01 Principles
The first thing.
A principle is not a rule someone hands you. The word remembers what we forgot: principium — the beginning, the first part, the first thing. "First principles" is a quiet tautology. You are already being told: go back to the start.
Latin root
principium — a beginning, the first part
From princeps: primus (first) + capere (to take). The first to take hold. Principle already means first. "First principles" is the tautology that points at the problem: we stopped going there.
Purpose
You are about to decide by copying.
The default move is analogy. Find the best existing example, improve it one notch, ship it. Fast, defensible, and quietly wrong: it can only ever reach one step past what already exists.
The category leader solved it. Copy their model; improve it one notch.
Their success was based on copying the previous leader, refined.
Each layer looks like evidence. The stack is confident and brittle.
Analogy's ceiling: a slightly better version of what already exists.
Pattern
Each 'why?' costs something.
Decomposition is the method. Each rung asks: why is that true? The claim either survives — it resolves to a measurement or an axiom — or it cracks open and reveals what was borrowed underneath it. Select each rung to descend.
Because they proved it works — revenue, retention, the product-market fit signal you want.
It might not be. Their market matured. Their timing is past. Their constraints shaped the solution.
It isn't — unless the underlying condition is the same. What is the underlying condition?
Now we are asking the real question. Is that pain present? Do you have that distribution? If yes — derive from it. If no — the copy is wrong.
An axiom or a measurement — the one thing that holds without borrowing from something else.
You know you have hit bedrock when the next "why?" produces a measurement (observable, testable) or an axiom (irreducible — denying it requires using it). Everything above is derivable from this. The copy cannot get here; it stops one step above.
Problem — Potential
Two ceilings. One is load-bearing.
Analogy reaches one step past the existing. Principles reach what does not exist yet. The asymmetry is not about effort — it is about what is possible at all. You cannot copy your way to a build the copy did not contain.
The same decision, reasoned two ways:
Notion, Linear, and Figma all have community — it drove 40% of their growth. Build it.
One step past what already exists: a community feature that looks like theirs, built for a market you haven't verified shares their conditions.
The underlying condition for community-led growth: high-frequency use + user-generated value + a switching cost from peer connections. Do our users use the product daily? No — once a quarter. Do they generate value others need? Rarely. The condition does not hold. Community is wrong. The root permits: instead build a benchmarking tool — the one thing our users actually want to share.
Something the copy cannot reach: a benchmarking tool, built for the condition that is actually present.
Decision: Should we build a community feature?
Perspective
AI made copying free. That changes the edge.
AI is the ultimate analogy engine. It interpolates everything ever written — instantly, fluently, at near-zero cost. Every surface reasoning move is now free. So the human edge collapses onto exactly what AI cannot do for you: derive from the root, decide what is true here.
The appreciating skill: as the cost of cognition falls, copying commoditizes. Root-derivation is where the remaining edge lives. Not because it is harder — because it is the move that produces what does not yet exist.
The kill condition for this page: if root-reasoning never yields a non-obvious move that analogy would have missed, it is intellectual theatre. A reader who runs the decomposition and reports no difference falsifies the claim.
Progress
The build from bedrock.
Once you stand on the root, you build up. Not by copying the next layer — by asking what the root permits that nothing built yet has expressed. The decomposition tool below is the gauge: the page works when you can fill all five rungs on a real decision and name the different move.
Your dig
Run one decomposition on a real decision you are about to make.
The page works when you can name all five. If you reach bedrock and the build differs from your copy default, the method paid.