Editorial series · 15 weeks · Week 5

The Brain Behind the Magic: LLMs in Plain Trade Language

Roberto Ocampo September 16th 2026 Course / Fundamentals

Last week I gave you the 5 Pillars: the protocol for asking an AI for anything and getting it right. But I never told you what that "AI" actually is, the thing you're asking. Today we pop the hood. Not so you walk out knowing how to code — so you know, once and for all, what engine you're actually dealing with, what it can really do, and what you should never ask it for.

It's called an LLM, short for "large language model." Sounds like a lab term. Let me translate it into your trade.

What it actually is, without the jargon

An LLM is a text engine. You give it words, it hands you back words. That is, at bottom, everything it does: it read an enormous amount of text — books, manuals, conversations, contracts, emails — and learned the patterns of how one idea connects to the next in your language. When you write to it, it isn't "looking up the right answer" in a database the way a calculator looks up a result. It's predicting, word by word, the most reasonable continuation of what you gave it, based on everything it read before.

That explains two things at once: why it's so good at writing, summarizing, and translating — tasks that are, literally, "produce the most reasonable continuation of this text" — and why it sometimes hands you something that sounds perfect but is wrong. It didn't lie on purpose. It found the most probable continuation, and this time the most probable one wasn't the correct one.

The trade analogy

Think of the LLM as the best-read employee you've ever hired, who has never once set foot on your job site. It read every trade manual, every template contract, every style guide that exists on the internet. Ask it almost anything about the trade in general and it'll hand you a solid, well-written, correctly structured answer.

But it doesn't know you already promised Mrs. Miller you'd finish by Friday. It doesn't know your gravel supplier raised prices last week. It doesn't know that particular client complains when an email sounds too formal. All of that — the specifics of your business, your client, your week — you have to hand it yourself. That's why Pillar 2 (Inputs) from last week isn't an optional detail: it's literally the difference between the well-read employee who's never been to your site and one who knows exactly what ground they're standing on.

What you can actually expect

  • Writing. Emails, proposals, descriptions, posts — anything that starts blank and needs to become organized text, the LLM does fast and well, especially if you gave it all 5 Pillars.
  • Summarizing. Hand it a ten-page contract, a three-week email thread, notes from a call, and it hands back the essentials in a paragraph. This alone, for a business owner who has no time to re-read everything, already justifies the engine.
  • Translating. Between languages, but also between "levels" of the same language: turning your quick shorthand note into a professional email for the client, or a contract full of legal terms into an explanation your client actually understands.
  • Restructuring. Taking five loose points — the ones you jotted down in the truck, between jobs — and handing you back a proposal with the order and tone a client expects to see.

What you should not expect

  • That it knows your business without being told. It has no access to today's prices, your client history, or what happened on the job site yesterday. If you don't hand it over as an input, it doesn't exist to the model.
  • That it's never wrong. It can sound completely confident and still be wrong — especially with exact numbers, dates, or very specific details you didn't give it. Always check the numbers before you send them to a client.
  • That it remembers last week's conversation unless you hand it back as an input, or unless you have a system — like the ones we build at Ocampo-Infra — that remembers it for you. That's what we're covering next week.
  • That it decides for you. It gives you options, drafts, analysis. The signature, the final call, the responsibility to the client — those stay yours.

The engine isn't the business

Here's what I hear most from business owners once this actually clicks: the relief of knowing they're not dealing with something magical, or something coming to replace them. They're dealing with a very well-read text engine that does exactly what you ask it, at the quality of what you gave it. The engine doesn't replace your trade judgment — it multiplies it. You're still the job-site foreman; the LLM is the crew member who drafts fast and never gets tired, but needs clear instructions or it gets lost.

And this connects straight back to last week's 5 Pillars: now that you know the engine "predicts the most reasonable continuation" from what you handed it, you understand why every missing pillar is one more chance it guesses wrong. It's not a moody AI. It's a pattern engine that was missing information, and it filled the gap with the generic.

A trade example

A painter emailed me recently with a familiar complaint: "I asked it to write me a proposal for a 3-room job and it gave me something generic, with prices that aren't mine." That's the engine working exactly as it should — without the inputs (his prices, his materials, how he charges for labor), it filled in with what's common across the trade in general. Once he gave it his own numbers and his own quote template as inputs, the next proposal came out ready to send. The engine didn't change. What changed was what he fed it.

This week's practice

Take a writing task you do often — a proposal, a follow-up email, an answer to a common question — and run it twice. First, give it nothing but the bare instruction. Then, give it your own prices, your own tone, a real example of yours as a template. Compare the two results side by side. That's where you'll see, with your own eyes, the difference between the engine alone and the engine with your trade built in.

Next week we take the next step: how to make the AI actually remember your business — your prices, your past quotes, your manuals — without repeating yourself every time. That's called RAG, and it's your company's memory living inside the machine.

Want to build it with us?

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