Automating your first workflow

Twenty minutes every evening

Sofia sells jams and liqueurs in a small online shop. She gets ten to fifteen orders a week, each one in an email. Every evening she opens the emails, one by one, and copies the date, the product, the quantity and the town into a spreadsheet. It takes twenty minutes. Always the same.

One evening she thinks: "This is always the same thing. Can't it be done by itself?"

It can. But not the first time, and not without her looking.

What a workflow is

A workflow is a repeated task described in three parts:

• Trigger: what makes it start. An email arrives with "New order" in the subject.
• Steps: what is done. The AI reads the email and pulls out the date, the product, the quantity and the town.
• Result: what ends up done. A new row in the spreadsheet, and a warning if any detail is missing.

Another example, with no email at all. Pedro writes the association's newsletter. During the week he jots down the news in a note.

• Trigger: every Monday, at 9am.
• Steps: the AI reads the week's note and writes a draft, in the tone of the previous newsletters.
• Result: a draft waiting for Pedro. It is not sent to anyone.

If you can describe a task of yours in these three lines, you have a candidate for a workflow.

First by hand, three times

Before automating, Sofia does the workflow by hand with the AI, always with the same request:

I am going to paste an order email. Pull out only these details and return them on one line, separated by semicolons, in this order: date; product; quantity; town. If there is more than one product, make one line per product. If any detail is missing, write MISSING in its place. Do not invent anything. Email: [paste here, without the customer's name, full address, phone and email]

Notice that she pulls out only what she needs for the spreadsheet. The customer's name and address stay out: that is one less piece of personal data passing through one more service (Level 1, lesson 5).

Doing it by hand three times, with three different emails, serves to:

• Discover the odd cases: the customer who orders two products, the one who does not give the quantity, the one who writes "a dozen".
• Fine-tune the request: each time you correct the answer, you add to the request the rule that was missing.
• Learn what a good result looks like: so that you recognise it at a glance later.

When the request works three times in a row with no corrections, it is ready to run by itself.

Two ways of automating, today

This is how it looks on 6 October 2026; your screen may differ.

1. Inside the AI tools themselves. Some assistants already let you set up a scheduled task: you write the request and choose when it runs, once or every week. The result appears in the conversation or as a notification on your phone. It suits Pedro's newsletter, a weekly summary or a reminder with prepared text. Look for an option with a name like "tasks", "schedule" or "scheduled actions". Not every tool and version has it: see the Current state page.

2. Connection tools. To join different programs together (the email, the spreadsheet and the AI, as in Sofia's case), there are tools made for it, such as Zapier, Make or n8n. They work with blocks of the type "when an email like this arrives, send it to the AI with this request and write the result in this sheet". Level 3 has a lesson just about them.

What you take from this level

• Advanced prompting: role, steps, examples, format and self-critique. Iterate part by part, saying what stays.
• Documents: attaching is sending. Ask for the page and the exact sentence, and check the sums in a spreadsheet.
• Instructions, projects and memory: three different things. See and clean the memory from time to time.
• Research: open the sources. An important fact needs the official source or two independent sources.
• Creating: five pieces for asking for images, you write the text, consent always and label what was generated.
• Automating: by hand three times first, and a person reviews everything that goes out.

What comes next

Level 3 (Advanced, no code) is for people who want to go further without programming: real automation with Zapier, Make and n8n, rules for using AI in a team or a company, how to judge whether an answer is good, how AI works on the inside, and ethics and the law. It is now available: open Level 3.