listed 13,775 community workflows when we checked on 27 October 2026. Many of them are AI agent examples you can import and adapt. Before you use one:

• Check who made it. Templates come from n8n and from the community. Read the description and the nodes before importing.
• Check what it connects to. Look at every credential it asks for. Remove tools you don’t need.
• Check the node versions. Older templates may use the deprecated agent settings mentioned above.
• Add approval. Many templates are built to show what’s possible, not to be safe in your business. Add human approval to anything that sends or changes data.
• Test with dummy data. Run it on test records before it sees real customers.