Musk Wiki

From Instinct to Algorithm

NextHow Musk Makes Decisions

From Instinct to Algorithm

One method, hardening over two decades. The thing Musk now calls his first-principles habit starts in 2003 as a gut instinct he uses to reject projects: they just can’t close the economic case. It ends in 2024 as an ordered, numbered procedure he says he runs “as a mantra.” In between it keeps picking up sharper edges. In 2013 it gets a public name, a good framework for thinking … first principles reasoning. In 2015 it gets a learning metaphor, knowledge as a semantic tree. In 2016 it gets its sharpest negation, reasoning by analogy named as the mistake. In 2021 it gets a spoken core: sets your axiomatic base, and then you reason up from there. Then it settles into a fixed five-step form. The arc isn’t a method discovered once. It’s one instinct calcifying, stage by stage, into a framework, then an algorithm, then a mantra. The same move shows up at every step. What changes is how hard and explicit it gets. Related ideas live on Asking the right question and The engineering algorithm.

Summary

The shape is crystallization. The same underlying move runs from the earliest datapoint here (2003) to the latest (2024): refuse the inherited framing, rebuild the problem from physical fundamentals, reason up. What changes isn’t the move but its degree of formalization. In 2003 it’s tacit, a gut read he can’t yet name. By 2013 he can name it as a framework. By 2016 he can name its opposite, reasoning by analogy, as the error to avoid. By 2021 he can state its spoken core as a general procedure. By 2024 it’s a checklist with an order he insists on, recited like a mantra. A loose disposition becomes an explicit algorithm.

Two threads run through the arc and harden with it. The first is the negative half, his standing warning against reasoning by analogy, which sharpens from a throwaway clause in 2013 (“as opposed to reasoning by analogy”) into a named mistake by 2016. The second is the set of physics tools he reasons with, above all “thinking in the limit”: push a variable to its extreme to find the optimum. Those surface as explicit, repeatable techniques by 2021. The endpoint is striking. A man whose method began as an un-nameable instinct can, twenty-one years later, hand you the steps in order.

ℹ️ One method, four degrees of hardness. Read as isolated quotes, the 2003 “economic case,” the 2013 “framework,” the 2016 “analogy is the error,” the 2021 “axiomatic base,” and the 2024 “algorithm I run as a mantra” can look like five different ideas. Read as one arc, they’re a single move recorded at five stages of formalization. The instinct never changes: boil the problem down to fundamentals, ignore the consensus framing, reason up. What changes is how explicitly he can state it. In 2003 it’s a thing he merely does. By 2013 it’s a thing he can name. By 2016 he can name its failure mode. By 2021 he can describe it as a general procedure. By 2024 he runs it as a fixed, ordered checklist. The two-decade trend is from instinct to procedure, the method becoming legible to its own user.

2003 — the instinct, before it has a name

The earliest datapoint of any kind here is also the earliest sighting of the habit, already operating but tacit. At the 2003 Stanford eCorner talk, at the very start of SpaceX, Musk rejects two then-fashionable space ideas. Not on engineering grounds, but by refusing to let a project exist unless the numbers close:

“the economics don’t make sense. They just can’t close the economic case.” ↗

“Closing the economic case” is the whole method in embryo: quantify the problem from the bottom up, and let the numbers, not the fashion, decide what’s worth doing. But in 2003 he has no name for it. It’s a contrarian read he applies by reflex. This is the ground floor of the arc. The move is already there, a decade before he can call it anything.

2013 — the framework gets its name

The 2013 TED conversation is the earliest long-form interview here, and it’s where the tacit instinct becomes a named framework. Asked how one person innovates across cars, rockets and solar, Musk reaches for physics and gives the line most often quoted as his definition of the method:

“Well, I do think there’s a good framework for thinking. It is physics. You know, the sort of first principles reasoning.” ↗

And then the canonical phrasing — the definition itself, which already carries the negative half of the method as a trailing clause:

“boil things down to their fundamental truths and reason up from there, as opposed to reasoning by analogy.” ↗

Crucially, he scopes when the method is needed — ordinary life runs on analogy, and the physics approach is reserved for doing something new:

“But when you want to do something new, you have to apply the physics approach.” ↗

This is the turning point. The same move he applied wordlessly in 2003 now has a name, a domain (physics), a definition (boil down … reason up), and a scope (something new). The instinct has become a framework: nameable, teachable, repeatable. Everything after this is refinement of a thing he can now point at.

2015 — the method as a way to learn (“a semantic tree”)

The January 2015 Reddit AMA catches the habit in its rarest form. Not as a way to reason (2013) or build (2024), but as a rule for how to learn. Asked how he acquires knowledge, he gives the metaphor he’s now most associated with on the subject:

“View knowledge as sort of a semantic tree – make sure you understand the fundamental principles before diving into details.” ↗

It’s the same fundamentals-before-details instinct as the 2013 boil things down … reason up, here pointed at learning rather than design. Knowledge has a structure (trunk → branches → leaves), and a detail only sticks if it hangs on a foundation already secured. What this adds to the arc is breadth. By 2015 the method is no longer just how he attacks an engineering problem; it’s how he thinks anyone should learn anything. The framework is generalizing outward from the workbench into a way to teach.

2016 — the negative half, named as the error

By 2016 the throwaway 2013 clause (“as opposed to reasoning by analogy”) has hardened into a named mistake he warns against directly. On the Q1 2016 Tesla earnings call, pressed on whether Model X’s troubles predict Model 3’s, he refuses the analogy and states the failure mode outright:

“It’s always tempting for people to reason by analogy instead of first principles. That would be the mistake of assuming that anything to do with the Model X production has bearing on Model 3.” ↗

He illustrates it with one of his most vivid images. The same result can be reached by wildly different means, so the question is always which means the problem actually requires:

“You can kill a fly with a thermonuclear weapon, you can with a MOAB, with a cruise missile, with a machine gun, or a fly swatter. The end result is the same, but the difficulty is considerably more significant from one to the other, and the collateral damage is considerably more significant.” ↗

This is the arc’s negative half coming into focus. In 2013 analogy was the thing first-principles was merely opposed to. By 2016 it’s the mistake, the specific cognitive error the method exists to prevent. The method has hardened enough that he can now define it by what it is not, and catch the error in others in real time. He restates it even more flatly in a 2022 tweet (“the fundamental error is reasoning by analogy, rather than first principles”), and on First principles he makes the same point.

ℹ️ The negative half is part of the hardening, not a separate idea. A method is only as sharp as its account of how it fails. In 2003 there’s no failure mode at all, just a read he trusts. By 2013 the failure mode is a trailing clause (“as opposed to reasoning by analogy”). By 2016 it’s the headline, the mistake, with its own illustration. The sharpening of the negation tracks the sharpening of the method: the more explicitly he can state what reasoning-by-analogy gets wrong, the more explicitly he can state what reasoning-from-fundamentals gets right. The two halves harden together.

2021 — the spoken core, and thinking in the limit

The 2021 Lex Fridman conversation (#252) gives the method its plainest spoken statement. It’s the same move the framework names, now described as a general procedure for “really any walk of life”:

“let’s boil something down to the most fundamental principles, the things that we are most confident are true at a foundational level, and that sets your axiomatic base, and then you reason up from there.” ↗

And the constraint that grounds the whole thing, physics as the one rule that cannot be bent:

“I’ve met a lot of people that can break the law, but I have never met anyone who could break physics.” ↗

The 2021 version states the axiom-first core (boil down, set your axiomatic base, reason up) more explicitly than at any earlier stage. It also surfaces a distinctive physics tool he says he uses constantly: thinking in the limit. You scale a variable to an extreme to find the optimum. What if our volume were a million units a year? If the thing is still expensive at that limit, then volume isn’t the cause, the design is. The same talk closes on the problem-framing companion to the method: the question is the really the hard part … if you can properly frame the question, then the answer, relatively speaking, is easy. By 2021 the method is no longer just a framework he can name. It’s a procedure he can walk through, with named tools he reaches for inside it.

2024 — the algorithm, run as a mantra

The arc closes on the method at its hardest: a fixed, ordered checklist he recites. The 2024 Lex Fridman conversation (#438) gives the most concrete first-person version. It’s a procedure he says he runs as a mantra, whose first step is to attack the question before the solution:

“I have this very basic first principles algorithm that I run kind of as a mantra, which is to first question the requirements, make the requirements less dumb.” ↗

The ordering, question first and then delete before you improve, produces his sharpest engineering line:

“the most common mistake of smart engineers is to optimize a thing that should not exist” ↗

The fullest public statement of the same five steps is the 2021 Starbase tour. Walking the factory floor, he lays out all five in order (make the requirements less dumb, delete the part, simplify, accelerate, automate) and insists the order is the whole point:

“First make your requirements less dumb, your requirements are definitely dumb.” ↗

“possibly the most common error of a smart engineer is to optimize the thing that should not exist.” ↗

This is the endpoint of the hardening. The 2003 instinct that could only reject a project (“the economics don’t make sense”) has become, twenty-one years later, a numbered procedure for building one. Its step one (“question the requirements”) is the problem-framing move and the first-principles reasoning applied to specs. And the recited form (“as a mantra”) marks it as a settled discipline rather than an open-ended habit. The instinct has fully crystallized into an algorithm.

ℹ️ “Framework” and “algorithm” are the same method at two hardnesses. The 2013 framework for thinking and the 2024 algorithm I run as a mantra are not two methods. They’re the same method named at the two ends of the arc. A framework is a way of thinking about a problem; an algorithm is a fixed procedure for acting on it. The distance between them is exactly the hardening traced across this arc. The open-ended physics instinct of 2013, applied for another eleven years to rockets, cars, and factories, sediments into a checklist with an order he refuses to break. The engineering algorithm is what first-principles reasoning becomes when a person runs it for two decades: the reasoning habit, set hard.

What the arc reveals

  • One move, increasingly explicit. From 2003 to 2024 the underlying move never changes: refuse the inherited framing, rebuild from fundamentals, reason up. What changes is how explicitly Musk can state it. An instinct he merely uses (2003), a framework he can name (2013), a learning rule (2015), an error he can name (2016), a spoken procedure (2021), a recited algorithm (2024). The arc isn’t a method discovered but a method hardening.
  • The negation sharpens with the method. The standing warning against reasoning by analogy tracks the method’s own crystallization: a trailing clause in 2013, the mistake by 2016. A method gets sharper precisely as its account of how it fails gets sharper.
  • The endpoint is a checklist. The most striking fact of the arc is that a habit which began as an un-nameable instinct ends as a numbered procedure he can hand you step by step, and recites as a mantra. The engineering algorithm is the reasoning habit in its hardest, most procedural form, and “question the requirements” is the seam where the two meet.

Connections