Two weeks ago we talked about what artificial intelligence actually is: a very fast helper who read every manual for your trade but never once set foot on your job site. This week we get practical: how do you talk to it so it actually does something useful? The answer has a technical name — prompt — but the idea behind that word is one you've known for years, even if you never called it that.
A prompt is a work order. Full stop.
Think about the last time you sent someone on your crew to do a job right, without having to fix it after. That wasn't luck. You told them what to do, with what, and how you'd know it was done. "Cut the lawn to two inches, edge the sidewalk, bag the clippings, and haul them out before five today" isn't a casual sentence — it has a task, a standard, and a deadline. That's why it works.
Now think about the opposite order: "fix up the yard." You come back two hours later with no idea what you'll find. Maybe it's perfect. Maybe they just swept the entryway. The problem wasn't the person doing the work. It was the order.
The exact same thing happens with an AI, just faster and without the confused look on its face. An AI is never going to ask you "sorry, what do you mean by fix up?" It will assume something — anything — and hand you a result with total confidence. If your order was loose, the result will be loose too, just written with a confidence that can trick you into thinking it's right.
Here's the before-and-after I want you to remember.
Before — five loose points, exactly as you'd jot them in your notebook before getting in the truck:
That's what most people paste into a chatbot: the note exactly as it came out of their head. And the result they get back is just as disjointed as the note — because the AI has nothing to fill the gaps with.
After — the same information, turned into a work order:
"Act as my quoting assistant. Using this information — backyard, 400 square feet, artificial turf installation, concrete border around the perimeter, client: Mrs. [last name], Miller Street — write a professional quote proposal in English. Include: a two-line job description, materials and labor broken out separately, an estimated installation timeline, and a one-year warranty note. Format: headed list, ready to paste into an email. Do not invent prices; leave the field '[price]' wherever a dollar amount belongs, because I'll fill those in myself."
Notice what changed. You didn't add new information — the same five facts are still there. What you added was structure: who the AI should act as (your quoting assistant), what it needs to produce (a proposal, not a reply), what format it belongs in (headed list, ready to paste), and an explicit rule about what it should NOT do (invent prices). That last part matters as much as the rest: an AI without limits fills gaps with whatever seems plausible, and "plausible" is not the same thing as "correct."
That example took fifteen seconds to write and saves you twenty minutes of drafting the proposal by hand — and, more valuable still, it hands it back in a format you can paste straight into the client's email without rewriting a word.
A lot of people stay away from this because they think it takes "knowing computers." It doesn't. A prompt isn't code — it's everyday language, the same kind you'd use to leave instructions for a new hire on their first day. The skill this takes isn't technical: it's the same skill you already use to direct people. It's called giving clear instructions, and some people have it more polished than others, but everyone practices it daily with their crew, their family, or their vendors.
The only thing that's different is that skill now pays off much bigger, because you explain something to an employee once a day, and you explain it once to a well-instructed system and it obeys you a thousand times, in two languages, at nine at night, without getting tired or billing you overtime.
When something comes out wrong with an AI, before you blame the machine, check whether your order had these four parts. It's almost always one of these.
One — the role. Tell it what role to play: "act as my quoting assistant," "act as the editor of my posts," "act as whoever drafts follow-up emails to late-paying clients." Without a role, the AI answers generically, like an employee with no assigned position.
Two — the inputs. The real information from your business: measurements, names, reference prices, policies. This is where what you already know about your trade comes in — the part the AI doesn't know going in.
Three — the output format. A list? Running paragraph? A table? Ready to paste into an email or a text message? If you don't say, the AI decides for you, and it almost never decides the way you would have wanted.
Four — the limit. What it should NOT do: don't invent prices, don't promise dates, don't use a certain tone, don't go past a certain length. This is the part people forget most, and the one that saves the most headaches.
These four parts are, without naming it yet, a preview of a fuller framework you'll meet next week: the 5 Pillars I use to fire off any task, from a quote to an entire system. For now, these four are enough to stop getting generic answers back.
You don't need a new tool or a single dollar spent. Take the next piece of writing you were going to do anyway — a quote, an ad, a difficult email — and before you sit down to draft it yourself, write the work order first: role, inputs, format, limit. Paste it into whatever AI assistant you already have on hand and compare what comes back against what you would have written in the same amount of time.
Most people who run this exercise once stop talking to an AI like it's a search box. They start talking to it for what it actually is: someone you're handing a work order, and the clarity of that order decides everything you get back.
See you next Wednesday, when we put a full name on this method.
This is exactly what we teach —and build with you— in Cohorte 2026. You walk out with a finished AI project that Ocampo-Infra can run for you. Grab the free guide and claim your seat.
Roberto Ocampo · ro@ocampo.ai · https://ocampo.ai