Let me make the case.
Donkey Kong, 1981. The jump arc, the barrel physics, the way Mario's momentum carries him off a platform if you mistime the jump... those aren't accidents. Those are tuning decisions, made by a person with a specific theory of how a player's body would feel a falling character. Miyamoto sat next to programmers and iterated on numbers until the feel was right. The programmers typed. Miyamoto decided what they were typing toward.
Super Mario Bros, 1985. The decision to put an enemy right after the first pipe so the player learns to jump by dying immediately. The decision to make the first screen teach the entire game's vocabulary without a tutorial. The decision to let Mario's jump height scale with how long you hold the button. These are architectural decisions about how information flows between a machine and a human nervous system. If you think those decisions don't count as coding, you don't know what coding is.
Zelda, 1986. A battery-backed save system at a moment when every other game used passwords, because Miyamoto decided the player's time was worth protecting. An overworld designed to reward exploration instead of punishing it. A dungeon system that teaches you the rules of its geometry in the first room. Architectural calls.
He made these decisions without writing code himself. He made them by understanding the machine, the medium, and the human at a level most of his programmers could not reach. When you play a Miyamoto game and it feels right, that feel is the output of a design process that operated above the opcodes and above the data structures, at the level where the whole system coheres or doesn't.
The industry calls that "design," and files it in a separate drawer from engineering/coding.
Three levels of coding
There are roughly three levels at which coding happens, and most discussion of coding only recognizes one.
Level 1: syntax. Loops, types, recursion, the standard stuff. How you write a correct sentence in the language. This is what bootcamps teach, what LeetCode measures, what junior interviews filter for. Most computer science graduates can do this.
Level 2: flow. What you do with Level 1. Formally correct and good versus formally correct and terrible. Choosing the right data structure. Knowing why the academically beautiful solution takes twenty minutes and the ugly pragmatic one takes three milliseconds. Most computer scientists fall off here. The ones who don't usually do it by instinct without being able to teach it.
Level 3: architecture. Macro decisions. Full awareness of all the consequences of each call before you make it. Why your node system is trigger-pull-based and not event-pushed. Why your data layer returns synchronously even though everyone expects async. Why you chose a flat table over a pretty schema because the retrieval pattern matters more than the shape. Level 3 is where systems either cohere or silently fall apart two years later, and you usually can't tell which it is until it's too late.
The levels are not mutually exclusive. Truly good coders work all three. A decent engineer lives at Level 2 most Tuesdays and gets dragged into Level 3 on Thursdays whether they want to or not. Nobody is pure anything. And Level 3 judgment usually grows out of Level 1 time: the architects whose decisions hold up ten years later are, overwhelmingly, people who spent a decade at the keyboard before they stopped needing to. Miyamoto is a genuine exception, not a template. The point is not that Level 1 is beneath anyone, but that the industry stopped at Level 1 and called the map complete. Likewise most developers operate and discuss Level 1. Some discuss Level 2. Very few can or enjoy discussing Level 3, because most have never had to make Level 3 decisions, they've worked inside architectures other people made, and they've learned to pattern-match on the surface without understanding or caring about the choices underneath. Concepts like OOP, functional programming, procedural design are used more as symbols of belonging and tribalism than anything. Most devs pick their hammer and then treat every problem as a nail.
Miyamoto operates almost entirely at Level 3. He doesn't operate at Level 1, he delegates it, and has for nearly fifty years. And he has produced more technical innovation than most programmers do in an entire career, because Level 3 is where the innovation actually lives and Levels 1 and 2 are where it gets implemented. When the Level 1 and 2 people are willing to implement it, anyway.
But the industry can't measure Level 3. It doesn't even have vocabulary for it that isn't squishy. "Architect," "principal engineer," "senior staff", these are job titles that gesture at Level 3 without describing the thinking it requires. Meanwhile the hiring pipeline is almost entirely calibrated to Level 1. Whiteboard coding, LeetCode, "implement a red-black tree." All Level 1 measurements. Miyamoto would not pass a modern coding interview.
Okay, confession time
I've been setting you up. Miyamoto was bait. Partly honest bait, I do think the case holds, but bait.
Watch what you just did. Some of you flinched when I called him a coder. Some of you translated it silently to “designer” and kept reading. Some of you are still waiting for the retraction. Fine. Now swap the name. Would you have flinched if I’d said it about Gunpei Yokoi? Or about any other random person standing on the same blurred line without a famous name to shield them?
No. You wouldn't have. Because the flinch is only reserved for people the industry has already decided are important enough to argue about.
The industry lets Miyamoto be a genius. It does not let him be a coder. Those are different containers, and the border between them is patrolled harder than almost any other line in this profession. Designer is allowed to mean taste, vision, intuition, feel. Coder is reserved for the people who type. A definition as narrow as narrow can be. Not architect, not systems thinker, not person-who-decides-what-the-machine-should-do, no. It is: "Person who types". The entire conceptual territory beyond the keyboard has been recast as "design" and quietly removed from the coding conversation, because coding as a word has been collapsed to its most accidental layer. Philosophically speaking.
That collapse is the measurement regime in miniature. The industry decided decades ago that coding means Level 1 activity, and anything above Level 1 belongs to someone else, someone softer, someone less technical, someone whose contributions are nice but don't really count when the chips are down. The work Miyamoto actually does - the architectural decisions that determine whether the system coheres - has no home inside that definition. So it gets filed under "design," which is the industry's word for "important but not coding."
But Miyamoto is the easy case. At least the industry fights over him. At least there's a border to patrol, a category to put him in, a name on the credit. The harder case is the one where even that bare minimum doesn't happen.
The actual person I want to talk about is Gunpei Yokoi, and most of you have never heard his name.
Gunpei Yokoi
Yokoi was hired by Nintendo in 1965 as a toy maintenance technician. Not an engineer. A guy who fixed the machines that produced hanafuda cards. He had an electronics degree from a mid-tier technical university. No programming background. No industry pedigree.
He noticed that the president of Nintendo kept walking past his workbench and looking at a wooden extending arm Yokoi had built as a hobby. The president asked if Yokoi could turn it into a toy. Yokoi did. It sold millions of units. Nintendo's transformation from a playing card company into a toy company started with a bored maintenance guy and an extending wooden arm.
Yokoi kept going. He designed the Game & Watch in 1980, a handheld LCD game that used cheap calculator displays instead of proper graphics hardware because the calculator displays were mature, cheap, and understood. He designed the D-pad that same year as the input for Game & Watch, the four-direction cross you are holding right now on every controller you own, because every controller since 1980 has copied it. He designed the Game Boy in 1989, shipping it with a grayscale screen at a moment when every competitor was releasing color handhelds, because the grayscale screen had battery life measured in double-digit hours instead of single digits and Yokoi understood that portable meant battery first, color second.
The Game Boy outsold every color handheld competitor by an order of magnitude and kept doing so for a decade.
Yokoi had a name for this philosophy.
Kareta Gijutsu no Suihei Shikō. Lateral thinking with matured technology… except “withered technology” is a better fitting translation, as that is how the industry has treated his inventions time and time again.
Take cheap, unsexy, mature tech that the industry has written off, and find a use for it that the bleeding-edge people can't match because they're chasing specs instead of systems. Withered tech is cheap, well-understood, forgiving, and interoperable. Bleeding-edge tech is expensive, brittle, under-documented, and obsolete before you ship. Yokoi built Nintendo's entire handheld division on this philosophy, and every other Japanese electronics company spent the 1990s trying to figure out why their expensive devices kept losing to Nintendo's cheap ones.
The D-pad. The Game Boy. The entire handheld category as it exists today. Those are Level 3 engineering calls made by a guy who started as a maintenance tech and never wrote a line of code.
The mythology machinery kicks in, and becomes the proof of why Yokoi's case is worse than Miyamoto's. Miyamoto walked into Nintendo in 1977 with a pencil and a guitar. That is the origin story that gets told, and it gets told because the industry knows what to do with it. Pencil-and-guitar is the artist-auteur container. Steve Jobs in a garage. Woz soldering in a bedroom. The industry has a filing cabinet for the creative young person who showed up with a dream and bent a company around them. You put Miyamoto in that cabinet and everything he touches afterward gets read as authored work, because the container tells you to read it that way.
Yokoi walked into Nintendo in 1965 as a maintenance technician. That is also the origin story that gets told, and the industry's filing cabinet for it is much smaller and much worse. The narrative that attaches to him (if at all) is "lucky janitor." The boss happened to walk past his bench. He happened to have whittled an extending arm as a hobby. He happened to be around when calculator screens got cheap. Every Yokoi achievement, under this reading, becomes something that happened to him rather than something he did. Street smarts at best. One-trick pony at worst. Whereas every Miyamoto achievement, under the pencil-and-guitar reading, is authored by definition. Same company, same era, and the mythological machinery produces opposite results.
His achievements are engineering achievements, full stop. The D-pad is in every hardware history textbook. The Game Boy is a canonical case study. Withered-tech doctrine gets cited in product strategy talks at companies that have never heard his name. The work made it through. The attribution didn't. His achievements were absorbed into "Nintendo" as an institutional voice while Miyamoto's were personalized as "Miyamoto, the legend". And he gets the border fight: designer or engineer, which container, which credit. And Yokoi? well, he gets no fight at all, because there's no one left arguing on his behalf. His work is namelessly scattered across the engineering canon. His name is in the footnote. The canon and the footnote are not connected.
That is a worse failure than the Miyamoto case. With Miyamoto, the industry at least knows whose work it's miscategorizing. With Yokoi, the work is categorized correctly, the person is written off as lucky, and the attribution is just gone.
The regime got stronger
The industry's blindness to Yokoi isn't an accident. It's structural. The industry only measures Level 1 reliably and Level 2 occasionally, and has no vocabulary for Level 3 that isn't vibes. The Yokois and the Miyamotos get classified as "designers" or "visionaries": nice words, soft words, words that imply they don't really belong in the engineering conversation... while the Level 1 people who implemented their ideas get the engineer title.
Then AI coding tools arrived, trained on every scrap of Level 1 code ever pushed to GitHub. Everyone read it backwards.
People thought AI would make Level 1 coders ten times more productive and leave the architects behind. It's closer to the opposite. AI does a growing slice of the Level 1 work now, and that slice is growing faster than the industry's ability to readjust to that change. Level 1 coders worry for a reason, and their anger is rational.
When I say Level 1 coders are worried, I do not mean "about to get fired." Or rather, they might be, but that's not a technical prediction, it's a prediction about middle and upper management, and middle and upper management are more likely than not to not understand the difference between the levels in the first place. Plenty of excellent Level 2 and Level 3 engineers will get laid off by executives who think they subscribed to a three level genius and actually bought a Level 1 LeetCode grinder. That's a management problem, not a measurement of the engineer.
Level 3 people are having the opposite experience. Level 3 was always bottlenecked by Level 1 execution. You had the architectural vision. You knew the shape of the system. Getting it built required either becoming fluent in Level 1 and 2 yourself or finding people who were and convincing them to execute your vision. Both are expensive. Both are slow. A Level 3 thinker with a verifier and a harness can now ship what used to take a team, because the Level 1 machine is happy to produce, happy to be rejected, happy to try again ninety times until the output matches the specification. A human Level 1 coder would not have tolerated that process, rightfully so. The machine does.
The people who were always architectural... the Yokois, the Miyamotos, the people whose value lived above the opcodes... are not being erased by AI. For now. They are being unlocked. The measurement regime that couldn't see them is losing the ability to constrain them, because the thing the regime measured is no longer the thing that decides who ships.
What Yokoi would say
Yokoi's withered-technology doctrine was never really about cheap hardware. It was about recognizing that the industry was measuring the wrong thing. The 1989 handheld market was optimizing for color screens because color screens were measurable. Battery life was also measurable but harder to fetishize. Portability-as-a-system, the whole question of whether a person would actually carry this thing in their pocket and use it for six hours on a train, was not measurable at all. Yokoi won because he solved the unmeasured problem and his competitors solved the measured one.
AI coding is withered technology in exactly Yokoi's sense. It's mature pattern-matching applied at a scale the industry didn't take seriously because it doesn't look like what the industry thought AI would be. It's not HAL 9000. It's autocomplete. Flawed autocomplete. And the people who figured out what to do with that early on are eating the lunch of the people who were waiting for the real thing, just like the people who figured out grayscale-plus-battery ate Sega's lunch in 1989.
AI is very good at the accidental work. The essential work, deciding what the system should be, how it should cohere, what gets built and what gets cut, is still entirely ours. For now. Whether that stays true in two years, five years, ten, I don't know, and neither do the people telling you they know. The deciding layer may be a permanent human province or it may be the next thing to fall. Anyone confident either way is probably selling something. Right now, today, the machines collapsed the cost of the typing layer quite a bit, and they have not meaningfully touched the deciding layer. Work from that. Watch the line. Move if it moves. This is my best guess.
The Level 1 coder arguing that "real engineers write hand-crafted assembly" right now is the 1989 Sega engineer arguing that "real handhelds have color screens." In both cases, they're not wrong about the craft. But they're probably wrong about what the craft is for.
The door
If you're a coder reading this and you're angry, I understand. But the worst thing you can do right now is double down on Level 1 signaling. The LeetCode grinding, the "real engineers do it by hand" posture, the insistence that craft means typing is the direction that gets you nothing, because the machines do that pretty well by now and are going to keep getting better. Sorry.
The door is Level 2. Level 2 is where the judgment lives. Knowing which ugly pragmatic solution to ship instead of which beautiful academic one. Knowing when to break the pattern. Knowing why the performant answer is wrong for the context and the slow answer is right. Level 2 is taste, and taste is exactly the thing the models don't have and can't fake. For now, anyway. A coder who levels up to Level 2 is safer than one who doesn't. A coder who tries to defend Level 1 turf by insisting the craft is the typing is doing the one thing guaranteed not to help.
And for the Level 3 people watching all of this: LEARN Level 2! Then - and only then - this could be your decade. Build the verifiers. Build the harnesses. Build the machines that reject everything that isn't aligned with your goals and ship whatever survives. I am confident you can do this. Yokoi figured this out in 1980 with a calculator screen. You can figure it out in 2026 with a language model. The tool is different. The method is the same.
Yokoi is dead. He died in a car accident in 1997, two years after leaving Nintendo to start his own company. The industry he built is worth hundreds of billions of dollars. His name is in a footnote.
Here's to Gunpei Yokoi.
P.S.: A+ to those who found my Brooks nods.
But what do I know. I'm just an artist.