A model for AI development

FIRE

Focus · Intent · Reality · Execution — the four elements that make AI work catch.

An agent's output is only as good as what fuels it. Three elements aim the work — Focus, Intent, Reality. The fourth, Execution, fires it. Drop one and the work doesn't catch: it stalls, or it burns the wrong thing.

IGNITION · ELEMENT 04 / 04

FOCUS agent · attends INTENT developer REALITY environment EXECUTION

The failure you recognize

It compiles. The tests pass. It's elegant.

And it's wrong.

It built the wrong thing, or against a fiction, or it lost the plot — or it did the right thing and ran the command that couldn't be undone. The model reasoned fine. The work didn't catch — or it caught the wrong fuel.

The model

Three elements aim. One fires.

Combustion needs four things at once — fuel, heat, oxygen, and the chain reaction. So does AI work. Focus, Intent, Reality are the grounding — they aim the work. Execution is the act — it turns a grounded aim into a change in the world. Remove any one and there's no fire.

ELEMENT 01 — FOCUS

Focus

The heading. What it's attending to now.

grounding — the agent

ELEMENT 02 — INTENT

Intent

The target. What success means.

grounding — the developer

ELEMENT 03 — REALITY

Reality

The terrain. What is actually true.

grounding — the environment

ELEMENT 04 — EXECUTION

Execution

The act. Carrying the work out.

the spark — the agent acts

The claim

The four elements are not alike — and they fail in different ways.

Realitycostly Observable. Pay to look. Scales with money — and it's getting cheaper. ↘ trending toward free
Intenthard Can't be retrieved if never expressed. Scales with discipline, not compute.
Focusfragile Gets worse as you spend more — more context, more room to get lost. Scales with discipline.
Executionirreversible The only element that changes the world. Aim is recoverable; a fired shot is spent. Scales with care — verification and reversibility.

Reality scales with money. Intent and Focus scale with discipline. Execution scales with care.

ELEMENT 01 / 04 — Focus

Aim — the heading

What the agent is attending to, out of everything it could. It can fail when everything else is perfect: rabbit holes, forgotten constraints, over-engineering, drive-by changes, quitting early — or never stopping.

"Focus gets worse as you spend more."

More context is more room to get lost; more autonomy is more chances to drift. The resources that buy Reality grounding actively erode Focus — they pull opposite ways. It's the agent's discipline, but you design the conditions.

ELEMENT 02 / 04 — Intent

Aim — the target

What you actually want. It lives in your head — partial, mutable, pre-verbal. The agent reads the shadow you wrote down. The gap between what you meant and what you said is where it fails.

"It built the wrong thing, correctly."

Hard, not costly: humans under-specify, assume shared context, change their minds, and don't know what they want until they see what they don't. A better model can draw intent out of you — it can't manufacture what you haven't formed. Author it: task → goal → constraints. The highest-leverage move is to convert intent into reality — encode it as a test.

ELEMENT 03 / 04 — Reality

Aim — the terrain

What is actually true. It invents an API that doesn't exist, breaks a contract it never read, assumes a config it never opened. We call it hallucination and blame the model.

"Hallucination is un-paid-for observation."

Reality is observable — so it's costly, not hard: effort buys success, you just have to pay. Every bit you don't let the agent observe, it will invent. And this is the element that's getting solved — use the tools, but don't build your career on it.

ELEMENT 04 / 04 — Execution

Fire — the act

The first three aim the work. Execution fires it — it turns a grounded plan into a change in the world. And it can fail when your aim is perfect: the right idea, carried out carelessly.

"Grounding is recoverable. Execution isn't."

Execution is the only element that changes the world — so it's the only one with irreversible stakes. A wrong belief about Reality costs a re-read; a wrong Execution can cost the database. Failures look like the destructive command, the half-applied change, the missing verification, the careless tool. The discipline isn't capability — models act fluently — it's care: verify what you did, prefer reversible steps, and confirm before anything you can't take back.

The diagnostic

Which element failed?

When an agent fails, don't ask "is the model good enough?" Ask which element was missing — each has a different remedy.

Looks like…Missing elementRemedy
Lost the plot — rabbit hole, forgot a constraint, over-built. Focus Decompose · checkpoint & re-state · prune · define "done".
Built the wrong thing, correctly. Intent Tighten the spec · restate · show an example.
Built against a fiction — wrong API, broke a contract it never read. Reality Give access · verify before acting · lean on contracts.
Did it badly — destructive command, half-applied, can't undo it. Execution Verify after · reversible steps · confirm destructive ops · smaller increments.

FIRE isn't four features.
It's a discipline.

Three elements aim the work; the fourth fires it. As models improve, Reality grounding approaches free — the enduring craft is articulating intent, defending focus, and executing with care.

Focus you defend. Intent you author. Reality you observe. Execution you verify.