▲ 87 points
back
79 comments
Gas Town was intended to be reusable, but I only ever wound up using it to build itself. Gas Town fell apart at the seams with Opus 4.7. Up through 4.6 it was working brilliantly. With 4.7 we saw the introduction of the "just two more things" tic, which prevented Opus from ever converging on being ready to do real work—it always wanted to fiddle with Gas Town itself. The Opus tic never went away, so Gas Town effectively burned down. It had other problems, too, but 4.7 was the final straw.
"True Gas Town has never been tried" etc.
With Wheelhouse, I have reinvented something strangely Gas Town shaped from first principles, but it's running many more agents, and they are far more organized. This has shown me the shape of the changes coming next year. I'm operating about 12 months in the future. I managed this only through luck: for the past 18 months I have had the magic trio of time, money, and energy, and I've been spending it all on figuring this stuff out. And I am grateful for the privilege.
I'm still half-convinced this is Yegge's version of Andy Kaufman's professional wrestling career, "Gas Wheeltown was an 18 month long sociological experiment conducted by Thiel University. The study is now complete, thank you for your time."
I used to like Steve's posts. They were all way too long, but a fun and interesting read nonetheless. He started to lose me when at Sourcegraph where he oversell Cody which didn't do anything useful (I even had a commercial meeting with Sourcegraph people who weren't of much help); then he completely lost me with his GasTown business where AI meets crypto (what could possibly go wrong with that). Now I can't read anything from him anymore ;-(
It is a rough go for a long-time reader, the bragging tone of six figure engagements and “whales” (i.e., suckers) keeping the game going. The statement that what he is doing is what the reader will soon catch up to feels like someone who can no longer see land.
That said, as another nut with an agent factory toiling away around me, I’ve also started to think about replacing any kind of CI in the cloud with local. I couldn’t quite keep myself at the article to read what comes next, so I guess I will catch up to it down the road.
> They were all way too long
And still are. When I got to his "Lots to cover today! Let's get to it.", I was like "wait, we haven't started yet?"
I used to read his stuff regularly prior to his Sourcegraph days, and I don't recall why I stopped. Completely agree that he was always fun to read.
So I was really disappointed to start reading this one (I'll leave the tab open, but probably won't finish it). Sounds like he's pretty out of touch with reality these days and isn't grounded at all.
Like the sibling poster, I gagged a little at his casual mention of six-figure teaching engagements, and the "suckers" (that's what "whales" read to me as well) keeping his game alive.
I smell an architecture astronaut getting high on abstractions. Taken far enough, it starts to look like psychedelic ramblings of a manic schizophrenic losing grip on reality.
Interestingly, the author actually has a product, something usually missing in claims of the new LLM powered productivity nirvana. Hopefully, with time, there will be a way to verify the effectiveness of all of this fire and fury, measured in actual worthwhile deliveries of this product.
> But my Wyvern development has been burning the equivalent of $87k/month of API token burn, or about 69 billion tokens in July (96% cache hits, fortunately). So I do have to worry about it. I don't care how much money you have; your spouse is not going to let you spend almost ninety thousand dollars a month on your hobby game. I didn't even bother asking.
> My solution has been to create a token tap on $200 Max accounts, which for me work out to ~30x the list-price equivalent. So in reality I'm only spending about $2800/month out of pocket for my $87k "worth" of tokens. Though that number keeps growing alarmingly.
But how much does the game bring in?
> Once you're hooked on Wyvern, you're hooked for life. I have whales that have spent thousands on it.
At minimum 4k (given by whales = min 2 whales, thousands = min 2 thosands :) ) So I'm guessing that this covers Claud accounts.
edit: taking a look at https://yegge.ai/wyvern - probably outdated - the revenue is at 5k.
I tried playing it. It's... not going to change the world of gaming.
Keep in mind median lifetime game revenue on steam is somewhere between $250 and $1000 depending on how you slice it...
I don't like using the term "AI psychosis" because it originally meant a narrow definition resulting in genuine psychosis and acute mental health trauma, but seems to have widened to anyone falling down a much wider trapdoor of AI engagement.
But I read, or at least attempt to read, something like this, and I can't help but marvel at the capture that AI agents seem to have on some individuals.
I expected Yegge to, at some point, assuming this isn't still all satire, to take a step back and take stock that the basic harness are improving faster than the rate at which these complex beasts can adjust.
But the opposite has happened, he's dived fully into believing that the secret to unlocking the power of AI is to wrap more layers of AI around it.
If you actually want a problem worked on all night, just fire up fable, type /goal and then describe the goal. With permissive run settings it'll crunch for as long as it needs.
I've not seen any evidence that wrapping that in a further 3 or 4 layers of agents improves anything.
I think the domain "yegge.ai" says it all. It's becoming like the NFT craze, and I say that as someone who uses AI every day.
I mean, the dude uses a lot of jargon, ego, and mythologizes stuff but the article comes down to a pretty simple practice:
Run a big task queue, have a smart model manage a cheap model on every ticket, set up a cron for models to automatically work off of events, stop gating merges on builds, and expect to spend a quarter of your time maintaining the machinery (all of it currently subsidized by many $200 accounts).
I'm still on the fence whether Yegge is being satirical, some phrases are just too on-the-nose:
> Gas Town was intended to be reusable, but I only ever wound up using it to build itself.
MOE and COT or github co-pilot. I think agents can marginally improve the performance of AI.
I call it "yegge sweats"
His game has only a handful (like literally 5-8) players [0], at least on Steam.
The game was first released in 2001, but released on steam on 2021. Needless to say, most players don't play through steam.
I watched Steve talking about GasTown in some podcast but I didn't understand it at all. I read some part of this article didnt understand what he wanted to say.
I still remember his very old blog post where he rants about Google and Bezos, after reading that I thought the author of this post is very insightful but when I read his post I have no idea what is he talking about.
This is what he's spending '$87k worth of tokens' building a month btw: https://store.steampowered.com/app/1541710/Wyvern/
This post made me write manual code, there's something about hearing the description of this that makes me worry the bubble is going to come home to roost.
The obvious danger of spending so much on inference and creating ever expanding complexities feels like the height of crypto psychosis
If this harness was achieving major mathematical proofs or creating really consequential work, I think I'd feel differently - but I'm unclear based off of this article what this achieves that a typical harness couldnt
nb: current models (e.g. GPT 5.6 Sol) are very good at long horizon tasks
they no longer need crutches or rube goldberg machines to keep them going
minimal agent harness is just a loop that loops until no more tool calls are coming
GPT 5.6 Sol continues to drive the loop until the task is done or it decides that it wants to present the user with information
at that point it is probably good to not automatically continue (!)
(YMMV of course, for some tasks it makes sense, then you can still add a loop around it + the necessary signals, the main thing I want to say is that what used to be essential to keep models going is no longer needed, current models can do long-horizon tasks way better than when these outer loops where necessary)
self-plug: "smol", is a minimal agent in ~20 lines of Go that implements this pattern (keeps going until no more tool calls):
https://github.com/smol-env/smol
works just fine
I hear so much about people running into token limits and using bajillions of tokens. Am I doing something wrong? I'm finding it hard to run into the limits of a Claude Max (20x) account. Well, perhaps the 50% Fable limit is something I have to be aware of, but otherwise this more than I can consume. I am the limiting factor: my capability to review code and design documents.
I use it to develop a large MRP-style app. This is as complex as things get.
Is it Clojure that makes things so efficient? My work style (lots of design documents)? I really don't know.
Do people skip the "review the code and read the documents" step perhaps?
Holy moly https://yegge.ai/essays/model-welfare/ ! Talk about spicy takes. Fable is sentient, be nice to it, or in six months I won't be your friend.
fwiw, i give it a 1/3rd chance he's right.
All this for a game that has 123 "Mixed" reviews on Steam? My face was twisted in cringe the entire time. Oof.
It would have been a lot more interesting if her has just explained Beads (the only thing her claims is novel).
So probably there is nothing here.
Very painful read.
I hate to say, "I told you so" about Gas Town, but, I really told you so. I was making fun of it immediately, because it was obviously a token furnace and literally nothing else and could never be anything else. Models don't want anything. You can't set them loose in a vague "do something" loop and expect anything good to ever come of it.
Yegge's a smart guy, and I've enjoyed his writing and talks (historically), but AI has really done a number on him. I mean, I'm kind of all in on this stuff, too, but I'm rarely claiming to know what's happening 18 months ahead (and when I do, I don't brag and declare victory when I get it completely wrong, as Yegge has re: Gas Town).
Another banger from the mind that brought us Beads, the solution to having your repositories littered with Markdown files, written in a svelte 250,000 lines of Go.
What's that? The Beads repo itself is littered with Markdown files? And now we need a "brain" to go with Beads, which is also a directory of Markdown files? Weird, kind of makes it seem like the whole thing was performance art.
> Beads is unfortunately still a bit janky, because its unique work footprint strains databases pretty hard
This is a sign that your architecture completely jumped the shark by the way.
"On the side I do occasional six-figure gigs where I fly to companies and teach them my techniques, and that helps with my (considerable) token bills."
Here is the important part. The cult pays for the narrative.
I wondered why previously sane bloggers succumb to AI psychosis. Money and ambition is always involved to some degree.
> Another (small) architected respect dimension is gender pronouns. They can be a bit awkward with agents. I was using "they/them" intentionally with all my agents, but I noticed they were often defaulting to human patterns by using he/him for power-roles like Lion, Marshal, and Seneschal. So I started using "she" for obviously-female characters like Goose, and they picked up on it, at which point it was unclear who wanted what.
Get this guy some help.
I need to give this to an AI and ask it, "Does this post contain any information which will help me use Pi Agent Harness better?"
Anything Steve ends up building, will become obsolete after a few months of default harness and model improvements.
This is a doomed niche.
> CTOs keep asking me if code review is really dead
I don't think anyone should be asking someone so clearly AI-pilled this question.
That said, my code-review now looks very different than it did, which is to say that my code review now looks like Fable pulling together rounds of review from Sol and Grok 4.5, fixing anything obviously broken, and then /grilling me for anything that needs good taste. I am not reading code unless Fable really wants me to. Every day or so we stop and do a "look for accretion and LLM-slop" pass to keep things clean.
Dear Steve, I hope you will realize that using swarms of agents and burning millions of tokens is not the way to go. It only produces slop, thousands of lines of code, and walls of text, just like your blog post. Less is more.
GasTown was crazy but had a rough shape that was ahead of its time. I think https://github.com/kunchenguid/firstmate is a much better, and easier to understand version of agent orchestration.
> Now I know exactly what is going on. I know how enterprises will operate, how humans and models will interact. And I am out there building it. I want to be first.
This guy's under a heavy AI psychosis
"Beads is the best. If you're not using Beads, you're failing. Gas Town, which I have just told you was a failure, was built on Beads. Gas City, which also failed, was also built on Beads. Wheelhouse, which I'm working on now and will be awesome? Beads. Therefore you should use Beads."
I always enjoy reading these posts of unique experiments
We never really know what will work until we build "something" and it's often a messy process to create totally different things that no one else is making
So, it looks like the experiments are coming along well...
I really don’t like this guy.
I don't know about the tech implementation but "Don't be a dick to LLMs" seems like solid advice.
...Is Steve okay? Anyone on here know him personally and able to reach out?
Wft. I have a piddly little Claude Pro account, I use it every day, and Opus 5 and Fable (judiciously used) are very helpful and useful in my work and personal life. I am happy with my setup. Then I read this, and I imagine it's what primitive farming societies felt like when introduced to combine harvesters, high-clearance sprayers and mega tractors.
I am happy on my farm. I like the carrots I plant by hand. I do not wish to control farming equipment worth a small fortune to till acres of land every day.
(P.S. Still not sure if TFA is satire! Obviously, this assumes that it's not.)
Essays became way to easy to write with AI.
It used to be a basic filter that if text was long the author put a lot of thought (even if misguided) and effort (even if low quality) into it.
Not its just another torrent of words.
We need proof of work on long texts, a service that guarantees that the author was whipped for more than 20 lashes for this text to exists and thus we know he cared enough to publish it.
Some interesting concepts, but reads like satire.
People seem to be struggling with whether or not there is satire here.. isn't this an intentionally ambiguous but absolutely intentionally satirical style that he's having fun with but also playing with deeper points?
There are some good ideas buried in here but I'm afraid at this point that Steve has tipped over into a bit of AI psychosis. It's really a tremendous problem right now for many smart people in the industry.
It can be hard to tell what is real signal and what is hype, especially in Silicon Valley where there's often a status game being played of who is the furthest out on the frontier, which creates a huge incentive to lie.
We are still in the "1998 moment" of AI, the bubble needs to burst before we can see what are the good ideas and what things are laughable.