Your first automation

An automation is just: when this happens, do that — with an AI step in the middle to do the thinking. We'll build a real one: a new enquiry comes in, AI summarises and prioritises it, and the right person gets a tidy notification.

The shape of every flow

Almost every automation is three blocks in a row:

Once you see this shape, every automation looks the same — only the three blocks change. We build these in n8n (the tool behind Flowsmith), where each block is a "node" you connect with a line.

What we'll build

A new-enquiry triage:

• Trigger — a contact form submission (the same forms gateway you met in Lesson 2).
• Brain — AI writes a one-line summary and rates urgency (low / normal / high).
• Action — a tidy email (or Slack message) to the right person, urgency in the subject.

Result: instead of raw form emails, you get "High · Acme wants a quote for 500 units by Friday" — sorted before you even open it.

Building it

You won't wire this by hand — you'll describe it to the agent (Lesson 1's loop) and check each block as it appears. The pieces:

1. Add the trigger. A Webhook node gives you a URL. Point the form's notifications at it (or forward gateway submissions to it) so each enquiry starts the flow.
2. Add the brain. An AI node takes the form fields and runs a prompt (next section) that returns a summary + an urgency word.
3. Add the action. A Send Email or Slack node, using the AI's summary in the subject and the original message in the body.
4. Connect them trigger → brain → action, and switch the flow on.

The AI step

The brain is only as good as its instructions. A solid starting prompt:

You triage incoming sales enquiries.
Given the enquiry below, reply with exactly two lines:
1) URGENCY: one of low / normal / high
2) SUMMARY: one sentence (who, what they want, any deadline)

Enquiry:
Name: {{name}}
Email: {{email}}
Message: {{message}}

The {{name}} bits are placeholders n8n fills from the form. Tune the wording the same way as Lesson 1 — try it, read the output, adjust, repeat.

Test it — as the person submitting

Test the form the way a visitor uses it, not just the way you built it. The fact that your notification email arrives does not mean the person on the other side knows their message went through — those are two different things.

1. Submit the form as a real visitor would — open the public page, fill it in, press send.
2. Confirm you get feedback on the page. After a successful submit the visitor must see a clear confirmation ("Thank you — your message has been sent."). No confirmation = a broken form, even if the email arrives: people resubmit or give up. The forms gateway shows this thank-you automatically when you use its client.
3. Then check your side — the notification arrived, summary and urgency correct.
4. Watch it flow — n8n shows each node lighting up and what it passed on. If a node goes red, that's just an error to read: copy it, paste it to the agent, fix, retry.

Good habits

• Keep a human in the loop for anything outbound. Auto-summarising for your own inbox is safe; auto-replying to customers should wait until you trust it — start by drafting, you send.
• Start with one trigger and one action. Add branches (e.g. "if high, also text me") later.
• Test with real-looking data, including a messy enquiry — that's where AI steps earn their keep.
• Always show the visitor a confirmation. The form's job isn't done when the email lands — it's done when the person who submitted knows it sent. Show a thank-you on the page (and/or a thank-you page), every time.

What's next

You now know the shape of every automation. From here it's just swapping the three blocks for your own case.