Leave it as a plain flow
The steps are the same every time, the input is structured, and a mistake is obvious. If a checklist describes the whole job, keep the checklist.
n8n
n8n AI agents
Most n8n flows don't need an agent. The ones that do are where someone reads a message and decides what happens next.
I'm an n8n AI agent consultant, and I work mostly with UK teams. For a new build I start in n8n. The version that covers Zapier and Make as well is in from automation to agents.
Agent node or not
When a flow should get an agent node
A plain n8n flow is enough when the path is fixed. A form lands in a sheet. A payment marks a row paid. A status change posts to Slack. You can write every step down before the first item arrives. An AI agent node would only add a decision you didn't ask for.
Add an agent node when the next step depends on the content. Someone has to read the email, choose a route, pull a few facts together, or write a first draft. The agent can do that reading and that draft. A person still approves anything that gets sent.
Add an AI agent node
The next step depends on what was actually said. A draft or a shortlist is useful, and a person approves it before anything leaves the building.
Controls
How to run one safely
Three controls, on every agent. They're there so a person catches the mistake, not a customer.
Approval gate
The agent can read, draft, and queue a change. It does not send the message, post it, or write back to the other system until someone approves. The gate sits before anything leaves the building, on the step you would not want to undo.
An alert when it fails
A failed run, a stall, or an answer the agent isn't sure of raises an alert. It goes to a channel someone already watches, with the input and what the agent was about to do. A log nobody opens doesn't count.
A short runbook
Written so someone else can fix it. They can stop the flow, see what it last did, and do the job by hand. It covers where the workflow lives, what it may touch, who approves, and what broken looks like.
My own site
The form that died quietly
This one is my own site. The contact form had been forwarding each enquiry to a Zapier hook. Zapier was paused. The form still accepted messages, and nothing came through. It stayed that way for months. I noticed only because I went looking.
I rebuilt it in n8n. Every enquiry is emailed to me and logged to a sheet. If the run fails, I hear about it. A flow with no alert fails silently.
FAQ
Questions I get asked
When should an n8n workflow get an AI agent node?
When the next step depends on reading the message. Routing an inbox, drafting a reply, or pulling facts out of an untidy email are the usual cases. If a checklist already describes the job, leave the agent node off.
What should stay a plain n8n flow?
Anything with a fixed path. Form to sheet, payment to invoice, stage change to Slack. For a new build I use n8n for that, not Zapier. The steps don't change, so an agent has nothing useful to decide.
How do you keep an n8n AI agent safe?
An approval gate before anything leaves the building. An alert when the run fails. A short runbook so someone else can fix it. The agent can read and draft. It does not send on its own.
Do you work as an n8n AI agent consultant in the UK?
Yes. I work remotely, mostly with UK teams. If you want an n8n AI agent consultant, bring one workflow. I'll tell you whether it needs an agent node or should stay a plain flow.
Get in touch
Start with the $300 audit
Bring me one n8n workflow. I'll tell you if an agent node belongs on it, and what the gate, the alert, and the runbook should cover, before anything is built.