Building Your First AI Automation

The best first AI automation is the one that solves a real, recurring problem you have right now. Not a grand system — just one thing you do manually every week that you'd rather not.

We're going to build a content summarisation workflow: paste in a long article, get a structured summary with key points and action items, ready to forward to your team.

What you'll build

A reusable prompt template that turns any article or document into a clean summary — with a specific structure your team can rely on. No code. No tools. Just a prompt you can reuse in 30 seconds.

Step 1: Define the output format

Before writing a single word of prompt, decide what good output looks like. The clearer your definition, the better your prompt will be.

• One sentence: what this is about
• 3 key points (maximum)
• Any action items or decisions required
• Source link or reference

Step 2: Write the system prompt

A system prompt sets the context for every conversation. Think of it as the standing instructions you give before pasting in any content.

You are a research assistant for a small business team. When
given an article or document, produce a structured summary
using exactly this format:

TOPIC: [One sentence describing what this is about]

KEY POINTS:
• [Point 1]
• [Point 2]
• [Point 3 — maximum 3 points]

ACTION ITEMS: [List any decisions or actions needed, or
write "None" if this is purely informational]

Be concise. No filler. If the content is not worth
summarising, say so.

Step 3: Test with real content

Copy a real article you'd normally forward to your team. Paste it after the system prompt. Run it. Check: does the output match your format? Is it actually useful? What's missing?

Step 4: Make it a reusable template

Save the system prompt somewhere accessible — a Notion page, a text file, your phone's notes. The workflow is: open Claude, paste the system prompt, paste the article, read the summary in 10 seconds.

Step 5: Scale it (when you're ready)

Once you've got a reliable manual workflow, you can think about automating the paste step. Tools like Make.com or Zapier can connect to Claude's API, watch your RSS feeds or email, and automatically summarise and deliver content to Slack or email — all without code.

What to build next

Once this is working, the next logical step is customer query triage — same idea, different prompt. Or weekly report generation. Each new automation follows the same pattern: define output, write prompt, test with real data, refine.