Ten years of experience is, mostly, ten years of accumulated analogies. You’ve seen enough situations that a new one triggers a “this looks like that” reflex before you’ve finished reading the brief.
The uncomfortable part: AI does that reflex better and faster. It’s read everything, remembers all of it, never gets tired. If experience is mostly analogy, experience alone is a depreciating asset.
The turn: reasoning from fundamentals doesn’t depreciate the same way.
• First principles = breaking a problem down to what’s actually, provably true, and rebuilding from there instead of from whatever worked last time.
• It’s slower than analogy.
• It’s the one part of “experience” AI can’t just absorb by reading more of the internet.
• This isn’t a pitch to abandon experience, it’s a pitch to stop treating it as self-sufficient.
What Is First Principles Thinking? (vs Reasoning by Analogy)
Analogy and first principles solve the same problem differently. Analogy wins most of the time, not because it’s lazy, but because it’s usually right, and always faster.
• Analogy sounds like: “Last time we needed to grow, we hired sellers, so let’s hire sellers.” Answered inside the meeting.
• First principles sounds like: “What’s actually constraining growth right now?” Might take a week. Might land somewhere nobody expected, maybe it’s not headcount at all, maybe it’s a broken handoff between marketing and sales.
Why analogy usually wins anyway:
• It’s cheap and, most of the time, good enough.
• You don’t want to first-principles every Tuesday decision, reasoning your coffee order up from thermodynamics is how you never leave the house.
• First principles earns its cost on decisions expensive to get wrong, where the obvious analogy has started feeling too convenient.
Why this matters for AI specifically: serious AI programs teach how models actually work, tokens, embeddings, why hallucination happens before teaching any tool. You can’t question an answer whose mechanics you don’t understand; you can only accept or reject it on vibes. That’s the design behind the first month of Scaler’s School of Business: understand the mechanism before you trust the output.
Why Experience Can Work Against You Now
Pattern-matching gets rewarded right up until the pattern breaks, and nobody rings a bell when it does.
Seniority has a specific side effect: it makes your assumptions invisible to you.
• “We tried that in 2019” sounds like wisdom.
• Sometimes it’s a five-year-old data point treated as a law of physics, because the person saying it has enough authority that nobody asks what actually happened in 2019, or what’s changed since.
Jeff Bezos’ version of this, the proxy trap:
• A business reaches for a process, survey, or precedent as a stand-in for the outcome it actually cares about.
• Over time, the proxy quietly replaces the thing it was meant to represent.
• “We ran the survey” becomes the win condition, instead of “we understood the customer.”
Why it stings more for senior people:
• The more decisions you’ve made a certain way and had them work out, the more that way starts feeling like a law instead of a habit that fit its moment.
• Nobody sets out to stop questioning their own playbook, it just gets expensive to keep questioning something that keeps paying off, until it doesn’t.
Bottom line: this doesn’t make experience worthless. The recall half (“I’ve seen this before”) is exactly what AI commoditises fastest. The half that doesn’t commoditise: knowing why the pattern worked, and spotting when the current situation isn’t that pattern anymore. Worth a proper root cause analysis on your own most-repeated decision, not just a gut check.
The Canonical Case: Rebuilding Rocket Costs from Raw Materials
Musk and SpaceX, told once, tightly, then straight to a business example, since rockets are a lousy analogy for whatever’s on your desk this week.
• Early 2000s: Musk quoted ~$8 million for a single rocket, traditionally sourced.
• Instead of negotiating against that price, he decomposed it, what’s a rocket actually made of?
• Aluminum, titanium, copper, carbon fiber, priced as raw materials, that stack came to ~2% of the quoted cost.
• The other 98% wasn’t physics. It was assumptions: single-use hardware, fully outsourced manufacturing, a supply chain priced for how rockets had always been built.
The takeaway: that gap between materials cost and industry price is the whole method in miniature. The price wasn’t a law, it was a habit dressed up as one.
Translate to a desk job, what’s your department’s own 2%-raw-materials number hiding?
• The vendor quote nobody’s re-priced in three years.
• The headcount ask justified by “that’s just what it takes” nobody’s broken down what “it” is.
• The “we need a bigger budget” that’s really “we’ve never audited what the current one is doing.”
One note on outcomes: SpaceX built reusable rockets and cut launch costs, the part everyone remembers. Less remembered: it worked because the 98% gap got interrogated line by line, not because someone had a hunch. The hunch is free. The interrogation is the actual work.
The 3-Step First Principles Method (With a Business Example)
Three steps, in order, skipping to “rebuild” without the first two just gets you a differently-flavoured version of the same assumption.
1. Deconstruct — list every assumption buried in the decision, including the ones too obvious to write down. Especially those.
2. Interrogate — for each one, ask: is this a law, or a habit? Price it out if you can. A law doesn’t move no matter who’s asking. A habit just hasn’t been questioned by anyone with standing yet.
3. Rebuild — reassemble the decision using only what survived step two, not whatever felt closest to the original conclusion.
How leaders practice this: in Scaler’s Online PGP in Business & AI, this exact method gets applied to live build-vs-buy, pricing, and headcount calls from week one — with a senior cohort whose job is to find the assumption you didn’t notice you were making.
Worked example, build-vs-buy on a $60K/year internal tool platform:
• “We don’t have the engineering capacity” — Habit. Nobody’s checked; it’s a guess based on how busy the team felt last quarter.
• “Buying is faster to ship” — Partially a law. Probably true for v1. Whether that speed justifies the ongoing cost is a separate question.
• “The vendor will maintain it better than we would” — Habit dressed as law. Nobody’s compared SLAs against a realistic internal on-call rotation.
• “$60K a year is basically nothing at our size” — Habit. Nobody’s run that against three years, compounded, vs. a one-time build.
Rebuilt verdict: buying probably wins on speed to v1, but the cost argument was doing emotional work, not logical work, and the maintenance claim was never tested. Same vendor, different (defensible) decision.
Note: framing the problem correctly before running this, instead of starting from “should we buy X”, is most of the real work. Our problem statement examples piece and MECE framework breakdown both help when the assumption list gets messy.
Using First Principles to Question What AI Tells You
Not an AI-doomer section, not an AI-hype section, a QA procedure.
• AI outputs inherit assumptions the same way a junior analyst does, from training data, from how the question got framed.
• The output sounds confident regardless of whether those assumptions hold in your specific situation.
Example: an AI recommends a 15% price increase, backed by a clean elasticity model.
• First-principles review = decompose it: where did the elasticity number come from?
• If it’s category-level data or a different market segment, the recommendation is a habit wearing a law’s clothing, a plausible number from an adjacent context, presented as if it describes your customers specifically.
The real trap: the output arrives with the same confident tone whether the assumption is rock-solid or borrowed from somewhere irrelevant. A junior analyst usually hedges a little. A model doesn’t hedge unless you make it, so the interrogation has to come from the reader, not the system.
Bottom line: operators who rebuild decisions from fundamentals (pricing, build-vs-buy, what deserves automating) are the ones AI amplifies. Operators who accept nicely-formatted output are the ones getting replaced by whoever’s doing the interrogating. Our analytical thinking piece is the natural next read if this is a muscle worth building deliberately.
Building the Muscle: Practice Routines for Leaders
No program required to start practising, just actual practice. Three weekly routines:
• One assumption audit per big decision — run the three-step method on whatever’s on the table this week. Only the decisions expensive enough to justify the hour.
• “Five fundamentals” journaling — for any recurring problem, write the five things you’re certain are actually true, no hedging. Most people can only confidently name two or three, that gap is informative on its own.
• Red-team your own proposal — spend fifteen minutes arguing against your own pitch as the most sceptical person in the room. Can’t find a real weakness? You haven’t looked hard enough.
Honest limit: solo practice has no contestation built in. You’re both the arguer and the only checker, a conflict of interest even done rigorously. The real gap structured environments fill: reasoning pressure-tested by a senior cohort with no reason to let a shaky assumption slide.
Ten years of experience is an asset exactly once it’s paired with reasoning AI can’t shortcut. That pairing is trainable.
Two ways to build it deliberately: the Online PGP in Business & AI for the structured, cohort-based version, live cases, a senior peer group, a syllabus instead of a search history, or our case study method piece to see how that pressure-testing actually works before committing to anything.
The FAQs
What is first principles thinking in simple terms?
Breaking a problem down to its fundamental truths and reasoning up from there, instead of copying what’s been done before.
What is the difference between first principles and analogy?
Analogy adapts existing solutions; first principles rebuilds from fundamentals. Slower, but capable of genuinely original answers analogy can’t reach.
What is a famous example of first principles thinking?
SpaceX questioning rocket prices: raw materials came to roughly 2% of the quoted cost, which is what made reusable, self-built rockets look possible instead of reckless.
How do leaders use first principles with AI tools?
As a filter: decompose the AI’s answer, test its underlying assumptions against your actual situation, and rebuild the recommendation before acting on it.
Can first principles thinking be learned?
Yes! And it’s a practised discipline of questioning assumptions, not an innate talent. Structured problem-solving training accelerates it considerably.
