Stack

The 4-Tool Automation Stack I’d Build From Scratch

If I were starting a one-person business again, I wouldn’t begin by collecting AI tools. I’d build one small system.

Four tools is enough. The point isn’t to automate everything. It’s to remove a few pieces of repetitive work without creating another system that needs constant babysitting.

The stack

1. n8n — automation layer.

2. A model provider — extraction, classification, reasoning or drafting.

3. Airtable or another structured database — somewhere useful results can accumulate.

4. One output tool — email, CRM, CMS, Slack, Notion, or whatever the business actually needs.

Why the model shouldn’t be the architecture

AI is useful inside the workflow, but it shouldn’t automatically become the center of the system. The automation layer should coordinate the source, the model, the database and the final action.

The database matters more than it looks

A lot of AI workflows produce an answer and then forget it. If you’re doing research, lead generation, content planning or monitoring, the valuable part is often the accumulated structured information.

That lets you avoid processing the same thing twice, compare old and new results, add human review, and see where something failed.

What I’d automate first

Not writing. Not social media. Not a collection of autonomous agents.

I’d start with information collection.

Something like: source → n8n → filter → AI → database.

Now the second run adds to a body of research instead of starting again from zero.

The hidden test

Every tool in the stack should answer one question: what work disappears because this exists?

If the answer is unclear, remove it.

A real example

For software research: Reddit / YouTube / product discussions → n8n → AI classification → Airtable → research brief.

The useful result isn’t “AI summarized Reddit.” It’s that complaints, use cases and buying language start accumulating in one place and can feed comparison articles, product reviews and positioning.

Bottom line

I’d rather run four tools I understand than twelve tools I barely use. The stack should get more complicated only when the business forces it to.

Start with one repetitive job. Make it disappear. Then decide whether you need another tool.

If n8n is the part you’re unsure about, start with the hands-on review →