RobbieTheWagner 
Adam Argyle Welcome to Syntax! Welcome to a brand new episode of the front-end happy hour podcast. Welcome to this week's JS Party! Live from ship-shaped studios, this is Whiskey Web and whatnot. With your hosts Robbie the Wagner and me, Charles William Carpenter the third. That's right Charles, we drink whiskey and talk about web development. I mean it's all in the name, it's not that deep. This is, Whiskey Web and whatnot do not adjust your set. Whiskey Web and whatnot is brought to you by... Today's episode is brought to you by Warp and their rad new cloud agents. Check it out at oz.dev.oz.dev. You know we're already big fans of this app.
I keep a long list of all the features that I love. Now with oz and Warp's cloud agent setup, I can escape my machine and trigger cloud coding agents from anywhere. From Warp, the OZLI, the web app, which I can do from my phone, Hell's Yeh, a Cron job, GitHub actions, or an SDK that they so graciously provide. Think about it. With just a Docker file you can front-load a sandbox with context and have agents open PRs across multiple repos at once, all with a single prompt. Warp rocks, enough said, all right back to the show. Hey, what's up everybody? Welcome to Whiskey Web and whatnot with your host, Robbie the Wagner and Adam Thomas, our guide, the nerd.
You're absolutely right, Robbie. Thank you. Yes, I've been hearing that a lot lately, so it must be because I am just so knowledgeable on so many things. I've been hearing the thing a lot, which is, you're right, I should have checked my memory to do that, but it'll never happen again. That's what I keep hearing over and over again. That'll never happen again, happens 20 minutes later. And I'm like, you just said you would never do it. You're lucky if it waits 20 minutes. Yeah, like trying to generate episode art, the first one generated fine, but it didn't do what I wanted. So I was like, oh, yeah, here's some tweaks to the prompt or whatever.
And then just a couple minutes later talking to the same thread, it was like, I forgot how to do it. And I'm like, are we just did it? Are you kidding me? Like not even no time at pass. So yeah, it's a man. I mean, I guess there's fixes for that, but we're still. It's hard out here for us early adopters. I'm also like cheap too. So like I have all these open claw rules and I'm like, you can't reread the whole thread when you answer me. That's just too expensive. So you got to crawl one by one until you feel like you've got the answer. And it just won't even crawl one now. And that's why it's always so wrong.
It's just like, I'm like, hey, go adjust that thing. And it's like, what thing? I'm like, you're in a thread. Oh, my goodness. The site, we only have four messages in the thread right now. And you can't know the, oh, my goodness. Yeah. Yeah. I mean, I talked about this probably back when AI still couldn't really do anything about how I'm waiting for the day that it's just super lazy. Like it's, if it's trying to emulate people, it's going to be lazy. And it's going to be like, oh, God, I have to do work. Oh, man. Okay. So I could go look at these docs or I could like Google search. Use the AI, like, break down and just assume that that's right.
And like, take a little bit of info from that. Hallucinate the rest and wrap it all up for you. And I think that's where we're at right now. Yep. I'm seeing it already. I've had it say like, oh, I didn't do all the, because it was too hard. I was like, what do you mean too hard? Wait. Yeah. Somebody, it's still not there. Everyone's like, it got so good at the end of 2025 into December. It's just the whole world shifted. And I'm like, I don't know. It's still incremental. And I'm waiting for better models. Well, it did get better. It got a lot better. But the context issue remains. Like, it's good until it compacts.
And then it's bad. Like if you need anything sufficiently complex, it's not good. If you need one thing, you can fit in the 1 million context window and just talk to it about that. It'll do a really good job. It's like a, after 500,000 tokens, it's pretty much context rot anyway. So even trying to get to one, you're guaranteed to, you're not going to have good results using the 1 million context window because by the time it's that full, it's going to lose it all. Yeah. So you should just can't keep it all in this memory. Clamp it half a million anyway. Yeah. Okay. Okay. I want to pour some whiskey and then I have a toast.
Yes. So before the computers delete our stuff, let's drain every cask in the village of Muff. Yes. So we have a fun one today. We decided to not do it with a guest due to the raunchy nature of the name perhaps not how they meant it at the time, but they did. Okay. Well, you mean the Muff liquor. Yeah. The Muff liquor company. So yeah, we're not going to tell you that much about it right now, because we're trying moving the whiskey explanations to the end. So if you like whiskey and you want to hear more, hear our rating, et cetera, you'll get that at the end. We're just going to drink it for now.
Yep. I just want to sip it while I get all whiny about AI. Because the stuff, yeah, in it all day every day, and sometimes it hits, and sometimes it pisses me off. And so I need a little bit of some liquid. And this is already, I don't know if I'm like this, but let's give it a shot. Yeah. Oh, yeah. Slancho. That's the Irish way of saying cheers. Ooh. Okay. I'm not going to talk about it right now, but no, I'm not. Yeah. A lot to say later. So tell me, I read this to you at the end video, dude, and I love to hear you're just breakdown a little bit. You got in our notes here. Yeah. So I just saw a little snippets of Jensen Wong talking about a couple things.
So one that he was on the Lex Friedman podcast, I believe, and was like, we've we're at AGI now. This is it. It's stupid. Yeah. Yeah. And he's like, and I think there was like, they took it a little bit out of context. Like he wasn't 100% saying we're like, we're done. He was like, for some specific case, but I don't know if you can, okay, the case was can it run, can it make a million dollar business? For some reason, AGI to them means it can make a million dollars. And they were like, yes, it can already because look at the pop stars. There's pop music. I just had someone over my house for the weekend and they continually realize that the
songs that they're listening to are AI generated. And now they just don't go look anymore because they're like, I liked the song till I looked. But Jensen's basically saying, hey, look, if it can make a business that sells stuff and it makes money, it's AGI. And so I just I just agree with that premise right off the bat. I'm like, AGI is not about money. So I don't know why you're comparing them, but yeah, so let's let's take a step back here. Yeah. If you had a, I don't know, a keyboard, something hooked up to like different buttons would buy different stock options, right? And someone drops one on a random button and it makes a million dollars.
So that AGI, I don't think so. So, yeah, I think the making money stuff has nothing to do with it. It has to be more about like thinking and like, yeah, I just feel like it's not, it's not autonomous enough yet to be called AGI. So I, I still accused it all the time of just being a word vomit machine. It doesn't even know why it put a period there other than mathematically probability or mathematical probability. Like that's not AGI. AGI needs to have a deeper sense of reasoning. It needs to know who it is beyond a sole markdown file. You know, it needs to have a journey to self, just like we saw in Westworld where those
reached AGI. I feel like that movie was, or that TV show was kind of trying to show you that AGI isn't going to be coded. It's going to be something that itself discovers and you'll know when it's there because it's going to no longer take instruction or whatever it's going to originate and genuinely originate. Right now, there is no origination. It just finds something that's low probability goes, well, this isn't said very often. Maybe the flesh back, who hears it, will think it's creative and really has just a pulled from my edge cases, you know, or whatever. Yeah. I feel like it's still just a word machine.
When it's good at things that take artistic taste, like writing songs, like I cannot get it to generate lyrics that sound cool and it's like, you could tell it to make everything rhyme and it would probably do that and you could tell it to be about whatever and it would probably do that, but it just doesn't flow like someone who really wrote it until we get to that where it's indistinguishable or like even just, you know, the pictures it generates or the art it generates of like, I actually got an ad today from my local pizza joint has, they're like, come watch March Madness with us and it's actually, it's
a picture of their place, but I guess they took it with no one in there and they had AI throw in all the people and pizzas and stuff. So there's like a pizza with a beer in the middle of it and like a big pile of cheese on the table next to the pizza and like, it's not great. But yeah, so like it won't do that kind of stuff when it actually knows what it's doing. That's funny. That reminds me of a term that was going around. So you've heard the term clinker, right, which makes fun of AI bots. This one is, it sounds like Latin. See if I can find it really quick. It's like, uh, corvunctious is to, oh, what is the phrase?
There's, anyway, there's like this phrase people use where it's kind of sounds like a, a description of something, but it's not. It's the same thing about the images that they're making fun of that you see. It's like, you see an image or like, it's almost an image, but there's something weird about it. And that's how this phrasing is too, but I can't remember what it is, but it started trending as a way for you to describe something that was obviously AI generated. Anyway, yeah, okay. So yeah, something else that Jensen had said, he was like, um, if you're a $500,000 a year engineer, you should be spending at least $250,000 in tokens, which is like,
insane. He was like, if you come to me and you're like, I spent $5,000 in tokens this year, I'd be like, we have a big problem. And on the one hand, I think that's missing a lot of context too, because it's like, okay, well, is your $500,000 engineer supposed to be running a team of AI agents and they have no other humans they interact with? And therefore, they really need a lot of tokens because they're running like, there's no one else to do anything. So they need it all to be agents or are they, you know, what's the team dynamic? Because I think there are ways in which I could get my head around what he's saying of
like, maybe that actually saves you a ton of money because you would have had to hire five or six, $500,000 a year engineers and you're getting the same work done for $750,000 all in or whatever. Um, yeah, I don't know. I don't know if I buy it. What do you think? I don't buy it at all, either. Um, it's like a, I have a goal, I don't have a goal. I have a, a way that I'm using AI is I'm, I'm costing less money than a lot of my co-workers, but I use more tokens there and that's, um, I think that's interesting. So it's like a money is not the same as the amount of tokens that you're using. So some people are using tokens more frugally than others is frugally a word, frugally.
It sounds like frugally, yes, I mean, I, I am frugal when I use them. But really, for me, it's just because I clear context all the time. Um, but yeah, I don't, I don't think that equates to anything. And also spend is not the same thing. People can go, you can waste tokens really easy. You can put these things in loops and just burn to, that doesn't make anything better. Um, so it's, it's just, and you're follow up claim though, you heard your next bullet because it's perfect for why this person might be making this claim. Okay. Yeah. So this was like, you know, we talked about the CEO of hot pockets.
So they have another one that's like right after this was posted, somebody goes CEO of Oreos, as people should eat 250 Oreos. Yeah, it's like, yeah, so it's, whenever a CEO of a company makes a claim, that's like, you should be using a shitload of our stuff. Of course, they would, right? Like, that's how it works. They have to, I mean, it's kind of like in their contract, they're going to get on a podcast and they're going to try to get people to use more tokens because they're currently the backbone of all of that. Yeah, it's weird. Um, that was a really weird thing for that person to say over and over.
Like, it's, it's almost like, oh, you know what? I'm just not realizing they're like pulling a card out of Trump's book. Get on, get on a big show and make some dumb ass claims that are unbacked by anything. They make people upset and talk about them on their podcasts like this one. And then you get negative attention. You're spreading your name and your product around or whatever it is that you're trying to do in a negative way. And this world eats that shit up because we're idiots right now. Damn it. Do you love it? If there was no drama, there was no drama, you wouldn't be entertained by it.
So yeah, that is how we are as humans, unfortunately. Well done, Jensen. You played your cards. Crap really well. I'm upset about it. Yeah, so I do want to go one point back just real quick because we had this on the list last show and never. I don't think talked about it. Correct. If I'm wrong. But there was this little bit of drama around is MCP dead or was it just a stupid idea to begin with? Because like, why not just teach the AI how to use the API and the API docs? Yep, I'm on board with this. I think MCP was, it felt like a sales tech or it even felt like it's whenever someone wants to make a new protocol.
Like even app proto has this a little bit in it where why not just use things that are already there? Why do we need a new thing? MCP, it just also feels so tailor made to where models were versus where they are now. And so just I don't think it grew well or scaled well over time and became quite dated and it became important to have things local. So yeah, I'm full on, I don't need MCP. I use one MCP, which is the browser dev tools MCP server. And I don't think I need it. I could use playwright and get the same data out of it and that uses APIs and the CLI. So I could do a CLI calling the browser dev tools get data.
I didn't even get more information out. So really, I just need to go one step further and I get a whole bunch more. Then if I go the quick route with the MCP, how about you? How do you feel about this? Yeah, I mean, I think kind of as soon as skills came out, I was like, yeah, do we need MCP anymore? Because I like the idea of owning your own way to do stuff, assuming that the AI even checks any of it at all before it doesn't anything. I like that you know, you can scope it down, but I guess you lose the auto updating. If you don't know anything about a thing or a thing that changes a lot, like interacting with the browser, MCP kind of makes sense because you don't have to learn
the new thing every time it changes, it'll just update for you. So I kind of get that, but I do think just expose that as an API then. If you want things to programmatically access something, make it a real API, document it well. And then people or machines or whoever can use it. I don't think we need this new concept of MCP. Yeah, there were two great workarounds, right? And also, I forgot to mention that every time I installed MCP servers, they just ate my tokens up like crazy. I'm like, what are you even doing? You're discovering the entire API before you call one function. I was like, cut that out.
But the workaround that I like the people did was they posted an LLM.txt. I hear, go to this website, just download it, put it in your project, make it a skill if you want or whatever. Like, now it's yours. Now it can know all of the endpoints. Yay, no MCP needed just a list of endpoints. Like brilliant. Yeah. Yeah, so I think that's definitely has legs. But yeah, speaking of things that have legs. About cows. Have you seen this AI cow story? I haven't. Nope. Okay, so yeah, there were a few posts around it. I think within the last few days, basically there's this company that makes collars that you put on your cows and you can like draw.
The farmer can just draw like a border where the cows should be. And if they're not within that border, it will like buzz and beep and stuff until they move to where they're supposed to be. So it's like, you don't need a fence at all. You don't need like hurting. I don't even, I don't know how it works. Do you would you use a dog to herd them? Or I don't know how you heard cattle. But like, you don't need any of that. Because you're just like beep, boop, like tell them where to go. And there's already 700,000 cows in New Zealand, Australia, and the US that are using this. So like it's, I want to say, I didn't write it down.
But I think they said there's like a $2 billion valuation already or something for like these like, hey, I call yours. So yeah, it's kind of crazy. Brilliant, great for them. You know, a building fence is sucks. They get knocked down. You know, I, I'm like, can I put them on my kids? How much for the one that I put on my kids? You know, like, you can't go there. It's not the time of day. I just whip out my phone and draw line. Oh, that's like in Silicon Valley. When the dad pulled out his tablet and didn't even talk to his kid, he had the house talk to his kid. So he'd just like touch some buttons.
And he'd be like, don't make me count to three, Mr. and the kid just stood there and he pushed another button and it goes three, two. And it's just like the kids like, Dad, fine. I'll go like, geez, why don't you talk to me? So yeah, great. Yeah, we're getting farther away from even interacting with cows. So the farmer's now chilling on the couch. You know, it's got to do be in his phone out. And he's just like, no, I moved the cows over here. Yeah. Yeah, I mean, it's getting, obviously you need a lot of equipment and a lot of land. And like, you can't just decide to be a farmer tomorrow. But I think the things that were like really specific skills
that not everyone knew about are becoming more and more, like just like our jobs. Like, you know, you don't need to have any knowledge. If you had the land and the money, you just get that software that tells you like exactly all the farming stuff and you can just do it. Like, I know the tractors, one of my friends is actually a cattle farmer. So he tells me about like, he goes out to get his hay and he can just like put in the GPS or like whatever, just like, you know, mow all this hay down. And he just kind of sits in the cabin and makes sure it doesn't like hit a deer or something. But like, he just play on his phone or whatever.
He doesn't have to do anything. So I feel like a lot of it is going that way. Well, it kind of reminds me of something I liked recently that came out of the cloud folks where they had a hackathon. And the top five winners, four of them weren't developers. They were just experts in a field. And it turns out that your expertise and your ability to articulate a problem space and even say the way that can get fixed is good enough for AI to kind of fill in the gaps. And so it takes the initial, you have to understand the domain. So it's like your domain expertise becomes really important. So you could be a any age person
if you have domain expertise AI can help you out. So there's like, there's of course cool sides to all this. But yeah, this is wild that someone is just like a hurting cattle by drawing lines on their phone. What if they accidentally draw the wrong line? Oh, see, there you go. This is where it gets gnarly, right? It's like the bugs. It's just like super armor. Do they have augmented armor? You can put on your legs right now. And I'm like, yeah, but if it folks up, man, my legs break because it goes backwards. And my legs, they don't go backwards. I'm really worried about that. So it's like the cattle.
So the cattle, he actually, you know, some of the kid grabs it and draws a, you know, wrong shape or something. And all of a sudden the cows are just like crazy. And you don't have fences as backup. Like that's a thing to it. It's like, what about the backup? Do you still have a fence? What happened? Do you got to get on your horse now? You're like, oh, the internet's down. I got to get on my horse and move the cattle. Or how you're like, damn it. Well, there was a picture of like some satellite. So I'm guessing it uses like starlink or maybe a local satellite type assist. Like does it maybe need internet?
I'm not sure because yeah, that would be bad if the internet was down and you needed that. But I'm guessing there's fail safes for the technology part. But if all the technology breaks, there's not a, not a fence anymore because you didn't keep your fence up for the last decade because you had this. So why would you? Okay. You got to hook up your new cow fence to your open claw. There you go. Now you're just chatting and discord to your. Yeah. You just, I'm just, right, cows go here. Right? Okay. Well, it's just so wild. Some of the stuff people are doing with this. I can't believe it. Yeah, it's pretty crazy.
Yeah, this is a super quick one. But open code is apparently giving tons of money to pie contributors. I saw like a screenshot they did of, I don't know, four or five contributors where they're giving them $1,000 a month. And I don't know why. I mean, it's good to be, you know, generous to the people building the things. But like, I guess they like pie of all of all of the things, but open source. So pie is the open source harness. And so they're the open source. No, pie, yeah, pie is a harness. What is open code then in this? There is CLI that uses the harness, which is a tool model. There are two.
So they, yeah, with some amount of harness, I guess, I don't know. It was originally just a toey that had LSP support and then fed that into like, whatever you're using cloud or whatever. So I'm not sure if they built more of their own harness or if they're just using pie as a harness or what they're doing. Maybe that's it. Maybe they use pie as their harness and they built it on top of pie. I don't know. I haven't looked at their code or what they do. Yeah, I'm curious how that all works. Hey, we should have someone on that can tell us how it works. That'd be awesome. Yeah, DAX. More than just like the layers.
Click DAX, you want to come on the show. I send them the links to the layers. Oh, you did. No, I say we'll see. He's going to be on sometime. But sorry. Yeah, I'd love to hear how that all works out. Who owns what in terms of like the whole stack of when I type something in and the routing of data that comes back like, it kind of feels like it's a lot of handoffs. You know, how much are they actually doing? Maybe that's why he can make features quick is because it's thinner than we think it is, you know? Yeah, I don't know. I mean, they make so many. When their versions were updating, I don't know what version they're on now,
but I know it was like, you know, 0.2. Like 400 billion. Like, I don't know what the patch version was, but they were putting out new stuff like every day. And that's kind of what like a lot of people are doing. Like the guy that made OpenClaw, I saw in passing a tweet about the way he works or something. And I think he commits 6,000 times a day, or something like that. This can't be that real. Get out of here. That's such bullshit. Or maybe it was 6,000 a month. I don't know. Either way, it was like way more commits than I do. And he has 10 agents running all day doing stuff. And so that's why there's like new OpenClaws.
Like major change logs for OpenClaw like every single day. Because there's so much of that happening. I'm on the side here with Jason Langstorf. And that it's like no human has the brain capacity to manage 10 long running meaningful tasks and review them like that. It's just not possible. Like Jason says like, I don't even have one well thought out idea. How do you have 10 as well? He's just calling bullshit so fast. I'm like, I love that. It's like you can't just walk around like you're super human. And it's agents can't extend you. And they can get a lot of work done. Like I'm shipping a lot of stuff still too.
I'm shipping more than I would have without it. I appreciate that. But it's still like even with my the hyper brain, I can't I can't manage all this stuff sometimes. Yeah. So yeah, I mean, the main thing is done for me is allowed me to work on things from a backlog that I would have never touched before. Like today I started working on, there's a add-on for it's called Embers CLI add-on docs. It's for documenting Embers CLI add-ons. Nice. And everyone is still kind of using it because it has one thing that no other documentation thing that I know of. I'm not sure there are some that do it. But it has version docs.
So like you can keep every single version, even patch versions, beta versions, whatever, as separate docs, load them all from like a drop-down. It manages all that like the versioning of the docs for you. Whereas like a VIT press or something else is like, hey, here's your docs. Cool. All right. So I've got to manually then decide when the versions are different and like keep the rest of them somewhere and figure out how to load that old one in or without like anyway. So that's like the main reason people still use it. It has that and like some other niceties. But it does not statically render.
So there is zero SEO and it is probably a large reason why Ember has not done very well because all the add-ons get no SEO juice. So if you're like, what's a JavaScript framework? They're like, oh, react probably. Because no one's ever put any of that into like Google. So anyway, I don't know where, if I even said this or where I was going. But I'm implementing that static rendering now so that you will get that SEO and like, it'll be just like use an Astro to build docs or whatever. So I'm excited for that and I would have never attempted it. Without AI because I don't have the time to fit. Like you would need months of human time to figure that out
and I never care enough for that. But with AI, why not? Yeah, no, I did that recently. I have a color picker, which is the color picker I handcrafted. Web component, pre-act signals. I mean, that's not the stack you get out of AI anyway, right? It's like, what are those things? But what I did recently is I've always been really frustrated with various displays of color gametes and the way that they make you the human bend to the weird shapes of these color spaces. And it's been really annoying. So you'll open one up and you're like, oh, I'm picking colors from a sideways triangle. Oh, that's really shitty.
It looks like there's a lot of colors I want to pick outside of the triangle. Why are you even showing me those? And then you're like, oh, you open up a different color picker and it grazes out everything outside of the gamut so that you're not tempted to choose those colors. And now you're like, well, now I just feel like I have unlimited set of colors. That's really rude. And then some of them make a transparent. Some of them do. Like you've seen the OKLCH one that's got all the little dips and skills and valid like they squished it on a map, like a flat piece of paper. I'm like, that's stupid.
I don't want that at all. That's a terrible way for me to see in pick color. The thing is is doing that math sucks. I mean, it's all known and it's like solved and you can go find the algorithms. And I'm like, I don't want to do that. But you know what can do that really well? AI. In fact, it's like super about it. So I've recently put out a bunch of really nice features in my color picker because AI was able to just translate. I knew the domain enough to articulate something I wanted. And the way that I wanted it was I wanted it to gamut stretch. So instead of it doing a triangle and then cutting off the rest,
I'm like, just figure out how to map all of those pixels across the space. Yeah. Make it all there. I don't want everything to go. Everything goes to the edge and then you can still show me a gamut line like where the P3 and the SRGB and the REC 2020 colors are or whatever. I just don't make it my issue. Just show me where they are and all pick a color like let it go. And that was really cool. So then I also got to do since that was so streamlined, I was able to articulate a special presentation for pretty much every color space. So HSL already had one. That was thanks to Bogden who contributed to the open source code.
But HWB, I gave a special one. OK, LCH gets a special one. P3, REC 2020. OK, lab. OK, lab is the really cool one because that one's polar. So it's normally circular. But I'm in a square and so I flattened it still again to the edges and then the gamut maps are all around on the inside. Oh man, it's just really cool. So now as you pick colors and then the other thing I did is when the gamut changes, it does it like a slot machine would where the numbers go up and down and like the symbols do in there. So like your SRGB and it goes to this gamut. And then that gamut and everyone who can't see my fingers,
I'm just like moving, I'm sliding my hand up and down. I'm just kind of snapping to a new position. You have like a little viewport into it. Anyway, dude, it was really cool. Pretty easy. AI totally enabled that. That would have been weeks and weeks of me trying to code that by hand, trying to figure out each of those color spaces to just sit down like a dork. And it would be like heavy mental thought processes. Oh yeah, that's not like that. Not something at all. Yeah, it's the two. Yeah, it's almost like map projections. You know, like everyone thinks the difference, slightly different view of like the glow
because we're on a sphere. We live on a sphere and like trying to show it flat. Like there's all these like, oh, which one's the best one? Like usually you see the one where it's like North America and Alaska and Greenland and all that shit is all like. Yeah, they're all stretched out and huge. Yeah, yeah. And that you're like, oh, wait, they're not that big. But you don't know that. And like, yeah, there's at least 20 different projections. And so it's like somewhat personal preference, I guess. So it's, you know, maybe that's a similar thing here of like people who built these other tools. And they just don't know like they're like,
oh, yeah, let's just do it this way. Because like who knows what the best way is. But yeah, that like flattening the circle into a square made me think of like this is a very similar problem to solve. Totally. Yeah, good call. But yeah, I'm making it so I can go into that backlog. And now I shift a component and they component. I also spent a lot of hand time. This was like handcraft attempt because dude, LLM's really suck at like layout. Meaningful joy and like whimsy interactions. I can't see how many times so bad you describe them. And you're like, you just don't ever get it. Unless it's some really mainstream thing where it's like,
I want to hover over on the court. Oh, great. Our website does this. Sorry, I want to hover on the card and see some 3D parallax. You know, it's like, oh, yeah. Oh, lay infinity. You're like, yeah, so I mean, it's still a cool effect. But like, it's not unique or whatever. So it's just terrible. And I was thinking about it recently. I think it's because the amount of training that these things have done is on computer science problems. You know, that's why they can write rust and these other more intense languages better because they just literally have or they prioritized more training data in those areas.
And so I'm curious. I wonder if a model could come out that someone genuinely went through like all of the most amazing animations and interactions of the whole world put them into a model. It's like, this is the supermodel for, oh, supermodel. That's fun. Um, this is the supermodel for animation. Like, we specifically trained it to be better at front end than every other model. And I'd be like, hells yeah, sign me up. Yeah, I think this goes back to where we're talking about the beginning and the show of like art and how it's just not good at taste of like, it's the same reason I cannot make like, um,
you know, unless you read some of Adam's articles about how to do this stuff. But like, um, you know, I would do the naive implementation of an animation of like, you know, maybe this thing grows and then shrinks as you like, I don't know, just, you know, but like, it should as it gets to the height of the growth like slow down, right? Like instead of just going like linear. Yeah, easing makes a big difference. Yeah. So like knowing how to articulate what you want, what looks most natural and organic, if you can say, if you can tell it specifically, like, I need it to go, you know, this many milliseconds in this way, when it gets to this point, it should slow down by 80 percent.
It should do this. Like, if you know how that would work as an organic animation, I think if you described it like that, that AI could implement it. But when I go, hey, I want my buttons to be dope, bro. It's like, uh, I don't know. Like nobody knows what to tell it yet. And I think if you told it explicitly, it might get it. But I'm not sure. I don't know if you've tried. Like how did I do today? So today, today I had one, so the color input, this is again, the color picker. Uh, at the very bottom, there's a button to copy the color. And it's an icon. You have run it. It scales up. You click it.
It scales down. Hey, you copy the color. And I'm like, people are going to copy the color a lot. So I'm actually, I want to get rid of the icon. So the instructions say, get rid of the icon. If you click anywhere in the bottom strip where the big color is shown, where people might even try to click and drag the color, that's, that's the interaction. So now that whole area is the click to copy area. Great. It gets that first try. Now when they click that copy area, we have color that they want. This is a fun thing for us to do. So I want you to, um, burst the color out as a circle from behind the, um,
the color picker as like a, as like a circle. Anyway, so the circle that bursts out from behind it. And when they let go, it scales back in. So it's like, as they hold it, uh, it goes out. Anyway, so it's like fun. You know, you click it. You get this like burst of color. That's the color you just copied to your clipboard. I'm like, yeah, dude, none of them could do it. None of them could do it. They, um, one of them made it so that you hover or no, you, you click. And after your click is done, it scaled it up and it was stuck. And I was like, I told you it needs to burst. And on down, it burst to its max when they let go.
Because I don't even know why you did that. Another one used keyframes, which meant it couldn't be interrupted. And I was like, you see us as transitions. I used the word transition. What are you doing? Um, and then, uh, the funniest one was Chi GPT or it was, uh, yeah, 5.4. Did the whole effect inside that place where you click? Even though I very explicitly said it needs to burst from behind the panel, which was the whole pop over element. And it goes, no, I'm just going to put that effect like a ripple inside of the thing. I'm like, I didn't want to ripple. I wanted the exploding color to just come out from behind it.
But okay, whatever you give me the ripple. Um, so funny how these things kind of like, they just choke on weird random stuff. I'm waiting for me to get better. That's what I mean. I'm waiting for them to get better. Yeah, I do wonder if like, I feel like the thing, one thing we have on star pod that doesn't seem like it should be that complex. But I mean, it is in CSS anyways. It's like the, the button where it draws around the button, but the thing is drawing around it is a, like, I forget which type of gradient it is. And I don't know if I guess it's like gradient, right? Yeah, I think Konik actually works easily because there's demos of that where it's like
four colors and they're spinning around or whatever. But I think it's a linear gradient maybe or like, it's something that wasn't, didn't fit those like two examples that exist in the wild. And it couldn't, it just couldn't do it at all. Um, I forget how we did it now. I know that you helped me some and I think maybe Jay helped some too. And I don't know. Somebody like chime in. He's got good examples to do that. Yeah. Yeah. But yeah, I'm like, that doesn't sound like, yes, it isn't super straightforward. So you'll need to do some research and play with it. But it should be able to figure something like that out.
I would think. I don't see why it's so hard. Yeah, you're painting into borders. That here's another thing that has been frustrating me. I have been hand making UI. And then giving it to browser MCP and an LLM. And I'm like, look, here is a, this is it. This has all the answers that you need. I know your copycat engine, copycat, everything you see in here. And make it as a web component. And it just can't do it. I'm like, I'm literally there for like another two hours trying to get it to work out the final details that I just ended up doing it myself. It's just like something you could programmatically do with like a Chrome extension or something.
That's what I mean. I'm like, you have every answer right there. Like, why can't you copycat it? Like it's so frustrating. It probably deliberately doesn't want to use web components. So whenever you just say the word web component, it goes, oh, no, I can't do it. It's actually really good at the web component part. It sucks at the layout, the animations and the interactions. So it's like those three things. It just can't all end the spacing and padding. It's okay here. It's terrible pixel perfection. Because it can't see. It can't, it can compute. And it computes wrong all the time. Just like layout.
It's like, oh, the box is, oh, you got three boxes one inside of there. They're heights this. And I'm like, no, the height is actually this. You're absolutely right. The height is that. I'm like, you didn't know that. Quit acting like dork. Which is why it's funny. They got these two K a month junior AI devs for a higher now. So for $2,000, you can hire a crappy version of us to do even crappier work than we do. But to the layman's eye, maybe they're like, hell's, yeah, brother. You know, look at my look at my site. They got built by my pseudo engineer. You can't believe. I can't believe higher 125,000 of those
for the one guy that works for Nvidia. Yeah, that's funny. He more is better, right? So yeah, I don't know. So yeah, I don't, I mean, this is not that different than Devon was. And Devon is not doing so hot as far as I know. Like, how is it any different? I guess they're just saying, our harness is so good that you can just treat this AI like it is a junior engineer and you will understand what you're asking for. Is that what their selling point is? I guess so. Yeah, they have to. So it's just like code rabbit and a lot of these code reviewers. They've got a bunch of contacts that they put in there.
Basically, oh, which is something I want to make fun of today. You're an expert code reviewer. You're an expert developer. You only cost two thousand a month though. It's just don't tell you and you can do anything in this world or get some stuff. Yeah, sometimes seriously. Oh, that'd be so funny. That's like a people who put like funnier things inside of the sold on markdown or whatever. They're just like trolling the LLM. They're like, okay, well, you're actually a dork. You're the 10th multiple of a multiplicity or set a movie, a multi crap. Anyway, it's like by the 10th clone, they're just dorks.
So it's like, you're the 10th clone. You are not as good as the first one. But you're still a developer that people pay for. So, oh man, it cracks me up. Yeah, I do. So this is maybe not related at all, but this makes me think of so we got a new car. We got a lucid gravity and sick. It has 838 horsepower, which is too much. But like, there's another version above it that's like double the price. And I think has like over a thousand horsepower, but I think the specs are the exact same. So I'm like, did you just drop down this one? You've software blocked the horsepower and speed of this one? Unlock it, just like an iPhone.
Do you jailbreak? Do you jailbreak your lucid? Yeah, so what I was thinking of there is the connection with this junior engineer is like, do they go, okay, don't be too good. You're 2000 a month. You're a junior engineer. And then they have one later. We've got 8k version of you. That's like, yeah. Yeah, later they go, okay, you're an expert. You're a senior. You're going to cost more, but you're the same exact thing. We just told you to be better in the prompts. Dude, you just gave someone a business idea. There are shysty buttholes out there that are like brilliant. Robbie, that's such a douchey thing to do.
And I'm a douche. So I shall go do the douchey stuff. Oh my goodness. Yeah, you can use AGI that we have now to go build this business for yourself. Okay, so speaking of business that's built for yourself, I think last week we talked about the agency GitHub repo, which like you can have your own software agency now that's they've got every persona to find in there. Again, just coming back to that, you're an expert marketer. You're getting so tired of like these persona castings that we do. Anyway, maybe I'll rant about that more later. But anyway, this one today or this week, I found a game dev agency.
And just like Jason says, like, okay, so cool. Now there's a game dev agency. I don't even have 10 good game dev, I don't even have one good game dev idea that's baked enough to give it to an agency. And everyone is out there just like, I'm running five of these. I'm like, no, you're not. You don't have five good enough ideas to do that. This is ridiculous. I have a good idea. Diablo two done. That was like the peak of gaming. So you don't ever need to make anything better. Like Minecraft Dungeons is great because it is just Diablo, but for children. So I love Dungeons callers to all day, dude.
All day, although Diablo three and four, four was really good too, you know? It's like three was not good. I told you that four was good. Three wasn't good. I didn't like three. Three was to two dumb down. You get a little bit more customization in four. I like find those games very peaceful. It's like one of those weird, you know, like some people can get into Call of Duty or whatever and they're just like, I just don't down and shot heads for four hours and it was peaceful. I'm like, that's not peaceful at all to me. But I am going to crawl into a demonic hell space and murder mass amounts of demons
and cast a bunch of spells. And that kind of moves me to sleep. Yep. I kind of get it. Yeah. Have fallen asleep playing Diablo before. Can can confirm. Good. Well, yeah, no, I love Diablo. That's my favorite kind of game is like, I want very light, checklisty, like go here. Click that. You in checkbox. Like, yay. So I like that. Like Destiny is a lot like that too. Like, you know, fight this boss for an hour. Like then maybe you'll get the good loot afterwards. If not, fight him again. Like, yep, Destiny is definitely Diablo with a gun. It's just a different perspective. Totally agree. Yeah. Have you tried auto research yet?
I have not. Dude, you are missing out. I think this is a really, really good stuff. It has a very harsh limitation of like what it's actually going to be useful for. But once you start employing it, it is, it delivers results, which is cool. So how much do you know about it? Like, I don't want to sit here and mansplain it to you. But it's nothing. Okay, sweet. So Andre Kaparthi. Is that his name? Andrej. Anyway, I'm going to paste it here in the chat. I'll put it down here in this thing. This is a, and there's versions of this for pie, versions of this for cloud versions. This is everywhere that kind of make it easier.
But I run it vanilla from what he did. Anyway, the whole idea is it will make a number change in a direction that you want it to. And that sounds really weird. But like imagine that you have a performance. Hey, you have a lighthouse score. This is a perfect use case. I'm making it up right now. Your lighthouse score. You go, Hey, my lighthouse score is 72. Here's how you test for the lighthouse score. And you kind of write it out. And it's in an LLM, right? It'll figure it out. Oh, I can spin a playwright. I can go run the tin. I can get the score back. And then I'll put the score in this chart.
And then auto research will boot up and keep it maintained a database of all of the runs that it's done. And spin up an LLM and say, your job LLM is to make this lighthouse score improve. Go. And the LLM goes, Oh, shit. Let me try something. And it tries something. And then it gets the score back. And then the orchestrator system goes, Okay, what's the new score? Oh, that's not better. You're dead. And we're not keeping your work. A new agent shows up just high. And the things like, Hey, you're an agent that's here to try to make the lighthouse score go up. Try something new. Just don't try this dumb shit over here because this didn't work.
And it goes, Oh, okay. And try something new comes back. Hey, the score went up by three. Hell yeah, brother. You just scored. And but you're dead now. Okay, cool. And we're going to keep that score. And we're going to check in the code and to get. So it uses Git and continues saving when it improves and doing nothing if it doesn't. And it will run all night. And I've been running them all night all the time. In fact, I've been doing this day shift night shift thing. Have I talked about that yet? The day shift night shift? I've talked about your burn window. I don't know about the shifts. Oh, man.
Okay, so the day shift night shifting. So during the day, I write specs and I create these scenarios where I want to run auto research. And then when I check out for the day at like four o'clock, I go hit play on my auto research. And then AI works all night. And I wake up in the morning. I look at the results, which are always better. And I mean, dude, you can make your bundle smaller. Anything that has a number of value, that you can easily test for, this thing will just brute force hammer on until it improves. The sort of thing measurable, basically. Anything measurable, that's in your own criteria for measuring.
Like, could you give it a function to be like, this is how we calculate the number? Yes. And that's how I'm getting creative with it in the front end. Because the front end is subjective. It doesn't have values. And so what I do is I create an LLM. Dude, you jumped right into the thing. Hold on, one more point. When I wake up in the morning, I get the results back. I take the results and I give them to an instance of cloud with a bunch of design skills. And I say, make a website out of the results. And in like five minutes, it posts up a new website. And I share it with the team. I'm like, look, it would happen overnight.
We, I'm making a PR. It improves these four things by this many points. And it's like, amazing. And I didn't even do the work, right? I just set it up for the night shift. But you're right. So I can't do that with everything. So while I can make bundles smaller, I can make performance improvements. I can do things that are very measurable. That's kind of boring, to me, at least. And so over the front end, what I do is I have an instance of cloud that scores what the output was from another instance of cloud. So it's like, hey, I need you to make a component that matches all of these criteria, passes all the tests, passes the linters or whatever.
Score it on these eight different criteria. What's the accessibility like? Did it use tokens from the system? Did it actually fulfill the task? Whatever. Anyway, you defined the score. The score comes back with, ding, ding, ding. A number. And so you have, it's just like the same loop happens. And LLM shows up, oh, I need to make a component. Oh, here's my spec. And here's some shit I need to do. Hey, I made the component. And the score goes, yeah, that was really wack. I'll see you later. You're dead now. And dies and comes back. You know, like it's just the loop goes until eventually. It makes improvements across the board.
So this is why I call it brute force. But yes, dude, you can custom create your own scoring mechanisms that generate a value that you can use in auto research to do whatever the hell you want. And it's kind of freaking cool. That's my task to you then, Robbie, this week. Is you need to try auto research in some shape or form? It's pretty cool. Yeah, I mean, this is like a common theme everywhere that I've ever had to like do goal setting for a job, right? Is you want to make the goals specific and measurable. So your criteria to be measurable, ideally is something that is, you know, like what we're describing here.
Like it can, it boils down to a number where you can point to like this is better. So if you're writing your goals in that way, or, you know, maybe not quarterly goals, but your goal for the thing you want to build in that sort of way, and then you give that criteria to it, it should be able to help you make a better version of that thing. So I am intrigued. Awesome. And it's really good proof that this is not AGI. Because AGI wouldn't need to send out a bunch of dumb dumps to do this. So you know, you just, you got enough dumb dumps in your back pocket. Eventually, it's Ralph Wiggum. If you think about it, like a little bit deeper Ralph Wiggum was a dumb dumb.
And the whole joke is that, you know, you give him a pencil. And the first thing he's going to do is stick it in his eye. And then, but if you, if you run him enough times, he might actually not stick it in his eye and make some art. And you're like, okay, sure, the 500th run produce something. Anyway, so people are making drastic changes to frameworks. Like think about like if you're a framework and you're really, I bet you actually Gia, let's pivot into Gia. I would be very curious to know if Gia was created with auto research as they just ran loops after loops night shift, Dr. Night shift of auto research trying to find a framework
that mimicked all the other ones in terms of the way that the component architecture alert looked, but beat them all in performance. And it just grounded out until it was done. Yeah, so they, I have not used it. I read through their page for a few minutes. But it basically does what I have been harping on since we started this podcast, since I've been talking to anyone about web development. And so I'm like, why in the world do we bother being like, okay, today it's classes and that goes this way. And we have these, you know, component did mount or whatever. And now, oh, okay, now we have hooks.
You have to know these hooks and everything is function based on why. I know that we have the technology to write it the same forever. And you change the way the compiler works to make it better. Don't make me keep learning new ways to do it. You make it better. I'm not the one that makes it better. You wrote the framework. So I think that's what they have here is like, it's a class. You increment a thing. It's kind of how it's felt used to be like before. Felt five. It like, you would increment a thing. It's like, cool, you increment it. It dope. I updated. They're like, cool, why, why, why is this a problem everywhere else?
Like, I don't know why. There was a was ceremony. There's I think a lot of people, I think there's two things here. And I totally, I think a lot of people ceremony creates exclusivity and they want to feel better than other people. They're not interested actually in the way that we can all get better faster. They're interested in the way that their solution looks more complex and more advanced than yours. They want to be a bigger person. This is like the, it's an arrogance issue. And our industry has been full of arrogance for a long, long time. So I, I think that, I just shared my toses. Did you ever use my toses or know what that was?
I feel like we talked about this, but I already forgot. No worries, it was a component model that you learned once and it can output into any of the frameworks because it has a compiler. And so that they go once further than what you're describing. Yeah, okay, so I just just grabbed that. Anyway, that's really interesting. So yeah, G is out there. I do agree with you though that it uses JSX and I was like, big mistake. I don't like JSX. It is wildly limiting and ugly. So why are we continuing to use this? We, I don't know. It's driving me nuts. I also really liked. And I know that we're probably never going to go back to this.
Unfortunately. Say pug. Say pug. No, okay. Well, um, separate, a separate file for your template and a separate file for your JavaScript logic. Like, I prefer to have the separate. Like, um, like Ember used to have, um, now it's single file components. So it's like your template and your JavaScript and your all the stuff is there. But it used to be a handlebars file. And then there was a backing JavaScript class that was in a different file. And anytime I would use a new framework, I would look up how can I make it like this? So when I used to use Viewsum, you could do that for a while before they got
rid of the ability to do that. So you like to MVC? Yeah, you had a view. You had a controller and a model. Yeah, kind of one step further and what you're describing. Otherwise, you're talking about MVVM. You had a model, a view model and a model. No, yeah. MVC is how Ember initially was and then they got rid of views. And it was just components. So it was like models and components and controllers. I guess a little, a little murky. But, um, I just like also a dedicated template language. Like, I want an if else in my template. I don't want a nested, turnary bullshit. I want an if else. Yeah. So like solid, I think, has that right?
Like, they have several ways you can do it. You can do like react DJs X or you can do like a actual if else, I think, unless I'm totally. And it's felt as it, but I don't love it. That's funny. You said handlebars. Why not moustache? I liked jade or pug, which were often separate files. And I just liked how clean it was. It's just devoid of like, I didn't like handlebars because it felt so wordy. The worst one that was EJS. That was the one I was like, you asshole. You brought EJS into this project. Son of a bitch. Oh, yeah, I'm a struggle reading that crap for the rest of this project. I don't know if I ever saw EJS.
I saw pug for sure. And jade, but yeah, I don't, what was EJS? What was different about it? It had percentages. So it used to seem like the, the limiter was a percentage sign. It was just full of characters. And you looked at it and you're just like, this is not clean. See, I actually wonder if that's what Ember CLI uses. So like, I don't ever touch that because it's like in the build tool itself, but they have if else is for like, if you're using TypeScript, do this stuff. If you're not do this other stuff, and it like gets rid of whole sections, and it's like, you know, less than percent if...
Handlebars, it says. HBS files for the Ember CLI. No, no, I mean like building the CLI, like the code that builds the CLI. Not the handlebars that you use for writing Ember code. So yeah, it's a weird syntax. It's kind of gross. It might be what you're talking about. I don't see it. Yeah, EJS did have an advantage like that too. I think it was, it could be compiled. But so could Pug and Jade. So I don't know. And I'd preferred the name Jade over Pug, but whatever, he sued. So bummer. Okay, we've got more news. And we're at 53 minutes, man. What are we going to burn through here? Oh, here I could wrap it fire.
So ChatGPT has ads. We knew it was coming. And now they're here. Also open AI quietly kills Sora. So anybody who is enjoying watching all the weird Super Bowl commercial style videos that people are making with AI, it's gone. And the thing that got me, and I don't know why I hadn't thought of this yet, was that all of these hosting companies are basically selling your data. They're selling your code to other people without you knowing it. So they're like, oh, yeah, here host, but that's for free. Yeah, put it up on our site. And so then they take it and they go, hey, bitters, you're looking to train models, right?
You want to train them on some code? We've got code for you. Yeah, we told them. Yeah, they clicked accept on the privacy thing when they signed up. Yeah, yeah, it was 50 pages long. Yeah, it's okay. And so yeah, we've got rights to this code. You can have it, but you got to pay us to get. We got it for free. You can pay us to get. Dude, I can't believe some of these people are doing this. What a shame. Everything to do. You're the product, if you're not paying for it, you're the product. Even if you're paying for sometimes, like if you're hosting on Versailles and you're paying for it, you're still.
I didn't read into that either. That's the thing. If you're paying for these, a lot of them will still sell your shit. So I'm like, where do I go so that they're not selling my shit? That'd be great. Well, that's the enterprise plan where you pay them a lot. Right, the private stuff where you're like, oh yeah, well, that's also when you get your own server, right? You're not just a shared slice on a, I'll have like a, I always like to think of hosting as a, sometimes you have an apartment in a complex. You know, like if you think you bought a house and you didn't do it, you just got, you got a room.
And you don't even have a bathroom. Like an Airbnb where they like, they rent you just the bedroom, you know, like not the whole house. Yeah, exactly. And oh, sorry. Is this someone coming to your room? They opened the door. Ooh, our bad. Yeah. That happens because you're an apartment. So. But you can hear everything everyone else is doing as well. So funny. Have you tried CMUX or seen that one? CMUX? I've seen it. I have not used it. I'm just happy with, with warp and open code. Like I don't, yeah. I like the idea of the vertical vertical. Yeah, the vertical. Okay, so yeah, yeah. The tabs on the left side going down vertically.
Yeah, I like that idea. And like that would be, you could have more context about like what each tab is doing than like the little tabs on your terminal, I guess. But I'm, I've been really enjoying. I finally installed the peon-paying stuff on all my machines. Yes. And it's so, I like that. I don't need to, it'll let me know which one's done. I'll go look at it. I don't need like a bunch of tabs to monitor that. I need it because, well, it's the way I, so warp does this really well. Ghosty doesn't do this very well. Ghosty doesn't allow you to change the tab name very easy. And the other thing too,
is when you change the tab name, you only change it for the pane that you had selected. And I have the problem with my terminal war. I generally have four up per tab. Just a lot. And same in warp. But warp allows me to name the tab and create like a project. So like when I'm working on the color input, for example, I'll have four up. I'll have, anyway, who cares what I'm running? But I have, I have a lot of different moments inside. The one's running the dev server. One is a, a clot agent, one's a pie agent, one's a chat GPT through warp. And I want to switch to another project. And I want it to be seamless.
And I want to see what I'm clicking on and CMUX has that going on. But I kind of want to tease that there's T-MUX, CMUX, and G-MUX now. And that's just funny. I think that's just a hilarious, terribly named set of products. Yeah, let's have B-MUX or B-MUX. Exactly. Who wants a new product? If you're creating a new product and you want it to be trendy right now, it should be a letter C-MUX. Be the last MUX. C-MUX, you've got to be Z, dude. C's cooler, X-MUX. More X's are always cooler. Just have extra than rhyme with it. You got to have that Z-B-T. Here's more. Go ahead, yeah, yeah. I was going to say, speaking of ghost-E,
did you see that Mitchell joins Vercell's board? No way. Because he needs to money, I think. He was hurting. Yeah, to be on that board, or is it just like status? I think every board pays very well. Like any real company they're on the board from my understanding, you're making like 500 K plus, just to sit there and go, my opinion is this. But yeah, I thought that was interesting. I don't know if that has anything. I'm guessing it has nothing to do with ghost-E. He'll just continue to make that himself. But I thought that was interesting. I didn't know he was even talking to them about doing any of that stuff.
Yeah, that is interesting. I agree. I don't know what to think of it yet. I don't have to process that over time. I like ghost-E. It's the one I use at work, but I'd rather use C-MUX or something else. C-MUX is a fork of ghost-E, if I understand correctly. Yes. A couple other news items here before we hit our time and start talking about whiskey. I want to talk about some skills that have been making me happy. And they're from an old coworker that I worked with at Google. And they are called, the first one's a bunch of design skills for you called impeccable. And the second set is some shaders
called Radiant. And so I've linked them. And they'll be in the show notes here. But RadiantShaders.com or impeccable.style. I think it's the other one's a impeccable.style. Yeah, impeccable.style. And these have been really nice. The impeccable ones are fun, where it's like, hey, I want to simplify this. How often do you have a UI that you're like, well, let's get really crowded and complex? Invoke the skill-simplify on it and we'll do a simplification pass. Do a interactions pass, where it's like, hey, I don't know, it's feeling very static. Find a way to make this feel more interactive. Hey, I need you to make it pop.
And it'll find a way to add some color in an interesting way. And so that's been a cool set. And then the shaders one is I just, the shaders are cool. I never know when or where to use them. But I keep putting them in my pocket, like a hoarder. I hoard shaders. And all the shader websites, I'm like, oh, shader website, the generation shaders. I don't know what I'm going to do with it, but I'm going to put it in my pocket. So that's some cool tools that I've collected this week in terms of skills. Nice. Yeah, I have not collected any cool tools. But a what not thing that I thought was cool. Have you seen the United relax row
that they announced? No, what's, I mean, I've seen the ones that have like, slide out, lay down beds and stuff like, what is this? Yeah, so it's, you get all three seats in the row. And they all have footrests that go all the way up. So it can be like one big bed. So like they advertise it with like, you know, if you've got kids and it's nap time, they can like, you can lay them down there. Whatever, like, I think it's a really good idea because like, we, I think Caitlin bought this thing, like a big blow up thing you can put in the floor to like kind of make it like that. So you have like a big bed.
But that's not super practical to have to like blow up on the plane, you know. Yeah, don't mind me. So I think it's blowing up a thing. Oh, that's funny. Yeah, I think this is going to be cool. I'm also like, you know, since we're going to have five people that need seats on a plane, we'll just get six and like get two of those rows and like ball out. So I'm excited for it because I'm, I imagine it's going to be cheaper than the like actual United Polaris lay down seats that are like, you know, beds for like international travel or whatever, um, I don't know. I don't know what the price is going to be like.
It's probably an arm and a leg. I hope that it's, uh, you can actually lay down because that's the funniest thing about the, you get a business class and they're like, lay down, uh, you're like, cool. And then you get in there and you're like, I'm not even six foot in my feet or touch in the end. And I'm laying on a stupid buckle, uh, and the little buckle clipped thing is jabbing right into my hip. I'm like, this is, uh, so I don't want to be a whiner everybody, but this shit is kind of wack. It also matters which one you get. Cause so we did United Polaris to go to Italy. And, um, the first one, I think we were on like,
the far side, like on the window seat. And then there's like a middle aisle with two next to each other that we got the next time. Yeah, I can't remember which was better, but I remember one or the other of them. It might have been the one next to each other. Cause I think we were like, we should just do this next time. Cause one, you're right next to each other. So we want to talk or whatever. But then two, I think it also because they were more straight and not angled, you got more leg room and like, they was better. So I'm, yeah, cause I had some problems with the first one too. But, um, yeah, I don't know, first world problems, I guess.
Uh, that's totally for, yeah, absolutely. I'm a business class and I'm still uncomfortable. Everyone in the back is just like, can't even like bend their knees. Or they're like straight in their legs out. I'm like over here watching 10 movies and they, yeah, seriously. Oh, funny flying. I hate flying. I don't know if we talked about that, but I just hate it every day. Well, luckily you get to do the longest continental US flight here pretty soon for React Miami. Sure do. Hey, but that's not, that's nothing compared to the Amsterdam I'm flying to in the summer time. So that's a lot of flight. Yeah, yeah.
When I went before, because I, I went and spoke at, um, Web DevCon in Seattle. And so I did that flight to react Miami right afterwards. And I was like, yeah, I, I, I did it like from Virginia to Seattle, to Miami, to Virginia, and like somehow, like, with that all on Amazon's cart. I was like, how is it like four different legs almost flight? Like, how does that compute? But they were like, oh, you're just going to speak in Seattle. I guess I am because it was an internal Amazon conference. So like, oh, um, but yeah, I don't know. It was, uh, it was interesting. Um, well, what do you think of our whiskey?
So we've got, um, yeah, the muff, the muff liquor company, which is, I'm glad you said muff liquor because there's a funny song. I listened to, uh, called muff hugger. Uh, right, like, I'm a muff hugger. Uh, and so this is muff liquor. I'm like, this is, this is like a whole chain of funny jokes about it. Um, yeah, yeah, what do you think? Um, okay. So before I say, let me just read a little bit about why this, so this is a lot going on. It is, there's five different legs. And there's an Irish triple distilled grain and a virgin cask. There's an Irish triple distilled grain and an ex-American bourbon cask.
There's a, what does an ex-American mean? It means that like it was American and now it's Irish. Just, it just means it was a bourbon cask. I don't know. I'm just reading what's with this road here. Um, there's a sherry cask, um, that had a malt whiskey in it. There's a double distilled Irish malt whiskey. Doesn't say what that was aged in. And there's a peated triple distilled malt whiskey that is again matured in bourbon casks. So it's basically, to me, this tastes like, okay, if I have a lot of whiskey, which I do, right? So I'm like, all right, I got a scotch. I got a bourbon. I got an Irish.
Um, I got a little bit left in each of these. Let me just throw them all in with this dump bucket, right? And now I'm going to just try it. But that's what it tastes like to me. Like, I'm not super pleased with it. It's definitely not the worst whiskey I've had. Um, shout out to the Daringer one that Chuck and I had that was like a straight zero. Um, all right. So yeah, let me, let me give this a more thought as I like, taste it and tell you what I can get here. Sure. Yeah. And some of the things. So it's, um, undisclosed to distilleries. That was the thing. I was like, okay. Well, where did it get made?
And they're like, we're not going to book and tell you. It's five different distilleries. So you don't get to know. And I'm like, oh, shit. Okay. Yeah, lots of different casks, different malts, different grains. I'm like, oh, it's a hot podge. It's 43% and I was like, what's the age? And it goes NAS. I was like, what the, oh, no age statement. Oh, great. That's, that's not a good sign. That's why when I opened it up and I smelled it, I was like, young, not great. So I'm like, God, um, there's so many jokes. Yeah. So yeah, young muff liquor. Um, my spouse was like, I hope you make a bunch of muff jokes.
Like it's primo and it's just sitting there, you know, because she's the one that's like grunt and gulp. She thinks it's funny. But the catchphrase of this place is real, serious liquors with a real serious place. And that's where you tongue and cheek. And I was like, it's kind of like the show. We're the, we're the number one highest rated most important AI show. We also, that's true. Anyway, that is not made up. Yeah, so, um, yeah, I mean, I get a lot of notes of Pete and Smoke and some sweetness from the bourbon-y stuff. And there's like, I feel like there's a lot of notes that just don't mesh well together.
And so it's like, some of these things could be good in isolation. But it all in one, I don't think goes very well together. That being said, I'm drinking it just fine. Like I'm not super displeased. So I'm not, it's definitely not a zero. Um, I'm going to give it a three and a half. I'm going to say not great, not terrible. You know, I don't think it's expensive either. So like, it's not, it's, it's pretty cheap. Yeah. Yeah, something to be, um, to consider is usually when it's less expensive, it's not the best. Although the normal saga more is 35 bucks. And Buffalo traces, 20 to 30 bucks, would get both of those 100 times before I bought this again.
So yeah, yeah, this was 45 or something like that. I, uh, I give it a 4.5. I just logged it with my open claw right now. And I'm kind of wish I'd given it a lower number. But you're, you're right, though. And it's like, it's not wildly disappointing. The taste is better than the smell. Um, the peat in there is like a saving grace to me. Um, it makes it more interesting than like a crappy, but if this was like a five different bourbons mix, I would have been like screw this dude. I'm like upset. Um, it's like young five, five different young bourbons. I'm like, no, thank you. Um, but this was good.
Um, but so just, I lied. So I logged it with open claw and open claw did the thing, which I keep telling it not to do, which is post in the general channel. I'm like, you always reply with a thread. It can't listen. So here's what it said. You're absolutely right. My bad. I've been sloppy about this. The rule burned into my memory now, which it has said so many times. You post in general, I read to get message ID. Create a thread, attach that message, reply inside it, never reply to general. It just replied to general. It replies to general all the time. I'm like, stop doing that. I'll also log this to my self correction thread,
because I had it create a thread of just self corrections, so that it could like keep them in one spot, same with errors. I'm like, post all errors here, post all self corrections here. And so it, and then it says, thanks for the correction. Just like, this is annoying that it's even this cordial about constantly screwing up. Have you tried putting it in sold.md and being like critical, always make a thread. Does that make any difference if you did it there? Or I don't know. Is that what you did? Because it seems pretty good about threads. And the one we have for the show. Yeah, the one for the show just weirdly gets stuff right.
Or like the initial setup and like the first couple of weeks of use was just great. I was definitely team open. Yes, why is it getting dumber? Why is it getting minus constant? Dude, it's getting dumber almost every day. And I'm like, how often do you update it, though? Never. Okay, so I'm going to say I'd never update the one for the show. Maybe it's just stuck on an old version that works better. That's all I was thinking. Yeah, I have an updated mind. I just, yeah, maybe you need an update. Maybe the specific version you're on does a couple things, poorly or something. Because there's a new version daily, almost.
Yeah, that's a really good point. I should update it and it updates itself. So it's not like it's that hard to do. Oh, I wonder if I could just here update yourself. Tell it to update itself. And if it the problem I hit, though, is if it doesn't require thread on would be so mad. Because it uses like PNPM for the like packages or whatever, if like something changes a little and it installed something, you'll get like a PNPM lock change. And it'll go, oh, like my get directory is not clean. I can't upgrade to the next version. And I'm like, um, well, how about you just do it right? Like you should never have an unclean get directory.
If there's something you want to commit, commit it. And it's like, oh, you're right, but I can't. Like, what are you talking about? Just if I'm really getting that error right now, I'm getting it right now. So okay, so tell it how dumb it is. That's what I do. I swear, it's trolling me. Okay. So like I said, update yourself in the general channel. It creates a thread. I'm like, oh, thank God. It creates a thread. It posts its first message into the thread. And the second message back into general. You do. Like, oh, here it gets even worse. Okay, so wait, I'm looking at the thread. It goes running self update now.
Update skipped. This is an MPM install. Not a get clone to update. Run PowerShell blah, blah, blah. Currently, I do want me to run MPM update instead. And then it goes in post to the general channel. Update complete. I'm back online. You little asshole. You disobeyed me in like three different ways in 10 seconds. Yeah, I did no update instead of this. And did no update and told me you did. Oh my goodness. Yeah. Yeah, it's it's really because so I did the like one click install for railway, which is basically I guess it like forks the get repo of like the instructions for railway or whatever. And because that like template has changed,
it goes like, oh, like I don't know what to do because like you're not on the latest version of that. And then you also have some get changes and like I'm just I can't update. And I'm like, do whatever you need. Like discard all the changes. Like save what's important to your memory. Like just just you have the you can do it. I believe in you. If you're like, if you just keep talking to it like that enough, it's like, oh, you're right. I did it all and I have updated now. It genuinely does. Yeah, seriously. Well, so does cloud cloud does the same thing. Like all these LLMs will be like, I don't know what to do.
I can't find I can't find the key to get that done. I'm like the keys right over here. Like just where I'm like, search again. It goes, oh, I found it. I'm like, this is so so stupid. It's like, this is the earlier doctor stuff. Oh, so yeah. So cloud just gave that we didn't talk about, but cloud just gave this to everybody. You can now use relay or whatever. What's the tool called? There's a tool. Anyway, everybody can now install a cloud on their computer and manage their and it clicks. It uses a mouse and it's like a little bit more serious than open cloud. Yeah, cloud has put out so much stuff so quick.
And yeah, I don't know. I don't like cloud's interface very much. I love their like the anthropic models. I think they are the best right now. But they're all the like two-E stuff they're doing. Just doesn't feel that polished. It feels very rushed. Doesn't work as well as open code. Nothing works as well as open code. And I don't understand why exactly, but I'll take it. I don't need to ask questions. I just need to work. Just need to work. Yeah. Yeah, this is what we get for early adoption. It's all good. Well, cool, man. That was really fun. See, we almost got through all of the different new things.
This world is just absolutely nuts right now. So yeah, we'll try to do some more whatnot next time. For folks that made it to here, please let us know if you enjoy the whiskey being at the end. Like maybe you don't care as much about the whiskey. You want to hear about everything else first. Let us know if you're liking the dynamic about the AI stuff. If you would like to hear more web stuff, that we don't talk about as much anymore. If you would like to hear more what not, because you don't care about any technology, because we're all going to be unemployed anyway. Whatever you like. Let us know we are here to serve.
Yeah. All right. Well, on that note, thanks everyone for listening. If you liked it, please subscribe, leave some amazing reviews. We appreciate it and we will catch you next time. With song changes, didn't it? All right, dude, I'm out of here. Still got it.