The Prompt Was Never the Problem
Every few weeks someone in my family or a friend asks me for help with AI. What tool should I use? How do I write a better prompt? Why doesn't it do what I want?
After enough of those conversations, the pattern became impossible to miss. Almost nobody was actually stuck on prompting. They were stuck one step earlier — on describing what they wanted.
The request would start confidently and then dissolve. "I want it to organize my client stuff." Organize how? Which clients? Starting from what — a spreadsheet, an inbox, a pile of PDFs? Ending in what? The moment those questions landed, the person would realize they hadn't decided any of it. They'd been asking the model to make a decision they hadn't made themselves.
That's the whole thing:
You can't ask an AI for something you don't know what it is.
No prompt template fixes that. So I stopped giving out prompt tips and built the material I actually wanted to hand people.
Two pieces
The workshop — A Jornada de Luke
A full session framed around Luke: someone who uses AI every day and is still stuck at the surface. He knows what a prompt is but. The image won't obey. Every AI behaves differently. He wants to automate the repetitive parts and build agents, and he can't get past the first step.
It works through the parts that actually matter:
- What the thing is — AI vs. generative AI vs. model vs. app, which get used interchangeably and aren't the same.
- Why it behaves the way it does — it predicts rather than understands, which is why it's non-deterministic, why it never says "I don't know," and why garbage in is garbage out.
- Tokens and context — what the context window is, why long sessions degrade, and why one task per conversation beats one endless thread.
- Getting it out of your head — turning implicit intent into something explicit, where the gaps become visible.
- Splitting a problem into named roles — instead of one giant request, pieces with a role, a responsibility and a boundary.
- C-T-C-F — Context, Task, Criteria, Format. The structure that turns a vague ask into a specific one.
The order is deliberate. C-T-C-F sits near the end, after the thinking work. Handing someone a prompt framework before they can describe their goal just produces a well-formatted vague request.
The exercises — Missão Padawan
The workshop alone doesn't stick. Reading about clarity feels like acquiring it, which is a trap. So the second piece is five exercises done against your own real task:
- Explain it to a puppy — say it in plain words, no jargon. If you can't, you don't have it yet.
- Draw the flow — what goes in, what happens, what comes out.
- Split it into roles — break the giant into named pieces.
- Write it in C-T-C-F — twice. Version A the way you'd have written it before, Version B structured. Comparing them is the lesson.
- Feel the weight of context — see what actually happens as a session fills up.
Exercise 4 is the one that lands hardest. Putting your old attempt next to the structured one makes the gap obvious in a way no explanation does.
Who it's for
People who use AI and sense they're scratching the surface. No technical background needed — there's no code in either piece.
If you already write structured prompts and think in terms of context budgets, this isn't aimed at you. It's aimed at the person you keep explaining this to.
A note for English readers: both pieces are in Brazilian Portuguese, since I wrote them for the people who kept asking me. The ideas travel, but the pages don't — yet.
What I learned building it
The instinct is to lead with tools and prompt tricks, because that's what people ask for. It's the wrong order. Tool-first teaching produces someone who can copy a template and is stuck again the moment their problem changes shape.
The uncomfortable part of teaching this is that most of the work isn't about AI at all. It's decomposition, explicit thinking, and knowing what "done" looks like — the same skills that make someone good at briefing a contractor or writing a ticket. The model just makes the absence of those skills expensive and immediate.
Which is the actual reason it's worth learning. AI is an accelerator: it multiplies what you already have. If what you have is a vague idea, it will multiply that too, very quickly, into something that looks finished and isn't.