The Field Guide Nº 01–04

The Four Layers of Building With AI

Getting real work out of an AI comes down to four things, each one wrapped around the one before it. Start in the middle and build outward.

04 Loop keeps it going
03 Harness the machinery
02 Context what it can reach
01 · Prompt
The words

Each layer wraps the one inside it. Tap a frame to jump.

01
Layer 01

PromptHow you ask

The words you use. Ask clearly, say what kind of answer you want, and show an example to copy. Same AI, far better answer.

Like briefing a sharp new hire. The clearer the ask, the better the work you get back.

Vague“Write about our product.”
Clear“In three sentences, tell a busy plant manager how this saves them time. Skip the jargon.”

Skip itVague asks get vague answers, shaped nothing like what you pictured.

02
Layer 02

ContextWhat it can reach

What the AI can see and get to. The facts and files you put in front of it, plus which tools and data you let it use. It can only work with what it can actually reach.

Like the difference between asking someone cold and handing them the file and the keys first.

No access“Is this invoice overdue?”
Given access“Here’s the invoice, today’s date, and our net-30 terms. Is it overdue?”

Skip itIt fills the gaps by guessing. The question was fine, the facts were missing.

03
Layer 03

HarnessEverything around it

The system built around the AI that makes it actually work. It runs the tools, checks the results, keeps things safe and on track, and hands off work to helpers. Not the AI, and not the tools, but the machinery that drives them.

Like the office around a new hire: the power, the printer that really prints, the safety rules, and the manager checking the work.

Just the AIDecides what to do, but nothing actually happens.
With the machineryRuns it for real, catches mistakes, and won’t let it send the wrong thing.

Skip itGood ideas that never safely turn into real actions.

04
Layer 04

LoopKeeps it going on its own

The outer cycle that lets the AI keep working on its own, turn after turn, on a schedule or until the job is done. It picks the next step, checks its own work, and stops when the goal is met.

Like the difference between someone who nails one task and a teammate who owns the whole job, week after week.

One and doneDoes the job once, then waits. If it trips, it’s stuck.
Runs itselfKeeps going, fixes its own mistakes, and only asks for help when it truly needs it.

Skip itGreat demos that never turn into something you can rely on.

Putting it together

How they stack

Each layer only works if the one inside it does. A loop is useless if the machinery is broken. The machinery is useless if the AI can’t reach the right facts. And none of it matters if you asked the wrong thing. So start in the middle and build your way out.

01 Prompt› 02 Context› 03 Harness› 04 Loop