I feel like this should have had a warning like the one on the Epiphyte business plan in Cryptonomicon.
"EXTREMELY SERIOUS WARNING: Unless you are as smart as Johann Karl Friedrich Gauss, savvy as a half-blind Calcutta bootblack, tough as General William Tecumseh Sherman, rich as the Queen of England, emotionally resilient as a Red Sox fan, and as generally able to take care of yourself as the average nuclear missile submarine commander, you should never have been allowed near this document. Please dispose of it as you would any piece of high-level radioactive waste and then arrange with a qualified surgeon to amputate your arms at the elbows and gouge your eyes from their sockets."
I really fundamentally do not understand what problem Gas City solves that is not already solved by normal subagent orchestration patterns. If you want to call your main LLM session the "mayor" and have it delegate its work out to planners and coders and reviewers and QA and so on, this is already a thing you can do! If you want to do this in a reusable way you can create skills and subagent definitions and use /commands, etc. Why do we need hundreds of thousands of lines of opaque Go code to accomplish any of this?
I listened to his podcast on Pragmatic Engineer. I don't think he specifically addressed what it solved, but he talked about shifting the Overton window in regards to what's possible with AI agents. I'm not arguing that he actually accomplishes this -- just noting that his goal seems to be less "create something useful" and more so "create something that gets people's attention and maybe gets them to thinking about AI in a different way".
Cynically, he published a book on vibe coding recently, so he may just be grabbing attention as some effort to boost book sales.
I'm trying to make the phrase "AI DDOSing" happen.
ex. someone's GitHub repo with a ton of code and a README written by AI claiming fantastical features not present in the code.
Or, more subtle someone, "self-DDOS'ing via AI" - thats for when "LLM psychosis" is too strong, i.e. for "I went too far down a rabbit hole with the interactive chatbot for a month and now I have 1M LOC and 95% test coverage and an app that I don't understand"
I quit my job at Google in 2023 and have spent 2.5 years working on an LLM-based agentic app.
To me, this looks like an unfortunate self-AI-DDOS'ing by someone with even more runway than my seemingly infinite runway.
It's well-meaning, like, in 2030 I'm fairly sure we'll have a meta-layer and simplistic "here's a bug, read files, edit, fix" will seem slow/strangled. But he's at least a couple years ahead of the models, and whatever metalayer exists won't have the bizarre UX model.
This is true, without people trying, we wouldn't know what doesn't work yet - for all I know he's cracked something big and in a month we'll see the first AI-built operating system (I'm not being sarcastic)
Claude Code released agent teams in Feb. that was the first of the major players to have separate process/context agent orchestration that I know of. Who did it before that of the major IDEs?
If I want to create a furnace into which I can shovel tokens (ie: money) I dont think I can do it quite as elegantly as Gas City.
Its novel and funny, but the hype around agentic coding is bad enough for some engineers to think this represents the pinnacle of current software development practices.
I can't tell if this is the Piss Christ of the software (I hesitate to call it "development") world, or if we are watching a GG Allin show. It's sobering to watch people uncritically take cues from these Steve Yegge artifacts as though the software development enterprise had always been like this.
He's very transparent about his addiction to agentic coding. He's doing it for the giggles as far as I can tell. Trolling and being a personality. I don't mean that in a bad kind of assessment. He's said as much.
Why would it be an addiction versus obsession? Getting obsessed with something and building things around it, even if it's nonsense, is artistically genuine.
It could be either. The bigger critique that people have is that the entire thing is an obvious LARP. None of this is "built for enterprise" or deployed in common ops. This is a zero-risk, zero-judgement project where vanity PRs are encouraged so that another agent can take credit for the vanity fix.
That's fine, I also have side projects that do absolutely nothing useful. Most people are taking issue with the deluded "it's all so obvious" rhetoric used in Yegge's writing.
I don’t know. I have coding so much I got into technology management a decade ago. But I can’t read myself away from personal projects, it’s addictive in a way.
yeah that's true. i used the word addiction because i remember reading him using that exact word. and including tokens and llms alongside the rise of gambling and such.
> So you should never just have one coding agent managing a piece of infrastructure. Not even for a low-stakes part of your business. You should always have at least two or three working together on a little crew.
I know why people think this, and I went down this route early on, but I’m not sure this logic follows.
Teams of agents, even adversarial ones reviewing work, are prone to the same types of mistakes as the problem they’re attempting to fix. In fact, mistakes at the adversarial/approval layer are so much worse, because the next stage assumes it was correct, and the error cascades. Maybe enough agents and tokens and context bring the probability of “correctness” closer to 1, but what is the judge of what is correct? what if that criteria is hallucinated?
I’m not sure this concept has been proven. I’ve certainly been unable to prove it to myself, and I’ve tried very hard.
Is this just urbit all over again? Making up a bunch of terms and abstractions to make a mundane thing seem novel? At least in this case it seems more like psychosis than grift. In any case, I will be interested to see if this tooling yields any actual results, doesn't seem like it so far (except maybe more tooling lol).
This is so misguided, I don't even know where to begin. If you think that throwing more tokens at a problem with your agents cos-playing whatever fiction you cook up is the road to reliability, robustness and good design, you deserve all the misery that comes your way. If you want people to take this seriously, ground your ideas with rigorous data to prove that this works better than the state of the art. Until then, this is just irresponsible planet burning, token burning propaganda.
I don't know about any ready made explanation, but it is basically multi-agent orchestration, but with lore and role-playing created by a mind deep in AI psychosis. You know how agents can call sub agents? That, but with dozens of different layers and calling things different names for extreme anthropomorphization beyond mortal comprehension.
I recommend going deep into the Gas Town rabbit hole, it's really funny and a bit worrisome. He has like a federated system for random people's "towns" to connect to and he drains his wallet on multiple Claude accounts to build software with virtual polecats.
You know there are software which are so marvelously monumental, they inspire me to improve in my craft. Among them in no particular order
- The Witcher 3 (honorable mention, for all its jank, Skyrim)
- Sublime Text and vim
- Krita and Procreate
- Early Google Chrome
- Redis and Postgres
Seeing "steve-yegge.medium.com" on the HN frontpage averts me by reflex. Like, I'm suddenly inspired to learn more about agriculture. Or contracting, I heard that could be lucrative. Heck, I'd even go full hipster and open a bookshop in this economy.
Steve, you probably don't give a shit about my opinion but I just want to know from which Jamaican zip codes do you source your supply.
The bucket on an NPC's head while you rob them blind is my favorite example of a shining success of the line-of-sight system while simultaneously being a spectacular failure of basic AI.
The other nasty bug involving pixelization that we did manage to fix before shipping, but that I unfortunately didn't save any video of, involved the maid NPC, who was originally programmed by a really brilliant summer intern, but had a few quirks:
A Sim would need to go potty, and walk into the bathroom, pixelate their body, and sit down on the toilet, then proceed to have a nice leisurely bowel movement in their trousers. In the process, the toilet would suddenly become dirty and clogged, which attracted the maid into the bathroom (this was before "privacy" was implemented).
She would then stroll over to toilet, whip out a plunger from "hammerspace" [1], and thrust it into the toilet between the pooping Sim's legs, and proceed to move it up and down vigorously by its wooden handle. The "Unnecessary Censorship" [2] strongly implied that the maid was performing a manual act of digital sex work. That little bug required quite a lot of SimAntics [3] programming to fix!
Speaking of physics, the giant striking you with a hammer and launching your body into orbit.
And of course, can’t forget the classic Skyrim pathfinding jank. The number of enemies that would go aggro only to get stuck on a bush or a slight change in elevation while you stand there at a distance peppering their face with arrows until they die.
Your examples are from 10 or more years earlier that Skyrim's creation engine.
That is to say, the older engines could have been limited by hardware requirements, or maybe decoupling physics from fps is an innovation that appeared between 2004 and 2011. Or maybe they are also jank.
Notably, the source 2 engine (2015) decoupled physics from fps (as I understand).
You answer your own question! Not remotely to the same extent. Quake, for instance, gave a small advantage for jump height [with high FPS]. Skyrim would outright break.
Also, you've listed three generations of the same family; goldsrc, the child of Quake, predates Skyrim/Creation by at least a decade. Of those, Source would be the timely match. Not even close to the same amount of jank. Just... no. You aren't tricking me into writing lists.
I don't really intend to be critical of Skyrim, like many: I love the jank. It's expected. It's a Bethesda game.
Came here to say that a lot of these LLM posts make me feel like I was hit by a hammer and I can't understand the world anymore. Thankfully, the HN comments confirm that this is as insane as I thought it was.
I think it's because LLM output tends to be snippets of truth, connected in such a way as to be subtly false. It leaves you disoriented because it is in the uncanny valley of truth: you know it's not right, but can't put your finger on why it is not right. It's made worse buy its asymmetric nature: it takes seconds to generate pages of words and hours to figure out what is wrong with them, for little to no gain to the reader.
"The uncanny valley of truth". That's a brilliant way of putting it. Thank you.
> It's made worse buy its asymmetric nature: it takes seconds to generate pages of words and hours to figure out what is wrong with them, for little to no gain to the reader.
I think you've just formalized the best argument for the Real Butlerian Jihad we're ever going to get.
Deploy packs of Gas Cities with Gas Country! All of that inside our new Gas Planet, where multiple sovereign Gas Countries can cooperate or go to war with each other! All hail our Great Comrade Polecatius of Gasoland, the greatest coding leader of Gas Planet.
"EXTREMELY SERIOUS WARNING: Unless you are as smart as Johann Karl Friedrich Gauss, savvy as a half-blind Calcutta bootblack, tough as General William Tecumseh Sherman, rich as the Queen of England, emotionally resilient as a Red Sox fan, and as generally able to take care of yourself as the average nuclear missile submarine commander, you should never have been allowed near this document. Please dispose of it as you would any piece of high-level radioactive waste and then arrange with a qualified surgeon to amputate your arms at the elbows and gouge your eyes from their sockets."