← Writing

MBA

First principles: a real lever with a real bill

"Reason from first principles" is the most quoted and least examined phrase in the founder canon. It is a genuine cognitive tool that occasionally produces a decade of advantage. It is also expensive enough that using it on the wrong problem is how smart people reinvent the wheel, badly, at great cost. The skill is not the reasoning. The skill is knowing which problems deserve it.

Two ways to know a thing

Aristotle defined a first principle as "the first basis from which a thing is known" — the bedrock fact that is not derived from anything more basic. Reasoning from first principles means decomposing a problem down to those facts and rebuilding upward, refusing to inherit any step you have not checked. Its opposite is reasoning by analogy: this new thing resembles that known thing, so treat it the same. Analogy is fast, cheap, and correct most of the time, which is exactly why it is dangerous the rest of the time.

Elon Musk's battery example is the cleanest illustration, whatever you make of the man. The received wisdom was that battery packs cost about six hundred dollars per kilowatt-hour and always would, because that was the market price. First-principles reasoning ignores the market price and asks a different question: what are the batteries made of, and what do those raw materials — cobalt, nickel, aluminium, carbon — cost on the commodity exchange if you bought them directly? The answer was a small fraction of the pack price. The gap between those two numbers was not a law of nature. It was an assumption everyone had inherited and no one had re-derived, and inside that gap was a company.

Why analogy fails exactly when it matters

Analogy is compression. It reuses a conclusion from a reference class instead of recomputing it, which saves enormous effort. The failure mode is subtle: it silently imports every assumption baked into the reference class, including the ones that no longer hold. When you say "SaaS companies price per seat," you inherit not just the pricing convention but the entire context that made per-seat sensible — a context that may not describe your product at all.

First principles earns its keep precisely where the inherited assumption is both load-bearing and wrong. That is a narrow target. Most assumptions you inherit are load-bearing and right — they encode real constraints, and re-deriving them just wastes a week to arrive where you started. The tool is a scalpel for the rare case where the crowd is confidently mispriced, not a hammer for every decision on the desk.

I think it's important to reason from first principles rather than by analogy. You boil things down to the most fundamental truths and then reason up from there.Elon Musk

The asymmetry that tells you when to use it

Here is the decision rule I actually use, and it is an asymmetry rather than a slogan. Testing an assumption is usually cheap. Rebuilding on a wrong new answer is usually expensive. So the question is not "is this assumption true?" but "what does it cost me to check, versus what does it cost me to be wrong?"

That reframing sorts the work:

  • Use first principles when the assumption is expensive to accept (it shapes the core of the product, the pricing, the architecture), the payoff of being right is large, and the crowd's answer is inherited rather than re-derived. High stakes, load-bearing, suspiciously unquestioned.
  • Don't bother when the decision is reversible and low-stakes. There, the convention is almost certainly fine, and the time you would spend re-deriving it is the real cost. Copy the best practice, ship, and move on.
  • Test before you rebuild. The cheap version of first-principles thinking is not tearing down the whole structure — it is running one small experiment against the load-bearing assumption. Buy the raw materials at spot once, on paper, before you start a battery company.

The optionality lens makes this concrete: prefer the reversible, cheap probe of an assumption to the irreversible commitment of a full rebuild. You want maximum information about whether the crowd is wrong for minimum sunk cost, and only then, if the probe pays off, the expensive reconstruction.

A worked decomposition: the pricing you inherited

Take an assumption most software founders accept without a second look: you price per seat. Everyone in the category does it, the billing tools default to it, and investors expect the metric. That is exactly the profile — load-bearing, inherited, unquestioned — that earns a first-principles pass, so run one.

Decompose what a price is actually charging for. In principle a price captures some share of the value the buyer receives, ideally scaled to how much they use. Per-seat pricing assumes value scales with the number of people who log in. Now check that assumption against your specific product rather than the category's. If your tool is one where a single administrator does the real work on behalf of a whole team — most of the "seats" never log in, or log in to read — then per-seat is not capturing value at all. It is a tax on the exact thing you want, adoption, charging the customer more precisely as more of their colleagues benefit. The inherited convention, re-derived, turns out to be pushing against your own growth.

This is where the discipline earns its reputation, and also where it is most dangerous, so apply the cost asymmetry. Do not rebuild your billing system on this insight. Run the cheap probe first: model three pricing structures — per seat, per unit of work done, flat platform fee — on a spreadsheet against your actual usage data, and call five real buyers to ask which they would prefer and why. That is an afternoon and a few calls. Only if the probe confirms that the inherited model is both wrong and expensively wrong do you pay the real cost of re-architecting how you charge.

The whole move fits the rule: decompose the assumption that is load-bearing and inherited without proof, test it cheaply, and rebuild only if the test pays. Applied to the one pricing question that shapes the business, that is leverage. Applied to every decision on the desk, it would be a way to never ship.

Where first principles becomes a trap

The phrase has become a costume for contrarianism, and the failure modes are as reliable as the successes are celebrated. Three deserve naming.

One: Chesterton's fence. G. K. Chesterton's rule is that before you remove a fence you do not understand, you must first learn why it was put there — because the reformer who cannot explain the fence is not yet qualified to tear it down. Most "best practices" are fences. They look arbitrary until you hit the specific failure they were quietly preventing. First-principles reasoning that skips this step does not liberate you from received wisdom; it rediscovers, painfully and at your own expense, why the received wisdom existed.

Two: it is expensive, and most invocations are rationalisation. True first-principles work is slow and effortful, and bounded rationality is real — you cannot decompose everything, so you necessarily run on trusted priors for almost every decision. When someone claims to reason from first principles on every topic, they are usually doing something cheaper: dressing up a contrarian instinct in prestigious language. The tool is rare in practice even for the people who genuinely have it.

Three: we only count the winners. Musk's re-derivations that worked are canonical; the confident first-principles bets that were simply wrong — the founder who "re-derived" a regulation that existed for a reason, or a market that wanted the conventional thing — do not get TED talks. The method does not guarantee the answer. It only guarantees you own the answer, right or wrong, which raises the stakes as much as the upside.

So decompose the assumption that is expensive, load-bearing, and inherited without proof — and copy the convention everywhere else. And before you tear down the fence, make sure you can explain, in one sentence, why someone built it.

Sources

  1. primaryAristotle, Metaphysics and Physics — the notion of a first principle (archē).
  2. primaryRené Descartes, Discourse on the Method (1637) — methodical doubt and rebuilding from certainties.
  3. primaryG. K. Chesterton, The Thing (1929) — the parable of the fence.
  4. secondaryElon Musk, TED / interview remarks on first-principles reasoning (2013 onward).
  5. secondaryHerbert A. Simon, work on bounded rationality — why we mostly reason by heuristic.
  6. secondaryCharlie Munger, on "circle of competence" — knowing where your re-derivation is trustworthy.