Seven years ago the world’s foremost running scientists got together, crunched the numbers, and determined that the sub 2 hour marathon will be achieved–let me check the models again–in 2032. There was a chance it might happen before, but it was slim. A 5% chance for 2024!

Last weekend Yomif Kejelcha of Ethiopia finished the London Marathon with a time of 1:59:41. He finished second. Kenya’s Sabastian Sawe won the race at 1:59:30.¹

Which goes to show: we’re good at predicting what is possible, but very bad at predicting how soon it can happen. My heart says to keep aiming for a better world but my gut tells me it will get much worse sooner than we imagine.

Fourteen years ago if I wanted to have a button do a little scale and opacity animation in Javascript (I’d have preferred a pure CSS solution but go with me here), I’d have added the GSAP library and written, by hand, something like:

gsap.to(".cta", {

  duration: 0.667,

  scale: 1.2,

  opacity: 0.5,

 ease: "power1.inOut"  

});

In 2026, during the Slopothecene, in tech companies that rely on AI, push AI on its workers, and where performance is measured by AI tool adoption, you’d now write a prompt for a Large Language Model. You’re not expected to write code anymore; you’re expected to, natural language, talk your way through it and get an agent to create it for you. Something like:

Scale up the CTA to 120% it’s normal size and lower the opacity to 50%, over three quarters of a second. Use an in Out ease in for the animation. Do not talk about goblins. Please don’t talk about goblins or ghouls. Don’t make mistakes.

Then you wait a while, hope the service is up, have your credit card charged a few bucks, and then pray the result does what you want. Otherwise repeat. Keep your credit card ready for more token charges.

And the above prompt isn’t even a joke. OpenAI’s Codex literally has a system prompt that says: “never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures unless it is absolutely and unambiguously relevant to the user’s query.” That means when you prompt it for code, deep inside it’s being told to not talk about goblins while it’s trying to output a javascript animation.

This is better, apparently.

As AI companies continue to lose billions not only will this not get better it will also get considerably more expensive. And lest you think you can just write the javascript the old fashioned way, the former will not be marked with AI provenance. And as we established, that provenance is important as your performance, and job, is tied to shipping code via agents and hitting certain thresholds. You will be ranked on leaderboards.

Nothing bad can come of this.

Twenty-eight years ago, during much simpler computing times, I briefly got into emulation. After a few years, once Nintendo started putting emulated games into Animal Crossing on the GameCube (and in Metroid Prime if I remember correctly), once they started porting old games for the GameBoy Advance, once Microsoft launched XBox Live Arcade, which had many arcade ports, I lost interest. There were enough games as it was and the old ones were getting “official” releases when, at the time, a lot of emulators still had problems with accuracy.

On the underpowered computer I had then I liked NESticle (Thanks Shitman!), and MAME, and played with Snes9x, but mostly I loved ZSNES with that floaty mouse cursor and general jankiness. It wasn’t perfect, support for many games was lacking, but it’s how I first played Final Fantasy V (didn’t beat it), Secret of Mana 2 (didn’t beat it), Terranigma (didn’t beat it) and a bunch of other games I had missed or were never released. I think I tried to play Front Mission but I don’t think a translation patch was released until much later.

Since then emulation has improved drastically. Some emulators claim full accuracy, they mimic Cathode-Ray Tube televisions, and even run on specialty hardware. Into this environment returns ZSNES as Super ZSNES.

It still has that very outdated ZNES user interface, it doesn’t go for full accuracy and instead tries to add game specific enhancements, it allows for uncompressed audio, upscales textures, adds 3D height maps to mode 7 effects, and has a random falling snow effect for no reason. Games with special chips aren’t supported, it’s missing netcode, it has many bugs, and it is janky as fuck: I love it.

I’m not a fan of the default upscaling, it looks like a muddled mess to be honest, but one of the emulator’s Key Features is “No Vibe Coding. Classic development style.” Hell yeah.

Last October, zophar (of famous emulation site zophar’s domain) had an interview with the developer. 30 years on and I just learned that the creator of ZSNES is Canadian which means it’s officially pronounced “ZED S N E S”. 

1. It’s worth noting that the 2hr mark on a marathon has been beaten but in an unofficial capacity in a non-competitive race specifically designed to beat that mark. That it was done on a normal street race, and twice, is amazing.

Related Links

Martin Galway C64 Music (github)

Speaking of old consoles and ‘classic development’, video game composer Martin Galway released source file for some C64 music in pure assembly.

Listen to This

If ZSNES wants to make real enhancements have it replace the already weird and great Secret of Mana soundtrack with its composer’s, Hiroki Kikuta, arranged version: “Secret of Mana+”. It’s probably the most bonkers arranged version of a game soundtrack from that era, a weird prog rock and new age amalgamation with some unusual sample selections (phone dialing as a theme throughout? Is that a boat engine?? Bird calls over water splashing, sure why not.) I love it:

Thanks Shitman!,

sometimes