1046: The Pi Updates are insane

Scott, Wes, and CJ break down the latest Pi updates, including durable long-running agents, Codemode, and MCP support, then debate whether pull requests are dead in favor of issues. Plus Supabase picks up Turso, Screen Studio 4 drops, JavaScript Composites bring value-based equality, and a stack of new AI models from Google, OpenAI, Cloudflare, and more. Show Notes 00:00 Welcome to Syntax Weekly 00:46 example.com gets a facelift Oliver Dunk’s post IANA’s reply 04:37 The death of pull requests Sindre Sorhus’ post Dave Kiss’ post SWC’s post Andreas Kling’s post Yusuke Wada’s post Wes’ NiimBlue PR JSON Alexander 12:46 pi v1.0 24:14 Brought to you by Sentry! 25:08 Turso Joining Supabase 28:13 Screen Studio Updates Screen Studio 4.0 upscaling 33:24 Svelte 3.0 Release 35:32 Effect 4.0 38:14 Alchemy Cloudflare’s Programmatic configuration 42:59 Foldkit 44:10 OpenAI Dev Day Announcements Dots Rahim Nathwani’s reply regarding Luna Plugin Extensions Sites in ChatGPT 56:40 Cloudflare Clef Decision Model 01:01:25 Gemini 4 Argon Google’s announcement post 01:09:04 Flux 3 Image Model 01:11:50 Claude Mods 01:13:44 Muse Gadgets Nat Friedman’s post 01:17:27 Nvidia Lyra 2.0 World Model 01:27:30 Game Recomps and Decomps 01:30:54 JS Composite Type 01:35:00 RFC 1149 Auction First implementation of the Carrier Pigeon Internet Protocol 01:37:49 eslint cssicorn 01:39:15 Syntax Meetup Hit us up on Socials! Syntax: X Instagram Tiktok LinkedIn Threads Wes: X Instagram Tiktok LinkedIn Threads Scott: X Instagram Tiktok LinkedIn Threads Randy: X Instagram YouTube Threads

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Scott Tolinski?

  What's up everyone? It's another Syntax Weekly. We're coming at you with a whole ton of stuff. Last week was a wild week for updates. Tons of stuff got updated. We're going to be talking about both software libraries as well as AI stuff, what got updated, what got released, what all the new hotness is, stuff with Pi, SvelteKit, OpenAI released stuff, Effect, all kinds of crazy stuff. My name is Scott Tlitsky. I'm a developer from Denver and with me as always is West Boston CJ Reynolds. What's up guys?

• 0:32Wes Bos?

  Man, hell of a week. Stoked to talk about everything that's going on because some super exciting stuff.

• 0:39Scott Tolinski?

  Yes, super, super duper exciting. And man, guys, I'm so excited to get into all of it. And I know CJ, you had something up first, which was example.com. I need to get schooled on what the heck this thing is because I saw this in the in the show notes here. I got no idea what you're talking about.

• 0:57Speaker 02

  Yeah, let's get into it. So if you're not familiar, example.com is actually a live website. And I believe it's owned by the IANA, the same people that handle domain name registry, like the top tier one, not individual companies, but you know, the in charge of all that. But they host example.com and it's supposed to be used in things like documentation. And you've probably seen before, like if you're looking at Dev Docs or, I don't know, something, it has a link to like example.com when it's talking about websites or URLs or whatever else, or you might see a fake email like [email protected]. Well, example.com is a

• 1:39Speaker 02

  real domain. And they keep it live for those very purposes for use in testing and documentation and monitoring and stuff like that. But it finally got a facelift. I would show you what it looked like before, but the web archive is currently down, so I can't. But it was pretty, pretty simple before, like now it has multiple languages listed. It's nothing crazy. I will say that I first learned about this from Oliver Dunk. So they actually reached out to the IANA, asking what happened? Why did you like, why is there new content? And they replied to his email that it's a placeholder that really no one should be accessing, but

• 2:21Speaker 02

  people still do. So they actually updated what it looks like to help with that and like clear up confusion if somebody actually goes to it. It's really funny. Wes is the top comment here. How many CIs did this break? Because people probably use example.com, maybe even as like a health checkpoint or like

• 2:39Wes Bos?

  Yeah, I'm sure people like, like curl it and then check the like byte output to like as like a sanity check that they are getting the real internet.

• 2:48Scott Tolinski?

  Yeah, that's

• 2:49Wes Bos?

  And I wonder how many it broke for them. And is it, I can't get to it right now. If I visit it, if I visit example.com, it just says like DNS issues, which maybe they broke something.

• 3:00Scott Tolinski?

  They broke it in Canada.

• 3:02Speaker 02

  No WWW, just example.com?

• 3:04Wes Bos?

  Yeah, straight up example.com DNS probe error.

• 3:08Scott Tolinski?

  They don't want you to have it in Canada.

• 3:11Speaker 02

  There's be a Canada issue, yeah.

• 3:12Wes Bos?

  Yes, unfortunately.

• 3:15Speaker 02

  So I thought this was fun because I know I've been to example.com a lot because I create a lot of tutorials over the years. And so, yeah.

• 3:23Wes Bos?

  I think it would be hilarious if they just start publishing all the emails that get sent to example.com. You know, because like everybody's, everybody is sending all of their test data to example.com. That's exactly that's they say if you need to put in a domain name for something, don't put in something assuming it's not going to go somewhere because we've all been there where you accidentally switch your prod keys or something and then you're sending real emails. And so what you do in that case is you use example.com knowing that that email will never go anywhere. But they should just as a as a joke, or as like a data collection endpoint, maybe they can go full agentic. Just start publishing everything that people send to example.com.

• 4:06Speaker 02

  Yeah, that's it exactly. So I went to DEFCON this year. I'll have to find the paper because I don't remember the exact domains, but there was a hacker that did a presentation. They bought a couple of innocuous domains, very similar to example.com, but they actually listened to all of the incoming emails. And they reported like the number of API keys, the number of like just random stuff people were sending to these domains that they thought nobody was actually listening at. Yeah. So yeah, be very careful of the domains you use and tests and stuff like that for your internal secret data.

• 4:35Scott Tolinski?

  Yes.

• 4:36Wes Bos?

  All right. All right. Let's talk about the death of pull requests. This is something that is kind of unfortunate, but I'm seeing it. I saw it probably what four or five times last week alone. And what is happening is prolific open source maintainers are simply just coming out and saying, like this is Sinjai Saurus is due to AI, I have disabled external pull requests on all of my repos. So Sinjai is the like creator of, he probably has, if I were to estimate probably five, 600 NPM packages that he maintains and he's, he's like probably one of the four pillars holding up the entire NPM ecosystem. And

• 5:18Wes Bos?

  he's just simply closing external pull requests because open source as we know it was fun while lost. It'd been 15 years from me. I will still maintain projects and handle issues. Essentially, everybody is getting this SWCRS due to AI generated pull requests. We are no longer accepting external PR sorry for the change of procedure, understanding, right? Andreas Kling works on Ladybird Browser. Every open source project that matters will soon do this. If you've always wanted to get a PR somewhere, now is the time. It's just everybody is getting slopped. Hano is closing their pull requests as well, because it used to be that a pull request was golden because somebody would put some effort into

• 5:59Wes Bos?

  solving that issue, figuring out the issue and then sending you the code. And now that the actual implementation is essentially free, it's not the actual writing the code that matters. And you'll notice a lot of these people are not saying we're not closing issues because it's more important that people submit really good detailed issues of what the actual problem is. And then the the open source maintainer can stick an agent on it themselves, knowing full well that they're they can have a little bit more confidence in the code rather than these like drive by pull requests that people are getting slopped with.

• 6:36Speaker 02

  Yeah, I like the change ultimately because it, I mean, you could say that going only to issues, it's just going to be agents talking to each other. But I've had this issue with some of my open source repos where people make a PR and they didn't talk to me first. And then I feel bad because I close them like this. I don't need this. This doesn't align with the project. And I can only imagine how much worse it's gotten with people thinking that they're solving an issue because AI is is writing code. So issue first makes sense because then you can actually talk about it, decide what needs to change, what should be added. And then you could just throw an agent at the thread and it could figure out how to implement it. But with the discussion first, then you're more likely to get more quality submissions.

• 7:17Wes Bos?

  Yeah. You know what's another really frustrating thing is that people aren't contributing back to projects. They're simply just forking them and making their own version of it. So here I've got this NimBlue, which is I have this little Nimbot label printer, right? It prints out these little thermal labels and I absolutely love it. And the iOS app is kind of awful for it, right? So I figured out there is a whole open source, you can use it in Chrome. And there's a whole open source implementation of this printer called NimBlue, right? And I thought this is awesome, but it's not exactly what I want. So my gut

• 7:59Wes Bos?

  reaction was I'm just going to make my own version of it. My, so I says, let's stop. Let's see if we can make this version better because quite honestly, I'm really excited about something right now. I'm going to fork it. I can make my own version about it. And then I'm going to, I'm going to lose interest and there's going to be a thousand other. And if it seems like the guy behind this, this repo is like pretty invested in it. So like, it's so hard to get people that actually care about these projects and will continue to develop them. So I'm like, let's, let's open an issue, you know, let's talk about it. So I was like, okay, well, there's no label preview. So I have these little like pig labels that we print on. And I, there's a couple like gotchas here where it's like you can't print over the pig's face. You have to keep it in the bounds. You should

• 8:42Wes Bos?

  be able to preview it. And then also, there, there is an NFC tag in the labels themselves that the printer will read. So it knows which label you currently have. And there is an API for pulling all of the information about these labels, you know, size, printable area, etc. So I was just like, let's, let's add this. This is a great idea. And then like he didn't want to. He's like, we shouldn't, I don't want to do this, you know, like it's serverless. Okay, you can, you can, you can do it entirely from the browser. They could start blocking requests at any moment. I'm just like, damn it. Like, I really

• 9:15Speaker 02

  Yeah, yeah, you tried.

• 9:17Wes Bos?

  I really wanted to do this. So like stuck an agent on it, made a version that's 10 times better. But I tried.

• 9:23Speaker 02

  Yeah. Yeah. I think that's the issue though, is like, I like psych myself out of like, well, I don't want to wait for a response or they're not going to like my idea.

• 9:33Wes Bos?

  Yeah.

• 9:33Speaker 02

  And so, I mean, especially it's sad to say, but especially in the agentic era where I can literally just fork it, add the thing that I want. It works for me. I mean, it really is the era of like more personal software. But that also means there's less collaboration. There's less sharing. So it's unfortunate.

• 9:50Wes Bos?

  Honestly, though, like in my personal software is decent, but it's not going to be as good as if

• 9:56Scott Tolinski?

  Right.

• 9:56Wes Bos?

  80 people have input on like little tweaks and whatever, like, like this type of thing. And then I think that's going to be the rarest thing is the people that simply just care enough to continue on with a project because Every right now, every time I search for something, I find eight GitHub repos that have been touched four months ago and nothing since. Because everybody vibes up a version of it, they post it on Reddit, announcing blah, blah, blah, studio, open source version of X, Y, and Z. And then a month later, it's crickets, doesn't work anymore. And nobody's, nobody's interested because they're moved on to making motion graphics and Claude or something like that.

• 10:36Scott Tolinski?

  You know what? Paulo Paulo just says as a maintainer, Paulo works on Svelte kit and a number of things. The fact that that now every issue comes with a PR is nice, but the fact that every issue comes with a PR is a nightmare, which is hilarious because anybody who's been in this industry for any amount of time, like, yeah, you would have been so delighted if every issue came with a PR before. It's just such a weird swap that now they're just inundated with so much slop PRs or slop requests that you have to act as such like a gatekeeper of this repo. Otherwise, who knows what's going to get let in and, and therefore the, the most, you know, optimal solution there is just to turn off PRs. Yeah.

• 11:21Wes Bos?

  I made a Chrome extension a couple of months ago called JSON Alexander, which is essentially just like a JSON highlighter for Chrome. We talked about it on the show previously, but as soon as I released that thing, like the feature canon got shot at me and like, everybody wanted to add every feature under the sun. And I ended up just starting to ignore everything because I was just like, I am happy with the current status of it. I don't want to have to be implementing all, taking all your slop pull requests. And like, this is just a stupid little Chrome extension that has like 15 features in it, you know, let alone like Svelte, something that's actually real.

• 12:04Scott Tolinski?

  Yeah, you should have had like a bot that auto replied like an AI JSON Alexander just been like, it's slop.

• 12:10Wes Bos?

  Believe it or not, this is some good slop. Please leave a message.

• 12:17Scott Tolinski?

  Yes, exactly.

• 12:21Wes Bos?

  Oh, what a weird world we are in now. And even like, look at this, there's like a thousand, at the time, everybody was using it because there was no good alternative JSON formater after the like, one that everybody used went like paid. And now there's bazillions of them.

• 12:36Scott Tolinski?

  Yes, billions, yes, bazillions. All right, got anything else there, Wes?

• 12:41Wes Bos?

  PRs are dead. The next one.

• 12:43Scott Tolinski?

  Yes.

• 12:44Wes Bos?

  Pull requests are dead. Rest in peace.

• 12:46Scott Tolinski?

  Pull requests are dead and Pi is alive and well. Pi has hit version 1.0 and there's been a number of wild new features in addition to a new harness entirely. And we're going to be getting into all of that here. So Pi, if you don't know, is the minimal extendable harness. It's the harness that you can build and make it like you want. And because of that, it does not include things like MCP or many other things that could just be added on. Well, in a reverse of course, Pi has added MCP because today the MCP world has changed, right?

• 13:28Scott Tolinski?

  MCP is not what it was a year ago in terms of like, remember back when we were first getting into TypeScript, Wes, there was TypeScript and Flow and, and not that MCP had all these competitors, but it wasn't so ubiquitous and now it is.

• 13:43Wes Bos?

  No, it was like, it wasn't stateless and like MCP has been changed, what like probably six or seven times over the last year or so. And it's, it's finally at a spot now where, oh, six, seven, I can't stop saying it. It's finally at a spot now where it's like, it's actually really good.

• 14:00Scott Tolinski?

  Yes, well, Pi in MC, well, Pi always had MCP through extensions, but now it's, it's built into Pi. So wonderful. There, another cool thing is, is code mode. I'll talk about code mode in just a second. Extension support for virtual models deferred tool loading, so it doesn't load all like tool descriptions and titles, I believe, at, at the jump. Correct me if I'm wrong about that. Cash warming for anthropic models, mid-conversation system messages, a new TUI theme and full-screen mode by default. The big, big billboard features here are obviously MCP and code mode. Now, I, I feel like code mode is one of those things that like, you hear about, you don't necessarily know the benefit of.

• 14:45Scott Tolinski?

  And so I want to just break it down real quick, what the heck code mode is and why you should care about it being added to Pi. It's just a simple configuration turn on, but the way that your agent or Pi worked before is that the agent, you know, searches, it reads a file, it reads another file, it compares results, it does this and that, it writes a script, right? These are the things that your agent's doing with code mode. What it's doing is it's writing a small program and it's doing so inside of a small sandbox. So it goes off, it writes a program, it does its stuff. So instead of the agent kind of doing each thing, what it's doing is, is it's inside of a sandbox, it's able to do things like perform a search, read the files in parallel,

• 15:30Scott Tolinski?

  filter and combine results, and then get the useful results only, and then send that back to the agent. So ultimately what you're doing is a massive amount of context saving by the amount of unnecessary things that are taking place essentially behind the scenes of a general call. So that sandboxing is basically giving you a more efficient way of working with tools, MCPs, even things like Jev, right? You could run inside of code mode, you could have it go off and classify a hundred things or whatever, but that's not going to blow the context. The only thing that matters is what comes back from that.

• 16:10Wes Bos?

  Yeah, Cloudflare was like, I think the first to like popularize this whole idea of code mode, which was their, their MCP was absolutely massive. And what you had to do is create a tool for absolutely everything that you could do on Cloudflare. And then Cloudflare would say, okay, these are my tools. I will call this one. And then the response came back, okay, now I will check your dashboard. And then the response comes back and like, it's like, okay, well, these, these LMs are pretty good at like calling tools, but you know what they're, they're really good at? They're really good at writing and running code. So if you give them, essentially you give them the whole TypeScript definition of here is all of the functions you can call and all of the arguments. And then here is somewhere you can run it.

• 16:52Wes Bos?

  And then all you also have all of JavaScript available to you, right? You know, the entire entirety of the language to run and then just go nuts and then run it all. And then like Scott says, you just simply just get the answer back and then that goes, it's, it's much faster, it's much more flexible and it's not going to blow out your context. So it's great to see code mode really catching on.

• 17:15Scott Tolinski?

  Yes, people who don't know too much about context bloating in general, the more you fill your context with, obviously the more money you're spending, but also specifically the worse your responses are going to get because everything that's in context is being passed through the LLM. So therefore you might have a bunch of unnecessary stuff in the thing you're trying to do. So basically you don't want your context to be too big at any given point. You want just enough so that the agent can solve the task with enough information and knows how to access that information. And this helps that. So I'm pretty stoked about that because Pi is already great and it just got greater. Another thing that they just announced is Pi durable, which is a durable agent harness.

• 17:59Scott Tolinski?

  So Pi durable is a completely, I don't want to say completely different project, but it's not like Pi 1.0 feature. Basically long-running durable agents that can run anywhere. They have storage that storage contains the conversation log. So all user messages, model responses, tool execution responses. It stores tasks. These things are excellent because they can be marked as replay safe. They can be resumable. So that way, if something happens and the agent work stops, there's a crash or anything like that, it can just pick up exactly where it left off on resume. And if something is marked as replay safe, it will just replay it.

• 18:44Scott Tolinski?

  If not, it won't. You could run many things in parallel. So this opens up like better forking, it opens up better subagents.

• 18:55Scott Tolinski?

  It's pretty cool. It's kind of like Pi's version, like I said, it's a durable agent harness and it can run anywhere. You can run it on durable objects. You, you know, you, you can have this to build your entire agent system if you want. In fact, the Sentry has an agent called Junior, which is, Junior is like a Slack bot agent that has access to a bunch of stuff. And I think they're rebuilding it right now in Pi durable because this solves so many of their issues. So really cool stuff.

• 19:27Wes Bos?

  I'm trying to think like, like what, what is this useful for? Like, I understand, like you could like pause and resume. Yeah.

• 19:35Speaker 02

  Yeah, so I'm, I'm, I have a personal assistant I'm building that's basically like a more locked down, less give it all your data version of OpenClaw. And I'm building it on top of Pi. So I have a web interface where I can have different chat threads. And anytime I'm chatting in that thread, that actually fires up a Pi agent, passes it the context, but I had to manage everything myself. So I have my own database of message history. I have like try catches that are able to like launch an agent and if it fails, relaunch it and like re-inject the context. So like, all of this stuff, I basically had to build from scratch on top of Pi because Pi is really good as kind of like the core agent because it can connect to various AIs. It has a few good built-in tools.

• 20:18Speaker 02

  But with, I can imagine replacing basically half of my code base with Pi durable. So instead of having to re-architect all of the storage of messages and like launching of agents and making sure that when they crash, they reload and that kind of thing, just use Pi durable instead. And it basically handles all the, the extra details for me.

• 20:37Wes Bos?

  Oh, but like Pi itself doesn't have this then?

• 20:42Speaker 02

  I think of Pi, think of Pi as like one Claude instance, right? So like, if you wanted to write an, write an app around it, you have to add everything else. Like, um, yeah, so I don't know. Maybe if you have a better example, Scott, or

• 20:57Scott Tolinski?

  No, I think that's the, cause like typically Wes, what you're thinking of like, uh, for right now, like I'm doing everything with my prompt boy, which is also going to get moved to Pi durable. Because prompt boy, again, is like a personal agent system. And I, each of those messages has context. They have long conversation histories, this or that. Uh, they, they, they might have steps involved. And each time that I run something for my agent system, it's just kind of doing its process. Right? It's either running scripts and then calling Pi from the CLI or whatever. Um, this is more like a control system for all of that. And, um, another thing that it does really well is that it can allow you to have like multiplayer,

• 21:42Scott Tolinski?

  uh, coding sessions where multiple users are able to be a part of the same context.

• 21:47Speaker 02

  Oh, yeah.

• 21:47Scott Tolinski?

  And the same conversation history and use the same tools and the same process.

• 21:51Wes Bos?

  You know what's really interesting about this is like, you could theoretically, like pause a conversation and then move it somewhere else. Um, whether that is to a, an externally available, a computer to something maybe with a little bit more oomph behind it, you know, like imagine you were working on some, what would be something like that? Like a, like a big remotion thing. You're rendering a whole bunch of motion graphics. You're halfway through and you go, you know what? My computer's on fire. Let's, let's bring this thing up to like spin up some, some external thing that has a GPU available. Right? You could just move it. And, or you could like say, hey, CJ, I'm done for the day. Can you continue on? Here is my durable.

• 22:36Speaker 02

  Yeah.

• 22:36Scott Tolinski?

  Yeah. Yeah. And you're building any agent system that needs to, uh, be resilient to crashes. Right? Like

• 22:44Wes Bos?

  Yeah.

• 22:44Scott Tolinski?

  Client work, uh, customer work that isn't, is a, your app is built on top of an agent platform. Right? And a customer is doing something on your agent platform and that fails. The process fails, it crashes out or whatever. Right? To just resume and restart that would be, uh, I think like a really great thing for a, obviously for a engineering perspective, um, not having to write all that yourself.

• 23:09Speaker 02

  Yeah, I see it as being the base for a lot of people's agent tech workflows and apps. Like this is the thing that people have kind of been building from scratch themselves. And I think it's going to fill that niche.

• 23:20Scott Tolinski?

  Yes, yes. Here's I think, think of a Slack channel where your agent answers mentions from anyone. Then somebody opens a thread. The channel can be one conversation and the thread, a fork of the conversation at the message thread to the replies. Both run at the same time and neither blocks each other. So that is a really interesting thing, like the forking stuff I mentioned, like it would enable, uh, better forking and long running process and stuff like that. So yes, very cool stuff. Uh, Pi folks be cooking, so I'm, I'm stoked to see this. And like you said, CJ, my, my own agent thing, the prompt boy, uh, man, this is going to get a lot.

• 24:02Scott Tolinski?

  Now I've been doing all this stuff myself, so I'm, I'm very stoked to see this again. It's going to replace a bunch of my code. Yeah. Um, that's all I got here from Pi. Awesome new release 1.0 and durable, Pi durable. It's sick. You know what else is sick folks is Sentry, Sentry.io. Head on over to Sentry to solve your errors, your exceptions, to solve your slowdowns, to give your agents some dang context on how to fix your stuff. Because just because AI is writing it and they're writing a thousand if statements to check and type guard something and, uh, codex models, doesn't mean that that code is still ultimately going to produce no bugs from your users. Because what? I had a bug yesterday, guys, in my prompt boy website.

• 24:45Scott Tolinski?

  And it was just a stupid little, little website bug, but it's not going to be one that an agent's going to find. It was, I log in and the, the, the side wrapper didn't render. Why didn't it render? I don't know, but it threw an error in the console. That gets logged to Sentry. I have an agent, uh, ingest system that'll pick it up and just get that work done for me. So check it out, Sentry.io. They are the very, very best. Check them out.

• 25:08Speaker 02

  Hope. All right, this next story is super exciting to me.

• 25:13Speaker 02

  Um, cause do you have just like two favorite things? And then all of a sudden they become one and then it's just an even better thing.

• 25:18Scott Tolinski?

  Were you saying when two becomes one?

• 25:20Speaker 02

  When two becomes one, and that's basically, uh, Terso is joining Supabase. So if you're not familiar, uh, Superbase is, um, essentially a way to host a Postgres database, but with a whole UI on top of it and also real time on top of it, plus a bunch of other features as well. But Superbase was my preferred way of deploying Postgres databases. And Terso was my preferred way of deploying SQLite, LibSQL databases. And now they are one, now they're one single platform. I don't, it doesn't seem like this is a bad thing in any way. Basically now, instead of having to go to two different places, you go to one. Uh, in the post here, they say that, uh, for everyone building on Terso today, nothing changes. Terso database stays open source and we keep shipping new features.

• 26:05Speaker 02

  Um, so I think this is all around good news. Um, and essentially like if you need a database hosted and it's Postgres or SQLite, you now have basically a one stop shop for it. So

• 26:17Wes Bos?

  So I'm, I'm curious about this cause like, like will Superbase sit on top of Terso at some point? Or like, will there be a SQLite version? Or is this just simply like, uh, we are a database company. They are a database company and they acquired them for whatever reason.

• 26:34Speaker 02

  I, I don't think it's in there. They don't mention anything like that in the announcement, but I think we can speculate. Right? Because if you've ever used Superbase, it, it's almost like a CMS for the database in a sense. Like there's a web UI where you can create your tables and insert rows and add cloud triggers. And they also allow for, um, row level security. So you don't necessarily need a backend to like lock down rows in your database. I absolutely imagine that you could, they have, will get to a point where you swap out, uh, Postgres and use Terso behind the scenes instead. So like a SQLite, with LibSQL database as the backing store instead of Postgres, with all of the same real time features on top of it. I, I bet that's going to be a thing. Like, why not?

• 27:15Scott Tolinski?

  I bet that's, I bet that'll be a thing. Yeah.

• 27:17Speaker 02

  Yeah. But as far as I know, it's not in the announcement, but I, that's basically what Superbase does best. It's what Terso does best. So I could imagine the two marrying, doing something like that.

• 27:28Wes Bos?

  Yeah. Man.

• 27:30Speaker 02

  Yeah.

• 27:30Wes Bos?

  We've, I think this is our third story with a past syntax guest today. And we're only a half hour in.

• 27:36Scott Tolinski?

  Third story? No, yeah, we had Oliver Dunk. We got Mario and Armin from Pi. And then, uh, what was this last one? Sorry, uh,

• 27:46Wes Bos?

  The

• 27:46Scott Tolinski?

  Blanking for some reason. Terso, yes. Uh, and, and Superbase. We've had Paul from Superbase and Glaber from, uh, Terso on.

• 27:55Wes Bos?

  Wow. Holy smokes.

• 27:56Scott Tolinski?

  We talked to everybody, folks. We're going to be talking to more people too.

• 28:00Wes Bos?

  I've been trying to get Sindray Sora's on the podcast for years. And, uh, he's just not one of those guys that likes to go on podcasts, but like the, I like to keep trying.

• 28:11Scott Tolinski?

  Keep trying. Um, cool. Wes, you got something else here?

• 28:15Wes Bos?

  Yes. What do we, what do we got here? Um, Screen Studio. So this is kind of interesting. Have you guys ever used Screen Studio? This is like one of these apps where you record your screen and it does, this, this got really popular a couple of years ago, where there's like swoopy moves around.

• 28:34Scott Tolinski?

  Yeah. It always did too much for me, but

• 28:37Wes Bos?

  And zooms in.

• 28:38Scott Tolinski?

  I thought it looked nice.

• 28:40Wes Bos?

  Yeah. Yeah. It's, it basically took like these like screen recordings and allowed you to make like these nice little demos. Super easy. Um, and so it kind of like, kind of bit into like the lightweight video editing space, but then it also kind of built, bit into like the like loom space where I, I just need to show you my screen and like demo how something works. Um, and it's pretty popular in the like launch video. Um, like if a company's launching a new product, they'll use something like this. And I, it never fit into any real use case for me because either the like normal screenshot tool was, was fine, or I'm going full on in, in like a video editor.

• 29:23Wes Bos?

  Um, because it's like, it's kind of limited because there's like not multiple tracks. But that's not what I'm here to talk about. I just think this is really interesting because they got absolutely slopped with everybody making their own version of Screen, Screen Studio over the last year. You know, everybody vibe coded their own version of this type of thing. Um, and, and quite honestly, this happens with a lot of products where like the same thing with like, um, like a text to speech, no, speech to text. Everybody's making their own version of like, press a button and dictate it and, and transcribe it.

• 29:58Scott Tolinski?

  I haven't made one.

• 29:58Wes Bos?

  Put it in. Right? Yeah. See, if, if Scott's made it, you know that, that has happened. So with these products, you kind of look at it and you go like, okay, like, like what are they going to do? Right?

• 30:11Wes Bos?

  Um, and they released version four with like all kinds of interesting implementations. And a lot of the stuff was like, they have this like glass loop where you get this like cool effect. And essentially a lot of people will see your demo and say, What did you use to record that? Or what are you using to, to take your little videos? Right? And if everybody has the same one, then it starts to get played out and nobody cares about it anymore. So they said they spent like eight months building this thing and they added this glass loop, camera cutout, edit by transcript, screen split, face tracking. One of the really interesting things is they added sharper zooms. Zoomed in text is now sharper.

• 30:55Wes Bos?

  We trained our own AI model on Mac OS screenshots we generated. When you zoom in, text is now perfectly sharp. So I asked him, like, what are you doing here? And he didn't reply, I think, because he doesn't want people copying him. But I think what's happening here is that they are detecting the text and the font and all of this stuff. And they are literally repainting the text on top of the video so that you can zoom in infinitely and the text will always be crispy, which is always one of my biggest pet peeves. All kinds of cool features. And I'm not, I'm not just here to like applaud them releasing this, but I think it was really interesting

• 31:36Wes Bos?

  response to now that anybody can three or four prompt a version, like a kind of a crappy version of this out there, what do you do with your software? And it looks like these guys doubled down and made some super high quality software. And I love to see it, you know? It's like, well, everybody can do their own kind of mediocre version of this. But what if we do some crazy stuff like pay attention to text rendering and zooming in? And like lots of these little fussy features that makes a piece of software a joy to use, which you don't see a lot of anymore these days. So I applaud them for this and I think that they did an awesome job with it.

• 32:19Speaker 02

  Yeah, huge props. I think honestly, it's like a model that people should be following in this age of slop forks and agents, like, yeah, do some real research, like, like dig into your expertise and make some good stuff instead of just slopping out like copies of everything else. So.

• 32:35Wes Bos?

  Exactly. Yeah, that it kind of drives me nuts that like, people are like, well, I can just make this software myself in 10 seconds. And I've, I've said that many times over, but like, we're not going to stop there with the software, you know? People that actually have skills and, and this type of stuff are going to figure out, okay, well, what else can we do? What can, we also have this technology. What can we build? You know, can, could we re-render every single letter in a video? Unbelievable that you even thought you could do that, let alone actually ship it.

• 33:11Speaker 02

  Yeah.

• 33:12Scott Tolinski?

  Yeah, we need, we won't stop there with software, Wes. You accidentally coined another chant.

• 33:17Wes Bos?

  Don't stop there.

• 33:19Scott Tolinski?

  Software. There with the software. Yes, put it on a T-shirt, folks. Yes. Speaking of not stopping there with the software, Svelte Kit 3 is here. We talked about it when it dropped in the release candidate and it is officially out as of October 1st. It's spooky season and therefore it is time for the Orange Framework to drop a major version release. We talked about this before and all the major features, you know, charts are going up and to the right there and they have really removed one config file from our config file hell. So we now just use v dot config instead of v dot config and svelte dot config.

• 34:01Scott Tolinski?

  There's a small change in how they're doing aliases to use standard sub path imports. So again, removing configuration, leaning on the general standards.

• 34:14Scott Tolinski?

  And V variables got some neat little features, which I believe are Varla, Varlock like in terms of not, not Varlock entirely, but let me see, cause I actually read this a while ago and I forgot what these were.

• 34:30Scott Tolinski?

  I believe, see, did you know what these, the validation? Yes, you can validate your ENV vars using standard schema. So that is what I was talking about.

• 34:40Wes Bos?

  Oh, like typed environmental variables.

• 34:43Scott Tolinski?

  Correct.

• 34:44Speaker 02

  It's built in.

• 34:46Scott Tolinski?

  Really nice. So service workers are now less boilerplatey, which is great because I never liked doing those anyways. And then error handling, there's instrumentation dot, that is an instrumentation svelte file. Either way, that allows you to connect to any instrumentation. Paolo's in here to clean up what I'm saying because he worked on this. Remote functions are not quite ready yet, which is a huge bummer because I wish these things, I've been using these things for like a year now, believe it or not. Bear with us. Folks, I want my remote functions. I will, I will say that right now. I want them now. But no, no, shout out to the svelte team. The remote functions do absolutely hit. I've been using them for a while. They're great.

• 35:30Scott Tolinski?

  So that's one big thing. Another big thing that was released was effect version of 4.0. Folks have any, either of you used effect yet?

• 35:40Wes Bos?

  No.

• 35:40Speaker 02

  About it, read about it. Yes. Haven't used it. Yeah.

• 35:43Wes Bos?

  We had, speaking of another people that's been on the show, we had Johannes on to talk about effect and try to explain it to us. Cause I think that was the hardest part is like the, understanding the mental model of it. And that was, I don't know, probably about a year and a half ago. And I tried it. I, I tried, but it was just, it was just too much of a jump for me from that type of thing. But I think it's gaining crazy popularity right now because you don't necessarily need to understand that mental model as well as you did previously.

• 36:17Scott Tolinski?

  Yes, effect is falling into the same camp as Rust. People found it to be scary and because effects overly verbose syntax and its intensity when writing it, I looked at this and said, no, thank you. It's not for me. But now that I might not look at it, then well, it gets a little less scary for me or might not have to write it. I'm not going to have to write the star yield or whatever is going on in here. But I, you know, I can read the code. It looks good to read just now. I don't, I don't want to write it personally. Either way, svelte for charts go up and to the right, five times smaller bundle, 6.4 times concurrent task throughput and 86% less memory per fiber. Reason being is it was rebuilt from the ground up with a dependency free core.

• 37:02Scott Tolinski?

  So, and it went LTS too. So long-term support for version four here. Either way, dependency free core, I feel like is something that is just a wonderful thing, folks.

• 37:14Wes Bos?

  That's actually crazy.

• 37:15Scott Tolinski?

  Dependency free.

• 37:16Wes Bos?

  No dependencies all the way down.

• 37:19Scott Tolinski?

  Dependency free core.

• 37:21Wes Bos?

  Did copy and paste them in then? Does that mean, or? They literally wrote them all themselves.

• 37:26Speaker 02

  I mean, they had to re-implement some stuff internally, I'm sure.

• 37:29Scott Tolinski?

  Yeah. Look at this chart, the NPM downloads.

• 37:32Wes Bos?

  That's every single thing ever right now. Show me a chart that doesn't look like that these days.

• 37:37Scott Tolinski?

  Yes.

• 37:37Speaker 02

  Of all the agents downloading.

• 37:38Wes Bos?

  You know, except for like, like businesses revenue is the opposite.

• 37:43Scott Tolinski?

  Yes. Well, either way, I do have a project that I've been throwing effect on to have agents write it. And you know what? This isn't, this is again, this is a vibe coded slop that I'm not, I'm not, it's not a long-term project. I don't care about it, but I wanted to see what it was like. And so far the code results have been good in terms of, I'm not having to go back and tweak things or whatever. It just seems to work. Blah, blah, blah. It is using the effect V4, it is using it with svelte. And it's using it with something called alchemy. Folks, have either of you heard about alchemy? Look at this transition.

• 38:17Wes Bos?

  No. No.

• 38:19Scott Tolinski?

  Okay. So I posted on Twitter last week, I said, or maybe it was a week before that. I said, hey, like, what are people using with effect? Like, just because I wanted to try effect, but I wanted to, you know, use what people are using, use what the people like, any good repos. And the one thing that every single person mentioned to me was, oh, I use this and this with alchemy. Oh, I use this and this with alchemy, this and this and alchemy. Alchemy is, I think this is, you know, similar to like SST, I believe it was called SST, the one from DAX that allows you to, to.

• 38:57Wes Bos?

  Yeah, before their whole company got sidetracked with open code.

• 39:01Scott Tolinski?

  Yes, before the whole company got sidetracked with open code. Yes. Man, so alchemy is basically that works for everything.

• 39:11Wes Bos?

  Can you explain what SST was? It's for deployment, right?

• 39:14Scott Tolinski?

  Yes, it's for your infrastructure as code. Basically, instead of, you know, connecting this to that and doing click ops where you're clicking around to connect this to this, to this, or you're having to deal with like wrangler files and whatever in cloud flare, you're basically defining your entire infrastructure as code. And because of that, it makes it easy. It makes it portable. It makes it just generally like more deterministic in how things are going, right? You don't have a, like what? We, we've had like in the past, like Slack documents or Notion documents that have like process for things and this would, you know, eliminate some of that. So connects to all this stuff.

• 39:56Scott Tolinski?

  So I'm using it with cloud flare right now and it handles all kinds of connecting all of my services together to make sure that all of my things talk to each other, my data, my D1, my R2. I actually do have an R2 in this one. And I do have, yeah, I got D1, I got email, I got all this stuff and I'm doing it all with alchemy. And again, just so far, it's been nice and solid. No complaints. It just works. Infrastructure as code is one of those things that I always felt on like, felt like a bother to me until I implemented it one time and it's like, oh. Yeah, I get why this is good.

• 40:33Wes Bos?

  Speaking of infrastructure as code, and you just said that Svelte dropped a config file. Guess who dropped a new config file?

• 40:43Scott Tolinski?

  Who?

• 40:43Wes Bos?

  Cloudflare.

• 40:44Scott Tolinski?

  Okay.

• 40:45Wes Bos?

  Cloudflare.

• 40:45Scott Tolinski?

  Cloudflare.

• 40:46Wes Bos?

  Yeah, maybe about a year ago, they moved from the wrangler dot Tumble file, which like, kill me, hate writing.

• 40:53Scott Tolinski?

  Yes, exactly. Yeah.

• 40:54Wes Bos?

  Well, I hate. I could never understand how to copy paste all the Tumble stuff. They moved to wrangler dot JSON or dot JSON C. And now, last week, they announced cloudflare config dot TS, which is, I'm assuming they're doing this because you can programmatically configure your, your infrastructure as code, meaning that like previously with, with the JSON file or whatever, it's just static. But if you want to have an if statement inside of where something is deployed, then you would have to all of this, you'd have like six or seven different cloudflare config, or you'd be programmatically generating your wrangler file out the other end.

• 41:37Wes Bos?

  And it looks to be that they are solving that with a single cloudflare config TS file, which I very much welcome that kind of thing.

• 41:46Scott Tolinski?

  Or use alchemy and not be locked into cloud flares.

• 41:49Wes Bos?

  That's true.

• 41:50Scott Tolinski?

  And you get, you get the same thing, you get TypeScript, but you can deploy to Hetner, Fly, Railway, Planet Scale, Neon, Prisma, AWS, Cloudflare, and more.

• 42:01Wes Bos?

  That's actually not a bad idea. Maybe I will be using alchemy because then, yeah, you have, have an out. I.

• 42:08Scott Tolinski?

  Portability.

• 42:10Wes Bos?

  This is always a hard thing with like, a lot of the cloudflare stuff. Because I'm always like, like, don't build on something that is going to totally lock you in. Build on top of like, like whenever I use cloudflare like D1, I'll just use drizzle on top of that so that if I ever want to move to something else, it's really not a big deal to move from one to another, right? So I guess like this, using alchemy, it'd just be one step further in that type of thing. But I'll say, I don't care. Like moving from one config to another these days is so easy.

• 42:46Scott Tolinski?

  Yeah.

• 42:46Wes Bos?

  It's not, that escape hatch is so much easier now, these days where I'm not as guarded against being locked in.

• 42:54Scott Tolinski?

  Yeah. Wes, I got one more in terms of the effect stuff, but that's it. Foldkit was another thing people kept recommending to me. I did not bite the bullet on this because I do not need another framework here. And not another front end framework, but this is one that is built on effect. So the effect heads who are all going all in on effect are like, well, what if we put the front end framework in effect as well? And I mean, honestly, it's fine. If I wasn't having to write or read this code, yeah, I'm sure it's good. But it's also too new, the fact that there's a giant beta tag on here that like, I can't imagine building anything on this that's serious right now.

• 43:37Scott Tolinski?

  But it does seem to be quote unquote, batteries included framework.

• 43:44Scott Tolinski?

  Uses the Elm architecture. I don't know much about Elm, folks. I'm sorry, I'm that guy. I don't know anything about Elm, really. I know that it exists.

• 43:53Scott Tolinski?

  But I did not use this, but I heard so many people talking about this. I'm sure you're going to be hearing a lot more about it in like three months. So folks, you're here to hear, heard it here first if you haven't already, which I'm sure you have.

• 44:06Scott Tolinski?

  Actually, I think even like Fire shipped a video on this, so this might not be that niche.

• 44:11Wes Bos?

  All right, CJ, you got some news for us?

• 44:14Speaker 02

  Yeah, I think we'll get into like the AI stuff because OpenAI had a dev day.

• 44:19Scott Tolinski?

  Wait, what's AI?

• 44:21Speaker 02

  Wait, what?

• 44:22Wes Bos?

  I'm in.

• 44:23Scott Tolinski?

  We talk about web dev on this show, CJ. I don't think we talk about AI.

• 44:27Speaker 02

  It's gonna, it's gonna be quick. We're gonna, we're gonna be, it'll be quick. But also I have like one story, I think Wes, you have one not AI story. Should we do those first and then just do AI the rest of the time after that? What do you think?

• 44:38Scott Tolinski?

  Well, it's just.

• 44:38Wes Bos?

  No, I want to talk about the OpenAI stuff.

• 44:40Scott Tolinski?

  I want to talk about the OpenAI stuff.

• 44:42Wes Bos?

  I know you guys hate them, but I think there's some interesting stuff announced.

• 44:45Speaker 02

  Here we go. So last week, if you did not pay attention, OpenAI dev day happened and they basically just, it's like their regularly scheduled developer thing where they announce all the new stuff that we should care about. And there are quite a few announcements. One of the biggest ones was the announcement of dots, which are essentially agents that run on a computer. And if you have an OpenAI subscription, I think you need at least the max subscription or the pro subscription.

• 45:15Wes Bos?

  Yeah, I think a hundred dollars a month currently, there's a minimum.

• 45:19Speaker 02

  Minimum hundred bucks a month. Now you have an always on computer with an agent on it. And this was, it seems like this is very much a response to Muse agents that were launched by Meta. Because this gives you a very similar thing. Basically, your dot lives on a VM. It can install applications, it can run applications, and you can just ask it to do things. So that's dots. There's also GPT 6.1 soul that was released. So new version bump on GPT 6, it's slightly cheaper. Near Astra intelligence for the price of soul, which is pretty slick.

• 45:55Wes Bos?

  And I haven't been seeing the greatest reviews. I haven't tried this one yet, but haven't been seeing.

• 45:59Scott Tolinski?

  I've been trying it.

• 46:00Wes Bos?

  The greatest reviews. What do you think, Scott?

• 46:02Scott Tolinski?

  I've got stuff to say. You know, it's still annoying in the same way that OpenAI models have been annoying to me for a while.

• 46:12Scott Tolinski?

  I think that's really it is that like the things, you know, like when Claude 5.5 came out, it was such like, oh, wow, that you can actually talk to this thing. There's just some like really annoying quirks with OpenAI models in general that like, I find myself having to direct them way more. And it's not that I have to tell it to work because you don't want things to go off and work when you don't want it to. But I find myself having to really micromanage what it's doing in the directions it's taking and the approaches it's taking, rather than just like trusting it with stuff. It also still sucks at front end. It major sucks at front end. They haven't gotten that any better. I don't particularly like talking to it, not like in a personal type of way, but like

• 46:55Scott Tolinski?

  reading its responses or like trying to decipher what that the hell is trying to tell me.

• 47:02Wes Bos?

  I think that's why, like a lot of people are saying the like dots rollout doesn't have as much magic as either Grokbot or Muse had. And partially, I think that's because it was probably rushed out and it just like, isn't, isn't as exciting. And partially, I think it's just because of the like, the actual like soul that, that the model, for better words, you know, the soul that the model has of, of being able to talk back and forth like Scott just says, I don't like reading the responses that it gives you.

• 47:37Scott Tolinski?

  Yeah, I just found it made some really odd choices. So one of the things I'm trying to do is really standardize our dev process for local and what like to, for a number of things like, so that it's ultimately portable, ultimately easy to spin up anywhere to have, to connect the syntax off to whatever side project you're doing at any given point locally, databases, everything just works. And the choices that soul was making 5.1 was just really annoying to me. Like, let me go and modify your host file. Now I'm like, one, I didn't ask you to modify my host file. Two, like that's completely unnecessary for what we're doing.

• 48:17Scott Tolinski?

  And I just found like in that type of situation, it was just doing the classic OpenAI model thing of over engineering everything.

• 48:27Speaker 02

  Yeah, I could agree with that. But the thing is like, there are some things I'm building where I don't look at the code because they're one off projects, they're quick prototypes. And GPT 6 has been perfectly fine for that because I don't have opinions on it. But I feel like I'm in the same boat where like, if I have opinions, Claude is usually more aligned with like how I want to do things or it's more like developer centric. It's easier to understand what it's doing.

• 48:51Wes Bos?

  Yes.

• 48:52Speaker 02

  I could agree with that, but I use both models for, for different things, so.

• 48:57Wes Bos?

  One thing they did announce was their reaction to Jev, which it seems like what they did was they took GPT 6 Luna, which we said this last week, we're like, couldn't Luna just be the equivalent to Jev? And it seems like they just took Luna. They may have like put it on the like super fast speed because one of the benefits to using Jev is that it is fast as hell. You can get a response like we, I think we did a test when we were on live stream and it was like Jev is like 200 milliseconds and we did one with Luna and it was like two seconds, you know, like way, way slower.

• 49:38Wes Bos?

  And also the pricing was a little bit more, but they rolled out like a first class Luna API, probably as just a quick gut reaction to Jev so that you can now get deterministic output from Luna. It's still charging as far as I can understand Luna pricing, but this was really interesting to me is GPT 6 Luna, this is from Rahim on Twitter, is twice Jeff's price for uncached input, but if you have long instructions, you can put them in the first message and then reuse them at later calls for 0.0 like one cent per million token compared to half a cent per million tokens with Jeff. So that was really interesting.

• 50:19Wes Bos?

  If you're doing, especially for categorization and, and whatnot, you often are sending the exact same prompt.

• 50:27Wes Bos?

  Every single time with Jev. So it's, it maybe is not that far off from the pricing of Jev.

• 50:33Speaker 02

  Yeah, like, so I'm just imagining how that would work because with, with Jev, you basically pass it context and then a list of a bunch of questions. But I can imagine if with this setup with Luna, you almost like give it the context and then ask the questions almost one at a time. But now everything else above it is cached. So you're not paying for that input anymore. You're really just paying for the answer to like an individual question. So that's, that's pretty cool.

• 50:56Scott Tolinski?

  What a good time to mention clef.

• 50:57Wes Bos?

  There's image inputs as well, which is killer use case, right? Having, being able to pass it in visuals is not something Jev can do.

• 51:06Speaker 02

  Yeah, Jev doesn't support that. We'll mention that next, Scott. Yeah. So Scott mentioned there is this thing called clef. There's a couple more things we'll talk about from OpenAI, but yeah, clef is, did come out from Cloudflare. Wes, do you want to talk about extensions?

• 51:19Wes Bos?

  Yeah. So OpenAI rolled out, um, man, I, the, the names of these things is so hard for me to remember because they announced something. And then they announced an update to it and they announced an update to that. They announced plugin extensions. So they've had plugins for a while and plugins in ChatGPT are a combination of skills and MCP. Both MCP servers as well as MCP UI, meaning that, um, you could say like, hey, find me a pizza place is the example that they always give. And then instead of it returning a list of pizza places from your MCP, you can also return like some HTML and CSS that can be shown in there.

• 52:02Wes Bos?

  So I did a whole talk at JS, um, JS Nation a couple of months ago, being like, this is really exciting, but it still kind of sucks. Because they're just embedding these little frames into your chat window and it's kind of a crappy experience, right? Um, and the promise always was, well, that will get better. You'll be able to have sidebars and you'll be able to integrate it into the ChatGPT app. Um, and that's exactly what they announced last week, which is plugin extensions, which is you are able to control more parts of the ChatGPT app. Whether you have something in the sidebar, whether you have, I think probably the biggest use case is you'll be able to have an entire persistent

• 52:45Wes Bos?

  application open, similar to how you would have like a browser open to the right. And then you'll be able to talk to it and it will, it will continually update that UI rather than previously what would happen is you would simply just be regenerating a new UI every single prompt, which is like, kind of sucks. Even, uh, Sam Altman said in the keynote, he's like, we've, we've tried a couple of stabs at this in the past and they weren't very well received. And, and that, that is an answer to, yeah, it's, it's too limiting, kind of crappy. So I'm, I'm curious to see if this is the, the answer to people finally using more UI in it. Or are people just still going to existing websites?

• 53:27Speaker 02

  Yeah. So there are a couple of.

• 53:29Wes Bos?

  I don't buy it, but.

• 53:31Speaker 02

  Well, all I can think about now, like looking at some of their other announcements as well, is it feels like anybody that tried to build an app based on AI, like even something like lovable, which is like a vibe coding website, is just slowly getting, what's the word, sherlocked by OpenAI itself. Cause like all these little features, all these little apps are basically like, somebody started a company that could do that. But now OpenAI just released a, like a piece that you can now use as a part of your subscription. So like they're slowly eating away all of the little companies that started up to do these kinds of things. So a few of the other announcements, like I hadn't, I haven't used any of this stuff, but there's this idea of ChatGPT space. It's almost like a workspace where like everybody on your team can collaborate on stuff.

• 54:16Speaker 02

  But with that, you also have pages. So this almost seems like a Notion knockoff where like multiple people can all edit them at the same time and have access to AI. You also have slides. So now they have their own Google slides replacement. And like, there's also like a task board. So now it's like a linear replacement. So like all, all these things are just like, they're just slowly eating away the world of apps and software by hosting it themselves. And I, I personally am not using them yet, but I could imagine a lot of enterprise being like, well, we only want to pay one subscription. So we're just going to pay OpenAI our subscription and now we don't have to have like G suite and everything else.

• 54:57Wes Bos?

  Yeah.

• 54:59Wes Bos?

  So when is the, like the Microsoft Teams of AI gonna happen? Where you talk to anybody who uses Microsoft Teams at their work and, or you're like, oh yeah, like let's jump on a call. And then, then they go, we use Teams and it's because it just comes with their subscription and they, they have like, everything comes with Microsoft Teams, you know, and they, they pay for Microsoft Word and Teams comes with it. And like, that is for sure going to happen with the AI. And I think that they're trying to get ahead of it where being like, before Microsoft rolls out the like,

• 55:38Wes Bos?

  okay, you get Copilot with your PowerPoint license, then yeah, people can't use, don't want to, don't want to have to use it.

• 55:48Speaker 02

  Yeah. And then one other announcement, I didn't even know that maybe this already existed, but it's called sites in ChatGPT. And I see this as just like a lovable replacement. So it's a vibe coding replacement. You, uh, prompt it for a website and then it'll just be hosted within like your, uh, ChatGPT workspace. And people are creating like microsites, landing pages, that kind of thing. But now you can do it directly inside of ChatGPT.

• 56:11Wes Bos?

  Oh yeah. And they actually deploy them to Cloudflare. Did you know?

• 56:15Speaker 02

  Did not know that. Wow.

• 56:16Wes Bos?

  Oh, behind the scenes, somebody else deploys to, oh, Muse, Muse sites. If you, or they're called spaces inside of Muse. If you like make, you can make like a little app or something, right? And if you ask it to, to make it publicly available, it actually just deploys it to Cloudflare worker behind the scenes.

• 56:35Speaker 02

  Wow.

• 56:36Wes Bos?

  I see. I downloaded all the prompts.

• 56:39Speaker 02

  Oh, but speaking of Cloudflare, um, they also released their own decision model called clef. So basically if you, if you see the word decision model, just think Jev. Everybody's coming out with their own version of Jev. Cause Jev was like super hyped when it came out. And everyone's like, oh, we could sort of do that. So yeah.

• 56:58Scott Tolinski?

  It's a Jev off, folks. We got ourselves a Jev off.

• 57:02Speaker 02

  So Cloudflare came out with their own, um, and it's actually the, of all the benchmarks I've seen of Jev, clef is the only one. Again, these are like, trust me, bro, benchmarks, but, um, clef is the only one that actually it gets marked above Jev in some of these benchmarks, uh, which was interesting to see. I will say that it also is more expensive because I, it's, it's based on some kind of LLM. Let me see if I can get to the pricing. I think it was like four, like almost 20 times the price. Let's see. I can't find it in here, but they released their own decision model, very similar to how OpenAI released their own decision model.

• 57:44Speaker 02

  But again, it's, oh, I, yeah. So it's based on top of Quinn 3.827B.

• 57:49Wes Bos?

  Okay.

• 57:49Speaker 02

  So they essentially built on top of an existing LLM, but, uh, did some interesting things to make it faster and respond as quickly as something like Jev. So yeah, interesting to see.

• 58:03Scott Tolinski?

  Yeah. You were saying that it's, it's, you said it's way more expensive?

• 58:07Speaker 02

  I'm trying to find the pricing now. I think it might actually be on their, uh, documentation. Let me see if I can get to the page.

• 58:14Wes Bos?

  Oh, it's already on OpenRouter.

• 58:17Scott Tolinski?

  Yeah, it was released everywhere day one. I, uh, yeah, I downloaded this game and it's surprising that Cloudflare is getting the game, but then not that surprising. Given, you know, um, who they are and what they do.

• 58:30Speaker 02

  It's 24 cents per million input tokens and Jev is four cents per million input tokens.

• 58:37Scott Tolinski?

  Okay. Yeah.

• 58:38Speaker 02

  So.

• 58:38Wes Bos?

  And both of them are, output is free.

• 58:41Speaker 02

  Yeah, exactly. You're only paying for the, the input. I will say that, uh, clef also has a larger context than Jev. So Jev's context, I think is around, uh, 30,000 tokens. And, uh, clef's context is 65,000 tokens.

• 58:57Wes Bos?

  Okay.

• 58:57Speaker 02

  So bigger context, a little more expensive, but two or three percentage points smarter than, than Jev.

• 59:04Wes Bos?

  Yep. And it, I guess it really depends. Like this is why you need your own internal benchmarks for everything to see. As these things get released, you should be able to just stick this, these new models on it and see, is this actually better? Especially for the money. And I think that's why a lot of people were excited about Jev is that, oh, wow, this is significantly cheaper. And a lot, lot better than those dirt cheap LLMs that they are using.

• 59:32Scott Tolinski?

  Yeah. I really have the, the Jev bench open from, um, the last time that Jev was released. And there's one called Qyet, Q U Y E T 1.0 large, that's based on, uh, GMA 431B. And that seems to be crushing right now in the Jev bench. Very high intelligence score, zero, four, uh, four cents again per decision. So very cheap as well. I don't know about vision or anything like that. I haven't, but, uh, it is wild that this Jev just came out and the Jev bench, oh, a week ago, looked very different than it did today.

• 1:00:11Scott Tolinski?

  There was like one or two other things and now there's just hundreds of them. So, um, yeah, it's Jeff Timber is, is continuing in October. Yeah.

• 1:00:22Wes Bos?

  It just makes me think like, like how much more awesome stuff is just not being thought of right now? You know, everybody is just reacting to what everybody else is doing. And then Jev comes around and I realize there was some similar stuff previously, but a lot of people saw Jev and goes, yes, we actually need this. This is awesome. I'm so excited about this. And like, how much more awesome stuff is there out there? People like thinking about this stuff rather than just trying to make another version, a cheaper version, a slightly better version of what we have, right?

• 1:00:56Speaker 02

  Yeah. And also just like a week later, you're totally right. Everything's reactionary. I mean, I'm sure there are people that are, are building in stealth or like, you know, they're going to release at some point, but a lot of what clouds the news and everything else is everybody just reacting instead of like doing something like truly innovative.

• 1:01:16Speaker 02

  Cool. I got a bunch of other AI news. We're just gonna, um,

• 1:01:21Speaker 02

  Whirlwind, stop me if you want to talk about it a little bit more. First up, Gemini for Argon is soon to be released. It's, I do believe it's like, nobody can use it just yet. Because it's

• 1:01:34Wes Bos?

  There's some security researchers have access so far, but yeah, no, it's not like anybody, no influencers got it yet.

• 1:01:40Speaker 02

  Yeah, so it's limited access, but it's Google's frontier model. And the trust me, bro benchmarks put it slightly above Astra and Fable and Opus 5.5. So, once this gets released to the world, Google is, as far as I've seen, this is the first time I've seen one of the Gemini models ahead of the other AI companies.

• 1:02:04Scott Tolinski?

  Well, oddly enough, Gemini 3.8 Flash scored really high on deep SWE. So like, that was surprising when it did, but it is like the number two model on that, that benchmark.

• 1:02:17Speaker 02

  Nice.

• 1:02:18Scott Tolinski?

  Which I don't know what that entails, folks, I'm sorry. I, yeah, there's too much stuff.

• 1:02:22Speaker 02

  No, but I mean, I don't really reach for Gemini for anything, like maybe for like Nano Banana for image generation. So the fact that I already have a Google subscription and I'm paying for like the extra AI tier. Now we might get to a point where I can get rid of my OpenAI subscription or my Cloud subscription, but like, we'll see. Obviously, I obviously have to try it, but they did kind of the same thing as when like Mythos was coming around and, and some of the other newer models where they need people to try and use it and stress test it and jailbreak it first because it's really good at cybersecurity. And they're gonna basically ship a, a version of it that won't let normies do as much harm in terms of cybersecurity.

• 1:03:07Wes Bos?

  Yeah, hell yeah. Like, Google's been like slacking for so long on this type of thing. And I don't know, maybe CJ even knows this, I don't know if you do, but like, how long does it take to, to train one of these models? You know?

• 1:03:21Speaker 02

  Just, I've just heard like something like GPT-6, I heard like six months, I could be wrong, but that's the last number I remember hearing. But yeah, basically, it's so crazy to, because these models feel like they're coming out every other month. But that, the training for those models started half a year, maybe even a year ago. So it's a while.

• 1:03:39Wes Bos?

  Because we've just been sitting on our hands, waiting for Google to drop, like a good, like they're certainly capable of, of doing something like this, right? There, it's not like we're gonna live forever where they're not gonna have the OpenAI or Anthropic equivalent models. So, awesome. And like, the next one is, is Microsoft. You know, they announced several models a couple of months ago, and, but they didn't announce like their, their Fable or Opus 5.5 equivalent. I'm hoping that GitHub Universe is, is when they announce something, but you gotta think they've been, been trained in something and, and we'll get that out pretty soon as well. So I'm excited, I'm stoked to see this, especially because

• 1:04:24Wes Bos?

  like so many people currently access the internet through Chrome, Chromebooks, Google, you know? And I, I think there's still a huge opportunity for Google to like kick some ass in this arena.

• 1:04:40Speaker 02

  Yeah, it's interesting, like they haven't gone the way of Yahoo yet, like

• 1:04:44Wes Bos?

  No.

• 1:04:44Speaker 02

  Yahoo, like Yahoo used to be like, it was the homepage everybody went to and everybody used Yahoo Mail. And they slowly like, people still use it, but it's not what it once was. And in the early days of AI, it kind of felt like maybe Google would go that way. Cause I don't know if you remember Bard, like they, when OpenAI came out with ChatGPT, like they couldn't keep up.

• 1:05:05Wes Bos?

  Yeah.

• 1:05:05Speaker 02

  But it looks like they've had time to cook.

• 1:05:07Scott Tolinski?

  And they, you know what, how long before they dropped the name Gemini and named their next model like Google Model 1 or something stupid? Like, just the, the naming and product organization at Google is something to behold. Like, OpenAI, top 10, OpenAI is the Sony naming of AI products. But Killed by Google is still a thing where they just yeet projects out there. They yeet like eight different IDEs and tech setters and CLIs and who knows what? And eventually change their name and then standardize on one that's not as good. So I, I look forward to the next eventual rename of whatever their AI project is.

• 1:05:46Wes Bos?

  Simon, Simon Curtis says, according to Google folk, this is not even close to their internal model.

• 1:05:52Speaker 02

  Ooh. Like, meaning they have something smarter internally?

• 1:05:55Wes Bos?

  Maybe, like they, they probably do, but it's just like not affordable or like even possible to roll out at Google's scale. And like, they're just using that to distill into the, into smaller models, you know? Because like, even if you look at like Meta, the Muse API, like they're not, they're not using a killer model to power this thing. But for most people's tasks, it is totally fine.

• 1:06:22Scott Tolinski?

  Cause it's using all of your data.

• 1:06:24Wes Bos?

  Check their email, exactly. Yeah. But like, that's the thing, it's good at summarizing all of your data and being able to

• 1:06:32Wes Bos?

  like come back to you. So like, that might be where Google is like coming at it from this type of thing. It's like, we need something that's freaking awesome that you, we can put at the top of search results. That, that's the other thing is that they want to have a really good model, but it has to be fast enough that they can generate the result.

• 1:06:50Scott Tolinski?

  Yeah, I saw something that was like Muse, it was like a security researcher who was like Muse is building a complete graph of all of your connections and their jobs and their lives and all this stuff inside of Meta Muse. And I'm just thinking like, duh, what do you mean? It's like a security researcher needs to uncover that. That's obviously what it does, right? Like, that's why it's useful. Yeah.

• 1:07:15Wes Bos?

  Well, let me ask you this, Scott. All of these things come out, right? That you've got Grok Bot, you got Muse, you've got Dot. I'm sure Claude is gonna roll out their version of it. Who do you trust the most, if anything, other than the, the obvious answer of you should be self-hosting this? Cause like, I really want to automate some of my banking stuff, but just can't. I'm, I'm a pretty risky guy, but I don't think I'm doing my banking.

• 1:07:45Speaker 02

  Yeah, who do you trust?

• 1:07:46Scott Tolinski?

  Number, the, who would I trust the most? If, if you're saying I have to pick one.

• 1:07:51Wes Bos?

  Yeah.

• 1:07:51Scott Tolinski?

  Gotta pick one of these four to trust the most.

• 1:07:56Scott Tolinski?

  I, I'm going to pick Anthropic. I'm gonna pick Anthropic number one. I would probably, this is like, Wes, this is playing like, which hammer do you want to hit yourself in the hand with? I, I guess I would pick, oh, God, this is terrible, Wes. I would probably pick OpenAI 2. I would probably pick Grok 3 and Muse 4.

• 1:08:19Wes Bos?

  Where's the Google one? Because I guess they have, the Google one is called Spark or something like that? But like, they already have my email and that's probably one of the biggest things.

• 1:08:29Speaker 02

  Well, if you use Chrome, they have your browsing history. If you use an Android phone with Google Board, they have every key you've ever typed. So like, that's, that's where, that's the boat I'm in. I've already given up most of my life to Google, so I would, why not just

• 1:08:44Wes Bos?

  Can we just keep doing that?

• 1:08:45Speaker 02

  Keep everything there?

• 1:08:46Wes Bos?

  Yeah.

• 1:08:47Scott Tolinski?

  Yeah, that is kind of the thing. I did have an Android phone for a long time and yes, all my stuff on Google.

• 1:08:53Wes Bos?

  What? Probably, I have seven Google microphones planted around my house?

• 1:08:58Scott Tolinski?

  Yes.

• 1:08:58Wes Bos?

  Yeah. It's quite a bit.

• 1:09:02Speaker 02

  Yeah. So, okay, other AI news. So Black Forest Labs has released Flux 3, which is a new image model. I haven't used any image models from them. Have you, have you guys from Black Forest Labs?

• 1:09:14Wes Bos?

  I did the, I did the like, what if the web were felt video on the Syntax Channel? Where we were in real time generating the UI, like where you could literally click things and type and it was just a video output. That was using Flux 2.5, some dumbed down version of it that was fast enough to, to run in real time.

• 1:09:37Speaker 02

  Okay, pretty slick. Well, they have released a new model. And the cool thing about this is you can select specific sections of the, of the image to just regenerate that part. Now, I feel like we've sort of seen this, like, like Mid Journey has an editor where you can sort of do this. Or like, maybe you can do this in Photoshop, but like some of the examples here are like really slick, like basically just selecting the text on an image and saying, just regenerate this part, or selecting a specific spot. So it's, it's kind of getting to the point where simply just prompting gets you to the level of something like a Photoshop or a more in-depth image editor. So it's, it's basically saying like skip the dock boxes, skip the prompts and basically select the parts of the images that you want to generate.

• 1:10:22Speaker 02

  Also, just like the, the single prompt quality of some of these images are like really slick as well.

• 1:10:31Wes Bos?

  And this is great. I gotta cancel my Mid Journey. I, I've been back and forth to Mid Journey so many times. And this obviously has an API and I look at my Mid Journey and there's months that I don't even use it. And I'm paying 14 bucks a month for it. And I just think like, Mid Journey doesn't have the API. You can't just like pay based on credits. So I would love to just move, maybe I'll, I'll look at moving to something like this, because I only ever use it. The one thing is repeating tileable backgrounds. No models other than Mid Journey have the option to do that. And I make repeating backgrounds for every video I do and just as like a nice little wallpaper.

• 1:11:15Wes Bos?

  And that's my big use case for it.

• 1:11:17Scott Tolinski?

  Yeah, maybe Open Router is the, the answer, Wes. I mean, Open Router's got all this stuff and, and you can try different ones. And the, the pricing is good, so.

• 1:11:26Wes Bos?

  All right.

• 1:11:27Speaker 02

  Yeah, the other cool thing about this one is like reference images. So, I mean, I, I've tried to do this with like Nano Banana where you pass it a few images and are like, basically composite this into an image. Or like, put this outfit on this person. Yeah. So this can do that as well, but it, it seems to do a pretty decent job, at least in the example that they're showing. You can quote up to 10 different reference images.

• 1:11:49Speaker 02

  So yeah, pretty slick. Next up, Claude has released mods. We talked about this a little bit because this is actually how they, Claude launched the ability to read agents.md. It launched as a mod, but now in Claude, you can change how it behaves, customize the UI, swap in your own features. And it's, all of the mods are written in TypeScript. So have you guys tried these out yet or?

• 1:12:14Scott Tolinski?

  Nope, there's too much stuff to try.

• 1:12:18Speaker 02

  It's true.

• 1:12:19Wes Bos?

  I, I did look into it when they announced the agents.md support and that's how they did it. But I'm just curious, like, what other, like what else are people modding here?

• 1:12:34Speaker 02

  Yeah, some of the examples in the, um, just the, the X post here are like, it almost seems like they're, they're modding Claude desktop or just like the, the web UI to show different elements.

• 1:12:48Wes Bos?

  Oh, they're like changing the UI. That's what I saw. I was like changing the UI of Claude itself.

• 1:12:53Scott Tolinski?

  Yes.

• 1:12:54Speaker 02

  So this is Replay Theater where it basically shows you different steps that I guess the model took and it's like a separate window where you can like tab through each of the steps. I guess they, they, that was created as a mod to like tap into the history and be able to show it in a custom UI. Um, and then Token Weather adds your weather forecast to your Claude code CLI.

• 1:13:17Speaker 02

  So you can see at the very bottom, it says what the weather is while your output is happening above. So you have control of putting different items in different areas. So I could imagine people creating plugins that maybe do different calculations of like your usage, your context window, and then you can have that almost as like a heads up display in your, in your Claude code. So.

• 1:13:38Scott Tolinski?

  Oh, yeah. I see that.

• 1:13:40Speaker 02

  Yeah, yeah, yeah. Um, what else? Um, Muse Gadgets. So Muse is Meta's new AI agent thing we talked about last time, um, or maybe the time before, but this side project was, was launched by Nat Friedman and the idea is to take any open source ESP32 device and make it so that it can kind of be a, uh, external world way of interacting with your Muse agent. And so the site's pretty cool. They have a few of like different devices and instructions for like making your own. It can also run like on a Raspberry Pi. Um, this is fascinating to me because this is the kind of thing I am already working on, not necessarily to

• 1:14:24Speaker 02

  talk to Muse. It's like to talk to my own agents or this is very similar to like what Scott did with Prompt Boy. Um, and yeah, hack, hack your hardware.

• 1:14:35Wes Bos?

  Yeah, I got it running on my, what is this? RP 2350. This is a Tufty badge. Um, these are the badges that you see when you go to a conference and they have like, like full color screens or E Ink display or whatever. So I took the Muse API, which was meant for an ESP32. Um, inside of this is a similar chip from Raspberry Pi. It's not a Linux computer. It's not a Raspberry Pi. It's like, kind of like Raspberry Pi's, um, answer to what an ESP32 is, which is a dirt cheap, um, chip that has like Wi-Fi and Bluetooth inside of it.

• 1:15:17Wes Bos?

  And I popped it on there. It's really cool because the, the way that it works is that you just add the device to your devices inside of Muse and then that device can either bring in data into your agent. Like you could have like temperature sensors or, um, GPS or like, and you could literally any data in there that could bring that data into your Muse or you can use it as a way to display data onto your Muse. I guess kind of similar to the Claude code, um, hooks that we were, we were just talking about where you could say like, "Display on my Muse X, Y, and Z," right? Uh, so it, it worked. It works pretty cool.

• 1:15:58Wes Bos?

  Um, it's amazing that like I, it, it was written in C for ESP32 and in like 10 minutes I had it ported to MicroPython to run on this Raspberry Pi, which it's, it's cool because it still runs the original badge firmware as well.

• 1:16:16Speaker 02

  Nice. So that badge looks like you got it from a conference, but did you actually buy it from them as like an example to hack on?

• 1:16:22Wes Bos?

  I got it sent to me. So I did last year after GitHub Universe, I did a video on the GitHub Universe badge. Um, and the guy who runs Pimmarini, that's the company that makes these badges. He sent me three of their like upcoming badges, which I think are now out. One of them is E Ink. One of them is like a little LED matrix display and one of them is like a full RGB color video screen. Um, so he sent me all of those. And then I also did hear that GitHub Universe is going to have some sick ones this year as well. So you can just buy these if you like want something to hack on. Um, and you probably

• 1:17:05Wes Bos?

  see everybody grab the little circle ones from Waveshare, but this is also a really good one because it has four buttons on the front and it comes with a little game pad that you can put on it and it comes with a little accelerometer, temperature sensor, something like that, which is a pretty good deal if you're, if you're interested in getting into hardware hacking.

• 1:17:27Speaker 02

  Nope. I've got one more piece of AI news for you. Um, Lyra 2.0 was released. And so this is a world model released by Nvidia. And the idea is you can pass it a single 2D image and it will generate a 3D world that you can actually walk through and interact with from a single 2D image. Um, there's a bunch of like scene exploration here. So there's one where it looks like an office with like books and stuff. From that single image, it basically creates a world where you can walk in and see like the other parts of the house and, and stuff like that. There's another one that looks like the streets of some seaside, uh, town. Um, and then you can basically just walk the streets. Um, there's

• 1:18:10Speaker 02

  also one, like this giant robot in the desert. And then it just starts to expand into the, into the world. Um, so super cool from a single image and this is an open weight model. So they released it on Hugging Face to try out. I couldn't figure out what specs are required. You're definitely going to need some beefy GPUs to do stuff like this.

• 1:18:30Wes Bos?

  Yeah.

• 1:18:31Speaker 02

  But I, we're getting there. Like this is the, we're living in a simulation.

• 1:18:38Scott Tolinski?

  Yeah. Yeah, dude. My Vision Pro is going to be looking good. I'm going to have some, some new fake, uh, environments everywhere.

• 1:18:47Wes Bos?

  It's so cool. The, the 3D in the last couple of months has gotten so, so good. Like this is just one image. Um, but like, I've been working on one where I built an iOS app that takes LiDAR data and high-res images and then pipes them into like a 3D world. Um, and it's amazing. Like, like, this is so good at doing this from one image. Imagine how good your home tours are going to get when you can provide it good data about like dimensions of a room and, and high-res images of like the underside of a desk or something like that.

• 1:19:26Speaker 02

  This is it. Like it's basically Ready Player One. I don't, I guess that's a movie reference. Ready Player One was like a famous book about augment, like virtual reality.

• 1:19:34Wes Bos?

  Yeah.

• 1:19:35Speaker 02

  But yeah, like take any Google Street View scan, pair that with a little bit of LiDAR data and now you can literally inside of your VR world, walk around any town or city.

• 1:19:44Wes Bos?

  Oh, yeah.

• 1:19:45Speaker 02

  And yes, and like you're saying is like take home scans and, and like feed it into a model like this. Like we're basically going to get highly detailed, custom video game worlds that you can walk around in and exist in. So.

• 1:19:59Wes Bos?

  This is what I want for like real estate. Um, when we bought our house, there was like a 3D tour where like they would set up a camera and it would like spin around and take a million photos and then like, like piece them all together, right? And I saved that when we bought our house. I saved it because I was like, I want to be able to, if I'm not home, I want to be able to go to any room in our house now and move it around. And this was like a multi-thousand dollar camera. It costs lots of money to have somebody come and, and do your house. And then like, here I am a couple of years later where like I can literally just do this on my phone and I can have a better version of it. And we would talk about like AI staging photos. Like I want the opposite, which is like, give me the full tour. I want to be able to like go under the sink.

• 1:20:46Speaker 02

  See all the weird plugs that are messed up and like.

• 1:20:49Scott Tolinski?

  Those are the most fun parts of a house for sure.

• 1:20:51Wes Bos?

  Where's the robot that does that? You know, like strap a camera on a Roomba and give it legs and like, let, just like leave your house for a couple of days and then the, the little robot will just crawl around your house.

• 1:21:04Scott Tolinski?

  You think the real litter wants that? They, they don't want that?

• 1:21:06Wes Bos?

  No, no, they definitely don't want that.

• 1:21:09Scott Tolinski?

  You see this dry rot on the baseboards? Nope.

• 1:21:12Speaker 02

  Exactly.

• 1:21:14Scott Tolinski?

  All right.

• 1:21:15Wes Bos?

  I like that. It's actually something that is happening in construction right now, though. People will, when, once they like plum in an entire house, you know, you got your electrical, you got your plumbing, you have your gas, HVAC, all of those things, then somebody will come in and take photos of every, everything in your house. And then now you, you sheetrock over top of it. And now you know, if I want to be able to drill, like I was hanging something up above me. And I was like, shoot, what is there? I'm pretty sure there's a drainage pipe above me. So I can't, I can't drill in there. So I had to go back to old photos. And I found one photo or one video where I was just like filming around and I paused it and I could see what pipe was there. And I was like, damn, how sick would that be?

• 1:21:59Wes Bos?

  If I had just a 3D view of my entire house? So if I wanted to know where an electrical run is, or if I can hang up a picture here without popping a hot water line, that would be so sick. And it's already happening in, in construction, but like, give this to normies. Let everybody have access to this.

• 1:22:18Speaker 02

  Yeah, I would love that. Like, imagine that being part of your household package when you buy a new house. Like, you can see what's behind the walls and like, not have to guess.

• 1:22:27Scott Tolinski?

  Yeah. Oh, man. And then you can, but you combine that with the VR headset and you just, you see through the walls. You get X-ray vision through the walls.

• 1:22:34Wes Bos?

  I've seen that as well. We're like, guys who are, um, there's a guy, a guy using the Apple, what's the Apple goggles that you have?

• 1:22:42Scott Tolinski?

  Vision Pro.

• 1:22:43Wes Bos?

  Vision Pro. So, uh, an electrician is in the, he's downstairs and he's, he basically makes us a little circle on where he wants the plug to go. And then he goes up into the attic and it just goes right up into the attic. And he knows exactly where to drill to go, to go into the wall. There's no measuring or anything like that. There's Scott, he's got his.

• 1:23:06Scott Tolinski?

  Howdy.

• 1:23:06Wes Bos?

  Oh, my gosh.

• 1:23:08Scott Tolinski?

  Yes, I, dude, I use this like, five days a week now.

• 1:23:12Wes Bos?

  Do you really?

• 1:23:13Scott Tolinski?

  Yeah.

• 1:23:15Scott Tolinski?

  The OS has gotten so much better on it. Um, gosh, the, that's the number one thing that people don't understand about the Vision Pro, because it is a heavy piece of hardware. And that is the biggest downside with it. But this was not supposed to be a mass market consumer device. It like, there's no possible way at this size. The OS on this bad boy is so good, and it's going to continue to get better, that by the time that whatever form factor reaches, um, you know, consumers in general, that everyone's going to be using glasses, you're gonna be like, wow, this is dialed. Um, I use it mostly for virtual monitor, watching sports and stuff. So I'll have my big giant wall. Um, the, the resolution's

• 1:23:59Scott Tolinski?

  incredible. It doesn't bug my eyes at all. And then I can put on a sports game over here. I can, you know, uh, have my iChat over here. Either way, uh, I sit in the car while my daughter does dance for two hours, three days a week. Uh, my son does rock climbing for two hours, one day a week. And then while I'm doing that, uh, I'm rocking this bad boy out of my car to get a big giant ultra wide on my, my MacBook.

• 1:24:26Wes Bos?

  Have you taken it on a plane yet?

• 1:24:28Scott Tolinski?

  I have, yes.

• 1:24:29Wes Bos?

  I got the Zuck version that.

• 1:24:30Speaker 02

  It's the Quest.

• 1:24:32Scott Tolinski?

  You got a sandwich delivery, folks. I'm sorry.

• 1:24:34Wes Bos?

  Quest 3S is what I have. And I, makes me wanna puke. And it, it makes me exhausted. And every time I take it off, I'm just like, oh, but like, that's probably the difference between what a $300 and a $5,000 headset.

• 1:24:50Scott Tolinski?

  Yeah. The, the thing about the Vision Pro that's so good is the visual fidelity. The biggest downside is the weight, obviously. But the visual fidelity and the, the, like, what, what the, cause I have, um, a Meta Quest 2. The Quest, gosh, Wes, have you done the, the hand tracking for the, the Quest where you're like using your hands to point and click on things?

• 1:25:12Wes Bos?

  Yeah, I'm doing it right now. I'm beating up a couple nerds.

• 1:25:15Scott Tolinski?

  It's terrible.

• 1:25:17Wes Bos?

  It is. I'm just like, always confused. I'm like, how do I get that thing? How do I get out of here?

• 1:25:22Scott Tolinski?

  Right. And the Vision Pro, because it has the inner eye tracking, you're just looking at something. And my hands typically stay down. So your hands don't have to be up here. My hands are just down here. And I'm just going like tap, tap, tap, tap. And I'm looking at stuff. Um, and, or you can even do, you know, two, either way, it rips. It's so good. And I just can't wait for this as a form factor that other people can actually care about.

• 1:25:44Wes Bos?

  It's never gonna happen. It's like the segue. It makes 100% of sense.

• 1:25:48Scott Tolinski?

  No.

• 1:25:49Wes Bos?

  But no one is gonna use it because you look like a nerd. And like, maybe when they throw it on like glasses.

• 1:25:55Scott Tolinski?

  Yes, that's what I'm saying. Yes.

• 1:25:56Wes Bos?

  But then like, everybody's freaked out because all the pervs are wearing them in public.

• 1:25:59Scott Tolinski?

  No. Well, look at the new Meta glasses, which are much lower profile. Now, granted, those things are, are gonna be, uh, you know, a nightmare for Terms of Service in general. But people were losing their minds over that.

• 1:26:12Wes Bos?

  Scott's hate for Meta is so good.

• 1:26:14Scott Tolinski?

  It's deep, man. I don't like it. Yeah.

• 1:26:18Speaker 02

  I actually, it's so funny. I was looking into AR glasses this weekend and there are, there are AR glasses and not really even AR, almost just like second screen glasses, but they're, they don't have cameras on them. They're like the anti-creep kind of thing.

• 1:26:33Speaker 02

  And they're like sub $600. There's a few different, I forget all the brand names.

• 1:26:37Scott Tolinski?

  Yeah, yeah.

• 1:26:37Speaker 02

  But I was thinking of.

• 1:26:38Scott Tolinski?

  Or whatever.

• 1:26:39Speaker 02

  Yeah, Xreal One Pro or something like that. But I was thinking about getting them for like, using them on an airplane, like for watching movies or using my laptop.

• 1:26:48Scott Tolinski?

  I watched, I watched Robocop on the Vision Pro on my way to Italy. And it was so sick knowing that I could watch a guy get blown away in like a massive screen and not have to worry about little kids watching this thing.

• 1:27:01Speaker 02

  Yeah.

• 1:27:02Wes Bos?

  That's always the most awkward thing. You watch a movie on a plane and then it's like a inappropriate scene. And everyone's like, look at this pervert, you know?

• 1:27:09Scott Tolinski?

  Yeah, yeah, yeah.

• 1:27:12Speaker 02

  But I, I feel the same way. Like, I want glasses that aren't massive and like hard to wear and make you sweat.

• 1:27:19Wes Bos?

  Yeah.

• 1:27:20Speaker 02

  And you can wear them for longer. I think once, once we get that, that's, that's the point where I think more people will, will wear them. We'll get more adoption.

• 1:27:26Scott Tolinski?

  Yes. Guys, I got something.

• 1:27:29Wes Bos?

  Yes.

• 1:27:30Scott Tolinski?

  Yes. Have you all heard about the re-comps and de-comps that are taking place with video games? Have you heard about this?

• 1:27:39Wes Bos?

  I mean, no.

• 1:27:39Scott Tolinski?

  So because of AI, people are figuring out that it's getting easier and easier to decompile video games. These are video games that have been released for a long time. Then they're making modifications to them and then recompiling them. This video from Modern Vintage Gamer is, oh, it's a great channel either way. This video is so crazy. So let me, like, the, the scene has like blown up. So there isn't just like a couple of people doing this. There are people doing everything in this. And they're increasing, they're making them like, uh, 60 FPS and, and 4K or whatever. Like, they're making them run

• 1:28:19Scott Tolinski?

  really smooth, but they're also making them run on anything. People took full on games that are, have just been released or making them run on macOS just via recompiling them. It is wild. And I've also seen there's people who are like mix, mishmashing games or combining games together. Somebody put the NHL, uh, NHL skater inside of some haunted house video game I saw today. And they're just skating around the house in the game. It's insane, the stuff that people are doing. Putting Mario inside of Elden Rings, um, fixing bugs, long running bugs. And these, like, look at Diddy Kong Racing. It looks clean as hell. It looks smooth and fast. And it's wild. This, this scene is so interesting. And I

• 1:29:04Scott Tolinski?

  just heard about it from the Modern Vintage Gamer video. Um, yeah, it's cool.

• 1:29:10Wes Bos?

  That is, I accidentally got, gave our kids Mario Kart Black. Because I modded our Wii and I was like, oh, Mario Kart. So I, I acquired it and threw it on there. And I was like, all right, here's Mario Kart. And it like boots up and like, it was like really creepy and weird. And then I was like, what the heck? It's Mario Kart Black. And I looked it up and it was like this, like, well-known, like mod, like somebody decompiled it and totally changed all of the settings. And, uh, it was, it was weird. But now, I guess that's more common now.

• 1:29:45Scott Tolinski?

  Yes, this is.

• 1:29:46Speaker 02

  More evidence for like, we're living in the simulation or we're getting to the point of full simulation. So just like being able to have a random 3D world off of a single image. Imagine any game you've ever played and you can live inside of it or combine games or mashup games. I don't know.

• 1:30:01Scott Tolinski?

  Modify it. Yes. Uh, one I want to call out, man, these closed captions, get out of here. Uh, Ogre Battle 64. This game is a prime choice for recompiling and decompiling. This is an all time favorite game of mine. Number one on the list, Ogre Battle 64. This game cannot be emulated well. Every single version of emulation in this game has black scenes. It has bugs. It has missing elements. It lags. It's slow. It's built in a very odd way, apparently. And because of that, it's been so long since I've been able to build or play this game as it's intended. Even though I guess I have a Nintendo 64 and I do have a copy of it I could plug in. But like, man, I would be so stoked to play

• 1:30:42Scott Tolinski?

  Ogre Battle 64, you know, up sampled on my Mac, just right here. Give me that all day.

• 1:30:49Wes Bos?

  Beautiful.

• 1:30:50Scott Tolinski?

  That's all I got. That's all I got.

• 1:30:52Wes Bos?

  I got one more thing. This is nothing to do with AI and it's everything to do with JavaScript. Stage two composites being added to JavaScript likely stage two doesn't mean it's going to go through yet, but essentially if you want to compare an object in JavaScript, right now, if it is not the exact same object, if it's just by values, then an object with the same property keys is will not be equal to each other. And you'll have to use some third party library to check that the actual values, the keys of that object. There is now a proposal to add composites to JavaScript, which will essentially check the values of an object. If they are the same as

• 1:31:36Wes Bos?

  as another value. And this will work for objects, meaning that it will work for arrays, maps, sets, all of that other stuff as well. This is from Rob Palmer, which is he's on the TC 39 board. It's been moved to stage two.

• 1:31:52Speaker 02

  Nope. I'm curious, like, so there is, what is it called? Like, there's a comparison method that does deep comparison.

• 1:32:02Wes Bos?

  Yeah.

• 1:32:04Speaker 02

  What's that called?

• 1:32:06Wes Bos?

  Strict equals, was it? No. Yeah. What was that?

• 1:32:11Speaker 02

  It's on the tip of my tongue. Why can't I think of it?

• 1:32:14Scott Tolinski?

  What, what was it? Was it match?

• 1:32:16Speaker 02

  It's like deep, is it deep compare, deep clone?

• 1:32:19Wes Bos?

  Oh, deep clone. Yeah. What was it? Deep clone or deep copy would just make a copy of it. But when you are, oh, structured clone.

• 1:32:32Speaker 02

  Structured clone. That's it.

• 1:32:33Wes Bos?

  Structured clone, structured clone method of window interface creates a deep clone of a value using a structured clone algorithm, meaning that if you have an object and you want to make a copy of it, if the top level property keys of an object will simply just copy paste over, but if there's nested objects inside of that, they are still references and you can get yourself into hot water because you changed the copy, you're accidentally changing the original implementation of it. So that that's for creating new ones. But if you want to check if two objects are of the same values, now we are hopefully, we will be getting composites

• 1:33:12Wes Bos?

  for that. I don't know why this is its own like class and not just like a method.

• 1:33:21Speaker 02

  Um, I think be to essentially to override the double equals or triple equals sign. That's, that's the issue is like in JavaScript, there is no way for you to create a custom class that can do a different operation.

• 1:33:37Wes Bos?

  Okay.

• 1:33:37Speaker 02

  When, when you put the double equal or triple equal between it, right? But it always has like a built in way of doing the comparison. But if you look at a language like I know this from my days in C sharp, you can actually override the compares function. And so that way you can use the double equals on like two random classes you created. But then you implement the method that decides how to actually compare the properties and decide that they're equal. So I think that's what composite gives you basically composite on both sides of the double equal or triple equal sign can implement its own algorithm for determining if those two things are equal.

• 1:34:12Wes Bos?

  Okay. And I assume this also is if you are comparing one thing versus 100 other things, I'm assuming there's some performance gain of converting it to a composite first. And then, because like, like one way you can do this right now is simply just JSON stringifying an object. And then you can check if they are the same thing, same thing. So I'm assuming that converting it to a composite first has some performance benefits where if you are going to be doing multiple conversions of the same composite, you're not incurring that like stringify equivalent every single loop.

• 1:34:51Speaker 02

  Right. So yeah, this seems cool. Is this stage two or?

• 1:34:54Wes Bos?

  Stage two. Yeah. So like, don't get your hopes up just yet, but things are looking up.

• 1:34:59Speaker 02

  Nope. All right. I got one more.

• 1:35:02Wes Bos?

  Yes.

• 1:35:04Speaker 02

  It's pretty fun. So there is an RFC RFC 1149 and this was released in April of 1990, April 1 of 1990. If that gives you any clue to what we're getting into here. But this is a standard for the transmission of IP datagrams on avian carriers. I'll read you the memo. This memo describes an experimental method for the encapsulation of IP datagrams in avian carriers. This specification is primarily useful in metropolitan area networks. This is an experimental, not recommended standard. Distribution of this memo is unlimited. If you don't understand the words that I'm saying, basically it's IP over pigeons or birds. So

• 1:35:46Speaker 02

  the idea is there, it's a RFC for in a way of sending internet protocol packets over messenger pigeon, essentially. So obviously it was a April fool's joke back in 1990, but in April of 2001, there was a real life implementation of it by the Bergen Linux user group where they actually put a packet onto a bird and had it delivered and then decoded on the other side. And so why this is interesting is right now on Christie's. So Christie's is this place where you can buy art and like basically.

• 1:36:25Wes Bos?

  Auction house, right?

• 1:36:26Speaker 02

  Auction house. Yeah. And you, and usually stuff gets pretty expensive here, but the very first carry and pigeon, carrier pigeon internet protocol packet in a nice gold frame is for auction right now on Christie's. So if you want a part of history, right now, the bid is only 2,600.

• 1:36:46Wes Bos?

  Wow. And like, like this is, this was printed out and put stuck on the pigeon?

• 1:36:52Speaker 02

  I believe, yes. And this is the exact piece of paper that was transmitted via pigeon back in 2001.

• 1:36:59Wes Bos?

  Wow. That's not bad. I've seen, I saw people buy way worse stuff. Like when Twitter closed the San Francisco office, they had like a huge auction and they like sold the beer taps and everything and like the Twitter sign for way more. I would buy that for 2,600 bucks.

• 1:37:17Speaker 02

  I would too. There's still eight days left, but yeah, I mean, have a piece of internet history.

• 1:37:22Scott Tolinski?

  We all know how auctions work, folks. This thing's going to get to be $10 million by.

• 1:37:26Wes Bos?

  Yeah.

• 1:37:27Speaker 02

  You're going to snipe it.

• 1:37:28Wes Bos?

  Yeah. Yeah. You're going to like Matt Mullenwag or someone's going to come out and yeah, and buy this thing. Or like Steve Wozniak, someone, some, yeah.

• 1:37:38Speaker 02

  Someone that has money they can just throw.

• 1:37:40Wes Bos?

  Oh, well, I'll make my own.

• 1:37:42Scott Tolinski?

  Yes. That sounds great. Okay. That's very fun. CJ, I got one more thing and we can get out of here. How about that?

• 1:37:48Wes Bos?

  Yes.

• 1:37:49Scott Tolinski?

  Okay. Less fun, but still cool. Sendre Saurus did an ES lint CSS config. And this is brand new and it includes a ton of good stuff. So this is a powerful ES lint rules for CSS and I do use style lint, but I do find it to be a pain that I have to use style lint and this and then ES lint and then that and all this stuff. I like all of these things. So just a handful of great stuff. And again, the more we're building with AI agents, if you want your CSS to actually work well, having a good set of rules is really handy. So this is just a great little pre configured set of rules for your CSS. There's just a ton

• 1:38:33Scott Tolinski?

  of good stuff in here. No duplicate CSS selectors. Yes, obviously. Disallow duplicate font family names, just to reduce on, on bloat there. Disallow declarations after nested rules, deprecated CSS rules, required descriptors and CSS at property rules, just all, all this good stuff that you don't have to set up yourself. So shout out, this is a nice little package here. I'll get this linked up, but nice little CSS for ES lint stuff to wrap you out with. Wrap you out to wrap you. I'm wrapping you out.

• 1:39:07Speaker 02

  That's sick. I'll check it out.

• 1:39:08Scott Tolinski?

  Yes, do so, please. I will post a link for this in the chat. We'll have it in the show notes as well. Guys, is there anything else we got to say before we get out here? We do have a live meetup. Wes, you want to hit them with the details on that meetup?

• 1:39:19Wes Bos?

  Yes, we are going to be meeting in San Francisco. Go to syntax.fm/meetup on October 27th, 5 to 7 PM. If you are in San Francisco for any reason, maybe even GitHub universe, we're going to be there. You can come hang out with us, October 27th, 5 to 7 PM. Go to syntax.fm/meetup and grab your ticket. We are almost out.

• 1:39:45Scott Tolinski?

  Yes. We'll be there and it's going to be a ton of fun. We might even have some goodies. I got some little spooky syntax stickers right here.

• 1:39:54Speaker 02

  Oh, yeah.

• 1:39:55Scott Tolinski?

  Yeah, spookies. So we'll have lots of stickers and stuff and all kinds of who knows what we'll have. Who knows what we'll have. Yes.

• 1:40:02Wes Bos?

  I do know. It's going to be good. All right. Thanks, everybody for tuning in. We'll catch you later.