Claude Dispatch Update and Building Local AI Agents with Gemma 4

Brendan

Yeah, sounds like a good plan. Yeah. So for me, um, I'm trying to think. I guess it was like last. Thursday or Friday something. So it's been like maybe four or five days since I started working with it. Um. And it actually worked out pretty good. 'cause I was technically off of work and I was out and about doing stuff with the kids and I had time to kill when they were in the pool. And I was, you know, doing

Baird

Sure.

Brendan

And, uh, so I, I decided to just, you know, I, I already had dispatch set up. 'cause I think we might have touched on this in one of the, the other episodes, but you gotta, you gotta, the syncing wasn't actually that hard. It's, but it's basically. know, and this is probably good to just talk about it 'cause I didn't fully have an understanding either. But it essentially is, is it's, it's like a chat of its own that is dispatch and you can talk to it, you know, whether you're on your phone or your computer. I guess, again, I haven't even opened or used OpenClaw yet, but. From my understanding of how that works, you know, you can interact with OpenClaw via, um, a couple different things. I think people are using, um, telegram I think is the

Baird

Yeah. Yep.

Brendan

but, but an interface like that. So it's

Baird

Yeah.

Brendan

think that's probably what they're aiming for.

Baird

Yep.

Brendan

you know, with that in mind, it took me a few minutes to just kind of figure out, okay, what. This is cool. I have access to my computer I, I, I also wanna just say like, this is different than like some of the remote stuff

Baird

Sure.

Brendan

Anthropic has also added.

Baird

Yep,

Brendan

So I was trying to just get a grasp on, you know, what, what can I do with it?

Baird

yep.

Brendan

So I started was I've got all those scheduled jobs we've talked about before in co-work

Baird

Yep.

Brendan

and. Basically I just, I kept asking it questions, kinda like we did last week. Just ask Claude questions. What can you do? What can't you do? And. At first, I wasn't sure, is it going to, you know, give me some sort of an update from those, uh, you know, those schedule jobs. Am I gonna get updates there? So I asked it that question. Not surprisingly, the answer was no, unless it's something that I. Kickoff from dispatch. Now if

Baird

Oh sure.

Brendan

and I say, Hey, create a scheduled job that does checks my email or

Baird

Yep.

Brendan

um, it did say they would update me in dispatch there,

Baird

Okay.

Brendan

but I could still access if I knew a job was gonna run at say, like 3 o'clock PM then this is actually what I tested when I was at the pool with the kids is I knew I had one of my roll bar jobs running around 3, 3

Baird

Sure.

Brendan

like that. I just said, Hey, gimme the update on the rollbar analysis, and me that. And then I pretty much just said, okay, I wanted to push it, I guess, and see like what all could I do? then from there I just said, okay, now create tickets out of the two that you said were like the highest priority. Then after that I essentially just kept going and said, now, you know, take those two tickets and spin up two new Claude Code agents and work on 'em now. That's where so far it kind of stopped for me. I haven't actually gone back. Um, part of it came down to like. I still need to understand what this, what the, um, access I guess it has or like,

Baird

Sure.

Brendan

I'm not even sure the right way to put it, but what it was coming down to is like, have different repositories, get repositories, I know that my computer is like clawed and everything is all set up for one of them.

Baird

Yep.

Brendan

actually on a different one and I, I kind of lost track and then I got distracted by kids at the pool and stuff. But where I left off was. If I had got to, or, well, after I asked it to kick that off, it was asking for permissions and then I

Baird

Oh, mm-hmm.

Brendan

the wrong repository. I

Baird

Yeah.

Brendan

which repository to put it on.

Baird

Yep.

Brendan

so I want to go back and I, I'll report back once I figure this out. But I think overall it was, it was interesting. I definitely could have got more done, you know, then I did

Baird

Sure. Yeah.

Brendan

if, if I had more time and like, you know, didn't get distracted. But I can definitely see the benefit of it, you know, it's. And I, and I do wanna also play with , know, all the remote stuff too, which I haven't even touched.

Baird

Yeah.

Brendan

And I, I'm curious to see like how that interacts with dispatch. you know, there's definitely some power there to be able to, you know. Be wherever, be on the couch watching TV

Baird

Yep.

Brendan

something pops in my head. Or you get an alert and you, you aren't even at your computer. Now,

Baird

Sure.

Brendan

you have to have your laptop open.

Baird

Yep.

Brendan

I'm not at a point where I've got a Mac mini or a desktop of any, you know, any type of desktop

Baird

Yeah.

Brendan

But, you know, that, that's something I think I'll, I'll figure out at some point. But it, it's, it's a cool idea. Um, and I, I definitely wanna push it some more.

Baird

Yeah, and I've noticed that they now have dispatch in both Claude Code and Cowork. So like they've kind of unified it. So, I mean, I know cowork can actually code anyway, so I don't think it matters that much, but I think there is a, a better integration from when it first came out, when it was solely cowork based. So that is interesting. Um, yeah.

Brendan

too, but I, I hadn't really like, I, I'll put it this way, I saw that as well, but I haven't seen on my phone it reacts with the Claude

Baird

Mm sure.

Brendan

like, I don't know if it's two separate chats. I haven't checked that part out yet. Um,

Baird

Yeah. The interesting thing is it like, it looks like the same interface. It's just like located in both tabs of coworking code. So my guess is it's like the same thing, especially since it is a separate chat. It probably is just one, and you kind of get both worlds in one.

Brendan

Yeah. Yep. Yeah. And the other thing I really want to test out is, you know, what, what happens? So in that instance right there, when you've got it integrated now with Claude Code, I gonna see in as I would if I created a, created a new, I think it's called a session in Claude Code.

Baird

Sure. Yep.

Brendan

I gonna see a new session if I, from my phone tell dispatch to, you know, do that pull request I was talking about or whatever. Um, I'm, I'm curious. I, I'd like, I gotta figure that out. So I'll definitely report back on that next week. 'cause I'll, I'll mess around with it some more.

Baird

No, and that's actually an interesting thing that I've noticed in general with like cowork especially, but I think it's the same with Claude Code now is like your scheduled jobs don't actually sync between computers. Like if you do a scheduled job and you go to schedule like on a different computer, you don't see like it. And my guess is it's just running, you know, a local cron job essentially to do that.

Brendan

Yep.

Baird

And it's probably because it's an, I'm pretty sure it's an electron app, so it's probably just running it in the background there. And so that is kind of a, a interesting quirk to it that, you know, I think dispatch is starting to try to change that. So it is synced more places, but it's definitely something I've noticed with. Working with Cowork a bit.

Brendan

Yep. Yeah, so I mean, it's not a huge update, but. Definitely poked around. I think there's, there's a lot of potential there. Um, so I'm hoping to push it even more and, and report back next week on, you know, d different things I've been able to do, you know, 'cause I would love to get to the point where, even if I'm just like sitting at my desktop, right? Or my, my laptop today. And I'm just like testing around, you know, I, if I think there would still be some benefit to even having like the alerts on the phone I kicked something off in dispatch, but I'm doing something else, you know, I could see a benefit even to stuff like that.

Baird

Yeah.

Brendan

I, I know we're gonna Go ahead.

Baird

No, no, go ahead.

Brendan

No, I was gonna transition to the next topic, but if you got something to say about that, go ahead.

Baird

I was just gonna say, I, I think, and this is kind of the case of. Most of the stuff we do, it's also just like learning, you know, like building that into your muscle memory of like, Hey, I can have autonomous stuff going on in the background and your phone's kind of the portal for that, if you know. So it's like even if you're at your desk, if you have stuff going on in the background, it. Separate things a little bit like, so obviously you can still look at it on your computer, but it's also like you can see it in a different modality so that you have that always running. And so yeah, it's definitely kind of a, you know, with ai, for me especially, it's like. I have to keep pushing myself to like adapt to what's possible, and it's like reevaluate how I actually work on things versus just being like, no, I've always done it this way, so I'm just gonna keep doing it this way. It's like, no. Like how can I take a step back and change how I work to take advantage of the tools that are available?

Brendan

Yeah, that's a really good point. I mean, I, I've certainly noticed that with myself as well, and you know, even just working with the ag.

Baird

Yeah.

Brendan

Ai, it's still so new and you know, I, it's not like I had that much time to build up bad habits I guess, or

Baird

Sure,

Brendan

but you still have habits. And even one thing I've noticed, and that, you know, this is not related to coding at all, but I still default to doing a Google search

Baird

sure,

Brendan

you know, something like that something that I know, like in my head. The way I'm formulating it is already to talk to ai

Baird

sure.

Brendan

obviously, you know, Google and, and all of the search engines have AI at the top anyways. But, you know, one thing that I've end up finding is, especially if it's something that I really, I don't know how to ask for it, if I just go to the AI and ask that question, and dispatch would be a great use case for this too. Like, use dispatch to. Send out an AI agent to go do the research and come back with a report, rather than me doing the traditional wasting my time doing a bunch of searches and not quite finding what I want, and you know, stuff like that. So it's another just slight shift in the way you need to think about stuff. I think.

Baird

Yeah, from my perspective, you know, I. I'm kind of in the same boat where I still do Google searches, but I have started to shift where it's like I do a Google search almost to get that AI summary at the top because it's like I. To your point, it's like if you don't know exactly what you're looking for yet, if you go directly to like Claude to ask questions, it can help you. But it can also like, I mean, my Claude's opinionated now, it pushes back on everything I say. So I'm like, shut up, Claude. I don't wanna talk about, but it's also like just, you know, I always use Google as like a. Thinking processes well before where it was like, I'd ask a question, look at what's there, formulate, like what am I missing, and then ask more questions. And so I definitely still have some of that. But it, it tends to be like, sometimes I'll do a Google search, take what I find from the summary, then go into AI and like do a little bit more targeted type. And you know, I, I think to your point. Using AI to like go out and do some research on a topic could do that.

And, and I think that's the, the interesting part of, or an interesting part of the AG agentic AI stuff is like, how can you have it do things that you would've done yourself? And not even just in like the automation way, you know, like that, that's the part that I'm trying to explore on my side is like. I think age anti stuff is kind of cool, but I also think a lot of how it's being framed right now and used right now is like too hype driven. You know? It's like it's gonna do everything. It's gonna build everything and like. You know, like, that's cool. But I also have this feeling of, I actually enjoy doing some of this stuff. You know, like I don't want AI to do everything. I want AI to make the boring stuff easier so that I can focus on the cool stuff. Like, and you know, I think we talked about it or you know, offline we shared it. You remember Pulia, the

Brendan

Oh yeah.

Baird

build a company and.

Brendan

about it too.

Baird

Yeah, I actually was just listening to a podcast with the founder. A funny side note. The name is actually AI Slop spelled backwards and put together.

Brendan

that on a different podcast too. It wasn't

Baird

Yeah,

Brendan

was on, but yeah.

Baird

okay. Yeah, so that is kind of funny. It's almost an Nin joke because honestly like what I've seen there, well, the concept's cool. It feels like AI slot, you know? It's just like, let the AI run and do it. And so that's a bit of a tangent, but it really is, you know, like I've been fascinated by. Agents, but in a way that you have more control over them. You know, like how you can actually, it's less of it just goes and does things for you and more like it collaborates with you and it's actually a good transition to kind of the stuff I've been playing with on the local side of things. And I'll just kind of give a, a quick transition there too that ties into the cowork stuff and the dispatch stuff is that, you know, recently. Anthropic put out, I think it was a, the head of Claude Code or his attorney or whatever his name is, essentially put out a, a message saying, you know, during business hours, if you're on like the $20 a month plan, like you're gonna see way, you know. Smaller thresholds for your five hour stuff. And

Brendan

Yeah,

Baird

I mean, I, I noticed it right away, essentially, like I had, I think three or four different cowork processes that were running on a schedule, and I noticed like I couldn't even plan a full feature that wasn't even that complicated. Without using an entire five hour threshold in like 15 minutes, like it took no time at all. And I achieved pretty much nothing. And I mean, sure I could pay more money, but I also, you know, have five other subscriptions for AI tools that I'm using. And so I'm already paying, you know, a hundred bucks a month for everything I need. And it's like, I mean, honestly, and I'm curious to see your take on it too. It feels like since Anthropic has like taken off, it feels like the quality has actually like, come down a little bit. It like, it feels like they're stretching their resources so far that it's not as good as it was. And that could even be like, Claude Code has changed so much too that I, I just feel like there's some, like some frustrating things that I've seen with Claude right now. And you mean it's still good? You know, maybe other people aren't noticing stuff like that.

And it also could be a, if you're on the Claude Max, maybe that solves the problem too, but it's kind of made it so I wanna explore a different avenue and kind of as this was happening, you know, Gemma four came out and I don't know, have you heard much about Gemma four at this point?

Brendan

I saw some posts I haven't dug into it. It it's, Google's new

Baird

Yeah,

Brendan

what,

Baird

well, Gemma has been around for a while. It's been, it's essentially DeepMind. They've done like local models, like they've been focusing on small models and they've always been like relatively interesting, but not super useful. But essentially the new Gemma, they've gotten it relatively close to a frontier model. In, I think it's 31 billion parameters is like the biggest one. And which is, you know, I think what 19 gigabytes when you download it, which you can run on a computer like it, it struggles a little bit on my laptop, but they've also got like a 2028 and then like an 8 billion parameter one. So they've got a couple versions of it, but essentially like the top end version. When they benchmarked it, it was like ChatGPT or GPT-4, level intelligence on something you can run locally, which, you know, like ChatGPT-4 nowadays is relatively behind obviously. But for something you can run locally, it's like that kind of changes things. 'cause I've also done a little bit of research recently. I mean, actually a lot of research on agents lately, and one of the things I've seen is that. A lot of genic work doesn't have to be your, you know, true like state-of-the-art model.

Like you can use dumber models to do a lot of these like basic things. And so that kinda got me thinking like, can I, and like, sure, I could test out OpenClaw or Hermes or whatever and see if I can like get that to use a local model. You probably can, but I don't always work like that. I'm like, Hey, what can I build? And. Also because you know, I had some frustrations on the Claude Code side of things. I've got windsurf that I still use and I'm paying like 10 bucks a month still for it. So I was looking around in there and I noticed Kimi K 2.5 is sitting there for free, so you can just use it for free. And I've heard. Interesting things about it. And so I was like, you know, I'm gonna just try building, you know, kind of a age agentic framework, a little bit of like a Claude Code, but kind of a, yeah, I didn't know where I was going with it yet, but like, just something that I can start pushing into some of this type of stuff, but in a way that's more cost effective, you know? And see how far I can actually push. Local models and free models and, you know, open source models to see like what can it actually do? And, you know, it is been interesting.

I, I've been working on it for two nights, I think. So maybe like a couple of hours total. And I essentially have a full app running. It's got both a chat mode and in age agentic mode. And it doesn't have a background process yet. But the, the fun thing I've done, and I don't know if you've ever played this before, is. I actually go out and so like, I actually, you know, Hermes is actually, um, open source, you know, and so is like, we now have like the Claude Code, you know, iterations that people have created in Python. So I've actually taken those and you know, I take open source stuff and I don't have it like rip everything, but I start like building things with my opinion of what I wanted to do. And then I'm like, okay, look at this open source project. Just see like what are the things that it does well, you know, what are the cool parts of it? What are the things that would apply to what we're doing here? And it's like, make a plan of how we can actually build that in. And I know a lot of people will like, use tests essentially. Like they'll rip off tests and just throw them in and say, okay, build until the test pass. Like, I don't know if, if you've heard.

CloudFlare recently has built both an open next JS as well as, I think it's a CMS, called M dash, which is essentially like WordPress, but made in a way that can run on CloudFlare. And it was like completely, I mean, maybe not completely, but essentially like they ripped. The open source pieces to this, took the tests, put 'em in place, and then had AI build until it achieved something that could run on CloudFlare. And so, you know, what I've done isn't quite that far. It's not like taking the tests and doing it that way, but it is very much like evaluate the code and find the cool parts and then figure out how we can put it in what I'm working on. So it's been really interesting to see how far I can push it using local models and like, it's not super interesting yet, but right now I've, you know, hooked it up to some cps and also like tli. So the search, uh, provider is like, so it's like an AI search provider. And so like it can do research for me and put together reports. It can work on my local file system and it actually like put together a React app for me, like. It still needs some structure to be like, what does a good React app look like?

But that's another thing that I'm kind of playing with as I'm building it is like how, 'cause like there are some limitations, like the model, like the cool thing is it can use tools, which is a huge thing because it used to be like the smaller models were really. Not very good at tool usage, but

Brendan

Yep.

Baird

of the things that they did with this is actually make it so that it's pretty good at using tools. And so, you know, I'm kind of doing a like, and I, I will say I also put skills in it, so it kind of has like skills, like cloud skills and stuff. So it's got slash commands and all that. And what I'm trying to test is like if you have kind of an. Dumb model, so to speak, but you have this infrastructure around it that helps give it good context and good, like, you know, the skills are another piece to it where it's like it can actually use that. It also has a memory system to it. So like you can actually like ask it what you've talked about before and it can go back and look through your SI session history and stuff. And like, yeah, like I, there's still a lot of work to do to get it to a point where it's super interesting, but just like. Like I said, it's been like two nights that I've been working on it and I've gotten it far enough where I'm seeing capabilities that you see in Claude Code and Codex and like all of the more modern things that I've essentially built up in a couple nights, you know, and they're working relatively well.

Like the agent actually goes out and plans its steps and we'll like show you progress and all that type of stuff. And the cool thing is, is I'm doing it without. You know, Claude Code building it, or Codex building it. It's like using an open source model to push on it. And you know, it, it also kind of goes back to, um, I think we talked about it a couple weeks ago. There's this whole philosophy that like. Agents actually work better when they're working in a file system than when you try to put like everything in a database. And so that's another piece that I'm really focusing on is it's actually a, are you familiar with Tauri? T-A-U-R-I?

Brendan

No. Nope.

Baird

So it's kinda like, I mean, because this is cool, it's a rust based. Program essentially that it's like electron, but essentially it's built in rust and then it like has a bunch of native browser stuff built in. And so it's supposed to be really, really fast, which is kind of cool. And so, and again, like this is an experiment, so I'm like, Hey, let's push on something I've never really used before to see how far that can go as well. And so it is completely local, you know, it runs locally, it has a file system, like I have it, so it can have like a working directory to build. Stuff in as it goes and you know, it, it's really, again, like I think this is something I'll need to report back on as I push further on it, but it's been fun to experiment with like, what's a different path, you know? Because everything right now is so. You gotta use the state-of-the-art model, you've gotta use Claude Code. You've got, like, everything is very much in the ecosystem.

And so it's fun to see, again, like, I mean, we're vibing with AI code here, and so it's fun to, to do that and, and see how far you can push something without, I mean, I've, I've written maybe like three lines of code on it, you know, it's not like I'm sitting there pouring a ton in, it's just like directing where it goes and stuff like that. So, no, it is been pretty cool.

Brendan

Sounds really cool. Yeah. So just a couple questions I

Baird

Yeah.

Brendan

you, you mentioned Gemma four,

Baird

Yep.

Brendan

you also mentioned Kimi. So which one are you using primarily? Kimi.

Baird

So Kimi is, I'm using in Windsurf to actually build, so like that's the Claude Code replacement. Essentially Gemma four is the brain that's actually running the agents in the chat, in the app itself. So,

Brendan

okay. So the, the app, when you run the app itself, it

Baird

yep.

Brendan

Gemma. When you're building the app, you're using Kimi at the moment.

Baird

Yeah, so I'm using LM Studio, which I mean, I could actually swap out Gemma for any open source model I can run.

Brendan

Mm-hmm.

Baird

it just because I wanted to play with Gemma. That's what I'm using right now, but, and which is kind of cool, like as things get better, you're not stuck with anything. Like if you can run it on your local machine, you can download it and use it as the brain essentially.

Brendan

Yeah. Yeah, that was one of my other questions was gonna be like it. It's, I would expect, and I think you just answered this, but just to make sure I fully understood, like you can change, you've built this in a way that you can pick whatever local model is your

Baird

Yep. Yeah, exactly.

Brendan

that's part of the LM Studio thing.

Baird

Yeah, and the cool thing is, is like as long as you have it loaded in LM Studio, you could actually have like the option in the app to switch models as well. It's just you have to have 'em loaded and running. And I mean that's something that could expand too, where. You know, this could tap into like an open router or something like that if you really wanted to like move to a remote model. But for the moment it was just really, I wanna see how far I can push local only and see like what I can do without the limitations of a, you gotta pay for it because sure, I'm happy to pay for it, but I've got so much money invested in APIs. Credits all across the different services that I'm, like, I, I just, I wanna see how it works,

Brendan

Yep. Yep. You don't wanna pour your entire kids' college funds into AI. Research

Baird

especially on the, the agent side, because like I said, you know, I've done

Brendan

fast.

Baird

stuff on, you know, paid models and database environments on remote environments and like. I mean, even the best agents out there are still not great. You know, like there's still a lot of failing and like I always hesitate when I'm like, okay, I know I'm just gonna be burning cash here. I'm not gonna get a ton out of just like hammering on this. And so the cool thing is, is that this is something that if I can run it locally and I can get something working well. It should scale really well too. Like it should go to a state-of-the-art model and be amazing theoretically, if I can get it to run on a, you know, eight gigabyte model locally. So it's kind of, you know, what do they say? You know, constraints are, uh, key to creativity, you know, it's like how can you make things better with the constraints that you have? And so, no, it's been an interesting experiment.

Brendan

Yeah. That's really cool. I'm excited to hear where that goes. You know, in some other thoughts I have. Um. That you kind of touched on, but I've just, I guess things that I've seen that as you were talking about how you're building it, like it'd be cool to see you build this sort of functionality in as well. It's like I use Conductor for my, well, it's one of the harnesses I use for Claude

Baird

Yeah.

Brendan

and you can actually, I've. This is a quick update, but I've also started using Codex too

Baird

Oh, nice.

Brendan

now.

Baird

Yeah.

Brendan

and so that maybe could be a whole nother thing and I don't wanna get bogged down, but the point of bringing it up is that I have noticed when I'm writing, well, when Claude is writing code in there, I haven't noticed it with Codex yet, um, that it will switch models. Like they

Baird

Okay.

Brendan

something into conductor.

Baird

Sure. Yep.

Brendan

when it's writing the code, it's using Opus 4.6. When it's writing commit messages, it's using haiku or whatever, which is makes sense in that. Like, it got me thinking like you could probably build something like that

Baird

Yeah.

Brendan

where like if it's gonna, if you want to cut down on the cost, you, your thing could build, build with Opus 4.6

Baird

Yep.

Brendan

5.4 or whatever. But then everything else uses Gemma or you know, the free models.

Baird

Yeah. No, and that actually brings up another thing I was thinking about is that I also, as I do this and push further into it, I do want to test. Kind of the relationship there. Like if I have say, Opus plan a feature, how does Kimi do at implementing it versus if I plan with Kimi, how does Opus do with implement? Like is there a relationship where like if you just do the plan with a bigger model or a smarter model, can it help a dumber model implement? Better. And, you know, it, it could be like you're saying the other way around where it's like actually having the full implementation be the smarter model. But it, it's kind of an interesting experiment, I think to see like what relationship, you know, that has, and, and I do know that for a while here, like since Claude Code introduced, uh, just their planning mode in general. They switched over to Haiku for that as well, which maybe that answers the question is like, but I also know it's because it's faster, you know? It's faster, it's cheaper.

But that is kind of an interesting question is like, yeah, it's faster and cheaper, but does it do a better jet? You know, like, does it do or like, does the smarter model compensate for anything lacking? And so yeah, like that very Well, you know, I'm going against, against the grain here, so I'm probably, the question's already been answered. But, you know, part of this is very much just, I'm doing experiments here just to try to figure out like, kinda like what we were saying earlier. I kind of got stuck in the, I'm gonna use Claude Code or I'm gonna use Codex and I'm gonna like use all the stuff that's already out there. And I just kind of got to a point where I'm like, what else is there? You know, like, how else can we do this? How else can we push this to see what works and what doesn't? And you know, again, because it's taken me a couple nights so far, and I've already built almost a fully functioning application, it's not like I'm wasting months of my life. That I might throw away. Like, what's the worst that happens? I waste a week and I toss it. Who cares? Like it's not a big deal.

Brendan

Yep. Yeah, I mean that, that's the power of AI code and that's why we, we love doing it.

Baird

Yeah.

Brendan

Yeah, so I'm excited. You know, my, my last takeaway would be, expect there to be a name to this by the next week.

Baird

Well, I, I already have a name, but it probably doesn't work. I just jokingly because it's a local agent, I just called it Loki, but I'm sure like everybody's already used Loki, so if I ever push this further, it'll have to be rebranded, but for the moment, it, it's code name, Loki.

Brendan

All right, cool. Well, we will check in on Loki next week.

Baird

Sounds good.

Brendan

anything else you wanna add?

Baird

No, I, I think that's it. That's been occupying most of my time.

Brendan

Cool. All right. Well, yeah, I guess this is another episode and, uh, you know, like, and subscribe down below and, uh, we'll see you next time.