Building with Agents w/ Gerrit Hall & Taylor Savage

Rex Kirshner (00:02.264) Taylor, Garrett, welcome to the Signaling Theory podcast. Gerrit Hall (00:06.031) Thanks, Rex. Pleasure to be here. Rex Kirshner (00:08.462) Cool, so you guys obviously know this, but just for the audience sake, today we are gonna try something different. We're actually gonna take a step entirely out of crypto world, and I brought this panel together to talk about cloud code and the AI coding tools, and really how all of our lives have changed over the last six or so months is this introduction of, as Taylor put it last time I talked to him, the types of tools that we're always missing and... like have really just unlocked us as project managers and software developers and all of these things. you know, we'll build a little bit more context as we go on, but let's just do like 30 second intros of like who you guys are and like the types of projects that you are working on using these tools. So Taylor, why don't you kick us off? Taylor (00:58.082) Yeah, so I'm Taylor. I've spent most of my career in tech, in product, and most of that time building developer products. And my background's a software engineer as well, so very close to that world. And then most recently, I've started with a couple partners, an AI private equity roll-up fund. And so what we've been doing, what I've been doing, I'm the techie, my partners are the finance people. I've been building a lot of the platform along with an engineer, the two of us, building a lot of the platform and backend, frontend tooling agents to power our kind of service business platform. And so that's what I've been using these coding agents for. Rex Kirshner (01:43.478) And so sorry, just to be like a little bit more clear, correct me if this is wrong, but like what your firm is doing is buying up a bunch of businesses. And then the idea is like, okay, we need software to run the like conglomerate. And so you're using AI tools to build like brand new bespoke software that fits the companies that you guys are buying. Taylor (02:02.594) Yeah, exactly. I'd say in a nutshell and at the risk of maybe over simplifying for any private equity folks. But the concept of roll up has existed for a while now where you buy a bunch of small independent businesses as kind of owners retire, as they're looking to kind of scale and grow, provide them capital and then combine them together to build a bigger business over more territory with more capital behind it, with more kind of shared services behind it, et cetera. And a lot of strategy to date for these roll-ups has kind of hinged on both efficiency gains through scale and also financial engineering through scale. And the opportunity we see really is like, it's so much easier to build actually really good software and also build really good tools to help the people in these businesses do more work better. That like, it changes the opportunity set. when you're trying to combine a bunch of businesses. And we're specifically focused on service industries where the goal is not to replace the human, where like absolutely, we're looking at verticals in businesses where you need a human to do the job. is like an innately human type of business and role. Both like, because that's how I believe that these tools are gonna be best used and also because that's also a way to de-risk a lot of the acquisitions that we make. And so. where I've been building the tools for the people on the team to do the work that they're doing. Rex Kirshner (03:30.466) Cool, awesome. Garrett, can you talk a little bit about who you are as a developer and the types of projects you're working on right now? Gerrit Hall (03:38.147) Yeah, absolutely. So anyone who is watching Rex's podcast probably knows me best from my work with Curve Finance. I've been working there as a developer outreach dev rel type person for the past five years. And I've also anyone has been following me also knows that like no stranger to like a bunch of other random side projects that over the course of those five years would occasionally try to launch. Some went nowhere, some kind of went some places. But you know, about I'd say like four months or so ago, like me and a lot of other people started to recognize that basically, at least for the, within the niches of coding, it seemed as if artificial intelligence had passed the singularity. Now it might not be there in other aspects, but as far as coding goes, I think Rex, you and I bonded about this like late last year. We were just like, our minds were sort of blown at how good the like coding agents had gotten. So I'm sure we'll get the chance to share a lot of the stories about it, but as I played with it more and more, as well as kept in touch with what other developers in the community were doing and measuring how they were adopting and using it, I sort of realized that there is this massive vibe or paradigm shift, not to use the consulting speak on that, but this is real. The era of AI coding is coming. current generation of tools are very good, within a year or two, they're gonna be obsolete, which makes me think that within two years, it's going to be just like completely bizarre. And I don't think anyone knows what the ramifications are, but I'm like super excited to just dive head first in and be part of it. So I'm very excited to talk about all the various projects that I spun up and been working on that time, and hear from you guys as well. Rex Kirshner (05:30.062) Cool. All right, so production note, Garrett, you're like cutting in and out a little bit. Like we can kind of roll with it, but if you can, like you're uploading or downloading anything and you can maybe put a pause on that, that'd be great. Gerrit Hall (05:39.086) Here. Gerrit Hall (05:42.946) Why, yeah, let me try and close those tabs real quick. And then let me go cancel my node, which also helps sometimes turn my node off. One second. Rex Kirshner (05:53.134) Here, do your thing, just keep silent and I'll do my little intro and we'll run back in. Cool, thank you, Garrett. For myself, so computer science, back education, and then never once programmed professionally. I was always doing side projects and did some interesting things, but back in October, picked up these tools and realized, my god, this... enables a class of people that like have, all the right tools to be like good, like soft, software developers and platform managers, but like just don't want to get into the nitty gritty to do big things. And so I've got a lot of like little side projects right now. I spend so much time building like tools for my tools so that I have tools to eventually build a project. I wonder if that's like. a common theme, but the biggest thing that I'm working on is working with a nonprofit who has a website that was like sold to them by just some software vendor that built it once and is building like the same cookie cutter thing for every nonprofit. And it's really not cost custom to them. costs them like tens of thousands of dollars every time they want to make a change. And what I'm realizing is that like For me, what it feels like is AI tools allow me to sit as a product manager and sit customer facing and then have a team of engineers under me to actually build it. And so I guess that's kind of where I wanted to start the conversation, especially with Taylor, who I know has been at very major companies to mid-major companies of software developers. The meme that you see on Twitter or whatever is like, CEO fires all the junior devs, replaces them with AI three months later, has to get rid of AI and replace it with senior devs to fix everything. And like, on the one hand, I get it, like AI makes a lot of mistakes, like sure, sure, sure. But on the other hand, like, I'm just watching what it puts out. And at this point, it's hard for me to believe that like anything but like the top 0.00001 % of engineers are producing better code than this. And so... Rex Kirshner (08:13.1) I don't know, man, like what's both your vibes? We'll start with Taylor. Like is this kind of like creating junk code that eventually someone's gonna have to work out or do you feel like I do that? Like I have a team of somewhere between like 10 and a hundred like really eager Eastern Europeans that like don't really know what to do, but like if you give them instructions can like really run with it. Taylor (08:36.108) Yeah. Yeah. And I think I was in a little bit of denial when these things first came out of, of, uh, kind of overappreciating, you know, my own skillset and like a software engineering skillset and like, oh, these are never going to get good enough. Uh, and then they did, uh, I, I my perspective, and I think, uh, part of my early skepticism was that, how could these things possibly write good coding? You're going to have to review all the code and like, It's going to make mistakes and make the button that says create user, actually delete a user. And then you're going to have these broken apps as a result. From what I found, like that's not the type of mistake that they make at all, but they do require kind of architectural management. And that as the code base scales up, they get way harder to work with. And just like the nuts and bolts of working with a big complicated code base with a big complicated project, like you can't escape the problem of. You need to still, someone needs to have it all in their head or else it kind of gets out of control. and you're like, I personally have not gotten to the point and maybe this will come of just like fully trusting major changes to just let it fly and ship it to production. I'm not reviewing like every single line of code every single time I'm reviewing the systems changes that it's making and then testing for correctness. Rex Kirshner (09:34.936) Mm-hmm. Taylor (10:01.868) but I find that I've got, I've had to be it's they're awesome for zero to one projects. can one shot. Incredible, pretty incredible things. but as the system has gotten more complicated, I've had to manage it more tightly, I guess this is how I put it. Rex Kirshner (10:17.686) Yeah, are you you say you're not reviewing every single line of code? Like, let's just be honest here. I'm not reviewing any single line of code, are you? Taylor (10:27.988) I am reviewing, I am certainly reviewing lines of code that are security sensitive. So one, one perfect example of this is like a big part of the platform that we're building is exposing a bunch of API endpoints so that we can have a bunch of different individual tools that call a centralized API. And, it is more than happy to write you an API endpoint with no authentication on it. And you can, you can ask it to, you can put that in your, you know, cloud. MD file, you can put it in skills. can like, you can put a lot of guard rail, like, non deterministic agentic guard rails in place, but it still will so frequently just write me into a PM point and not put, authentication on it. So I reviewed that kind of code. Rex Kirshner (11:12.11) Hmm, yeah. Garrett, what's your feeling? Like, are you worried that you're building stuff that like, you're, that either you need to clean up behind it or that like, won't really be able to survive like a real stress test or like a real world use? Is that something you worry about? Gerrit Hall (11:29.976) I definitely shared that fear when I first got into it. And then I sort of realized after using it that if it has bugs, it's in most cases easier just to tell it to fix the bug than it is to like worry about it and go trawling through the code and trying to solve it. I've definitely gone more to the like take the plunge and let it do its thing. I definitely agree with Taylor's point that when the code base gets large and complex, it can, it sort of slows down in some ways. I've developed a few techniques that I've found are pretty helpful with combating it. So I have a few different systems of kind of like different levels of complexity. Like a really robust system of documentation is very helpful. As you know, like AIs are incredibly verbose and they have all these amazing plan files. And if you can kind of keep those organized into like sub projects within the task or like specific domain areas and then tell it to reference, like, you know, look through the last log of the last thing, five, 10 things you tried on this. for context before building the plan. Cloud codes plans are also very good, but I've actually found, this goes to your question earlier, that Codex is probably a slightly better coder. I don't think people like using it as much, myself included, because it's not as good at generating plans in the way that Cloud does, and it's a bit slower, and it's not as fun. But what I'd generally do at this point, if I'm dealing with a major system architecture thing, is I will have Claude generate the plan, then I'll send it to Codex, and I'll have Codex review it, and Codex is merciless, so they go back and forth. It slows things down a lot, because Codex will keep finding nitpicky bugs. But after five or six or 20, whatever, back and forths, Claude will then execute the plan, and it'll usually go perfect. So, I just bring that up to say that there are techniques that I found. Rex Kirshner (13:17.069) Yeah. Gerrit Hall (13:21.274) to deal with these larger and larger code bases. And I also think that keeping the code fairly compartmentalized is the other way to combat it. Rex Kirshner (13:31.619) Yeah. Like every single one of my instructions includes like make sure it's modular in there. But Taylor, what about you? Are you like pinging back and forth across different models or like how do you, cause like, I'll be honest, I started doing that and like, just, I don't want to pay for all of these. You know, I think you're paying like 200 bucks a month is already crazy. Like. Gerrit Hall (13:36.524) Right. Taylor (13:36.876) Yeah. Taylor (13:49.654) Yeah. Rex Kirshner (13:57.079) You actually have professional behind you, you can expense it, right? talk to me about what's in your toolkit. Taylor (14:05.225) Yeah. Yeah. So my, will use, found kind of similar. use Gemini Pro to do a lot of the, high level architectural planning. And what I've been doing there is keep an up-to-date sort of architecture document that the cloud code keeps up to date as it makes changes. And then piping that into my Gemini as a gem and then use that as context. When I'm thinking and planning a new kind of major feature or architectural change. And I'll go back and forth with Gemini on very long threads to kind of tease out, it what I want? Like, I don't like that. I like this. Like, let's try this. Let's try that. And then get to a point where I can say, okay, let's wrap this up and let's put it into a, a, a document, like an implementation document. And I'll hand that to Claude code planner and say, okay, here's the draft. Compare this against the actual code gaze. Cause obviously at that point Gemini doesn't have access to the code base. So it's kind of more. Rex Kirshner (14:59.086) Yeah. Taylor (15:01.707) abstract and then cloud code will come back with something. If it's a big diff, structurally from what I did with Gemini, I'll go back and forth. And if not, I'll have it fire off its plan. Rex Kirshner (15:12.246) And like, why you pick Gemini? Taylor (15:15.275) Um, I found it's very good with planning, um, very specifically, and maybe this is just placebo, but, um, we use Google stack for everything, uh, for our business. It's Google workspace and we're on Google cloud. And so I found it has good knowledge of the details of Google cloud. And I, I said, this is a side topic too, but a lot of the major challenges I've run into have all been cloud infrastructure related and not so much like the day to day code. Um, So it seems to be better at that that architecture. Gerrit Hall (15:49.394) Quick question, you're using the Gemini, you're not using the command line tool? Because I think the command line tool has, I've tried it, it's pretty good. Rex Kirshner (15:49.624) What about... Taylor (15:57.862) yeah, no, it's good idea. I have not been using the command line tool. I've been using just the Gemini app to do the back and forth. And I kind of like, I've kind of liked, I'll definitely try that. It's a good idea. I sort of like having like an abstract planner that like doesn't know the details of the specific decisions I've made in the code. And just be able to plan at that level. It's kind of helps. Yeah. Yeah. Yeah. Exactly. Yeah. Gerrit Hall (16:14.52) Yes. Gerrit Hall (16:20.76) Yeah, that used to be chat GPT for me. Rex Kirshner (16:22.972) Yeah. Yeah, well, speaking of tools, like this is something and again, the whole kind of thought behind this conversation is like, this is a pretty isolated, like isolating experience. Like I feel like I'm like a fucking wizard. Like I can create so much code. And then I'm sitting here and I'm like, I don't know if I'm doing this in like, a near childlike way. don't know if I'm ahead of the curve. I don't know how anyone else is using these tools. One of the things I wonder about is like, so when I got started, I downloaded cursor, installed the Claude plugin and then use Claude through cursor because I was like, it's cool to have the IDE. It's cool to be able to just click into files if you need them. But like I found that my sessions would get too long, cursor would crash and it just like wasn't dependable. And so now, like if I do, you know, command tab right now on another screen, I have literally nine terminal windows open all in cloud code. And I've decided to do that because terminal is the most stable application possible. But like, dude, every single day I'm on Twitter, I see people saying like, you really need to be using open code or I hear about this one called pie a lot. Like I'm not sure is cursor, is cursor still a something people are excited about or is that like a joke at this point? I have no idea. I just like want to ask you guys like how what are you using as you use these tools? So what Garrett, why don't you start us off? Taylor (17:44.554) Hehe. Gerrit Hall (17:52.195) Yeah, and what I would recommend if you want some good unfiltered signal about people who are using the tools, giving advice on which tools to use and how to really 100x max it, I would follow Banteg on Twitter, or now X. He was one of the earlier people to VibeCoding, and he's building some incredible stuff with this. And you could just see it when he launches a product. He's got one, for example, it's called Tacope. which is basically a Telegram interface so that you could move your Cloud Code session to your phone and it works between Cloud Code and Gemini, all these different ones. So you can go anywhere, have access, switch back. And he came up with this idea and within days was iterating and already on the 15th version release of it. it's like, Leonardo da Vinci a calculator and see what he can do, right? So as far as cursor, I was actually on cursor before I moved over to Claude code because for a long time they had this like cursor unlimited plan where if you're using the cursor homegrown or like auto, it lets them pick the model. They would just like give you unlimited credits for that. And it was great because like most of the tasks you didn't need like a heavy brain for. So you would plan it using one of the like heavier models which you had limited usage of. And then once you the plan in place, then you'd tell its auto agent to complete it and execute it. Because most of the tasks you're doing are fairly mundane, right? But then they stopped that plan. So I was like, okay, well, let me look somewhere else. I hate using IDEs and I tried Cursor's command line tool and it wasn't very good. Their ID is much better. But then that's around when I tried like, let me see what kind of juice I could max out of like on the like. not the most expensive tier, but I'd try Claude and Gemini and Codex and kind of like all of them in sequence. And eventually I just decided, forget it, instead of trying to figure out weird ways to juggle this, for a time I was experimenting with using Claude but piping the queries not to Claude but to DeepSeek, because DeepSeek's super cheap. And then I was just like, what am I doing? You say it's expensive, which it is, $200, but if you're actually hiring a dev shop of 10 to 100 people, Rex Kirshner (20:10.776) Sure, yeah. Gerrit Hall (20:12.507) It's like $200 is a bargain. If we're going to be replacing all the junior devs and organizations, like I shouldn't complain about $200 to do it. So like I said, like I'll bite the bullet. I'll pay it for a few months and see what I can get out of it. And the $200 plan is like really good because even if you like really hammer it like, like you, like I have 10 tabs open, I'm trying to constantly spin all the plates. Uh, and pretty much I'm like never able to actually like use up the entire plan for the week. So that's actually a really cozy place for me to be sitting in. I still have Kodak, so I'll probably get rid of ChachiBT at some point, but like I mentioned, I'll use that before. I think, I can't tell. I have like Gemini just because I have a Google account, I think. So like, guess I'll keep that one. But that's like more than enough, like AI agents for my life. Rex Kirshner (20:53.25) Yeah. Rex Kirshner (20:56.696) But so you're using just your like terminal window and then like CLI tools, straight, like kind of like raw dog in it, if you will. Gerrit Hall (21:02.383) Yes, yeah. Yeah, I have like a MacBook and I keep track of the activity monitor and, you know, like we have the IDE, would occasionally like, I'd look at it and I have four gigabytes of memory on my machine or whatever and like it's using 100 gigabytes of memory and I'm like, I don't even know how you're doing that, but you're getting forced quit right now. Rex Kirshner (21:21.474) Yeah. Yeah. Yeah. Taylor, what about you? Like, how do you actually use these tools? Taylor (21:28.726) I've been mostly in VS code with the cursor plugin. that's been my kind of daily driver. But as I've sort of rant, or sorry, the VS code with the cloud code plugin. and, and I've, I've used cursor a bunch and I've used, I've played with open code, played with kind of a bunch of the other ones played with anti-gravity. Cloud code just seems the most kind of straight down the middle. Rex Kirshner (21:44.142) Mm-hmm. Rex Kirshner (21:57.805) Yeah. Taylor (21:59.427) I like some of the ideas that like cursor and in gravity or implementing like some of the like debug mode and, know, some like a more explicit sort of planning mode. And there are some cool ideas in there, but I keep finding myself coming back to cloud code for now. But, what I've been working on the side is just building my own like IDE effectively. cause there's a bunch of things that I don't that like cloud code in the cursor area in the VS code ID is a little annoying about like, don't look at them. code files all that often. Why am I taking up the real estate and the memory and all of that? Also, I just don't like the way the tabs are organized. And I also want to make it really easy to have Git work trees so I can work on different branches at once. And I want to be able to connect it with my code base more tightly. So I want to let it run my tests explicitly and watch my console outputs explicitly. So I've been building my own IDE on the side to Rex Kirshner (22:32.364) Yeah. Taylor (22:58.42) to do that, to really have the workflow. I want shortcuts. I want exactly the workflow that I want. So that's been a fun side project. Rex Kirshner (23:07.864) Well, dude, I love that you said that because to me, that's one of the huge unlocks of like when I say cloud code from now on, like let's just, that's a generic term for all these tools. Right. But one of the huge unlocks for cloud code is so when I was trying to find a time for us, right. I went to doodle.com and like tried to create a meeting scheduler thing and on the free plan, you can only do 10 slots and you know, that's fine, whatever. But I was just sitting there and being like, I think I can make my own in. in minutes, you know, like I could be exactly customized. It could be to my own domain. could, and I just, I, I feel like we're entering this world where like the cost and the speed of software is effectively zero now. like, like the, imagine one year ago, you're like, yeah, I'm just like, kind of on the side building my own IDE is like, but now that that's yeah, like why not customize it for yourself? Why not? Taylor (23:38.472) Yeah. Taylor (24:01.267) done. Taylor (24:06.941) Yeah. Yeah. I know it's super fun. And I feel like always the best engineers spent a lot of time building tools for their tools. That was always a pattern. so, but like, who's got time for that? If you're not super efficient at authoring the code back then, but now you can have someone with a day job doing, yeah, you can, you can, you can afford to build tools for yourself, which is fun for people that like that. Rex Kirshner (24:14.317) Yeah. Rex Kirshner (24:23.117) Yeah. Rex Kirshner (24:34.242) Yeah. So I'd say like the biggest tool or the tool that I've been working on the longest since I started on these coding things was my thought was I want to be able to get these tools to externalize as much context as possible. And my original thought was, like maybe I want to switch from cloud code to a different agent. And so I want like all of the thinking to be out so someone can like pick it up from scratch or maybe I want. All the context to be out there so I can have another AI come in and say like, Hey, like review this guy's thinking and tell me like what's smart and what's not. And so I've built, call it like the AI context system, but it's essentially like a command of like save context and then review context that like tries to output as much as possible into files so that you can go cross session and, you know, you know, It passes the time. I, what I'm trying to get to is like, I've spent all this time building this system and like one of the traps that I a little bit worry about with Claude code is that like, lets you build these like hyper customized things that makes you really feel like you're doing something cool. And like, I'm not really convinced that my AI context system is. doing anything other than just creating like a ton of noise that, you know, it's not harmful, but like isn't really doing anything. And so I don't, know, that's not really a question, but more like, I feel like there's a sycophancy of these tools that is less like overt, like, you know. What an amazing job, it just, like because you can build stuff and it does outputs and it works, like you feel like you're doing something valuable and like, yeah, I can customize my own doodle and make it. And I did that, but like that, like what a waste of time, you know? So like, I don't know. mean, is that resonate with you guys at all? Or I don't know. How do you feel about, um. Taylor (26:41.511) Yeah. Rex Kirshner (26:51.97) Just like the ability to like let your mind chase down rabbit trails that like just because they exist doesn't mean they're good. Gerrit Hall (26:59.835) So I would say that at the end of the day, there's no escaping the fact that human beings are social creatures. And whatever superpowers we've all kind of acquired in the recesses of our man cave, where we're sitting and debugging things till midnight, none of this stuff goes anywhere unless we're actually doing the hard work of talking with other people, building partnerships with other people to collaborate, to put our superpowers together to start new organizations, talking to prospective customers and asking, would you buy these things? And yes, AI can supercharge those processes as well. But as I'm organizing all this, I generally organize this around what project am I working on with someone else, as opposed to what project am I working on. Rex Kirshner (27:43.17) Hmm, that's smart. That's interesting. Taylor, what about you? You have any reactions? Taylor (27:48.03) Yeah, no. I, I, I like that because I agree with that too. And actually that's, that's an area where I found the coding agents, they make some things easier, but they make some things a lot harder, which is like working on code bases with other people. Because you can produce so much more code so much more quickly that like, and the challenge always was when you're working on a team of engineers building on the same, like integrating your code together, agreeing on shared contracts and making sure that you're both writing to that, figuring out how to like divvy up the work. Now that problem becomes like. Rex Kirshner (27:59.417) Hmm. Taylor (28:16.169) 100 times harder and like more every single day you have to be doing that versus before you could like have a planning sprint meeting decide on the contracts and then like, you know, work for two weeks and then come back. So just like the amount of collaboration that needs to happen is just so higher throughput. I found when even with two people working on the same code base, it's just like we get we run past each other all the time and have to kind of circle back but Yeah, I do think for people whose mindsets are like engineer builder mindsets and the way our brain chemistry works, these things are incredibly addictive, like super addictive because it hits that reward center that we're used to having to like slog away for hours before you get the green test pass or before you get the like feature working. And now it takes a prompt watching it go. then so sometimes I feel like I'm sitting in a slot machine, just pulling the handle and winning every time. Rex Kirshner (28:50.882) Yeah. Gerrit Hall (28:52.493) Yes. Rex Kirshner (29:01.581) Yeah. Taylor (29:11.401) Which also makes me wonder if we're in the heat of this right now, but it's gonna stop being as fun. I kind of worry about this, because I'm having a blast. But sometimes it's like when you play a video game and you've already beaten the game and you've already maxed out your character and you just run around winning battles all the time. And eventually, I maybe I speak for myself, it's just like, okay, this is not fun anymore. Rex Kirshner (29:22.625) Yeah. Rex Kirshner (29:38.624) I am. Taylor (29:40.009) This isn't really, this is kind of just goofy. mean, code is obviously different because you can be building productive things, but I think we're just going to relearn that lesson that like, it was never really the authoring of the code. Anyway, that definitely created a barrier to entry and it made you like, made it a lot riskier to invest time and energy and effort into building stuff. And that like barrier entry went way down, which is great, but it's not really what. software, even software engineering was more or less like product development, product, building distribution, tech is. Rex Kirshner (30:16.62) Yeah. Well, you said two interesting things there, but like going back to the first thing about collaboration. So Garrett and I, at the end of last year, we tried to work together on a project that ultimately like got sunsetted because it just wasn't that good of an idea, which by the way, I think is like a really cool aspect of these tools. It's like, you can't hide behind, like just build the idea and like. It's probably a bad idea, like, least you know. But anyway, like I found it so challenging to work like with another person using AI tools because like I would run so far ahead. would say like, Hey, I just pushed a GitHub, like pull it and see what you think. like, Garrett would have no idea the types of things that I had worked on or why I did that. I would pull his stuff and then my agent would be like, all this is broken, I need to go and fix all of it. so, like, can you just talk a little bit about how you're working with, like, another guy, like, more sustainably and long-term? what have you figured out that, like, Garrett and I didn't really have to because we just ran into a dead end on the idea, but, like, yeah, how are you doing it? Taylor (31:31.367) Yeah, I mean, it all kind of comes back to just like good software engineering practices that have always been good, but now they're just like a hundred times more important to get right up front. so one big one has just been like building a layered platform. and actually we learned that lesson the hard way and then basically did have to had to do a whole rebuild of our underlying platform into like a really cleanly layered infrastructure layer, platform layer, SDK layer, individual services, like, so that Rex Kirshner (31:40.397) Yeah. Rex Kirshner (31:58.115) Yeah. Taylor (31:59.433) we can work on different things and have established layering in the middle. So that's been a big one. And also different specializations, I've noticed, are starting to emerge. Before you had backend and frontend engineers, because the API layer, the network between the client and the server, that was a very clear dividing line. And if you could agree on that interface, then you could have people working on the backend. writing up to that interface and feel in the front end, inspecting that interface and like that works. Here it's like, you can blow through that pretty easily. But one thing that like one clean separation that we have found is him working on the more like agentic layer. We're building agents, like operational agents to do work within the business and me working more on the like architectural infrastructure layer. And that, so like I'm building the tools. Rex Kirshner (32:32.908) Yeah. Taylor (32:55.728) and the way the data gets to the tools for the agents and he's building the agents and the agent system and like architecting what agents we should build and how they interact with each other. And so that interface layer has also been, that's kind of been our equivalent to the backend front end for how we've been working. Rex Kirshner (33:12.098) Yeah, interesting. Garrett, have you had any other luck working with other people other than me? Gerrit Hall (33:16.771) Yeah, I mean, I can't say that I've solved it, because this is entirely too new of a problem to know if anyone has actually solved this. I do think it could be, you know, people are looking for ideas, like Claude Vibecode being an app to solve the problem of, like, working in a cross-functional teams, make no mistakes. I will say that, like, you know, it's like an iterative process, like, working all efforts of working with teams are. The structure that I've been, like, most... Rex Kirshner (33:28.878) You Taylor (33:29.821) Yeah. Gerrit Hall (33:42.668) excited to try out as we're pushing forward with some teams is to have one main orchestration repo. This main orchestration repo is essentially responsible for all the project level intelligence stored as markdown files. So it's very easy for anyone to pop in with their agents and say, what's the business strategy here? What's the marketing strategy? What's the, you name it. And then within that orchestration repo have subsidiary, repositories, so you have your front end in one folder and the back end in a different folder. So then when people are working on the back end or the front end, they might only need to check out whatever they care about, or they could have all of it however they like locally. And those subsidiary repos aren't checked in the main orchestration repo, they live as their own independent thing. They don't have to be super cluttered with a ton of markdown files unless they're like highly relevant to it. So at least that's the structure that I've been trying to work with, allowing people then to kind of focus on their individual thing, see how it connects together. And even if I know no React, I can still say like, hey, like, you know, I couldn't solve this, but you know, Claude took a stab at it, it's filed as a pull request, you know, use it, don't use it, you know, go for it. Rex Kirshner (34:57.206) Yeah. Sorry, Taylor, do you want? Taylor (35:00.626) Yeah, no, yeah, that's interesting. I like that. I like the idea of having like just decentralized documentation repo. Yeah, the other thing similar to that that I've been finding is it's almost more important to get the architecture right up front. Like, of course you can go and it's made it really cheap to rebuild everything from scratch and just be like, okay, let's do it differently. But that's also a great way to sort of like waste a bunch of time that doesn't feel like you're wasting time because it feels like everything's happening so fast. like, Rex Kirshner (35:27.502) Yeah. Taylor (35:28.71) You can go down the wrong path and work on something for a week and then have to rip it all out. And obviously you can rip out a hundred times as many lines of code as you'd ever be comfortable ripping out in the past, but it's still a waste of a week. And so we've had to put way more effort upfront in like really thinking through how we want to architect it because it then gets built so quickly. You don't really have the like intermediate stages that you used to have where you could be like, wait a second. I think this isn't quite right. There's no time for that. Yeah. Gerrit Hall (35:58.032) Yeah, I a lot of fun. I had lot of fun around Christmas building out a small project. I started in Python and then partway through was like, actually, we probably need to move to NPM for this because of some random thing. And it's like, this could be months of work. I'm like, go ahead and do it, and five minutes later. Rex Kirshner (35:59.735) You know it's a... Rex Kirshner (36:14.144) yeah, dude, I love how I can't estimate how long it takes to do stuff. It's like, this is a three week project. like, okay, three weeks. Yeah. But you know, well, no, the engineers still take the three weeks. It's just they slack off for two and a half. then the full. Yeah. What's interesting about what you just said, Taylor is like, Taylor (36:18.033) Yeah. I know, the sandbagging club. Just that's how engineers work too, so it's no different. Gerrit Hall (36:27.513) Yeah. Rex Kirshner (36:41.162) I hear you, but I find myself almost taking the opposite approach. So with this project I'm doing for the nonprofit, like I built up what I thought they wanted from like a product requirements document. And I found that there's just like a lot of things that they want different, right? So I built like a little feedback button so that they can give feedback directly. It gathers all the metadata. It takes a screenshot of the page. Like, cool over engineer everything, right? And what I'm finding is that through all of these little changes, like the code is really migrating from what it originally was. And my thought, and tell me if you think this is dumb, but it's just like, let's let this happen. It's going to create not really a clean system, but it'll get the functionality that I want. And then once I'm like pretty confident that this is close to what we want, like I'm thinking about telling Claude, this is what we have. I want to build it again from scratch, knowing what the end state is going to be. Taylor (37:42.344) Yeah. Yeah. And I think that's reasonable. I mean, we've done multiple sort of like system level and module level rebuilds with exactly that. The, the, the architecture point to me, it was more when collaborating with someone else, if we haven't like really between the two of us in our heads, aligned on how the system is going to work. And then we both go off and start building. We ended up with two basically incompatible systems. Rex Kirshner (37:50.968) Yeah. Rex Kirshner (38:08.013) Yeah. Taylor (38:09.943) and we've also, mean, we've definitely wasted a lot of time on like big, hard migrations to try to merge them back together. And often we end up just throwing one away and saying, this is the goals, read the code, rebuild it in terms of this other system. but again, like, I think it's for me, at least I'm trying to keep myself honest on how much are we spending in terms of tokens and how much time are we actually spending in reality? Cause like, we haven't spending three months building a pretty big complicated system. Like that's not. instant. it feels instant while you're doing it because features feel instant. But to actually build up the big production ready system still takes time and you don't want to throw that away. And I also think that the the cruft compounds once you get to it, as the code base grows, like having having old stuff hanging around is just a step on a rake waiting to happen as soon as the Rex Kirshner (38:40.536) Yes. Rex Kirshner (38:51.022) Yeah. Rex Kirshner (38:59.053) Yeah. Taylor (39:05.903) you the agent finds it, you don't realize that's what it's looking at and it's old and all that. Rex Kirshner (39:09.816) Yeah, no, that's fair. Well, kind of in this realm of like tips and tricks that you've picked up, you know, starting with like working with other people, like what are some other things that have that like I... I'm assuming you guys are figuring this out too just by like every once in a while you see a YouTube video or tweet or read an article or talk to a friend and they say something that it's like, that's pretty interesting. And it just becomes part of your regular workflow. it seems actually like I'll give you the one that I'm thinking about right now is do you guys know this guy? I think like Steve Yeggy or Yej or something. Taylor, do you know his actual name? Yeah, he so. Taylor (39:46.311) Yeah, Rex Kirshner (39:50.783) I found him from Gastown, which like that, if you guys have want to like be interested, but like also just not be able to follow anything, read about Gastown. But he wrote another article about like six things he learned about working with agents. And one of them was like, no matter, every time the agent does something, I have it look over the code that it just wrote like five or six times. Taylor (39:55.875) ride. Rex Kirshner (40:17.312) And like, yeah, I'm burning a lot of tokens, just having it look at the exact same thing. But like, he says that he, he finds that he's really saving tokens in the long run. And so now like every time I do something bigger than like a, like a distinct bug that is fixed by lines of code that I can read. If it is even touching two files or more, I, I had literally sit there and just make it re, you know, I have a prompt that I copy paste. and it's just generic, please review the work that we just did and look for best practices, like make blah, blah, blah, blah, And, you know, that's the tough thing about these. Like you can't really tell if it's working or not. It just kind of feels good, but that has like become like one of the staples of how I work. So wondering like, are there any things that you guys have really picked up that like are just part of your workflow that maybe I can steal from you? Rex Kirshner (41:17.048) So Garrett, kick us off. Gerrit Hall (41:18.755) Yeah, I'm trying to think. guess as far as figuring out the best practices for workflow, I always have nine tabs open, right? Like you mentioned you do as well, because that's the maximum you have keyboard shortcuts for. It runs out of that. And I'd say with all the different projects I'm working on actively any given time, I just take different like... sort of like split test approaches, right? Like there are some where I'm like very rigorous about like all the commits. I look over everything five times before I check it in. There are some where I'm just like, I don't care about this project. It doesn't matter to me if AI screws it up. So I'm just going to give it complete free rein. And I'd say that like the kind of outcome of that so far has generally been that like the more trust and control I give to AI, the better things seem to work out over time. So like, you know, there's somewhere like it doesn't have SSH or GitHub or any keys at all. You know, I control that tightly. Like I have to review its commits before I do it. And there are just somewhere I'm like, you're in charge of this, like deploy it when you're ready, push it. And those tend to go a lot faster. So I think I'm like just going to ultimately like keep gleefully sliding down the slippery slope. Rex Kirshner (42:29.55) Do you use the dangerously accept all flag? Gerrit Hall (42:33.857) No, I haven't tried that one just yet. Does that- Rex Kirshner (42:35.992) But like, you know, if you launch Cloud, it's like Cloud and then dash dash, and I think it's dangerously accept all, then it never asks you for permission. just, yeah, yeah. Gerrit Hall (42:41.86) Right. Gerrit Hall (42:45.723) Never, so there's no, it just goes. I haven't done that. Although, I will say, okay, so two things. One is I set it up so my terminal gives me alerts, so it's not too tough to just look and see what terminals I need to pop to. And I will also say that I've been in this coding group for a bit, because we both are in crypto, there's a lot of really good devs. And I should have been paying more attention to this, because they were really starting all these processes six months, a year ago. And they're giga-cracked at the way that they approach this. What I haven't experimented with, that I probably should, is actually just integrating cron job scripts on my local machine to do as you're saying, actually run certain prompts, fetch the daily numbers, something like, daily tasks with that dangerously skipped permissions on. I probably should get in the habit of actually creating some cron jobs on my machine to actually run Claude prompts on my machine. That would probably be the next level of AI degeneracy that I haven't hit yet, I should. Rex Kirshner (43:54.627) Yeah. Rex Kirshner (43:58.169) Did you degenerate? Yeah. Taylor, what about you? Tips and tricks? Taylor (44:03.462) Most of the things that I've found like worked manually, I'm trying to build in this like little IDE I'm building for myself. I mean, one we've talked about already, which is like constantly having a document things for itself and keeping that documentation really organized. I'm forcing that in the IDE and like having it build basically a documentation center as it goes. I'm also using the IDE to kind of get Gemini and Cloud Code looking at the same stuff and being more... collaborative, versus me being the conduit for all of that. copy paste exactly. Like, having it summarize and then kind of feedback into itself. other thing, the other, the other kind of form factor that I'm building in for myself into this IDE is jumping between the tabs has driven me crazy. The cloud code tabs has driven me crazy. Cause I forget which one's which I've like kind of built little systems in my head of how I try to keep them organized, but, you know, and then someone will sit there with a plan and. Rex Kirshner (44:34.946) Copy paste. Taylor (45:00.07) It'll just sit there for a long time. I'll forget why I planned it. The code will have changed underneath it, like all that. So in the ID, I'm building out like branching so I can create work trees and have different branches going for different like projects at once. Um, but then also building like an inbox style approach to it. So I have the agents do their thing and when they need permission from me or when they finish something or when they want to propose a plan, they send an email to me and then I'm just watching my inbox. And that has messages coming in. It's got context. Gemini summarizes what the cloud code agent is trying to do so I can read the email and understand, or remember like, that's what I tried to get it to do. So I'm basically trying to recreate my job as a product manager. Just I'm product managing individual cloud code agents. And I'm sitting there just checking my inbox and responding to emails. Rex Kirshner (45:44.621) you Gerrit Hall (45:52.856) By the way, I figured out it wasn't too tough to get Claude to reply to emails. You can set some certain things in Google and give it access to your Gmail account, and you can just have a dedicated Claude that chat with by email. Rex Kirshner (45:52.918) Yeah. Rex Kirshner (46:06.498) Yeah. Do you guys ever like take that moment where you're like, huh, this thing has like complete sovereignty over my computer. And like it could like there's there's actual nefarious stuff if like Anthropic or whatever decided just to like grep through and find your social security number. Trivial. Trivial. Right. But then there's like the less nefarious and more just like, oopsie, that like Taylor (46:06.852) well. Rex Kirshner (46:35.276) I'm very aware that it could like completely wipe my hard drive. Gerrit Hall (46:40.409) I even mentioned that to Ed at one point. was like, and it like gave me a very reassuring reply. Rex Kirshner (46:46.286) I hope so. Yeah. Taylor (46:46.872) You Gerrit Hall (46:49.967) Yeah. Taylor (46:51.821) Yeah. Yeah. Rex Kirshner (46:53.218) So Taylor, on your IDE, is there any part of you that wants to try Gastown? What are your thoughts on that? Taylor (47:01.177) Yeah, it was an interesting read. And yeah, he was at Google while I was there. So I'm very familiar with this vibe and style. He's funny guy. the, do think that they're the limiting factor, at least in my own experience, working with these is keeping the project of what I'm actually trying to build in my head at once. And like, I wouldn't be able to run a gas town like thing because I want to build. Rex Kirshner (47:07.81) Yeah. Rex Kirshner (47:21.4) Yeah. Taylor (47:28.889) the feature and then I want to play with it and feel it out and see if it works the way I want it to work. And if not, go back and fix it and like iterate like that. And so just having something run like overnight would do too much. Like it's, too much for me to personally keep up with. And so it's not useful in that sense. It's a great way to burn a lot of tokens. same with like the Ralph loop, which has been funny to watch the, the, the whole like Ralph loop, viral live coding. Rex Kirshner (47:57.122) Yeah, yeah. Taylor (47:58.47) but I think it's also just a great way to burn a ton of tokens. And I have just not come across personally, I've not come across projects that like make sense to work at, at that way. Rex Kirshner (48:06.381) Yeah. I do. This brings me to like again why I want to have these conversations because I find myself scrolling through Twitter. Like at this point my feed has completely no more crypto at all. All it is is like horrible immigration stuff or like people talking about vibe coding tools and like I'm sitting here and I'm like I don't really believe any of you guys are building anything right. Like the best example is the Ralph loop like kind of like I got excited by it when they're like it increases the power of Claude Claude by a hundred X but I don't I had to build some tools and I'm like this is just the same experience as doing it yourself like you just maybe have to hit enter like a two more times less like I don't I don't so yeah I don't know I mean is that is that the vibe that you're getting from like a lot of the conversation is these aren't like real people that are trying to build things in the way that we are or am I in the wrong side of Twitter or I guess like, how do you guys feel about like the things that you, the way people talk about this? Gerrit Hall (49:15.59) I- I'd say I'm fairly skeptical of a lot of the conversations that I see, just because since the beginning, the conversation around AI has been really bad. I'm sure that you guys remember six months to a year ago, you'd log in and you'd see people being like, wow, here are 10 mind-blowing ways I'm making $10,000 a month. I tried one of them and it didn't work, right? I was like, I'm calling BS on this. I think at this point, if there are, this actually ties to your earlier question about like, Rex Kirshner (49:32.654) Yeah. Gerrit Hall (49:45.8) getting tips and tricks from YouTube videos because it's sort of like the Claude is good enough that that's not the limiting factor the limiting factor is my time and attention at testing the outcome Because there's like only a certain amount of hours for a day that I can like sit there and look at this and be like no like the shade of blue is wrong or this like didn't work and it's usually like at things that are so far down in the flow that like it will now take like Minutes before I can get it which actually like dramatically limits my testing time Rex Kirshner (50:04.14) Yeah. Gerrit Hall (50:15.003) And when you get down like that far down any branch, it's kind of like, yeah, it's just like a weird problem to have, I guess. So I guess like maybe, yeah, like maybe there are like tools that could plug in that would 100x Clawed's efficiency, which would take it from 15 minutes to five minutes to do, or one minute to do a task. But at a certain point, it's like, if that actually is real, we're gonna see it the next versions of all these AI models that are gonna drop in the next three, six months. I don't know, like, I mean, Rex Kirshner (50:39.478) Yeah. Gerrit Hall (50:44.907) I, the devs that I know that are cracked giga-chad devs and like reading how they're using it is like instructive and informative. but mostly like I found it's great. So I don't need to like pick up or spend too much time. but also, yeah, my, my feed has also become entirely a hundred percent AI stuff. So Taylor (51:03.618) Done. Rex Kirshner (51:03.712) Yeah. No, man, I just feel like everything I see is like here, like I same prompt to the different coding models of like some ball bouncing around a hexagon or like, look what this created in a one shot or things like Steve Yeggy, like look at this tool that I built to like build other tools, but he's clearly spending so much time on it that I'm like, what are you building other than just this tool? You know? And so I, yeah, again, that's just kind of. Taylor (51:31.837) Yeah, think this, I think that has always existed in the engineering world too, of like, it's really fun to like build fun stuff quickly for yourself. And like, that's, that's there. It's just, that's been amplified. And so that's what you, you hear and see, cause it's so much easier to do that. I mean, what I see like in real businesses, service businesses, like the, ability to produce applications is not the bottleneck whatsoever. It's like. Rex Kirshner (51:46.391) Yeah. Taylor (52:00.357) Is this the right problem to solve in the business? And can we train the team to use this tool effectively? And does it do a better job of like the system that they already have, which is in most cases generally fairly fine. And that's like a 200 times harder problem than cranking out more cloud code code more quickly. Uh, so it's just, yeah, it's not something that I'm focusing on because it's not even remotely like the efficient, my efficiency with cloud code is not remotely the bottleneck of my ability to add value to the organization. Rex Kirshner (52:05.176) Yeah. Gerrit Hall (52:30.107) Yeah, so I guess putting it in another way, like if someone's built an amazing application using Cloud Code, they're not talking about Cloud Code, they're talking about the application, right? Like a carpenter's not being like, hey, I just built an amazing house. Look at this hammer I used to build it, right? Like they're pointing at the house. Rex Kirshner (52:46.412) Yeah. And I don't, probably am like not supposed to share this. So I'll keep it like super vague, but I was talking to a CTO of a fortune, whatever 500, 100 company. And they were saying like, yeah, man, we have just used like cloud code to essentially like rip out all of our SaaS vendors and rebuild things like same functionality, just homegrown. And now we don't pay for like, like Google suite or like whatever. and I do, you know, that like, find myself the thing that I'm most bullish on right now is like, love my internal tools. love my AI context system. I have this whole like dashboard that is like for the first time in my life, I made a Mac OS app that like runs like a, you know, admin tools is like checking on all my projects and it has everything, right? But like the thing that I'm actually most bullish about for my career is building essentially like a time tracker app for a nonprofit because I think like what's interesting here is that like I can build tools for boomers in a way that is like really quick, really responsive. Like I'm in Los Angeles, I can come to you and we can talk about what you want and like we can move quickly and feedback. And yeah, I, to that point, like even bringing up cloud and AI probably is like counterproductive to that. Like it's more just about we can build what you need. Taylor (54:13.41) Yeah, that's the most fun and rewarding part, I feel like, of the whole thing. As much as I love building my own IDE, it's a means to it. Rex Kirshner (54:21.452) Yeah. Yeah. Yeah. All right. Cool. Last five minutes here. I would love to kind of continue to have these conversations. So like, let's like walk out on like, what is something that you are thinking about or like to figure out or is like causing you problems or like what's something that we can spend the next week just thinking about and come back and like start a conversation about things that are relevant to us and not just like vibe code our way through a conversation. So here, I'll give us a little bit of data here just to think about it. Gerrit Hall (54:57.915) No, it's good question. I'm just trying to think of what the bottlenecks I have, because all of my projects are blocked in various ways. And it's usually on me. It's like need to make it a critical design choice at some major crossroads, or I'm waiting for some customer feedback to bubble in, or waiting on other people. I don't have a good answer, I'm afraid. I'm just looking at my tabs here. OK. Rex Kirshner (55:20.942) Well, here, I've got an answer, which is like, I don't understand when is the appropriate time to like build out a skill to build out an agent or just to like give high level, like, please do a code review and check for like best practices and like. This is an area where I feel like I get really lost in my addiction of like, look at how cool and intricate this can be. I can say Claude launch eight agents. And then I get like the little dropdown and it looks really cool. And like all this stuff's happening and I don't, I don't really get it, you know? and so like, that's kind of the big thing that I'm trying to wrap my head around is like, Do I just trust the system to be essentially like AGI for coding and build on itself? Or is it really important to write skills, to go find skills, to bring them in and augment the system through these third party plugins? Taylor (56:23.265) Yeah. Yeah, the skills, the whole skills conversation reminds me of it's like the notion ecosystem. It's like everybody, everybody's lepping their templates because it's the most productive system of templates. And it's like, okay, for you, that's cool. Yeah. Yeah. Um, one that I'm thinking a lot about is, um, testing code, like testability of code bases. Cause that is like, to me, that's the gold standard of Rex Kirshner (56:39.202) Yeah. Taylor (56:54.775) both obviously just historically. mean, it's funny when I cut my teeth as a software engineer, it was like coming up through the test driven development, the TDD, BDD world. somewhere in the back of my brain, it's still like, that's where you start. And I think that like tests are almost like dramatically more, they're always important, but they're almost dramatically more important in this world because it's the only, but what I found is I found if I read way more test code than I do. production code because the agents love writing tests that they know they can pass. They write the 10 most trivial tests and then those it's like, wrote tests for you. Like that doesn't count. Rex Kirshner (57:29.11) Yeah, yeah. Gerrit Hall (57:33.499) What truth. Rex Kirshner (57:35.288) Well, not only that, if there's issues with tests, they're going to fix them. And that doesn't mean that they're going to like maintain what it was. Like sometimes they'll change the test so that what's happening passes the test. Taylor (57:46.308) Totally, totally, totally. But at the same time, I have, you know, I'm, especially now that we've got this running in a production cloud environment and like, you want to make sure it's really right locally and it mattered in CI and all that. so, yeah, having like a good test harness approach and like way to use cloud code to interact with that test harness approach in a way that is not as gameable. feel like that's something that would be really. Gerrit Hall (57:46.319) That drives me crazy. Rex Kirshner (58:15.554) Yeah, no, that's... Gerrit Hall (58:16.379) That testing is a very good point. I should have brought that up earlier when I was talking about best practices, I've added it at some projects at the very early stage. You're not allowed to commit unless it passes all the tests, and then go ahead and commit. And that works very, very well. But you super have to supervise the tests, because it will write a passing test. You'll be like, well, the test passed. And it's like, well, yes, it's because you just dropped a return statement. It's not testing the right thing. So that needs a lot of babysitting, but it's super rewarding when you get it set up. Rex Kirshner (58:47.446) Yeah. And the last thing I'll say is I really want to figure out robust ways to build confidence in myself that like data that's important isn't going to just get nuked kind of by accident when it's like, like I need to like push a migration. this migration is difficult. Like, let me just clear the database. like, it's much easier when there's no data in there. And like, it's like, Gerrit Hall (59:09.508) You Yeah. Rex Kirshner (59:16.866) I don't know. mean, I literally find myself like going into project directories and like copy pasting out like SQLite files or like JSON files and like saving it somewhere where the AI agent doesn't know where it is just in case something happens. And like, Taylor, I'd love to talk to you more about how you think about that in like cloud production environments, because that's like. Taylor (59:29.411) Yeah, yeah, yeah. Gerrit Hall (59:31.157) yeah. Rex Kirshner (59:40.642) data that matters and like you can't afford. So I don't know, like are you relying, are you building systems around that? Are you just relying on like snapshotting of the cloud systems themselves? Like I don't, I don't know, but I think that's so critical with these AI tools that like they'll just run away from you. Like they will run away from you. Taylor (59:58.147) Yeah. Yeah. And that's also why I couldn't use like a Gastown approach at this point, because I need to know what files it touched at a minimum, just to know if it didn't do anything weird. Rex Kirshner (01:00:08.724) Yeah. Yeah, cool. All right, well, I think that's some stuff to think about and hope you guys had a good time. I would love to do this again and just pay attention as you're coding and come up with things that you'd like to discuss or like tips and tricks for or just see if anyone else has solved in a cool way. Cool. All right, have a good weekend. Gerrit Hall (01:00:33.218) Absolutely. Taylor (01:00:35.534) Yeah, thanks. Doggone.