If I were starting automation again, here's what I'd do...

This is what I wish I'd known 3 years ago. I'm by no means a genius… just someone who's spent way too much time with these tools and figured this might help fellow automation noobs.

Four years ago, the only "automation" I knew about was our QA engineers running tests before releases. Honestly? It sounded boring. Just repetitive actions… login, create account, subscribe… to make sure our key user flows didn't break. Nothing glamorous, but it meant we could ship without someone manually clicking through everything.

Then LLMs entered the chat.

Suddenly you could string actions together AND include AI reasoning? My product manager brain started spinning. If this could work for more than just software testing, what else was possible?

As I dove deeper, I started following people in the content space who were using early models for SEO blogs, recipe generation, and other creative applications. The platforms they used - n8n, Make.com, Zapier - seemed to have endless integrations and could allegedly power entire SaaS products.

But here's what took me 3 years to figure out: I would have done things completely differently as a beginner.

Start With What You Know: LLMs

If you've been using ChatGPT or Claude for the past year, you can probably skip this. But if you're like past-me and got lazy asking the same questions repeatedly, start organizing your frequent prompts.

I use Claude Projects with specific instructions so I don't retype the same prompts constantly. My "PRD Helper" project asks me questions I haven't considered, ensuring my documentation stays clear and consistent every time.

The Anatomy of Automation

Before touching any tools, grab a pencil and paper. Every automation has three components:

1. Input — What triggers your automation?

• Weather data fetched daily at 6am

• A LinkedIn post URL you want scraped

• A Slack message that creates a Trello task

2. Action — What happens to that input?

• Transform data, write data, read and summarize data

• This is where the magic happens

3. Output — What's the end result?

• A message, a number, a list of URLs

• Whatever format you actually need

I will often sketch these things out on paper before diving into the tools.

Your First Automation: The Jacket Test

Let's build something simple: "Should I wear a jacket today?"

• Input: Weather data fetched daily at 6am

• Action: AI analyzes data and makes an educated guess

• Output: Message summarizing the decision

Yes, you could ask ChatGPT this every morning. But that's not automation…

That's just you typing the same question daily. Real automation feeds you the information automatically.

The Tool Trap (And How to Avoid It)

Here's my biggest mistake: Don't start with Make.com or n8n YouTube tutorials.

You'll download some .json file, upload it to n8n, get 7 days to figure it out, then hit a paywall and get discouraged. You're not ready for n8n yet.

Instead, start with Gumloop (not a sponsored link). Here's why:

1. Totally free with decent monthly credits

2. Best-in-class AI troubleshooting that teaches you to debug workflows

3. Just enough tools to build useful stuff without overwhelming you

Explore the nodes. Practice that input-action-output model. Get comfortable.

Level Up: The n8n Transition

After successfully running 5-10 automations in Gumloop, rebuild them in n8n.

Start with self-hosting on a Digital Ocean droplet (~$6/month) or use n8n cloud ($20/month). Setup takes about an hour with their guides.

This might sound repetitive, but it's crucial. n8n is more technical - you'll hit different problems, learn credential setup, and understand API access. ChatGPT becomes your debugging partner.

Only then should you start importing templates. Why? Because now you know what's broken, what needs credentials, and you can actually get stuff working.

The Automation Mindset

Soon you'll become an automation connoisseur, appreciating the elegant ones that solve real problems without breaking the bank.

At this point, use whatever works. I'll prototype quickly in Gumloop, then rebuild in n8n for cheaper long-term running. The tool doesn't matter - the mindset does.

When you face repetitive tasks, you'll automatically think: "How could I automate this?"

And that's when automation stops being a skill and becomes a superpower.

What's your biggest automation win so far? Hit reply and let me know—I love hearing what people are building.

Keep automating,

-Paul 

Did someone forward this to you? Sign up here for the newsletter.