💡 New here? Three words before you start
- Build — “build” in English. In our case, it means putting together something useful (a spreadsheet, a template, a mini-automation) with AI’s help.
- Directed — you’re in charge. You say what you want and check; the AI does the grunt work.
- Iterate — ask, look at the result, point out what to change, and ask again. This is the cycle that makes everything better.
🎯 Choose a small, real problem
The number one mistake beginners make is wanting to build their “dream system” right away. Don’t do that. Choose a a concrete pain point in your day — that annoying task that repeats every week and eats up 20, 30 minutes. Small, real, yours.
Why keep it small? Because a small problem has a quick win built in. You finish, see it working, and learn the whole process in a single cycle. That learning is worth more than the result itself—it’s what you’ll repeat later for bigger problems.
🔑 The central idea
A good first problem fits in one sentence and passes these three tests:
- •Repeat: you do this every week (or every day).
- •Bothers you: it’s tedious, mechanical, and makes you procrastinate.
- •Of course: you can explain what “done” means in one sentence.
💡 Practical tip
Make a list of 5 tedious tasks from your week. Cross out the ones that depend on lots of people or systems you don’t control. Is there one simple task left that’s yours alone? This is your first build.
Key concepts
🎬 Describe the result—“what good looks like”
AI can’t read your mind. It hits the target when you describes the target. "What good looks like" means saying, before you ask, what you expect to receive at the end: the format, o tone, o size and, if possible, a example what would be perfect.
Think of a bricklayer: "build a wall" could produce any wall. "Build a 2-meter wall with exposed brick and a niche in the middle" produces the one you want. The difference isn't effort — it's clarity about the goal. The clearer the “done” looks in your mind, the better the first result.
✓ A clearly targeted request
- ✓Specifies the format (“a 3-column table”)
- ✓Specifies the tone (“informal, for a client”)
- ✓Specifies the length (“no more than 5 lines”)
- ✓Show an example of what perfect would look like
✗ Untargeted request
- ✗"Help me with this"
- ✗Doesn’t specify the format or size
- ✗Leaves the tone in the dark
- ✗Expect the AI to "guess"
💡 Practical tip
Before making a request, write just one sentence: "In the end, I want ___, in the format ___, with a ___ tone." If you can’t complete it, you’re not ready to ask yet—and that’s okay. Clarity first, request second.
Key concepts
🧾 Provide context and examples
Context is what separates one result generic of a custom-made. AI is brilliant, but it only knows its own world — it doesn’t know yours. You need to feed it supporting material: an example of what already exists, your rules, your data.
Imagine asking a new employee to "reply to this customer" without showing them any old emails or company rules. They guess. Now paste in 2 emails you've already sent and say "use this tone": they get it right on the first try. AI works the same way — an example is worth more than an explanation.
💡 Practical tip
Before asking for anything, gather 1 example of “good” + your 3 main rules. Paste them into the chat with the phrase "use this as a reference:". That extra minute saves several rounds of corrections later.
Key concepts
🔁 Ask, Review, Ask Again (Iterate)
Here’s the secret many people miss: the first result is almost never the final one — and that’s okay. The magic isn’t in the first try; it’s in the cycle of improvement. You look at what you got, point out exactly what’s off, and ask again.
The trick is to give specific feedback, not vague. Instead of “I didn't like it,” say “it's too formal; make it lighter and cut the second paragraph.” AI doesn't get offended or tired—it adjusts right away. Two or three rounds usually take it from “okay” to “perfect.”
🔄 The iteration cycle
💡 Practical tip
When something is almost good, don’t start over from scratch. Say "keep everything, just change ___". This preserves what already worked and focuses the AI on the exact point — saving time and keeping it from “ruining” the rest.
Key concepts
✨ Your first result in action
There’s a huge difference between “I planned something” and “I have something that works.” Something running is worth more than a thousand plans. It doesn’t need to be beautiful or complete—it needs to exist and work, even a simple one.
Why does this matter so much? Because the real result gives you honest feedback that no plan can. You use it, see where it gets stuck, and improve based on what’s concrete. It’s also your first proof that you can build — and that confidence changes everything.
🧪 Try it now (5 minutes)
Goal: leave here with your first guided build—a small “assistant” ready to use. Paste the text below into Claude or ChatGPT, replacing the parts between < >.
Quero construir uma coisa pequena e útil pro meu dia. PROBLEMA (a dor que se repete): <ex: toda semana resumo 10 e-mails de clientes na mão> RESULTADO QUE QUERO (o "como é o bom"): <ex: um resumo de 3 linhas por e-mail, em tom informal> REFERÊNCIA (use como exemplo): <cole 1 exemplo do que seria perfeito> Faça uma primeira versão. Depois me pergunte 1 coisa que ajudaria a deixar mais "a minha cara".
How to know it worked: you receive one first usable version and the AI ends by asking 1 follow-up question. Then you just review it and ask again — you’ve just completed a guided build.
Key concepts
⚠️ Common mistakes beginners make
Knowing the pitfalls saves hours. Almost every beginner stumbles over the same three obstacles: make a vague request, trying to do everything at once e accept the first result without checking. All three are easy to fix.
✓ The right approach
- ✓Ask with a goal, format, and example
- ✓Break it down: one small piece at a time
- ✓Check before using the result
- ✓Iterate with specific feedback
✗ The three pitfalls
- ✗Vague request — “give me a hand”
- ✗Large scope — all at once
- ✗Don't verify — trust blindly
- ✗Starting from scratch with every adjustment
🎯 The cure for the three stones
Just reverse each mistake: vague → describe the target; big → break it into small pieces; not checking → always review before using. They’re the same habits from the previous topics. Repeat them until they become automatic.
✋ Before you continue — what's the best way to start your first build?
🎓 Module summary
Next module:
2.2 — 🧩 Building with skills: package your way of doing things into a "recipe" that AI repeats the same way every time.