1032: Stripe Buys OpenRouter for $7B

Scott and CJ break down Stripe's $7B acquisition of OpenRouter, the SvelteKit 3 release candidate, and their first look at TanStack Charts. Plus Zed's new Delta editor, stealing reasoning traces from proprietary LLM APIs, why dark mode toggles only need two states, and all 43 quintillion Rubik's cube states in your browser. Show Notes 00:00 Welcome to Syntax! 02:20 SvelteKit 3 Release Candidate 07:43 Video editor recs Casey Faris 09:58 TanStack Charts First Look TanStack Charts TanStack Charts implementations of every official shadcn/ui chart 14:24 Ponytail AI Tool 17:22 Zed Delta Editor Introducing Delta 23:44 Stripe acquires OpenRouter for $7B+ 29:27 Stealing Reasoning Traces from Proprietary LLM APIs 35:03 First look at Cursor’s GitHub killer 42:34 Dark Mode UX Discussion 46:40 Brought to you by Sentry! 48:34 Should AI respond in plain language? 52:57 Swamp Club Automation 01:02:37 Fight AI slop with this anti-slop plugin 01:03:39 Rubik’s Cube Permutations 01:05:02 Hackweek 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.

• 0:00Speaker 00

  Welcome to Syntax. In this episode, we're talking about all kinds of things. SvelteKit 3 is in release candidate. We're going to be talking about TanStack charts, taking a first look at that. Also Stripe acquired open router. We're going to talk about it, try to figure out why. All this and more today on Syntax Weekly. I am joined by Scott Tolinski. How's it going, Scott?

• 0:23Scott Tolinski?

  What's up? What's up? What's up? I am, uh, I'm, I'm feeling good, man. I released my, my video on Friday that I'd been working on really hard. CJ had been grinding on it as well. In the editing process, doing so much to make that thing perfect. So I'm feeling like I have a big weight off my shoulders. If you haven't watched it yet, I did a survey on mental health and AI coding. And the video dives into that and has a lot of interviews, including interviews from experts, as well as experienced developers and some really great perspectives. If you haven't seen that video yet, immediately following this, go and watch. It's getting a lot of really great reception.

• 1:07Scott Tolinski?

  So feeling very light today. And that, um, well, I'm feeling light because of that, but I'm also feeling heavy because it's Century Hack Week.

• 1:18Speaker 00

  Oh, you got the merch? I didn't get the merch. Nice.

• 1:20Scott Tolinski?

  Yes, I got the merch. It's Century Hack Week and I have chosen to do hardware. And, uh, I don't know if you, you probably won't be able to tell by this, this breadboard where I just have a single button and a microphone just kind of hanging out here. I don't, I'm bad at hardware. I'm very bad at hardware. So this is going to be an adventure for me. I'll just say that. I'm already experiencing the pain of dealing with hardware. How about you CJ?

• 1:49Speaker 00

  Yeah, I'm getting ready for Hack Week. Also, I'm super excited about the video that you put out. Uh, I'm proud of it. You did great work. It's, it's awesome. One of the best videos we've done on Syntax, so definitely check it out. But yeah, Hack Week, I'm making a Game Boy game. So last year I made a Sega game. That's actually a Sega game in the background that's spinning the Syntax logo. But the plan this year, Game Boy game. And if all goes well, we'll release it in a couple of months and you can buy a physical Syntax Game Boy game to play in your Game Boy. So yeah, it's going to be fun. But let's get into it. So I hear there's some SvelteKit 3 news. Scott, what's going on over there?

• 2:26Scott Tolinski?

  So in the last live, I believe I detailed that there was some action going around in Svelte 3 land. Well, the Svelte 3 release candidate is here, or SvelteKit release candidate is here. And this includes a bunch of really great stuff. There are some breaking changes, but as always, the Svelte team does an awesome job of giving you a migrate script. The last migrate script that went from, well, one, uh, there was a migration from, whoa, Sapper to SvelteKit and then SvelteKit V1 to V2 have all been wonderful and easy transitions. Uh, this one is guaranteed to be no different. And SvelteKit 3, uh, it's, it looks pretty dang sick.

• 3:11Scott Tolinski?

  So they got rid of a config file. We love the removal of config files. Everything now lives inside of vit config instead of svelteconfig.js and vit config. That was always confusing to me because some stuff lived in one and some stuff lived in the other. They also changed how the lib alias works. They're aliasing things with a pound sign instead of a dollar sign, reserving the dollar sign for other things. And it's now using the node native, uh, what was it? In package.json sub path imports instead of having it via the configuration there. So that's nice. And TypeScript configuration got more simple because of that.

• 3:53Scott Tolinski?

  And then the service workers apparently got a facelift as well. Now, the service workers I've used quite a bit in SvelteKit. And, um, we use one on the syntax.fm website and everything like that. So stoked to see that they got a facelift.

• 4:12Scott Tolinski?

  Cool stuff. I haven't used it yet.

• 4:17Scott Tolinski?

  And then also ENV variables. There's interesting stuff going on with there. So you can see that like, uh, envts lets you specify a little bit more. And then you can also use standard schema to validate your ENV vars and things like that. So ENV var is one of those things that I tend to use var lock for because just straight up dot ENV sucks. It sucks for a lot of reasons. So I'm really stoked to see that these SvelteKit folks are going to be taking this into account with their explicit ENV vars feature. And then also, check this out.

• 4:59Scott Tolinski?

  Shallow routing now happens with go to, okay? And it will, I believe, validate your routes existing. Go to instead of, do you have to use resolve? Paolo, let me know. Do you have to use resolve still with the go to? Resolve is a function that validates all of your routes and stuff like that. Error handling got better, apparently. I have not used this yet. And then also, uh, what's the, um, this is the big one to me. They put the big one down at the bottom here. Remote functions, uh, which have long since been an experimental feature and they still are an experimental feature.

• 5:43Speaker 00

  I want to see these go stable. I mean, I've been using them, but.

• 5:47Scott Tolinski?

  I was really hoping, yeah, that SvelteKit V3 would move these to stable. But if you know me, I've been using these in production for like a year now. So, uh, they're stable enough.

• 6:00Speaker 00

  Yeah. I mean, I haven't had any issues with them. I'm sure it's like lower level stuff or like weird edge cases.

• 6:04Scott Tolinski?

  Yeah. SvelteKit, uh, remote functions, for those of you don't know, it's like a true, kind of like, almost like an RPC style of working. You can just use functions. It makes a lot more sense than just the, uh, route based load function kind of ways of doing things.

• 6:23Scott Tolinski?

  Paolo says they are already pretty stable. Yes, in my experience, they have been as well. And other than that, just v8 and, uh, using roll down, all those types of things that go on behind the scenes. But I love a little cleanup. This is like a little bit of a snow leopard for SvelteKit in a little bit of ways. I don't know if there's any perf stuff going on here, but just niceties and cleanup. So shout out to the Svelte team for always improving SvelteKit. It's my favorite framework for a reason and it's going to continue to be. So, um, especially with the dang, uh, remote functions. Dang.

• 7:01Speaker 00

  Yeah. Yeah, those are dope. Yeah, happy to see it. All of this is cleanup stuff, right? Better ENVs, better imports, no config file. Love it.

• 7:11Scott Tolinski?

  Yeah. Yeah. And then the docs are over right next, .svelte.dev. If you want to take a look at the document. Oh, that's me.

• 7:19Speaker 00

  Oh, look at that. That's your, uh, is that the Svelte documentary on their homepage?

• 7:22Scott Tolinski?

  Yes, I'm in the Svelte documentary. Yes.

• 7:24Speaker 00

  Nice.

• 7:25Scott Tolinski?

  So yes, it's always nice going to the Svelte homepage and seeing yourself.

• 7:31Scott Tolinski?

  Sick. That's all I got here. Nice and quick one.

• 7:34Speaker 00

  Sweet. If you're just joining us, this is Syntax Weekly. We did, I don't think we mentioned it, but Wes is out today, so it's just me and, me and Scott.

• 7:43Speaker 00

  And before we get into our next stories, there's some chats that I'll call out. Taylor in the chat is asking, do you have any good video editor references or recommendations? I'm going to recommend Casey Faris on YouTube, if you have not heard of them, because they do DaVinci Resolve tutorials. So if you're editing video these days, DaVinci Resolve is really the way to go. It's, they have a totally free version. You can buy the studio version for a one-time purchase. And a lot of people have migrated off of Premiere Pro and, uh, Fusion, not Fusion. What's the final cut? Yeah, Final Cut. Migrated off of that to DaVinci Resolve and DaVinci Resolve Fusion. Cause DaVinci Resolve has all of that built in.

• 8:26Speaker 00

  It has like, uh, like audio, a whole audio section, a whole coloring section, a whole like graphic, motion graphic section. But Casey Faris, uh, really great videos. He has like a whole three hour course on DaVinci Resolve. A lot of the stuff I know about DaVinci, I learned from her. Do you have any video editing resources, Scott?

• 8:44Scott Tolinski?

  No, you know, I, I've been editing videos since I was like 12 or younger. And, uh, I was on Final Cut for a very long time. Then I was on Premiere for a bit and then I went back to Final Cut when Final Cut X came out. I was like the type of person who actually liked Final Cut X, even though I know a lot of people didn't. And, uh, I moved to Resolve. I want to say it feels like 10 years ago at this point. I moved to Resolve a long ass time ago. And I'm a huge, huge fan. I, you know, there's always things about it, but like, man, it's really sick and it just keeps getting better. There's some really cool stuff in Resolve. I actually, uh, Randy would be, like to know that I successfully used Fusion in the last edit to do like censoring and blocking out and stuff.

• 9:28Scott Tolinski?

  And I was pretty proud of myself to get into some of that node based editing and compositing a little bit there. But no, Resolve is where it's at. It's sick. And then there's a nice little MCP too, that you can communicate with your, your agents for sometimes or that type of thing too.

• 9:45Speaker 00

  Definitely. Yeah. DaVinci Resolve is the way to go. I too have been editing videos a long time. I used to use Windows Movie Maker for my little skate videos.

• 9:53Scott Tolinski?

  I used Windows Movie Maker for skate videos as well. Yes.

• 9:56Speaker 00

  Nice.

• 9:56Scott Tolinski?

  Yeah.

• 9:57Speaker 00

  Love it. Cool. All right. So let's get into our next story. And I'm talking about TanStack charts. So you might be familiar with TanStack Router or TanStack Query or many of the other things. If you head to tanstack.com, you can see that they have a lot of different libraries. They also have TanStack DB, which I think we talked about a while back, um, which is like reactive client first store for your API. It's a whole way of like structuring your API so that like your, the client does, does most of the querying. Really cool. They also have TanStack table. All of these libraries, they're, they're doing a lot of work in the space to make our lives easier, but they have come out with TanStack charts. Now, this isn't super new. It came out like at least a couple of weeks ago.

• 10:41Speaker 00

  It's still early alpha, uh, but it's a chart library that you can drop into your app and they have all kinds of charts. So they have donut charts and gauges and, um, uh, pie charts and line charts. And you can overlay charts on other charts and all, all the charts that you would need, they have a ton of them. And lots of really good examples. What's that?

• 11:02Scott Tolinski?

  I said, say charts.

• 11:04Speaker 00

  Charts, charts, charts. One of the cool things on their homepage is they have a tagline that says, make it look like your product. And so it says, same data and scales, CSS variables, themes, mark props, custom tooltips, your own render, or control the rest. That actually sounds like it was written by AI. But regardless, the screenshot shows the same chart with a different theme depending on, you know, whatever site you're running in. So highly customizable. And Tanner just announced a couple of days ago that he implemented all of the ShadCN chart catalog.

• 11:39Scott Tolinski?

  So Cool.

• 11:39Speaker 00

  Yeah. If you're familiar with ShadCN, they are like a UI library for React, but they also have ShadCN charts. Well, all of those charts have now been implemented in Tanstack charts. So if you do spin up a project, all of the charts that you're familiar with are already there for you to use, as well as a lot of extra stuff. The cool thing about this is it supports all the frameworks. You can see the dropdown menu. We've got React, Pre-Act View, Angular, Solids, Felt, and many of Vanilla and many others. So that's the cool thing about this versus something like ShadCN, which is very like React centric. You can, you can use this with pretty much any framework. And so, yeah, check it out. You can head to the examples page and they have all kinds of different charts you can dial into. You can get the example code to start from.

• 12:24Speaker 00

  So I thought this was slick. I've used a lot of chart libraries in my day and a lot of them are missing some things or don't work the way that I want them to. This seems to have a lot and it's just in pre-alpha, so I'll be keeping an eye on it.

• 12:38Scott Tolinski?

  I was trying to see a little bit of like what this is based on, because a lot of charting libraries aren't just like built from scratch. And I did find in the charts core package, a lot of D3 stuff, D3 array, D3 force, geo, hex bit. So it does seem like it is D3 based. And anybody who might be wondering like, why not just use D3? D3 sucks to use and having a wrapper around it is way better. So, um, yeah, I love a good charting library. I've been doing like charting library tutorials for a long time. So having Tanstack do a charts that's a D3 wrapper. It appears if, if someone out there works on Tanstack and wants to correct me, it does seem like it is D3 based. That's great for me.

• 13:23Scott Tolinski?

  Yeah, I'm very happy with this because there's, and I will say, there are some really wonderful Svelte charting libraries out there, but there isn't as much in terms of like fully, fully, fully chart libraries like D3 straight up. So yeah, cool about this. Yeah, very, very stoked on this.

• 13:44Speaker 00

  Yeah. And no, no issue with being based on D3. I mean, I think that gives you a lot of the fundamentals without having to do everything from scratch. And plus, if that is the case, there's probably a lot of D3 style visualizations that you could throw an AI, AI at to kind of adapt to use Tanstack charts instead.

• 14:01Scott Tolinski?

  So Yeah.

• 14:03Speaker 00

  Yeah, good stuff. So yeah, that's all I got. Check it out, try it out. What do you have for us, Scott?

• 14:08Scott Tolinski?

  Yeah, I have something that's fun. And if you're just joining us, once again, this is Syntax Weekly. We come at you every single week. Usually we got another guy here who's a, who's talking about some stuff, but he's not, he's not here today. So we're talking about whatever the heck we want to talk about. And so I'm bringing up the next one on Tap, which is called Ponytail. Ponytail has the best logo of all time. It is a ponytail old developer guy. And Ponytail, he says, he says nothing. He writes one line, "It works." Ponytail is supposed to be the AI kind of script to, it's a rule set to make your AI

• 14:51Scott Tolinski?

  coding agent write the least amount of code that works like a senior dev who's been paged at 3:00 AM one too many times. I think this is like the best branding and pitch ever. I have not used this. It's, uh, it's, it's new. I've heard good things. I will say, people have said good things about this thing, but I have not validated it. But I do want to just put this in front of you. Your agent reaches for 50 lines. The job needs one. And like, honestly, I love this type of thing. I'll have another thing in this, uh, in this episode of Syntax Live to talk a little bit about this same kind of concept here. We've seen many tools, whether that's cavemen or whatever, for reducing token costs.

• 15:35Scott Tolinski?

  This seems to be just like coming up with elegant solutions in your code, the way that an old school senior dev would. Again, there's audit, det, gain, help. I don't know how much of this is deterministic static code analysis stuff. I don't know how much of it is AI vibes or just prompting things, but this is the type of tool that I'm going to keep my eye on. And it just popped up and it's been one of those ones. I've had the tab open for a little while. So now the tab is open for you. And if you're out there, you use this thing, let me know how it works for you because I am super curious.

• 16:10Speaker 00

  Yeah. And it's mentioned in the chat, like I saw this when it came out. I don't think they had a landing page when it came out. But again, I think of like cavemen, these things that try to reduce the token counts. I have never reached for a tool like that. I've been pretty happy with, um, I'm using open code with a plugin that like prunes context and that, that keeps my token usage down. But it seems like this is the thing people reach for when they want to, when they want to keep their tokens down, want to

• 16:34Scott Tolinski?

  I think it's just keeping code down because you know, like, you know, we don't need infinite amount of like is records and extraneous solutions. And I have things that are not just for keeping your, your tokens down for the sake of reducing context, but keeping your tokens down or the responses down for the sake of like clarity and like maintainability and stuff. So either way, Ponytail, some people are saying this is a project that's been around. I'm the same guy that brought the joke tune thing to the show a few weeks ago and like it was new or something. So I occasionally am, I'm late on stuff, which is actually shocking to say that out loud.

• 17:13Speaker 00

  And that's okay, because I bet 50% of the people listening or watching to this hadn't heard of Ponytail. So we're here to, here to talk about it. So

• 17:19Scott Tolinski?

  Now it's in front of you. Yes.

• 17:21Speaker 00

  Yes.

• 17:22Scott Tolinski?

  CJ, what else you got?

• 17:23Speaker 00

  Yeah. So this is a new editor from the Zed team. And have you heard of this, Scott? It's Zed Delta?

• 17:31Scott Tolinski?

  Have I heard of it? I've been using it for the past

• 17:34Speaker 00

  You've been using it? Okay.

• 17:35Scott Tolinski?

  Two months, maybe. Uh, I got on like version like five of Delta, like point zero five of Delta. Yeah. Yeah. And then I got to give them feedback on it and, and, you know, work with it a little bit and I got a little private demo. So Delta is really sick. There's two things here. There's Delta, the product, and then Delta DB, both of which are sick as hell. Yeah.

• 17:57Speaker 00

  Yeah. The thing that really caught my eye on this was that there's a, the space in, in AI of collaboration. So like, let's say you're working on a team of people and you're using AI to build your product. It kind of seems like most developers all have their own little workflow. Like they have all their skills, they have all their, their setup for the things that they're doing. And really the main way people collaborate is like in pull requests. But that means that like some developers on your team may not be using the same types of skills and same types of workflows as you. But what caught my eye about Delta is you basically can create a shared thread that has multiple people in it that can, for one, see the AI working in real time.

• 18:40Speaker 00

  So I think this is also trying to be an answer to just the daunting amount of changes that AI makes and how you review those. And so this kind of puts it into a thread. So you can see those changes, but it's multiplayer. So multiple people can be in that thread and prompt the AI, ask for feedback. And so I definitely see this as something that teams would gravitate towards because that, that seems like a problem space that doesn't get a lot of attention is like, how do we create more reproducible, consistent workflows with AI and work on a team together? So that's the thing that caught my eye is this like multiplayer aspect.

• 19:19Scott Tolinski?

  Yes. It's, it's a, you know, when, when they first showed this to me, it was funny. It's right after I just had a long conversation about paired prompting with some folks at JS Nation in Amsterdam. So it was right after that. And when, when I first got into Delta, it was like, oh, this really just

• 19:38Scott Tolinski?

  identifies exactly that thing and distills it down into the right primitive, I think. Delta is really, really sick. And what's cool about it, too, is that you're getting like line by line diffs on which model was used for each line of code. There's some really, really cool stuff in here. You can go back and annotate things in the chat at any given point. So like if CJ chatted something and our response was weird, I could come in later, highlight that response and then annotate it and then like come back to it directly in the chat itself. There's just so many cool things about it. I was gonna pull up my Delta.

• 20:22Scott Tolinski?

  My version is outdated, though, so I don't know if they want me to show it on screen being outdated, but it's really cool. You can follow people, too, like in Figma. So I can like click on CJ and just like follow and watch and then give him feedback. Or we could discuss over like Cam, the different types of prompting that we're going to be doing. And then it's just really, really like, they took the idea of like, what if an AI chat and the evolution of code was rethought in such a way? So I'm a tremendous fan of this project and I can't wait for more people to use it. I think it's really sick.

• 21:03Speaker 00

  Yeah. And so last week, they just announced the private beta. So if you want to try it, signups are now available. I signed up yesterday, haven't got my invite yet. But they're releasing it for more people to try and give your feedback on. But like Scott mentioned, it's built on top of Delta DB. And what they were mentioning in their launch post is it essentially works off of like Git diffs. So like, even if there are people on your team that aren't using Delta or aren't using Delta DB, they're just using Git in the normal way. They can still collaborate, but Delta DB is what gives you all of these extra features that you get in the the Delta UI in terms of like threads and seeing like what model was used and different stuff like that. So yeah, exciting project.

• 21:48Speaker 00

  I'm excited to try it. I've tried out Zed before, but this would be one more reason to to give Zed a go as well.

• 21:54Scott Tolinski?

  Yes, it's really cool. And the Delta DB stuff, I saw some people, because we had Nathan on from Zed on Syntax very recently, go back and watch that episode. That was actually the day that Zed was dropped, or not Zed, Delta was dropped. So it was really great to talk to him about that. But what's really cool about the Delta DB stuff is some of the comments were like, oh, but now I have to buy into version control at like an app level or something like that. And you don't really, it's compatible with Git. You don't have to, you don't have, like, somebody can work in a repo, it just use Git and someone else can use Delta or whatever, and they can coexist just fine without goofing anything up. So really cool stuff.

• 22:37Speaker 00

  Nice. Yeah, so that's all I got for Delta. Do we want to get into this Stripe story?

• 22:42Scott Tolinski?

  Yeah, let me pull up, I don't want to pull up Yahoo here. I don't have anything to show on screen.

• 22:48Speaker 00

  There's a TechCrunch article, if you want to pull that one up.

• 22:51Scott Tolinski?

  TechCrunch, yes, I was just like, I don't want to.

• 22:53Speaker 00

  Yahoo Finance.

• 22:55Scott Tolinski?

  Oh, disable my ad blocker? No, TechCrunch.

• 22:57Speaker 00

  I think honestly Yahoo Finance or Tech Times might be fine. It's financial news because it's like acquisition. I don't know.

• 23:04Scott Tolinski?

  Folks, we're about to show Yahoo on stream. If you're sensitive to Yahoo, please turn away. Yahoo is about to be landing on screen in just a moment here. Fair warning. I actually, CJ, I use Yahoo for fantasy football, believe it or not. And it sucks so bad. I'm on a, I'm in a fantasy football league with a bunch of other tech people like Typecraft and some other folks. And believe it or not, I won the first season of fantasy football by only drafting Detroit Lions. That's right.

• 23:36Speaker 00

  Wow.

• 23:37Scott Tolinski?

  And yes, Feta said I use Yahoo on a weekly basis and it's going to start happening again soon. I'm not looking forward to it. Either way, Stripe acquires OpenRouter for $7 billion plus, turning model routing into a payments infrastructure problem. What is this AIS headline?

• 23:56Speaker 00

  Is it, though? Yeah, okay. Yeah, that's because that's my question, which is like, why would a payments company acquire a model routing company? There's probably reasons we can hypothesize, but I don't know if there's anything in the announcement that says that they're going to turn it into a payment gateway. Is that real?

• 24:13Scott Tolinski?

  No, you know what? I think Stripe is a smart company. I think they are always, in my mind, Stripe is one of those companies that does really interesting things, whether that is Stripe, what's it called? Stripe Press doing like really great books and stuff. They have a number of acquisitions. Most of them are, you know, finance based, whether that's like, they acquired Lemon Squeezy. Apparently they acquired Lemon Squeezy, which is a payment, payment, recurring payment platform. And they did try to acquire eBay at some point. Do you remember that, CJ?

• 24:58Speaker 00

  Stripe?

• 24:59Scott Tolinski?

  They put an offer for eBay.

• 25:01Speaker 00

  I didn't hear about that news. This is not a finance podcast, by the way. Like, I'm not up to date on all this stuff.

• 25:06Scott Tolinski?

  Yes, breaking news. Stripe tried to offer, oh, no, Stripe, not eBay. What PayPal? Stripe Advent offered to buy PayPal. Was it PayPal? I thought it was eBay.

• 25:20Speaker 00

  There was a chat that someone, someone was saying they they are in the talks to potentially purchase PayPal, which would be crazy because then they'd be all the, all the payments on the web.

• 25:30Scott Tolinski?

  Right, more than 40 or $53 billion. So not eBay. I thought it was something with eBay. My God, I am out to lunch on this stuff. Either way, OpenRouter, they host models. You can subscribe to OpenRouter, you can put money in your wallet there. And then what happens is is that you can choose the models that you're using, you get access to models from everybody. And that's, that's OpenRouter, folks. And OpenRouter is a sick product. We use it here at Sentry. I've used it here personally. I'm a big fan of OpenRouter. So to me, even though it's like, okay, how does payment processor fit with model routing or, you know, buying inference?

• 26:12Scott Tolinski?

  It doesn't really feel that weird to me for some reason. And I cannot put my finger on why it doesn't feel weird, but it doesn't feel weird. Just because like, yeah.

• 26:23Speaker 00

  Well, I was gonna say, like, if you think about like what Stripe did for the web, it's like all of these various types of payments they put into a single platform, right? Like you can add Stripe and now you got bank payments, you've got credit cards, you've got whatever else. And so they've, they've become the central place for payments. And that's kind of what OpenRouter is. It's the central place for AI models. And so maybe they're also trying to make sure that they can stick around in this new AI era. Like, I don't know why payments would go away or if some company could come around to take Stripe down, but maybe they're also differentiating like what they did for payments, they'd like to also make sure they're doing for AI models and access to AI, maybe? So.

• 27:04Scott Tolinski?

  Yeah, I don't know. You know what? By the way, chat says that GameStop was going to buy eBay. I knew it was somebody weird. GameStop was going to buy eBay. Wow. So also, any alternatives you'd recommend from Kevin? I would recommend OpenRouter. Honestly, I don't, I don't, this acquisition like doesn't hurt my opinion of OpenRouter in the least bit. I like Stripe a lot. I don't usually have any issues with Stripe. So for me, the Stripe acquisition here isn't like, isn't one of those things where I'm like, ah, shit, Adobe bought it. Now I gotta get off of it, you know?

• 27:43Speaker 00

  Yeah. Okay, so again, this isn't a finance podcast, but I am actually finding news that last month Stripe made a $53 billion offer to buy PayPal. So that was a thing. That was a thing. Okay.

• 27:57Scott Tolinski?

  Yes.

• 27:58Speaker 00

  I thought we replaced it saying it was GameStop.

• 28:00Scott Tolinski?

  No, GameStop tried to buy eBay.

• 28:03Speaker 00

  Oh, by eBay.

• 28:04Scott Tolinski?

  Put an offer for PayPal and PayPal.

• 28:07Speaker 00

  I'm following.

• 28:07Scott Tolinski?

  Wasn't PayPal part of eBay or am I losing my mind? No, PayPal?

• 28:12Speaker 00

  Yes, it was. Yeah, it was. Early on PayPal came out of eBay. Like, that's the online payments for buying eBay stuff. Yes.

• 28:20Scott Tolinski?

  Folks, come to us for your Business Insider news. I think we could give you a better headline than turning model routing into a payments infrastructure problem.

• 28:29Speaker 00

  Yeah, I don't, which I don't think is going to happen, but you know.

• 28:34Scott Tolinski?

  Oh, did you, did you see that? Was this, this Wingstop Inc. is down? So the Wingstop holders, your stocks are down. Just a shout out to, you ever know Wingstop?

• 28:46Speaker 00

  What is Wingstop? Oh, the food.

• 28:48Scott Tolinski?

  It's like hot wings. They have them in Colorado and I never seen them before I moved here. And I was like, what the hell is Wingstop? I've never been there, so.

• 28:55Speaker 00

  I've been before, but I am vegan now, so I no longer eat chicken. So.

• 28:58Scott Tolinski?

  Mmm.

• 29:00Speaker 00

  Yes.

• 29:00Scott Tolinski?

  Mm hmm. Yes, that makes sense. Okay, well, that's all I have here. Folks, I don't, I don't think this is going to hurt OpenRouter that much.

• 29:08Speaker 00

  No.

• 29:09Scott Tolinski?

  No, I think I'm like perfectly, in fact, you know, I feel like this might give them, not that they didn't have juice to stick around, but like, oh, OpenRouter was like fairly new, you know? So Stripe, OpenRouter, it's a marriage.

• 29:24Speaker 00

  There we go.

• 29:27Speaker 00

  All right, this next story I got for you is actually a research paper, but there's a couple of things I want to show. And I'd love to get your thoughts on this too, Scott, because the first thing that comes to my mind is just how cool this site looks like. Usually when a research paper comes out, it's just a blank, black and white page, or maybe you just go directly to the archive page. But they have, they bought a domain, stolen-thoughts.com. They have what I think is nice color choices. And then a really cool layout for looking at the results of their research paper. And then all of these like little reasoning blocks in the background, when you click on them, they kind of like explode on the page.

• 30:12Speaker 00

  So yeah, initial design thoughts before we get into what this paper actually is. What do you think, Scott?

• 30:17Scott Tolinski?

  Well, yeah, I mean, it reads AI to me, but it also does look nice. And I, listen, I will take this over some, oh, man, even like PubMed looks like ass to me. So I'll take this over anything like that, any day of the week, even if it does look AI-generated. So no, I think it's cool. No, and I appreciate this kind of stuff. I'm a visual person. I am dyslexic, so reading large amounts of text is hard for me, having pictures and colors and organization. It makes it easier for me like a child, so I am happy to see this.

• 30:54Speaker 00

  Oh, it's got all these like little design flourishes. Like there's like these, the cards are like slightly askew, like slightly rotated. And then there's some top borders that are like triangles that extend. And then just like a really soft color palette. I don't know, I thought it was cool. But the paper is called stealing reasoning traces from proprietary LLM APIs. So if you didn't know this, the reasoning response that comes back from, let's say like a frontier model like Opus 4.8 is actually encrypted. And the reasoning trace is where a lot of the work is done. And so reasoning models came out. And essentially, instead of a model instantly responding to your prompt,

• 31:38Speaker 00

  it actually creates this reasoning chain where it's thinking about how it will respond to your prompt before it actually shows you the response. And that reasoning chain is kind of the secret sauce because that reasoning chain can be used to fine tune other models and essentially kind of steal the secret sauce of some of these frontier models. But I actually didn't know this. When you're interacting with one of these frontier models, that reasoning chain is hidden from you. You can't actually, you can't read the reasoning chain. It's encrypted and then it has a signature on it. But what this paper showed is they can actually

• 32:18Speaker 00

  take the reasoning signature from a bigger model like Opus 4.8 and then craft an API payload that talks to a dumber model like Haiku 4.5, include that same reasoning signature, but just give it a prompt that says, hey, tell me what's inside of the thinking block. And the dumber model doesn't have the guardrails to prevent it from showing the reasoning trace. So it just like immediately dumps the reasoning trace. And so they showed this with a lot of different things. You basically get the reasoning trace from Opus 4.8, use Haiku 4.5 to actually read that reasoning trace out, and then did the same thing with like OpenAI models. So for instance, get a reasoning, an encrypted reasoning trace from GPT-5.6 Sol

• 33:03Speaker 00

  and then ask Luna, which is their smaller model, to tell you what the reasoning trace is. And then same thing with Gemini, use Gemini 3.5, pass that encrypted signature to something like Gemini 3 Flash. And the reason this is interesting is because these reasoning traces, again, like I mentioned, can be used to fine tune other models. And this is kind of potentially what we're seeing happening with some of the bigger open weight models. They're potentially using some of these reasoning traces from these larger models. So this paper shows also how they were able to like extract secret information from the reasoning trace that usually you wouldn't be able to get access to. And they kind of show how they did it with all of the different models.

• 33:45Speaker 00

  But with, another thing this paper shows is how similar responses are and how you can actually look at the response of a model and already kind of determine, or look at the reasoning trace and determining, determine which model that came from. So on the page, they actually have a game where they show you three different decoded reasoning traces. And you have to choose which of these you think is the given model. So this one is saying which one of these reasoning traces belongs to Claude Opus 4.8. And so the first one is, we need to solve problem. We need provide explanation and code, likely C++. I don't know about you, but I feel like I haven't seen Opus 4.8 use like short words like this.

• 34:25Speaker 00

  That's usually like OpenAI or ChatGPT that like really adds short words in its reasoning. This next one is, we need to understand the problem. We have in workers, each work, that could be Opus.

• 34:35Scott Tolinski?

  Yeah, it's kind of annoying like that.

• 34:39Speaker 00

  We need, we need to pick a subset of workers such that for every chosen worker, I think it could be B or C, I'm gonna go with B. It was C, so C was Opus 4.8. And then actually, that first one was from Kimmy K through. So you can, you can play a bunch of rounds, reading the reasoning chains and trying to determine what, what model actually created that reasoning chain. So yeah, I thought it was a fascinating paper, also really cool website.

• 35:02Scott Tolinski?

  Yes. CJ, we're getting a lot of chat, a lot of chatter about Grok Hub, which is what I'm calling it now. It's not actually Grok Hub. In fact, let me even share my screen. I, hot off the presses, CJ. Origin Code Hosting, which is Cursor Code Hosting, is here. Let me share my screen. And this is apparently Code Hosting from Cursor. So Origin begins, this had just happened moments ago. Origin begins rolling out today in early beta on all paid plans. We're starting with the Essential Design Agent Scales,

• 35:46Scott Tolinski?

  Pull Requests, Code Browsing, and GitHub Sync. So it looks like they are coming for GitHub here, which is basically where you can host your code. You can automatically bring in your GitHub repos. There's Pull Requests, yes, Pull Requests, Agents in every repo, App Extensions from the Cursor Repo, Settings. Okay, yeah, you know, thoughts on this? I mean, it's missing a lot from GitHub, obviously. It's a subset of GitHub's features. And it's a good for them to launch this on the day that GitHub is out for so long. GitHub outages have continued to go.

• 36:29Scott Tolinski?

  So I think that is the big sell, is that it's going to be, you know, it's gonna be that thing that can hopefully be have more uptime. But like, I use a lot of stuff on GitHub that I don't know if this is gonna be able to do. So we'll see, yeah, we'll see. I mean, I have it installed right here. I just signed up for it. This is what the dang thing looks like. And we're gonna see if this is actually any good.

• 36:55Speaker 00

  Yeah, there's a couple of things my mind goes to. So first of all, Cursor got bought by Grok, right?

• 37:00Scott Tolinski?

  Yeah, which is why I'm calling this Grok Hub.

• 37:02Speaker 00

  Okay. And we already know there were secret leaks, not leaks, but like the, was it the Grok code would just like upload your entire repo to an S3 bucket or something like that? And then they disabled. So here's the thing, we've already given all of our code to GitHub. And we somewhat, I was gonna say, we trust that they're gonna not train on our private repos, but we don't know that they're not doing that because they did train on a lot of code that's on GitHub. So that's, that's where my mind goes to is, okay, instead of giving all of your code to Microsoft, now give it to X instead.

• 37:38Scott Tolinski?

  Who's already kind of shown that they are not to be super trustworthy in that department. But hey, who is?

• 37:43Speaker 00

  But at the same time, like they have the Cursor IDE, which was like one of the first to be really good in this space. Now it has a lot of competitors and not necessarily everyone is using Cursor. But I don't know if you remember, maybe a year and a half, two years ago, Cursor IDE was the thing that everyone was using to write code with AI. So I think this is also them trying to like reposition and find, find spots where, or find things that can make it more enticing to keep using Cursor.

• 38:11Scott Tolinski?

  Yes. Cursor Origin, folks, if you're watching, I found a bug. This toast message does not go away when you click it. There's an X, doesn't go away.

• 38:19Speaker 00

  Can you say vibe-coded?

• 38:23Scott Tolinski?

  Yeah, I mean, hey, we need some, uh, we need some shakeup here. Obviously GitHub has been having some issues. And this product did just launch moments ago, so it like, there's some latency and stuff here, whatever. I'm willing to give this a shot and see how it goes. But I can't see myself lining up to use this, if I'm being honest. I just, maybe my own personal feelings getting in the way here. But I just can't see myself being like, yeah, let me just throw all my, my stuff on, uh, Cursor Origin. And I don't use Cursor, that's part of it, too. It's like, I don't, I don't use Cursor.

• 39:06Scott Tolinski?

  And so for me, it's like, okay, uh, is this like, that's the, the funnel is to try to get me to use all of their stuff? Cursor, Cursor Origin and all that, when like, just not my favorite. Um, and for Luke says, you may feel different in the next time GitHub goes down. Well, uh, this thing is currently struggling because GitHub down is down, it says. The performance is degraded because GitHub is down, so, um, we'll see. Yeah, I don't know. I, I would prefer, I don't know, I think there's like, who's the one? Pierre Computer has a, a solution here. Um, there, there's all kinds of different, different solutions here. But this does kind of feel like funnel to Cursor to me, and I just, I don't, like, I don't see the, like, the, uh, the pull there.

• 39:50Speaker 00

  Yeah. Looking at the replies, it looks like GitHub is down right now for some people, which is, I think that they timed this.

• 39:58Scott Tolinski?

  Been down all morning, yeah.

• 39:59Speaker 00

  Okay, okay. I don't even know, because I'm doing other things. Um, I, I can do my.

• 40:03Scott Tolinski?

  Trying to look at GitHub issues about another project that was broken and it was downloading to me.

• 40:07Speaker 00

  So yeah, I can do my work without GitHub a lot of times. But I think they, they tried to like meme time it, I think, like GitHub is down, we're just gonna release this and kind of get in on the bandwagon of GitHub being down, which I wish they would put more thought into. I wish the, you could close the, the popups, the, what, was it the little toast?

• 40:26Scott Tolinski?

  Yeah, that's just a little thing, you know, that's a, that's.

• 40:28Speaker 00

  But it's, it's a sign of a bigger systemic issue, right? Like, it's like pushing code that hasn't been tested and, yeah, I don't know.

• 40:36Scott Tolinski?

  What I wish is that like, this is like a little bit more forward thinking. I wish we would think beyond this kind of, uh, ChatGPT as chat bubble, whatever, and like new chat. This, this generic ChatGPT, uh, layout that we've all kind of just accepted as the thing. And that's the one thing I really liked about Zed's Delta, is that it doesn't feel like the same thing just done by a different company. So like, it takes a different approach, even though it's obviously apples to oranges here.

• 41:13Scott Tolinski?

  Yeah, hey, this is, this is new, hey, and, uh, you know, shout out to somebody to push GitHub, because they clearly need it. GitHub has been, uh, struggling for months now with the amount of code. And honestly, it might do, if this truly like gets people off of GitHub, um, what it might do is relieve some of that pressure on GitHub servers, so that it stops going down for me, you know? Yeah, I'll take that, yeah.

• 41:36Speaker 00

  Yeah. I mean, you could also put your tinfoil cap on and be like, okay, what if Cursor is generating bot traffic? Which is bringing GitHub down, which is forcing people to use it.

• 41:46Scott Tolinski?

  CJ, get really tinfoily of the, uh, see, I'm not even willing to go there and I'm not, I'm not a fan. So, well, we'll see, yeah, I, I'm excited for Pierre, I'm excited for anything that gives some, uh, competition here. I, I was a big user of GitLab for a long time, so again, I'm down for competition in this department. But again, like, if, if it's gonna replace GitHub for me, it's gotta do all the stuff GitHub does. It just can't have an AI chat on top of GitHub.

• 42:14Speaker 00

  Yeah.

• 42:14Scott Tolinski?

  And there's automations, cool, but like, the, the GitHub stuff is pretty dang good there, so, hey. Uptime is the big sell, as far as I'm concerned, and I don't know if that's right now. Like, that seems ridiculous to say. I don't know if that's right now inhibiting me enough in my day-to-day life to make me switch to something completely different.

• 42:34Speaker 00

  Right. Nice. Okay, so I, I have a CSS story. Um, and this, this, this blog post was actually published at the beginning of the month, but just came on my radar. But it's a blog post by Lea Varu, and the title is "Dark Mode Toggles: Two States Are Enough." And I've actually thought about this when I was implementing dark mode toggles. And they make a really good case for just having two states instead of three.

• 42:57Scott Tolinski?

  Yes.

• 42:58Speaker 00

  But the thing you might be familiar with is like, when you implement a dark mode toggle on your site, a lot of times you'll have three modes. You'll have system, which is use whatever the user prefers, whether it's dark or light. Then there's an option for dark, which is forced dark mode, and then a light option, which is forced light mode. And if, if the user chooses dark or light, it forces that, but if they choose system, it follows whatever their current system theme is. And, I mean, I kind of preferred that because sometimes I change my system theme and like, I want sites to follow along. But it's also not hard to just go back and toggle a site to light if I want it to be light. But this, this blog post makes a really good case for the user experience of it all, like,

• 43:39Speaker 00

  I'm more of a power user, like, I think about the different states and, and think about how sometimes I do want system versus like an exact mode. But they kind of get into the idea that for end users, non-technical users, or people don't care about this, there's a lot of user experience friction when you have three different toggles. And for some people, it's a little bit confusing as to what those three different things do. And so they make a really good case for just have two states, dark mode and light mode. When a user visits your site, default to whatever their system theme is and show that as the selected one. And then if they want to toggle it, they can toggle it. So essentially, you're now only keeping track of, uh, one piece of state, which is, has the user selected a preferred theme or not, otherwise just use their system theme?

• 44:23Speaker 00

  Which I like, and I think I'll start adopting. So had you seen this or thought about this, Scott?

• 44:28Scott Tolinski?

  I, I've thought about this a lot. I haven't seen this article specifically. But I've like more recently, and this is like a controversial opinion on my behalf. I more recently have the feeling of like, just use the system. Unless it's a theme, like, don't even give the option and just have it be a system toggle. It's very rare that I want everything to be dark and one thing to be light. Now, I recognize that other people want that control, but unless it's like a theme customization where you're choosing more than just lighter door, dark, it's like, I just, I just, I default to system. I like, I just want you to default to system. Well, it really bugs me when I arrive on a website and it's in lighter dark mode and it's not whatever my system's set to, and then I have to like go and manually change it.

• 45:13Scott Tolinski?

  Like, default should be system, absolutely. But I, I mean, I do agree with, I think this solves some of the concerns I had with that approach by not giving like users control or whatever. Um, so, hey, I think this is a good way to do it, if you're asking me.

• 45:29Speaker 00

  Yeah, and check it out, like, um, there's been a lot of converts after like reading this blog post. And there's an example at the bottom of how all of this works. So like, let's say a user arrives and their system theme is dark. When you, when they click toggle, it changes to light. If the OS switches to light, their preference is still kept because they clicked that button. And, um, if they switch their OS back to dark, it stays because they clicked that button. But then if they want to actually go back to dark mode, they would toggle it. So, yeah. I, I think it's simple enough, it, it makes it easier for end users and there's less state that you have to keep track of in your app as well, so.

• 46:04Scott Tolinski?

  Huh.

• 46:05Speaker 00

  Yeah.

• 46:06Scott Tolinski?

  Oh.

• 46:07Speaker 00

  Yeah.

• 46:10Scott Tolinski?

  Okay, sick. Uh, let me pull up my next one here, and I'm trying to think, I got several that I could do here, folks. I got, I got links all day I could get to. Here's a fun.

• 46:21Speaker 00

  Do, do all of yours back to back, but before that, do a Sentry read if you can, or work it into the, or do, or work it into them if you can.

• 46:27Scott Tolinski?

  CJ, would you describe this as a formal, uh, handoff here?

• 46:32Speaker 00

  Yes, we were a little bit, we broke the fourth wall, we're behind the scenes. You now know that I am instructing Scott to give you a Sentry ad read and do all of his stories.

• 46:40Scott Tolinski?

  Sentry, let me, let me tell you about Sentry, and then I'm coming up with this on my own accord here, not because, uh, I've been prompted to by CJ. And if CJ wants a good review, he will not be prompting me to do that again.

• 46:54Speaker 00

  Okay, noted.

• 46:57Scott Tolinski?

  Just joking, folks, I, uh, no, I need the prompting. Sentry, at Sentry.io, the perfect place to solve your bugs. And if we're talking about fixing bugs live, and then, and actually, I'll be talking about something that kind of fits in nicely with Sentry in just a little bit, is, uh, Sentry is really great for healing your software, because Sentry has a dope CLI. I almost said CI, because we've been talking so much about, uh, in the chat about CI/CD. Has a dope CLI for you to be able to connect to, to get, um, current bugs, to use their CER AI to be able to get to the root cause of a bug, because we all know those stack traces be looking ugly. And they're just sometimes when I want a dang thing to tell me, and I'll tell you about this in just one second, using plain language,

• 47:42Scott Tolinski?

  what the heck is going on? And I can get that information. So I love this stuff because, uh, Sentry is always having more tools. And another thing, I'm gonna connect it to another thing, Swamp Club, I'm talking about a little bit. I got Sentry Metrics connected to my Swamp Club, so that I can understand what's going on inside of my Swamp Club with my AI agents. And then I'm doing AI agent monitoring to understand how much all of my dang open router, uh, token cost is. I know Open Router has those features as well, but, uh, I'm using more than just Open Router, and I want to connect it all with, inside of my agents. So, check it out, Sentry.io, the perfect place to make sure your applications, your agents, your AI, any of that stuff is running smooth.

• 48:22Scott Tolinski?

  And it's the best. We love Sentry over here, and not just because syntax is presented by Sentry, but because we absolutely love them anyway. So, after that supernatural ad read, let's talk about plain language. I saw somebody talking about Cursor 5, and not just a one person, maybe people have been talking about Cursor 5 and how it's,

• 48:42Scott Tolinski?

  um, kind of being a weiner in the way it replies to you and being kind of annoying. And, um, I, I find that like, nine times out of 10, most AIs are being a weiner in the way that they reply to me. And it's actually very annoying when it gives you a whole, like, whole lot of words for something that it couldn't like, couldn't give you more simply, like, "The verification process encountered an unsuccessful outcome due to dependency unexpected in actual value." Like, who cares about like that? Tell me what the heck the problem is. So, uh, I saw some folks talking about different ISO standards. Apparently, these things exist for English, uh, they got some line height issues going on here, ISO.org.

• 49:27Scott Tolinski?

  Got some line height problems. Um, but ISO 24495-1:2023 plain language is an ISO standard for documenting governing principles and guidelines developing plain language documents. And there's a number of these different standards on how to speak plainly. And what I have gone ahead and done is added a one liner that is just says, always respond to the user in plain language using ISO and then this, this 24495-1:2023.

• 50:09Scott Tolinski?

  And it has made my life so much better. So instead of saying the verification process encountered an unsuccessful outcome due to dependency and expected actual values, it tells me the test failed because the actual values did not match the expected value. And the thing I like about this plain language standard is it does not remove technical details from things. So it's not obfuscating the issue, but it speaks to me in a way that I can actually like parse easily in less words. And it's not simplifying it. It's not removing that data, but it's using simplified language in a way that makes it all easier to say. And this again was a one liner. I'll paste it into the chat here.

• 50:53Scott Tolinski?

  It just so people can check it out and it has been effective for me. Like I put this in my agents dot M D and it's made my life. I, I have to, I have to respond to the AI, tell it to me in English or explain like on five way less because it's just telling me in simplified language and these kinds of standards exist. So, um, um, whether or not you're using this, this one in particular, 24495-1:2023 or not, uh, I think it's a really cool thing to do and it's better than like give me caveman speak or something, you know?

• 51:27Speaker 00

  Right. I, I actually saw this, uh, there was a thread on hacker, hacker news a couple of weeks back where they talked about this, but one, one user made a really good point. Sometimes plain language is too plain and so you might get worse responses if its reasoning chain is using plain language as well. So I think as long as you make sure that it's only responding in that way, its reasoning chain still does whatever it wants to do.

• 51:50Scott Tolinski?

  Yeah.

• 51:51Speaker 00

  Or you have like follow-up prompts, maybe you have just like a built-in plugin or skill that says rewrite this response in plain language.

• 51:57Scott Tolinski?

  Totally.

• 51:58Speaker 00

  Um, I think, so I haven't tried it yet, but I think I will. I'm building a, a home personal assistant right now with GPT five, six soul. And it is incomprehensible for things like, it actually, it was, it was running all weekend and it was just going off the rails. Like it's writing all these validators and like, I mean, that's good. If this were, if this were a production app, I'd be really happy, but like, I just want to prompt this thing already. And, uh, the, the language is crazy and I have to, it's so, so hard to parse. So I, I think I'll try this out to get it to tone it down a little bit, make it a little easier to understand.

• 52:30Scott Tolinski?

  I mean, my little thing says respond to the user using plain language. And so to me, I haven't noticed that causing a problem in the reasoning. Um, but then again, you know, I've been, I've had this on for about a week and a half now, and I have not seen any issues there. But you're so right about soul loves to be doing that stuff. I was working on my swamp, which we'll talk about in just a little bit here. Actually, maybe I'll talk about swamp next. I was, I was working in my swamp, a good little, uh, transition here. Um, I was working on my swamp and, uh, man, what happened? Oh, uh, it was like, I, I wanted to just use work trees and pie instead of, um, just stopping at one thing or branching all the time.

• 53:13Scott Tolinski?

  And it wrote a giant ass GitHub extension for Swamp Club instead of like basically doing all of the GitHub stuff. I was like, brother, this is just a pie thing. Like, I don't need you to, to, to write an extension. I didn't ask for that. It like, it comes up with the most asinine solutions to things sometimes where it's like, oh, user needs clearly to use, needs to use work trees and Swamp Club doesn't do it in this way. So let me write all that functionality. Like, come on, bro. Okay, Swamp Club. What is Swamp Club, folks? I just mentioned the word Swamp Club a whole bunch of times. Swamp Club at swamp-club.com is deterministic automation for AI agents. So the pitch here is really that if you ask an AI agent to do things multiple times,

• 53:56Scott Tolinski?

  regardless of if you have a CLI or scripts or whatever, it's going to take a different approach to it each time, or it's going to have to look up things or look for things or do whatever. The pitch for Swamp Club is that you build in deterministic code-based automations that run deterministically and are able to log what they've done in databases and stuff like that. So that way that your automations can respond to one another, they can pick things up. And a lot of people are using Swamp Club to do things like software factories or anything like that, or deploying like, oh, I, I have a deploy process that needs to run this and this and this and this and this, or a procurement process.

• 54:39Scott Tolinski?

  Like I want to procure a domain for this project and I want to get this all hosted on CloudFlare or whatever. Uh, you write all that in actual scripts. And then when I'm talking to my AI, it's like, ah, do this and then procure me a domain. Instead of it being like, oh, let me, let me figure that out and look up the CloudFlare CLI and do this or that, or the MCP, it's going to reach for its scripts, it's going to run those and it's going to do so much more effectively. I used Swamp Club to edit my wife's podcast using, uh, Python. And what it did is it analyzed, uh, both of our voice frequencies because it's a multi cam podcast. So the AI did not edit the podcast.

• 55:22Scott Tolinski?

  It used the two cameras and I said, this camera's lower voice, this camera's higher voice, and then do a multi cam cut between the two of them and then cut out the, the silence and then cut out the space after the introduction. And it did all of that with Python. And now since it's in my Swamp Club, it's always going to use that exact same automation to do it. Um, other ones, I've created a Swamp Club model for looking up analytics from my analytics dashboard because Swamp has a vault for your ENV vars. So like, if I have my analytics API key, that doesn't really make sense to go into the code base. It's just part of the automation. And when I'm working with the AI, like I wanted to look up the analytics.

• 56:05Scott Tolinski?

  So therefore it's always going to know to use the Swamp API key to go and do that. Um, it, it really rules. I'm a huge fan of this thing, swamp-club.com. And, uh, automations, there's a whole bunch of stuff in here. They have really great docs, obviously AI consumable docs. People have written all kinds of stuff. Their Discord is popping. And one of my favorite things about Swamp Club is this, they have, like they've gamified it where they have a leaderboard for who's spending tokens on what and who's publishing extensions. And then as you gain experience, you're, uh, increasing your account. So I'm a level 10 Merc Beast right now.

• 56:50Scott Tolinski?

  Um, I got about 50% more experience before I became a Blackwater Wraith, which I'm pretty stoked about. And, uh, I have a couple of published extensions, like one for logging my, uh, software factory metrics to Sentry at Sentry.io. Check it out. Um, as well as one for using my, uh, task tracker. So my Dex tax task tracker properly taking tasks and then moving them to the correct thing. And then again, for my plausible analytics. Man, uh, I think this is a delight to use. It's well thought out. The team behind this, this is not like some fly by night operation. They have like a, a really strong team of folks who have been, um, in the industry a very long time, specifically working in a infrastructure and, um, like things like

• 57:34Scott Tolinski?

  Terraform and that kind of stuff. So these folks are not like vibe coders who are just like, ooh, I wanna, I wanna do this. These are, these are folks who are used to building this kind of intense stuff. They are, they, yeah, yes. Um, if you recognize any of these words. In fact, I, I really want to get, um, Adam on the show on syntax to talk about Swamp Club because these guys are dialed. We just had a conversation last week and it was like really, really eye-opening to how they're using this thing. Um, yeah, this rules. Uh, nickname says it makes zero sense to me. Yes. Well, uh, think about it this way. Your AI just chooses at the point in which you tell it to do something, it's always

• 58:19Scott Tolinski?

  choosing in that moment, like what it's going to do. It needs to look up whether you have skills or scripts or whatever, it needs to look that stuff up. This just says, uh, procure, okay, here it goes. Let me run these scripts. So it's sick. And the extensions, the amount of like existing extensions are huge, obviously. And they have like a whole vetting process for them. Um, whether that's like all this AWS stuff. You hate AWS? Me too. Well, now you can, uh, Swamp Club it.

• 58:47Speaker 00

  Yeah, I think the simplest explanation for me would be take something that usually requires AI and really it would require AI on the first pass, but then every time after that, we don't have to use AI to do that thing because it figures out what are the endpoints that we call, what are the systems we need to integrate with, whatever else. And instead of being an AI reasoning chain of like doing all of that stuff, it turns it into just a script you can run that does all of those things back to back. So I think this is great. Yeah. Yeah. I do this kind of thing manually, but to have like a system around it to make sure that you're doing it more often. I think in the end, it's gonna like save you tokens, right? It's gonna save you AI usage because you're creating these reproducible workflows.

• 59:29Scott Tolinski?

  This is gonna save you a lot of tokens. Again, like somebody says, I just type into the box and what I needed to do. I mean, brother, you're wasting a lot there. Not that like, you know, we're all not like wasting tokens, but like you're, again, you're spending time telling AI to do this or do that, like AI CD into this directory for me, you know, like all of that stuff can be automated, any of the stuff that it's doing. In fact, I wanted to find a blog post in here because they do a really good job of writing blog posts in here and can't find it. It was from Adam and Adam basically took my boss, David

• 1:00:11Scott Tolinski?

  Kramer, his software factory loop thing that he was calling Garfield. He was like, I've made Garfield use like infinitely less tokens and it was really cool. I'll try to find this post while we're talking about something else, but the posts are really good in here. And the community, let me tell you, the community is sick around this thing. You can see they have 285 million automation events sent in the wild right now. So people are using this. Again, this is not like a little itty bitty thing. Cool team here.

• 1:00:46Speaker 00

  Yeah, I think so. One of the questions is, why not just script that then? I think this, think of this as just a skill you can call or a plugin you can call to take what you already do and then instantly turn it into something repeatable. Like you could probably do it without Swamp Club, but they also gamify it in a sense. I think they also give you access to models if you want to go on their like paid plans.

• 1:01:09Scott Tolinski?

  No. So the models thing, models is like what they're calling kind of their, their extension. So the model is like a model in this language, you know, the language model is kind of like my model for plausible analytics hits the plausible API and pulls down the data with these functions. If I'm asking for this or that, it's going to run that function and get that data and bring it in. And then again, you can combine these different swamp models together or workflows or whatever. What Swamp does that's different is it like beyond just having scripts, it has a way for these scripts to talk to each other, to

• 1:01:48Scott Tolinski?

  store data about their runs and to include the ENV stuff and the vault and security stuff separate from your code base because these are, this is your infrastructure and automations and stuff like that. I think it's sick. So I'm a huge fan of this.

• 1:02:06Speaker 00

  Nice.

• 1:02:07Scott Tolinski?

  Yes. CJ, do you have something that you can talk about right now?

• 1:02:12Speaker 00

  I only have one more and I'm gonna end the stream with it. So I guess whatever you got, keep it going.

• 1:02:16Scott Tolinski?

  Okay. I will only say that because Courtney was outside my window mouthing something to me just now and I have no idea what she was trying to say.

• 1:02:22Speaker 00

  We can end it. I can, I can go unless you have something else you want to show.

• 1:02:26Scott Tolinski?

  I can't. If it's urgent, she will text me and it's not a sandwich. I can tell you that folks, because that's never stopped her before.

• 1:02:33Speaker 00

  It is Hack Week and I want to get back to my projects.

• 1:02:34Scott Tolinski?

  Me too. Well, I'm gonna end it here then. I got one more from Dylan Mulroy, AntiSlop. Dylan works at CloudFlare and AntiSlop is his skills for dealing with slop. I have not used this yet, but I've been seeing a ton of chatter of people saying the PRs generated from this are removing a ton of their code in a good way. So a whole bunch of stuff in here, a bunch of rules and scripts deterministic and AI tools, skills to help you remove a slop from your code base. So yes, very cool stuff.

• 1:03:13Speaker 00

  Dope. Are these skills or plugins or?

• 1:03:16Scott Tolinski?

  This is a skill and some scripts.

• 1:03:18Speaker 00

  Skill. Okay. Nice. I'll try it out. I've been using the Sentry skills. There's like a simplify skill and then like a security review skill. That has removed most slop for me, but I guess this is a bit more targeted. So I give it a try.

• 1:03:31Scott Tolinski?

  Yes. Cool. Well, that's all I got there. Check it out. AntiSlop. I'm gonna be checking it out too today, I think.

• 1:03:37Speaker 00

  Dope. Last thing, Scott, are you a fan of Rubik's cubes? Do you know how to solve a Rubik's cube?

• 1:03:44Scott Tolinski?

  Ah, no, I'm terrible at them. In fact, I'm, I'm really terrible at them.

• 1:03:48Speaker 00

  Okay. Well, this site has every possible permutation of a Rubik's cube that you can scroll through. And they also have like nice little animations. So like as you scroll, you see the, the squares of the Rubik's cubes changing. But if you don't know, there are 4.3 times 10 to the, what is it? 19th power possible permutations of a Rubik's cube? And this site lets you just scroll through every single one of them, which I also thought was fascinating because I think that's like more than the number of known atoms in the universe. So like, how do you have a webpage that can like see every single one? But I

• 1:04:33Speaker 00

  do think they're like generated based on their particular index in the solve space. But I thought this was fascinating. First of all, because of the number and then also cool little animations to kind of scroll through the different ones. So yeah, that's it.

• 1:04:50Scott Tolinski?

  Sorry, I was muted. Courtney had come in to say hi. So try not to be disruptive, folks. Sorry about that.

• 1:04:59Speaker 00

  Sweet. But yeah, that's all I have. This was a quick one, but now we gotta get back to Hack Week. I'm making a Game Boy game. And I guess Scott, you can tell them what you're working on because we're doing this all week.

• 1:05:10Scott Tolinski?

  Yes, Hack Week.

• 1:05:11Speaker 00

  Yeah.

• 1:05:12Scott Tolinski?

  Oh, that's a sick shirt. Yeah. Is the hat a Hack Week shirt too? Yes. No, the hat is a University of Michigan Basketball National Championship hat from when the Wolverines won the national championship and basketball, which let me tell you, how many basketball games have I watched in the last year? Hmm, I don't know. But I did watch the ones where they won the national championship. So I get to buy the hat. I don't watch basketball. Sorry, folks. I'm building a little prompt boy.

• 1:05:42Speaker 00

  Oh, nice case.

• 1:05:44Scott Tolinski?

  Yes. It's a, this is my pre 3D printed case. This is the first attempt here. It doesn't hold together very well. Right now it's kind of junky, but I do have a little wave share screen that is a little touchscreen and an ESP device onto it. And the idea here for this device, folks, which I'm really excited to work on actually, is gonna be like a little note taker. Hey, this or that, whatever. And then it can pick up a Gentik work and start working on things, whether that's code. So if I say, hey, my site's down, it can go and start working on the site. Or if I say, figure out why my kid's

• 1:06:27Scott Tolinski?

  tire is not inflating. It can like be like, here are the possible reasons why your kid's tire is not inflating. So.

• 1:06:32Speaker 00

  Nice.

• 1:06:33Scott Tolinski?

  Yeah, that's, that's it right now.

• 1:06:34Speaker 00

  Oh, yeah. And I'm, I'm building a Game Boy game. It's gonna be syntax themed. There's gonna be puzzles and eventually if, if everything goes well, eventually we'll release it as a physical game that you can buy in the merch store for, for Game Boy. So.

• 1:06:50Scott Tolinski?

  Yes. OXIA says, so an audio recorder for mine? And let me introduce you to the, the hook of my project, which is this ain't your grandma's audio recorder, which is literally what I have as my hook in the, the description of the project. So more on that soon. It's gonna be dope.

• 1:07:11Speaker 00

  Nice.

• 1:07:12Scott Tolinski?

  Cool. Well, I don't have anything else. It's been real CJ.

• 1:07:16Speaker 00

  Yeah. Good times.

• 1:07:17Scott Tolinski?

  All right.

• 1:07:18Speaker 00

  Good fun without Wes.

• 1:07:19Scott Tolinski?

  Yes. Oh, man, it's so much more fun without Wes. JK, just joking, folks.

• 1:07:24Speaker 00

  Nope. All right. Talk to you guys later.

• 1:07:27Scott Tolinski?

  Peace.