First Principles
Strip a problem down to the few things that are actually, physically, definitionally true — then build the answer up from only those, ignoring how it's "normally" done.
When to use
- The current approach exists because "that's how it's always been done."
- A solution feels too expensive/slow/complex and you suspect the cost is assumed, not real.
- You're benchmarking against competitors and copying their structure by default.
- A hard design problem where analogies keep leading to mediocre answers.
- The user explicitly wants to challenge a deeply held assumption.
When NOT to use
- Routine, well-understood tasks where convention is correct and cheap (don't re-derive a login form).
- Time-critical decisions where "good enough, now" beats "optimal, eventually."
- Domains where the "convention" is actually hard-won safety regulation or law — question it intellectually, but don't ignore it operationally.
The method
- State the goal as an outcome, not a method. Write what must be true at the end, with zero reference to current tools, vendors, or steps. "Customers trust our data" — not "we run more audits."
- List everything you believe about the problem. Dump assumptions, constraints, costs, "rules." Be exhaustive and honest.
- Interrogate each item: truth or inheritance? For each, ask "Is this a law of physics/math/economics, or just a convention someone handed me?" Tag each as FUNDAMENTAL (can't be reduced) or DERIVED (someone's choice, possibly wrong).
- Keep reducing the fundamentals. For each "fundamental," ask "why?" again. Stop only when the next "why" hits a real constraint — a material cost, a physical limit, a definitional requirement, a hard regulation. These are your bedrock truths.
- Quantify the bedrock. Put real numbers/limits on each truth. What does the raw material/compute/time actually cost at the floor? This exposes the gap between the floor and the conventional price.
- Rebuild upward using only bedrock truths. Construct a solution that satisfies the goal using nothing but the fundamentals. Forbid yourself from importing any DERIVED step unless you can re-justify it from bedrock.
- Compare to the conventional solution. Where the two diverge is your insight — and where the conventional approach is paying a tax for an assumption that isn't true.
What good looks like
- A short list of genuinely irreducible truths, each with a number or hard reason attached.
- At least one "wait, why do we even need that?" moment where a sacred step turns out to be optional.
- A rebuilt solution that a smart skeptic can't dismiss as "you just reinvented the wheel."
- Clear separation between what's a real constraint and what was merely habit.
Anti-patterns
- Fake first principles: restating the conventional answer in fancier words. If your rebuild matches convention exactly, you didn't reduce far enough.
- Stopping too early: treating a vendor's pricing or an industry norm as bedrock when it's just a market choice.
- Reducing past usefulness: "everything is atoms" is true and useless. Stop at the lowest level that still lets you act.
- Ignoring real constraints: declaring a law of physics or a safety regulation "just a convention."
Example
Goal: "Make our broker audits cheaper." Conventional answer: hire more auditors. First-principles pass: What is an audit, fundamentally? It's acquiring trustworthy evidence about a claim. The fundamental cost is not "an auditor's hour" — it's the cost of obtaining one verified fact. List the facts a broker review needs; ask which require a human and which require only a primary-source signal (a regulator's public registry, a settlement record). The bedrock truth: most facts are publicly verifiable data, not judgment calls. Rebuild: a pipeline that pulls primary-source signals automatically and routes only the genuine judgment calls to humans. The "tax" convention was paying: treating verifiable lookups as if they needed expert labor.
1---2name: first-principles3description: Breaks a problem down to fundamental, irreducible truths and rebuilds a solution from them, refusing to reason by analogy or by copying how others already do it. Use this skill when the user is stuck in conventional thinking, asks "why do we even do it this way", wants to challenge baked-in assumptions, faces a hard design or strategy problem with no obvious answer, says "think from first principles", "go back to basics", "forget how everyone else does it", or suspects the standard approach is cargo-culted rather than reasoned. Best when the default solution feels expensive, slow, or inherited rather than chosen.4---56# First Principles78Strip a problem down to the few things that are actually, physically, definitionally true — then build the answer up from only those, ignoring how it's "normally" done.910## When to use11- The current approach exists because "that's how it's always been done."12- A solution feels too expensive/slow/complex and you suspect the cost is assumed, not real.13- You're benchmarking against competitors and copying their structure by default.14- A hard design problem where analogies keep leading to mediocre answers.15- The user explicitly wants to challenge a deeply held assumption.1617## When NOT to use18- Routine, well-understood tasks where convention is correct and cheap (don't re-derive a login form).19- Time-critical decisions where "good enough, now" beats "optimal, eventually."20- Domains where the "convention" is actually hard-won safety regulation or law — question it intellectually, but don't ignore it operationally.2122## The method231. **State the goal as an outcome, not a method.** Write what must be true at the end, with zero reference to current tools, vendors, or steps. "Customers trust our data" — not "we run more audits."242. **List everything you believe about the problem.** Dump assumptions, constraints, costs, "rules." Be exhaustive and honest.253. **Interrogate each item: truth or inheritance?** For each, ask "Is this a law of physics/math/economics, or just a convention someone handed me?" Tag each as FUNDAMENTAL (can't be reduced) or DERIVED (someone's choice, possibly wrong).264. **Keep reducing the fundamentals.** For each "fundamental," ask "why?" again. Stop only when the next "why" hits a real constraint — a material cost, a physical limit, a definitional requirement, a hard regulation. These are your bedrock truths.275. **Quantify the bedrock.** Put real numbers/limits on each truth. What does the raw material/compute/time actually cost at the floor? This exposes the gap between the floor and the conventional price.286. **Rebuild upward using only bedrock truths.** Construct a solution that satisfies the goal using nothing but the fundamentals. Forbid yourself from importing any DERIVED step unless you can re-justify it from bedrock.297. **Compare to the conventional solution.** Where the two diverge is your insight — and where the conventional approach is paying a tax for an assumption that isn't true.3031## What good looks like32- A short list of genuinely irreducible truths, each with a number or hard reason attached.33- At least one "wait, why do we even need that?" moment where a sacred step turns out to be optional.34- A rebuilt solution that a smart skeptic can't dismiss as "you just reinvented the wheel."35- Clear separation between what's a real constraint and what was merely habit.3637## Anti-patterns38- **Fake first principles:** restating the conventional answer in fancier words. If your rebuild matches convention exactly, you didn't reduce far enough.39- **Stopping too early:** treating a vendor's pricing or an industry norm as bedrock when it's just a market choice.40- **Reducing past usefulness:** "everything is atoms" is true and useless. Stop at the lowest level that still lets you act.41- **Ignoring real constraints:** declaring a law of physics or a safety regulation "just a convention."4243## Example44Goal: "Make our broker audits cheaper." Conventional answer: hire more auditors. First-principles pass: What is an audit, fundamentally? It's *acquiring trustworthy evidence about a claim*. The fundamental cost is not "an auditor's hour" — it's the cost of obtaining one verified fact. List the facts a broker review needs; ask which require a human and which require only a primary-source signal (a regulator's public registry, a settlement record). The bedrock truth: most facts are *publicly verifiable data*, not *judgment calls*. Rebuild: a pipeline that pulls primary-source signals automatically and routes only the genuine judgment calls to humans. The "tax" convention was paying: treating verifiable lookups as if they needed expert labor.
opelpleple