POINT 01 / 03 — REALITY
Reality
The terrain. What is actually true.
Source — the environment
A model for AI development
Reality · Intent · Process — three points of contact. Hold the work.
An agent's work needs traction. Without grip it slides — fluent but unanchored, busy but un-held. Two points aim each step; the third — Process — is the loop that keeps you in contact across many.
CLAMP · POINT 03 / 03
The failure you recognize
It compiles. The tests pass. It's elegant.
And then the next step undoes it.
Every individual step looked fine. But across many steps the work slid — and you never noticed because each handoff was plausible. The model reasoned step-by-step. The loop wasn't there to catch it.
The model
Borrow the word from the shop floor. A three-jaw chuck needs all three jaws to hold the work square — one slack and it spins out. Borrow it from climbing too: the three-point rule. Two contacts aim; the third is what keeps you on the wall while you reach.
POINT 01 / 03 — REALITY
The terrain. What is actually true.
Source — the environment
POINT 02 / 03 — INTENT
The target. What success means.
Source — the developer
POINT 03 / 03 — PROCESS
The loop. How the work stays held across steps.
Source — the rhythm you put in place
The claim
Reality scales with money. Intent scales with discipline. Process scales with rhythm.
POINT 01 / 03 — Reality
It invents an API that doesn't exist. Wrong signature. A style you abandoned years ago. It breaks a contract it never read. We call this hallucination and blame the model.
"Hallucination is un-paid-for observation."
Reality is observable — the code is right there; the test passes or it doesn't. So it's costly, not hard: effort buys success, you just have to pay. Every bit of reality you don't let the agent observe, it will invent. And this is the point that's getting solved — use the tools, but don't build your career on it.
POINT 02 / 03 — Intent
Intent 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: convert intent into reality. A test or a type is intent pre-compressed into an observable form — it moves a slice of what you want from the side money can't buy onto the side it can. When you re-explain the same intent a third time, stop explaining and encode it.
POINT 03 / 03 — Process
Reality and Intent aim each step. Process is what holds the work across many: the cadence of verifying, the size of the increment, the route back if the step was wrong. It's the only point that lives between actions instead of inside one.
"A loop you can recover from beats a sprint you can't."
Process is slow because it compounds. A single careful step doesn't look like much; a hundred careful steps in a row is the difference between work that holds and work that has to be done again. The failure mode looks like progress — five green commits, then a sixth that quietly broke the first — because no checkpoint asked "is this still the thing we wanted?"
Models are getting faster at each step. Speed without a loop just races you to the wrong destination sooner. Process is the discipline of making each step small enough to verify and reversible enough to back out of — so the next step starts from solid ground.
The diagnostic
When an agent fails, don't ask "is the model good enough?" Ask which point lost contact — each failure has a different remedy.
As models improve, Reality grounding approaches free. Intent stays your job — what you want, only you can author. And Process is the loop that holds it all together one step at a time. The enduring craft is the rhythm you build around faster and faster steps.
Reality you observe. Intent you author. Process you hold.