Musk Wiki

The engineering algorithm

NextVertical integration

The engineering algorithm

Ask Musk how to build something and he will hand you a numbered procedure. Elon Musk says he runs it “rigorously” on every engineering problem. It is not the same thing as his first-principles reasoning, which is about how to think about a problem. This is a method for building: how to strip a design, and the production line behind it, down to what is actually needed. Its slogans are some of his most-quoted lines: “the best part is no part,” “delete before you optimize,” “the most common error of a smart engineer is to optimize the thing that should not exist”. Behind all of them sits one stubborn claim, that the factory, not the design, is the hard problem. The production system is where the real work lives.

He laid out the whole thing once, unprompted, walking the factory floor on the 2021 Everyday Astronaut Starbase tour, all five steps in order. By the 2024 Lex Fridman #438 conversation he had compressed it into a “mantra.” The reasoning style it sits on top of is on First principles.

The five steps, in order

The order is the whole point. Musk insists each step must come before the next, because doing them out of sequence is the characteristic failure of a smart engineer: you optimize or automate something that should have been deleted.

1. Make the requirements less dumb. Attack the question before the solution. Every requirement is suspect, and the danger rises when a smart person wrote it:

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

No requirement is safe because nobody, however senior, is reliably right:

“Everyone’s wrong, no matter who you are, everyone’s wrong some of the time.” ↗

So he attaches a discipline to it. Every requirement must trace to a named, accountable person, never a department (the line itself is quoted on Part 1). Otherwise you spend years optimizing around a constraint some intern invented long ago and nobody now defends.

2. Delete the part or process. Remove it entirely before improving it, and delete hard enough that you sometimes overshoot:

“If you’re not deleting a part or process step, at least 10% of the time, basically if you’re not adding things back in 10% of the time, you’re clearly not deleting enough.” ↗

3. Simplify or optimize — and only now, third, because:

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

He blames schooling for it. Students are trained never to question the question, so they reflexively optimize a requirement that should have been deleted. He calls the habit a “mental straight jacket”.

4. Accelerate cycle time. Go faster, but only after the first three steps, never before:

“which is accelerate cycle time. You’re moving too slowly, go faster, but don’t go faster until you’ve worked on the other three things first.” ↗

His memorable warning against accelerating the wrong thing:

“don’t dig it faster, stop digging your grave.” ↗

5. Automate — last.

“And then the final step is automate.” ↗

He admits he has personally run the steps backwards “multiple times,” which is exactly why he drills them. His standard cautionary tale is the Model 3 battery pack. He automated a robot cell that glued fiberglass mats onto the pack, accelerated it, simplified it, and only at the very end thought to ask what the mats were even for. The battery team blamed the noise-and-vibration team; the noise team blamed the battery team. Nobody actually needed them. He deleted the whole “$2 million robot cell”. The sheer absurdity of that back-and-forth is what set off his Goldberg-cartoon / simulation aside.

“Question the requirements, delete the part”

On the Starbase tour he boils the first two steps, the most important ones, down to a phrase he keeps repeating. He is explaining why the booster has no dedicated attitude-control thrusters at all. He found he could steer it with the rocket’s own vent gas instead:

“Question the requirements, delete the part.” ↗

His 2024 Lex #438 “mantra” is the same first step three years on: “question the requirements, make the requirements less dumb”.

Everyone is chief engineer

The method has an organizational half. It only works if the people running it grasp the whole system, because that is what lets them catch a bad local optimization (the line, quoted on Part 1: “you really want everyone to be chief engineer”). His example: a team pours effort into shaving mass off the engine while ignoring a literal ton of unused landing propellant. Locally smart, globally dumb, and only a systems view catches it.

The factory is the product — manufacturing is the hard problem

The tour opens and closes on the algorithm’s companion belief: people over-value the design “Eureka moment” and badly under-value the production system.

“I think, currently a factory is underrated and design is overrated.” ↗

He puts a number on it. Designing the manufacturing system is “10 to a 100 times more effort” than designing the engine. Then he pushes it to the edge: “the amount of effort that goes into the design rounds down to zero,” quoted on Part 1. On Part 2 of the tour he says it flat out:

“The production system is the actual hard thing.” ↗

“The rocket design is relatively easy compared to the factory.” ↗

This is the engineering root of his build-the-whole-system instinct. If making the thing is harder than designing it, the manufacturing capability is the durable asset. Even the deletion discipline scales up to the factory. On Part 2 he halts work on the cargo door because it is not on the critical path to orbit, reasoning “Is the door necessary to solve the problem? No”. That is the delete step aimed at a whole feature instead of a single part.

The phrase is born — Gigafactory to “machine that makes the machine” (Tesla earnings 2013-2014)

He says it first, in the Tesla record, on the 2013–2014 earnings calls. That is six years before the 2021 Starbase tour gives the algorithm its five-step shape, and a year before Battery Day 2020 states it most compactly. The idea arrives with the Gigafactory. On the Q3 2013 call the factory itself becomes a product with versions:

“It’s like a version of the factory. So we’re trying to figure out what’s the right way to do version 1 of this Gigafactory.” ↗

By Q2 2014 it has become his explicit case for building custom equipment rather than buying it, for designing the machine that makes the machine:

“you really get to design the machine that makes the machine, not just do so with off the shelf equipment. So just everything about it is going to get a whole lot better.” ↗

And by Q3 2014 it has hardened into the thesis that defines the era, the one Battery Day and the Starbase tour later restate: production is harder than design.

“it is way harder to make the machine that makes the machine than it is to make the machine in the 1st place. Like the production of the car is way harder than actually I’d say the design of the car.” ↗

It is the same belief he states later at Battery Day 2020 (“designing the machine that makes the machine is vastly harder than the machine itself”) and on the 2021 tour (“the production system is the actual hard thing”). Here it is in its first mature spoken form, the Tesla-side root of his build-the-whole-stack instinct. These 2013–2014 calls carry no speaker labels, so the words are attributed to Musk by his first-person tells and his place in the Q&A; the per-quote case is on Tesla Earnings Calls 2013-2015.

The thesis hardens, then corrects itself — “the factory is the product” to over-automation (Tesla earnings 2016-2018)

The 2016–2018 earnings calls are the Model 3 “production hell” years, and they are the one stretch where he states the factory-is-the-product thesis at full strength and then walks it back. Through 2016 and 2017 the conviction only hardens. Tesla is “hell-bent on becoming the best manufacturer on Earth” with “more potential for innovation in manufacturing than there is in the design of the car” (Q1 2016). The machine that makes the machine gets a playful name, the “Alien Dreadnought” (Q2 2016). There is “the factory will be a more important product than the car itself” (Q4 2016). The machine that builds the machine is mostly software and “a very difficult thing for other manufacturers to copy” (Q1 2017). “speed is the ultimate weapon… if you can see the robot move, it’s too slow” (Q3 2017). In Q4 2017 it sets into what amounts to a River Rouge doctrine: “the Model T wasn’t the product, it was River Rouge… that’s really what will be Tesla’s long-term competitive advantage.”

Then 2018 is where he corrects himself, more plainly than anywhere else. Having pushed automation to the limit, he reverses: “we did go too far on the automation front and automated some pretty silly things” (Q1 2018). That is the seed of his later “humans are underrated” view. The corrected belief keeps the velocity-and-density core (“the path to manufacturing efficiency is velocity and density”) but rebalances man and machine and recasts production as a problem of software and people (“manufacturing at volume is mostly a software problem”; solicit “all of our associates, no matter how junior”). It sharpens a design-for-manufacturing rule too: make the designers feel the pain (“once you feel the pain… I was like torturing people with my terrible design. Now I know”), because “decisions made at the beginning of development program have massive implications for future CapEx.” Same algorithm, caught in the act of updating, which is what makes the era worth its own section. Ten of the twelve transcripts carry Musk’s speaker label; on the unlabeled Q3 2017 and Q2 2018 calls he is attributed by first-person tells and Q&A position (see Tesla Earnings Calls 2016-2018).

“Prototypes are easy, production is hard” — the maxim hardens into a refrain (Tesla earnings 2019-2021)

By the 2019–2021 earnings calls the thesis has stopped evolving. Now he just repeats it, almost word for word. The S-curve shape of a production ramp is stated as plain fact (“manufacturing follows the S-curve… people that haven’t spent a lot of time in manufacturing kind of think that once you have a factory, you can just sort of turn it on and it’s at capacity,” Q3 2020). The “prototypes are easy, production is hard” belief comes back three quarters running in near-identical words: “Prototypes are easy. Scaling production is very hard” (Q4 2020), “Prototypes are trivial. They’re child’s play… Production is hard, is very hard” (Q1 2021), “I’m fond of saying that prototypes are easy and production is hard” (Q2 2021). Each time he ties it to the same point about what Tesla actually pulled off: “the thing that’s remarkable is that Tesla didn’t go bankrupt in reaching volume production.” He restates the factory-as-moat claim with a rare explicit self-reference to the prior quarter (“A comment I made in the past is that I think Tesla’s long-term competitive strength will be primarily manufacturing… I am quite confident this will be what happens,” Q3 2020). The “machine that makes the machine” line shows up in its most recursive form (“we made the machine that made the machine that made the machine,” Q3 2020). And the factory turns into a living thing: “a factory is like a giant cybernetic collective” (Q2 2021) whose growth “goes as fast as the least lucky and dumbest of those 10,000 things.” The point of these years is the sheer repetition. The algorithm is no longer being discovered or fixed. It is being preached. Some quarters carry Musk’s label; on the label-light 2019 and Q2 2021 calls he is attributed by his opener and Q&A position (see Tesla Earnings Calls 2019-2021).

The Tesla statement — “the machine that makes the machine” (Battery Day 2020)

On the Tesla side, the factory-is-the-product thesis is at its plainest at Battery Day 2020, a year before the 2021 Starbase tour gives it its fullest form on the SpaceX side. Running through Tesla’s core competencies, Musk names manufacturing as the under-appreciated hard part and gives the slogan in its tightest form:

“I think manufacturing is underappreciated in general, and the difficulty of designing the machine that makes the machine is vastly harder than the machine itself.” ↗

He attaches the same ratio he repeats across the years, the factory at 10–100× the difficulty of the prototype, and uses it to explain why so many hardware startups die in the gap between a demo and a production line:

“It’s just at least 10 to 100 times harder to do the factory than the prototype, and that’s why you see a lot of companies out there or startups they’ll bring out a prototype, but they just can’t get it over the hump for who manufacturing, because manufacturing of new technology especially is the hardest thing by fa.” ↗

It is the same “10 to a 100 times more effort” and “the production system is the actual hard thing” belief the 2021 tour states for rockets, said here for cars a year earlier. In the Tesla context it is the engineering root of his build-the-whole-stack instinct: if making the thing is harder than designing it, the manufacturing capability is the durable asset. The whole point of Battery Day, in his framing, was the production of cheaper batteries at scale, not a single cell breakthrough. The cell-tech detail itself belongs to Drew Baglino’s engineering work, not to Musk’s mind model.

The 2019 statement — “the best part is no part”, two years early (Starship Update)

The record’s earliest dated statement of the algorithm’s core slogans is the press Q&A at the September-2019 Starship Update, two years before the 2021 Starbase tour lays out all five steps. Pressed by reporters on how SpaceX built a full-scale ship in months, Musk offers the scheduling heuristic he calls “management by rhyming”:

“if the schedules long it’s wrong and if it’s tightest right” ↗

The deletion slogan, in the form the 2021 tour later makes famous:

“the best part is no part the best process is no process it weighs nothing costs nothing can’t go wrong” ↗

And the un-design test — what he says he is “most impressed with” in a design meeting is what was removed, not what was added:

“undesigning is the best thing just delete it that’s the best thing” ↗

These are his September-2019 versions of the delete-first and “best part is no part” instinct, plus the schedule read as a design signal. They are the same slogans the 2021 five-step statement and the 2024 #438 “mantra” restate, here pointed at building the Starship Mk1 fast and said in front of the ship itself. The “recursive improvement on schedule … feedback loop did this make it go faster” reasoning around the same cues is the cycle-time logic of step 4, spelled out in prose at the September-2019 Starship Update. These are his own Q&A answers, not the reporters’ questions (details on attribution and verification).

Goal-first — define the objective, then judge the design (Mars Society 2020)

The October-2020 Mars Society conversation adds the step that sits upstream of the five. Before you can “make the requirements less dumb” or delete a part, you have to nail down what you are building toward. Asked to explain the design thinking behind Starship, Musk waves off the design question and insists on the prior one:

“you first have to say what what is the goal” ↗

“once you have what is the goal you can then measure the various designs against that goal” ↗

His point is that the comparison means nothing without a goal to measure against, roughly “why is one design better than another? what’s your goal?” It is step 1 (attack the question first) pushed one level up. The requirement is judged against the design, and the design is judged against the goal.

Pressed later on the methodology that lets SpaceX “innovate so swiftly,” he disclaims having one (“I don’t really know”) and points back to the goal. A hard enough objective forces radical innovation, because nothing less will get there:

“creates i think a good forcing function for radical innovation because in the absence of radical innovation we have no chance of meeting that goal” ↗

And the rate must be exponential, not linear, or the goal is simply not met:

“if we do not see something close to an exponential improvement in our rate of innovation we will not reach mars” ↗

This is the algorithm reasoned from the top down. The goal disciplines the requirements (step 1), and the goal’s difficulty is what makes deletion, simplification, and above all accelerated cycle time (step 4) non-optional rather than nice-to-have. It gives the 2019 “if the schedule’s long it’s wrong” rate-of-innovation instinct its reason: the goal is the forcing function. These are his own answers, not the host’s questions (details on attribution and verification).

“Ideas are trivial, execution is everything” (Tesla Shareholder Meeting 2023)

The 2023 Tesla shareholder meeting gives the sharpest ranking under the factory-is-the-product thesis. Asked whether Tesla might chase another efficiency idea, Musk first restates the standard slogan (“prototypes are easy, production is hard”; reaching positive cash flow is, roughly, “excruciating pain … at a level you cannot believe”). Then he ranks the idea below its execution:

“I find ideas to be somewhat trivial.” ↗

He drives it home with a metaphor, roughly “like the idea of going to the moon … getting to the moon is the hard part”. This is the “the machine that makes the machine” and “the production system is the actual hard thing” belief stated as a flat ordering: the design “Eureka” is cheap, the manufacturing is where the durable difficulty and value live. It is also the spoken May-2023 twin of the “I have a fountain of ideas … Innovation is not the problem. Execution is the problem” line he gives six months later, the same conviction, pinned here to the building method rather than to the storm of his mind. This is his own Q&A answer, not the questioner’s (details on attribution and verification).

Why it is its own concept (not first principles)

First principles is Musk’s reasoning habit: boil a problem down to physical fundamentals and reason up, with physics as the only binding rule. The engineering algorithm is the building method that sits on top of it, a repeatable ordered checklist for turning a design and a factory into the leanest thing that works. They interlock. Step 1 (“make the requirements less dumb”) is first-principles applied to specs, and problem-finding applied to engineering. But they are different tools, and Musk gives the algorithm a numbered form precisely because it is mechanical where first-principles is open-ended.

“Prototypes are easy, production is hard” — now self-cited doctrine (Tesla earnings, 2022-2026)

By the 2022-2026 earnings calls the production beliefs come out as long-held maxims he openly cites back to himself. “Prototypes are easy. Production is hard. I’ve said that for many years” (Q4 2024), with the why-startups-fail reasoning right behind it: “10,000% harder to get to volume production than to make the prototype… that is why there have not been new car startups that have been successful for 100 years, apart from Tesla” (Q3 2023). He coins a game-of-pennies cost model and works it out from per-part arithmetic (“It’s a game of pennies. It’s like Game of Thrones, but pennies,” Q2 2023), then dramatizes it as unsung heroism (“the heroes who got 20% of the cost out of a car,” Q3 2024). The factory-as-moat thesis recurs as a nested “machine that makes the machine” (“you have to copy the machine that makes the machine, that makes the machine,” Q4 2023; “the hardest Tesla product to copy will be the factory,” Q3 2024; “the factory is the product as much as the car is the product,” Q1 2025). And the S-curve ramp heuristic, the factory only as fast as its worst part, comes back almost word for word from Q1 2025 (“as fast as the slowest and least lucky component”) to Q1 2026 (“as fast as the least lucky, slowest, dumbest part in the entire 10,000”). Every quarter here carries an explicit Elon Musk speaker label (see Tesla Earnings Calls 2022-2026).