Atoms Don’t Have Reasons

Forward Deployed · Method

Atoms Don’t Have Reasons

Every forward deployed engineering page now carries the same bullet: build from first principles. Nobody has written down what that means inside a customer’s building. Here is what it means, why it breaks when used alone, and the two directions it needs beside it.

A wireframe building on a blueprint grid with three beams of reasoning, cyan from below, amber from the side and coral from above, each striking a different spot inside.

Apart: three lenses, three answers. Each is right about something.

  • First principles: up from the parts
  • Analogy: across from today
  • Axiom: down from the outcome
A customer’s building, with the fences someone put up inside it. Each direction of reasoning lands somewhere different. Slide them together: where they meet is the first move.

You walk into a stranger’s building with a laptop and a badge that says Visitor. By ten o’clock someone has shown you a spreadsheet four people touch every morning, and said the sentence you came to hear: we have always done it this way.

01 · The bulletEveryone says it. Nobody explains it.

OpenAI’s new deployment company describes how its forward deployed teams work in four bullets. The first is “Build from first principles.” Palantir invented the forward deployed engineer around 2003, and now every frontier lab has a team of them. Nearly all of them reach for the same phrase.

I went looking for the essay that explains it, and did not find one. OpenAI’s head of forward deployed engineering wrote a good piece in October about how his team works. The phrase is not in it. His team’s motto runs the other way, from the outcome down: start with the outcome, do what it takes, then scale. The best practitioner essay I read comes from the man who trained a generation of Palantir’s engineers. It describes watching live systems and mapping a business as its nouns and verbs, and never uses the words. A First Round roundup praises engineers who are “not doing a lot of pattern matching.”

In practice, “first principles” has come to mean one thing: do not walk in with a template. That is a good instinct. It is not a method. So here is the method.

02 · The principleWhat a first principle actually is

Aristotle called it the arche: the first thing from which something is known. You stop asking why when you reach something that does not need a reason. Elon Musk made the phrase famous by pricing a battery from the metals inside it instead of from what batteries cost. Strip a thing to the parts that cannot be reduced, then build back up without borrowing anyone’s assumptions.

It works beautifully on matter. A lithium atom has no reason for being in the battery. It has a price, a weight and a chemistry. Nothing about it was decided.

A customer’s operation is the opposite. Almost nothing in it is raw material. Every form, handoff, double check and odd spreadsheet is a decision somebody made, usually for a reason nobody wrote down.

Atoms don’t have reasons. Workflows do.

03 · The fenceThe reformer with a bulldozer

G. K. Chesterton gave the warning in 1929. A reformer comes across a fence in the middle of a road, sees no use for it, and wants it cleared away. The wiser one tells him: go and find out why it was put there, and then we can talk about taking it down.

First principles, used alone inside a business, is the reformer with a bulldozer. It reduces the morning spreadsheet to its parts, sees no purpose, and builds something clean. The clean thing ships, and quality quietly falls, because that spreadsheet was the only place anyone checked the data. The best practitioner essay tells nearly this exact story. An engineer kept defending a clumsy file format with reasons that kept changing. Only by watching her work did the team learn the format was her quality check. The fix was not a better argument. It was a viewer that let her keep checking.

The opposite mistake is just as common. Analogy, reasoning across from how things are done today or how the last customer did it, respects every fence, including the dead ones. It produces exactly the template the bullet was written to forbid.

Neither direction tells you what to build. Bottom up tells you what is possible. Across tells you what is true today. Neither tells you what should be.

04 · The third directionDown from the outcome

That is the job of the axiom. Write the outcome as if it were already true, and assume it. The claim is paid the day it arrives. The dinner is in your stomach, not just ordered. Then ask the only useful question: why isn’t it so? The evidence for why it isn’t so is the diagnosis. Fix that, and it becomes so.

  • First principlesBottom up. Finds what the fence is made of.
  • AnalogyThe current state. Tells you why the fence is there.
  • AxiomTop down. Tells you whether the fence is in the way.

I teach these as three directions, and the rule is to run all three at once, not one after another, and let them argue. Where they agree, the matter is settled. Where they disagree is exactly where to look. And the argument discovers something no single lens would: the first move, which is almost always the cheapest piece of evidence that would move the most. In the worked example I published in October, three lenses on a merger press release did not prescribe a system. They discovered that the two CEOs should call their top customers.

That is what “build from first principles” should mean in a stranger’s building. The bottom-up lens, held in tension with the fence and the outcome.

05 · The turnWhen building is free, direction is the product

Now change one input. Building used to be the expensive part of an engagement. It is becoming the cheap part. In that same October essay, OpenAI’s FDE lead tells of a team member asking what happens when Codex can produce, in one shot, what they had spent months building. His answer was that making themselves obsolete is the job.

Follow that to the end. If building is nearly free, the forward deployed engineer is not selling the build. He is selling the direction: which fence to keep, which to take down, and which thing to build first.

The machines change the method too. A human team runs the three directions in series, in meetings, and the loudest voice wins. A delegate can run all three in parallel, each lens its own reasoner, and have them negotiate overnight against the evidence gathered that day. What stays human is walking the building, seeing the fence with your own eyes, being the face the customer trusts, and deciding what is left in the bin the lenses cannot settle.

You walk into the same building. Your three delegates argued all night over yesterday’s interviews, and they agree on one thing: do not touch the morning spreadsheet yet. Call the two people who read it and ask what they would miss.

06 · PredictionsWhat I expect to see

  1. The bullet becomes a method.

    Within three years at least one major forward deployed organization publishes a named reasoning method in which first principles is one lens among several, not the whole instruction. The phrase on its own will start to read like “synergy.”

  2. Fence inventories come before builds.

    Deployments will open with a written list of every workaround in the operation and the reason it exists, before a line of code. Teams that skip it will keep shipping clean systems that quietly break quality.

  3. The outcome gets written first, as if done.

    Requirements documents give way to a finished-state artifact, a press release or a plain statement of what done means, and the whole engagement is reasoned down from it.

  4. Engineers are judged by their first piece of evidence.

    When code is free, the scarce skill is finding the cheapest thing that would change the plan the most. Hiring loops will test for it directly, the way Palantir once handed candidates an open problem and watched what they asked first.

  5. Small businesses get forward deployed.

    The method scales, not the headcount. One person with three reasoning delegates will give a neighborhood restaurant the kind of engagement only a large enterprise could buy before.

07 · Living itSmaller buildings, same fences

At Charleston AI we deliver finished AI work to hospitality businesses and local owners, which means I am forward deployed every week, just in smaller buildings. The method is the same. When I ran first principles on a restaurant’s food cost, it came apart in my hands: most of what everyone calls food cost is not food. It is diesel, distribution, storage and layers of margin between the farm and the plate. That is the bottom-up lens doing its job. It is also the exact moment a bulldozer would start swinging, and the exact moment the other two lenses earn their keep. I teach all three to my advanced students as one practice, run in parallel, and I wrote the worked loops up in Why Isn’t It So?, Rinse and Repeat and Settled, Resolved, Yours.

08 · The ledgerHow this could go

Already true
Every frontier lab runs a forward deployed team, “first principles” is on their pages, and no one has published what it means in practice. The labs’ own leaders say AI is making their months of custom building obsolete.
What has to happen
Building keeps getting cheaper until the scarce part of an engagement is deciding what to build, and these organizations notice that what they really sell is a direction of reasoning.
Where I am probably wrong
If operations turn out to be documented well enough that the fences explain themselves, if a model can read the systems and recover every reason on its own, then plain first principles is enough and the other two lenses are overhead. I don’t think that happens soon. Most reasons live in people.

Back to Monday, 8:10. The spreadsheet four people touch every morning is still there. Don’t tear it down. Don’t copy it either. Write down what the finished day looks like, and ask why it isn’t so. Then ask who would miss the spreadsheet.

Someone built that fence. Go find them.

Background

Author: John Rector

John Rector is a Charleston-based entrepreneur, author, and AI strategist. He co-founded E2open, the supply-chain software company acquired for $2.1 billion in 2025, and in 2026 opened Charleston AI, a 3,000-square-foot lab that helps people and organizations understand and use artificial intelligence. He is the creator of The Reality Equation — a lecture series, book, and curriculum exploring attention, prediction, and how reality is experienced — and the author of more than two dozen books. He writes and speaks widely on artificial intelligence, attention, and the future of human work.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.