Start Over.

A short history of expensive deletions — and why the second draft wins.

In 2023, Tesla deleted, by its own account, more than 300,000 lines of code from the software that drives its cars.

Not refactored. Not deprecated. Deleted: years of hand-written C++ logic, every carefully tuned rule for what to do at an intersection, replaced by a neural network that learned to drive by watching. Two years earlier they’d done something just as violent: ripped the radar out of cars mid-production, betting the whole stack on cameras alone, after shipping radar for years and building everything on top of it.

Millions of dollars of engineering. Years of accumulated cleverness. Gone, on purpose.

Elon Musk has a line for this: “Possibly the most common error of a smart engineer is to optimize a thing that should not exist.” The radar shouldn’t exist. The 300,000 lines shouldn’t exist. So they don’t anymore.

This essay is about the people who had the nerve to do that — to look at something they’d poured fortunes into and say the most expensive sentence in business: we’re starting over.

The walk out the door

In 1985, Intel was a memory-chip company being slaughtered by Japanese competitors. Memory wasn’t a product line; it was the company’s identity: they had invented the market.

Andy Grove, sitting with Gordon Moore, asked the question that should be carved above every boardroom door: if the board fired us and brought in a new CEO, what would he do? Moore answered without hesitating: he’d get us out of memories. And Grove said: “Why shouldn’t you and I walk out the door, come back and do it ourselves?”

So they did — mentally fired themselves, walked back in as their own successors, and killed the founding business. Intel became a microprocessor company, and the next forty years of computing ran on that decision.

Notice the mechanism. Grove didn’t have new information. He had a new vantage point — the outsider’s, the one with no sunk costs to defend. The strategy was always visible. The identity was in the way.

The $250 million admission

Apple spent most of the 1990s building Copland, the next-generation Mac operating system. Reported cost: a quarter of a billion dollars. Hundreds of engineers. Years of roadmaps promised to developers.

It never shipped. It couldn’t. And the bravest thing Apple did that decade wasn’t an invention. It was an admission: Copland was cancelled outright, and Apple went shopping for its own replacement, buying NeXT and rebuilding the Macintosh from the ground up on a foreign foundation. Every Mac, iPhone, and iPad since runs on the descendant of that humiliating, brilliant restart.

Steve Jobs, who arrived in that deal, later compressed the philosophy into eight words: “If you don’t cannibalize yourself, someone else will.”

The quarter billion wasn’t wasted, exactly. It was tuition — the price of learning precisely what the old architecture couldn’t do. Companies that refuse to restart aren’t protecting their investment. They’re refusing to collect on it.

We cast it all aside

The most underrated restart in engineering history happened in a bicycle shop.

The Wright brothers began, sensibly, by trusting the published science: Otto Lilienthal’s lift tables, the accumulated aerodynamic data of the era’s best minds. They built gliders to those numbers. The gliders barely flew. In 1901, Wilbur made the call that actually launched the aerial age: the data itself was wrong. As he later put it, “we cast it all aside, and decided to rely entirely upon our own investigations.”

They built a six-foot wind tunnel out of a wooden box and remeasured lift from first principles — two bicycle mechanics, re-deriving the foundations of a field, because the foundations had failed a test. Two years later: Kitty Hawk.

The lesson isn’t “distrust experts.” It’s narrower and sharper: when the artifact fails, interrogate the assumptions underneath it, all the way down. Most teams debug the glider forever. The Wrights debugged the tables.

Steel is better, actually

In 2018, SpaceX was deep into building its Mars rocket out of carbon fiber: the obvious, aerospace-grade, expensive choice. Tooling built. Composites ordered. Prototypes underway.

Then Musk did the math from raw materials up (carbon fiber at well over a hundred dollars a kilo, scrap rates brutal, performance degrading at cryogenic and reentry temperatures) and switched the largest rocket in history to stainless steel, a material the industry had abandoned for rockets decades ago, at a few dollars a kilo. The aerospace world laughed. Then it watched steel ships get built in a Texas field at a cadence carbon fiber could never have allowed.

First principles aren’t a vibe. They’re arithmetic done on reality instead of on precedent. Precedent said steel was a 1950s material. The arithmetic said precedent was confusing familiar with correct.

The pattern

Run these stories side by side (Tesla’s deletion, Intel’s identity change, Apple’s admission, the Wrights’ wind tunnel, the steel rocket) and the same shape appears every time:

1. The admission precedes the insight. Nobody finds the better way while defending the old one. Richard Feynman put the prerequisite plainly: “The first principle is that you must not fool yourself — and you are the easiest person to fool.” Every restart begins as an act of honesty, not of genius.

2. Sunk cost is tuition, not inventory. The millions spent on Copland, on radar, on carbon fiber: none of it transferred to the new design, and all of it informed it. You keep the learning. You delete the artifact. Teams that conflate the two end up curating museums of their own mistakes. (Kodak invented the digital camera in 1975, then spent two decades curating the museum.)

3. Restarts are smaller than they look. From inside, starting over feels like losing years. From outside, it’s usually the fast path: the Wrights spent weeks in the wind tunnel and saved a decade of debugging a doomed design. The slow option was loyalty to the broken thing.

4. The restarters were the most committed people in the room. This is the part the cynics miss. Grove, Jobs, the Wrights — none of them were dilettantes flitting to the next shiny thing. They were the people most loyal to the goal, which is exactly what freed them to be ruthless with the artifact. Loyalty to the mission and loyalty to the codebase are different loyalties, and every great restart is the moment a builder figures out which one they actually hold.

The famous counterargument is Netscape: the company that started over and died doing it, the rewrite that birthed the rule “never rewrite from scratch.” But look closely at what that rule was actually about: the cost of the second draft: three years of standing still while the market moved. The rule was never “the second draft is wrong.” It was “the second draft is unaffordable.” That price just collapsed. For the first time in the history of software, starting over is cheap — which means the only thing left standing between a company and its second draft is the honesty. The judgment got more valuable, not less. The excuse is gone.

(One footnote for the quote collectors: the most famous first-principles line of all (Henry Ford’s “faster horses”) was almost certainly never said by Ford. There’s no record of it before the 2000s. Which is fitting, somehow: even our favorite quote about questioning inherited assumptions is an inherited assumption that fails the check. Verify everything. Especially the things everyone knows.)

The question

Most organizations never get a Grove moment because nobody performs the ritual: walk out the door, fire yourself, and walk back in as the new leadership with no history to defend. Ask the only question that matters:

If we started today, knowing everything we’ve learned, is this what we’d build?

If the answer is yes — wonderful, keep going, you’ve earned the conviction.

If the answer is no, then everything after the “but” (but we’ve spent so much, but the customers expect, but the roadmap says) is the sound of a company choosing its artifact over its mission.

The builders in this essay all heard that sound, recognized it, and chose the other way. It cost them millions and their pride, in that order.

It was the cheapest thing they ever did.

— Haltere