Conor

So Weewa is implemented in Rust. Dialog APL is implemented in a combination of C and C, but it's it's closed source, so you can't see it. J is implemented in C, albeit a version of C that I would call macro C because there's like 10,000 macros and you basically code in the macros, not an actual C. BQN is implemented in C, although the initial version that I used was a JavaScript implementation. Tiny Apple is implemented in Haskell, and Cap is implemented in Kotlin.

Bryce

You know, we might have services that are like 99.999% reliable, but we do not have software that is 99.999% bug free. Even the best software releases have some amount of bugs in them.

Conor

It's called map in Parrot C. It's also called map and parrot python. And the CUDA C backend uses a make transform iterator, and the CUDA.compute backend uses an iterator.transform iterator. So it's you know basically the same thing. Today with my co-host Bryce, we chat about ArrayBox.dev, some parrot algorithms, the future of agentic software development, and more. Alright, what should we talk about next? Should we talk about we've mentioned ArrayBox a couple times. Technically, Parrot is still on the short list of things that we have yet to talk about. But you said ArrayBox. Okay, so ArrayBox is something that I started back in January. It's now been live arraybox.dev for it looks like since February 6th or 7th. So over a month now. We're recording this on March 10th. And there's been just under a thousand unique visitors. I made the mistake of when I initially launched it. I didn't really look at how they were tracking the visitors. And I I only wanted like unique IP addresses, but they were resetting it every 24 hours. So pretty quickly I hit like 1,500 visitors, and I was like, that seems a bit high. And when I asked for an explanation, they said, Oh yeah, it's like uh resets every 24 hours. And I said, no, no, no, no, no, no, no, no, no. Like, just I want once they visited, I don't want it to count. I want to know like over time how many people actually visited. And it said, okay, and I'll store the IP addresses. And I was like, no, no, no, no, no. Don't store people IP addresses. I don't I'm tracking nothing here. And they were like, all right, I'll do whatever hashing coded, so you don't know. And there's been just under 14,000 code evaluations, and it's pretty it's pretty fantastic. So the idea of the website is there's a bunch of online REPLs for different array languages. There's try APL for dialogue APL, there's BQN pad for BQN, there's weewa pad for weewa, etc. Some of these are better than others, and I basically just wanted a one-stop shop where you could code in all of these languages. And so that's basically how I started. I went to I went to Cursor, 4.6 Opus, or maybe it was 4.5 at the time, and said, I want uh like an online REPL, and at the time I thought this is how the way would have to be built. Tiny Apple is built in Haskell, and they have a a WASM back end, so that can be client. Like Tiny What? Tiny Apple. Tiny APL. But uh the creator of the language pronounces it tiny apple. So the six languages that I support at the moment are dialogue APL, BQN, J, Weewa, Tiny Apple, and Cap. That's K-A-P. And so five of those are Unicode languages, and J is an ASCII-based language. And so in in order of the how I'm looking at them, Wewa is implemented in Rust. Dialog APL, and and most of them, I think actually all of them are open source except for dialog APL. So Weewa is implemented in Rust. Dialog APL is implemented in a combination of C and C, but it's it's closed source, so you can't see it. J is implemented in C, albeit a version of C that I would call macro C because there's like 10,000 macros and you basically code in the macros, not an actual C. Although, you know, technically a macro is C code. BQN is implemented in C, although the initial version that I used was a JavaScript implementation. Tiny Apple is implemented in Haskell, and CAP is implemented in Kotlin. And so initially I thought that WiWA, because it's in Rust and they've got a Wasm compilation, you know, backend, BQN because I had the JavaScript implementation via bqn.js, and Tiny Apple because it is in Haskell and has a Wasm backend as well. I thought those three you could just build, you know, the JavaScript WASM things that can live in the browser. And the other three would have require a server. And so that's what I went about doing that, got it all working. I actually thought because Kotlin has something called multi-platform Kotlin or Kotlin multi-platform. I think that's what it is, KMP. And they're they have the ability to do stuff across you know different platforms. However, I couldn't get this working initially. And so that's how the website started. I launched it at the beginning of February. Three of them were client side, and three of them were server-side, and server side was a laptop sitting down beneath my TV.

Bryce

That is a course, you being a mature and responsible adult, have moved it to a secure, you know, cloud environment, right?

Conor

No, no, it is still, and and it's not just any laptop. It is my my wife's old 2012 MacBook Pro, which barely works. I mean, yesterday, I I think yesterday the so I have an ArrayBox server manager that basically runs and serves endpoints. And it was running for I think like 200 plus hours, which 24, so it was like over a week straight. And we this is also our media laptop, so it's HDMI plugged into our TV.

Bryce

So if you want Connor's Netflix password.

Conor

And and admittedly, when I when I set this thing up, at some point I said, you know, put this in uh you know a Docker sandbox and you know, make sure it can't be hacked. And when I launched it though, it as soon as I put it live, I I didn't launch it on the YouTube channel, so I just told a few people about it. And you were, I don't know, probably the sixth or seventh. And everybody that I told immediately started trying to go like run shell commands because APL, dialogue APL, J, and CAP all have the ability to run like you know, shell commands locally. And so I started getting all these messages of like, oh, I can look at the contents of this container, and I'm like, okay, that's great. And but like yeah, it didn't matter that much. But you pointed out to me when you so I when I told you about this, you immediately started trying to hack this stuff. You might be thinking, how does Bryce know Jay? Bryce does not know Ja, but AI knows Jay. So Bryce is like showing me that he's like looking at the contents of the container. But you asked or pointed out at some point, you were like, if these are impl you asked what they were implemented in, and I said J is implemented in C, and dialog APL is in C and C, but it's it's closed source. And then you said, well, if it's implemented in C, can't you get an inscriptum build running? And I was like, that's a fantastic point and idea. Why did I not think of that? Sure enough, I went to the J source code, they already have like flags for an inscriptum build. So I just had to build it. And there was actually another uh REPL online called the J Playground, and I actually just asked Cursor and Claude, like, what are they doing? And they said, Oh, they're actually using an inscriptum build as well. So I went and set up the and that's the thing, I don't know how to go and you know set up the and I remember actually several years ago I did a 10 or 8 or 10 hour live stream where I was porting the J source code to C, and I swear to God, I can go find it. I spent the first five hours of that live stream just trying to get J to build. Like this is what we're talking about in terms of like a hundred X like productivity multiplier. I now ask Cursor and Claude to do it. It not only do they go and do it, they get the inscription one working, and like 10 minutes later, I have my whole back end written. It's now client-side, it's no longer on the server. And so now we're down to two languages on the server. And I was like, well, if I can do it for Jay, I certainly should be able to build this from scratch. Because at first I was just trying to pull like executables from other places and not build stuff from source. But then I basically, once I had the website up and running and people were trying to hack it, I was like, all right. So I just said this is Kotlin multi-platform. Can't you build this and get me like a Kotlin native or like running in the browser? Sure enough, five minutes later it's working. And so now we're down to just dialog APL, which is the only one running on the server. And even that was a bit insecure. But I asked a couple folks about it, or I'm not they had told me about it, and so then I said, I told I asked Claude and Cursor, like, there is some kind of security thing that try APL uses, and that's called like Safe 3 dialogue. So it went and found that. It took it a while for it to iron out all the kinks, but it's just wild. Like, you know, you're you're a passer buyer, you don't know anything really about like J or these languages, but you using AI figure out basically how to hack my machine, but then you come along with an idea, and that idea like I take, send it back to the AI, and it's just like it is just wild that like I have wanted this kind of website like forever because I'm constantly for my YouTube videos, you know, testing out these esoteric array languages like Cap and Tiny Apple that like admittedly not many people have heard of, but there's some really interesting ideas in those languages, but the online REPLs are pretty bad, and like navigating the docs and like switching, it's just cumbersome and not the experience that I want. And now, like when I want to bring up docs and stuff, I just hit F1 next to a glyph and the the the code editor just slides to the left. The docs show up like on the same screen. It looks beautiful when it works. There are some bugs here and there, but it's just it's just wild that this is the world that we live in. And I just I thought it was so awesome that you you just like it reminds me of uh Bill Burr has this like Steve Jobs joke where he's like, What did Steve, what is the big deal? Why are all these nerds such a big fan? You know, what did he do? He has this bit. I'll cut it in here, but my impression of it is he's eaten an apple going, oh, here's an idea.

Bill Burr

Big smile, big smile, make it happen. Actually, nerd Jesus died in the last year, right? Steve Jobs. Yeah, he died, right? I know, I know a lot of nerds here tonight. I know you said. I didn't get it. I didn't get the big deal they made about that guy. What even he told other people what to invent? I want my entire music collection in that phone. Get on it! And then these poor nameless, faceless scientists gotta go in a back room and figure it out. Steve Jobs just walking by. I don't hear any thinking going on in there. Strutting around the office, eating some pretentious fruit like a pear. Throwing on ideas. I was reading a magazine the other day, turning pages, you know. I like to turn pages on a screen that aren't even there. Yeah, wrap your fans around that guy. See you in a year. Where are you going, Michael? Big, little, big, little, get on it.

Conor

And he's he's doing the gesture with his hands to like zoom in and zoom out of a photo. And so he's just this, and like Bill Burr's like a joke is that he's just this guy that would like walk around throwing out ideas and be like, I don't hear you thinking hard enough.

Bryce

Well, I feel like that's been like a large part of my value add to Nvidia is like every now and then I meet with like a wide variety of people, and sometimes I'm like, hey, why don't you try this or that? And then like somebody comes back to me like three months later and they're like, I spent the last three months working on that and it worked out. Or as is the case, somebody comes back a week later and they're like, Yeah, I looked into that. It's a terrible idea. That's like 90% of them. But there there is value.

Conor

That's the thing, is there is like extreme value in insights. Like I was about to say ideas, but you know, taking from Michael, it's like ideas that are like, you know, I don't know, what's the word? Like meaningful, you know, or worth keeping in in the words of Michael once again. It is like small things that I think about, just like experience, like UX things with my podcast player. It it's so nice now when like these things work. Like every once in a while you'll accidentally end up like hitting the hundred percent mark on a podcast that you already listened to, and I have this like history now. So like I don't have downloaded and up next. I just have a cue, and when you're you listen to something, it gets put in history. And the idea is you might be able to share like your weekly recap on social media at some point, but every once in a while you listen to something twice. You can just go swipe left, hit a delete button on your history, and maybe you don't even want to share it. That's why you don't want it in your history for whatever reason. You're ashamed. I listened to the reality receipts podcast because I saw a reel on Instagram on the Love is Blind season 10 finale episode because Shima and I watched Love is Blind. And, you know, am I ashamed that I listened to the Reality Receipts podcast? Clearly not, because I'm mentioning it here, but others might be. And they might want to remove it from their history before they they, you know, share some weekly recap. And stuff like this, you know? If more people had great ideas and were implementing them, we would all not be whining about the quality of the software that we use on a day-to-day basis, you know. Isn't have you seen that Jonathan Blow 10-minute clip where he just like tracked everything that was like broken with the software he was using over like a one or two-day period?

Bryce

No. But um, yeah, I can I can imagine. I can imagine.

Conor

And his point is that like we we are like desensitized to how awful software is. We're just used to like, you know, uh swiping up on the app to close it and then reopening it. Like we are so desensitized to like, oh, it's not working, I'll just restart it. Like and when it shouldn't be that way. It shouldn't be that way, you know? Like how why how why is it so hard to get things to work correctly? It's like it's like you know, the the complaints that you have about cursor when it was freezing, you know, it doesn't do that anymore. Why? Because you fix it, you fix it.

Bryce

Yeah. I mean, I I certainly, you know, it's it's it's funny because I do go through that exercise, but like almost only when I'm like using some open source library or like I'm using like one of NVIDIA's products. Like that that Google Doc that I showed you earlier with like a like you know, a hundred or two hundred to-dos and like some of those are things that are to-dos for the project I'm working on. Some of those are to-dos for like other projects, like because I'll like collect like a list of you know what's wrong, what could be better. And when you go through and you like actually do that, like it is kind of amazing just like how how quickly that list accumulates. Yeah, it's it can be it can be shocking. And yeah, I mean we take a lot of these things for granted. That's the reality.

Conor

Yeah, it is. And that's the thing, is like, even though the pod god app still has its issues, like it has less issues than the other podcast players that I used to use. Like the the cast box had gotten so bad that if you were just not on Wi-Fi, like the app would almost stop working. I think I I think I recounted that, you know, like you'd have to go to the downloads because the queue would just entirely disappear.

Bryce

Like it was it was why it wasn't network dependent, it was the Amazon Prime Video app for like the last week for me. Yeah, I cannot watch any content when not connected to Wi-Fi. I've checked all the settings, I have uninstalled and reinstalled the app, I have cleared my cache, I have like cleared all my data. It's like nothing, nothing I can do can fix it. Or I remember one time, right after I got my iPad, there was a period of time where like my iPad, the screen brightness setting was like always messed up. And if you set it to full brightness, like any video that you would watch would just like look completely wrong. And it was like that for like you know a month. And it's like, how does how do big companies in a scale, how do like things get out in the wild like that? You know? And the reality is that you know, quality control is just not, you know, we might have services that are like 99.999% reliable, but we do not have software that is 99.999% bug-free. Even the best software releases have some amount of bugs in them.

Conor

Yeah, this, I mean, this is something that I've really started to think about in terms of like how do you how do you develop in a uh you know, vibe-coded or agentic, you know, lead-driven world. And I was talking to Ashramansanellia, a fellow NVIDIA, about this the other day. And I was, I was, I was, I was thinking about this while I was like falling asleep, is like, what is gonna be the new, like, you know, there's TDD, test-driven development, DDD, domain-driven development. There's a bunch of these different ones. Uh, and I was thinking of the title of a talk, that's odd, for ODD, which is like outcome-driven development or something like verification-driven development. Because like, right now, one of the issues I run into with the podcast app is that sometimes I'll fix something and that'll break something else. But like, you know, I I know that there are some kind of Android, you know, UX, you know, it'll have some like workflow and it'll test stuff, but you know, I have to look into that because it's very tricky for like downloading an episode and then listening to it and it hits the hundred percent mark, and then it should automatically remove like testing, and so the point is is like even if we're not talking about like some hobby project, we're talking about work and we're trying to write some kernel that you know is r is is speed of light. And like are we just going to be like setting up, you know, you've we we have unit tests and we've got lints.

Bryce

We're all gonna become to some degree DevOps engineers and test engineers. And what I mean by that is that the importance of testing and verification will only increase. Of course, of course. You know, like like I think uh in if you think about it, in the past, if you had your own project, you would you didn't necessarily need to set up like code formatting or code linting rules or like all these various checks, because it was like just you working on it. But like my experience has been as soon as I have a project where I have external collaborators, then I have to set up all the code formatting and everything so that like everybody, everything is consistent across the code base. And I think it's like sort of similar here where it's like you have this external collaborator on any project, and like the thing that I spend my most of my the majority of my time thinking about is like how can I add more checks or verification or validation. And it's funny, I recall a conversation with Titus Wenders like in the early days of AI, where his proposition is basically you can't have the AI write the tests, that like the tests need to be written by the people, because if you have their AI write the tests, then the they will you know write tests that will you know basically just test what the code currently does, not necessarily what it should do. And I think time is like this is like in 2022, 2023. I think things have evolved a little bit now, and like obviously if you're writing tests, like you know, you're gonna use AI to help you write the tests. But I do think that like that's one of the places where I spend the most time thinking or like manually reviewing things, like in the setup of the tests. Because in an ideal world, I shouldn't really need to review the code because or sorry, I shouldn't need to review the code really for functionality. I like the functionality should all be verified by the tests, even the performance should ideally be verified by the tests. Maybe I review the code for like maintainability or style. But like I should know if I've got some AI written change, I should have a high confidence about whether or not that change works based on the the test results.

Conor

Yeah, I want to show something because it's like we're gonna, it's gonna we're just gonna be bootstrapping like using AI to set up these frameworks for testing and performance. And it's just it's AI all the way down, folks. But so I've got you know, we're probably gonna have to defer to the Parrot, you know, the parrot conversation that we've been saying we're gonna have for a while. But so Parrot, we released back in October, depending on where you're when you're listening to this, it was uh 2025. And we're chatting about this in March of 2026. And so we're in the middle, we released that on NV Labs, uh, which is the research GitHub repo repository, and we're in the middle of building, and that was so that was built on Thrust and Cub to CUDA C libraries. And we're in the process of building a Python version built on top of CUDA compute, which is a basically a kind of equivalent of the Thrust algorithms in CUDA C ⁇ . We've kind of brought those to the Python ecosystem. And so, you know, I've obviously been using AI to you know build this library and at some point, you know, uh working on a potential paper, and I went and asked it to build a table of you know the names of the different functions and then the implementations, kind of you know, not succinctly, not the full implementation, but a lot of these operations, here I can blow it up a little bit, are implemented in terms of other things. And so I'm not that old, Connor. It it it did this, uh well, I mean the text is a little bit small. It it one-shotted this. Like, and so it basically shows at the top of the table, and and there's there's a bunch of pages of this stuff. It's at the top it says it's called map. So for a unary operation, you know, what you call a transform in C, it's called map in parrot C. It's also called map and parrot Python. And the CUDA C backend uses a make transform iterator, and the CUDA.compute backend uses an iterator.transform iterator. So it's you know basically the same thing. But you can see that like I'm missing the IDiv from ParrotPython just because I clearly didn't get around to adding it. And on top of that, like when you go down to more interesting operations, so the the where function, which is One of my favorite we'll do a whole episode on where at one point. It is called where and indices in NumPy, if you're familiar with it. We've talked about it on ADSP before. Look at that. A little algorithm content for the uh the listener here. Where is a function that takes a Boolean array, so zeros and ones, one for true, zero for false, and it returns you the index that corresponds to the trues, to the ones. So if if you're given 10101 for a zero indexed implementation of where indices, it'll give you back zero, two, and four. And so the implementation for this in CUDA C is range with the size passed to it, dot keep this. So you're basically range is the equivalent of IOTA and you're doing it for size, which is the size of your array, and then you're doing a dot keep this where this is the mask of ones and zeros. So you can think of keep is basically a filter. A filter takes a unary operation and does a compaction based on whether the unary predicate returns true or false. Keep is the equivalent of that, but you're given a mask that is equivalent in length to the array that you want to do the compaction on. So another way of thinking of this is well, I guess copy if, where you're giving it the stencil overload. Listen, that's not exactly the same thing. But think of keep. Basically, another name for keep could be filter based on Boolean sequence instead of unary operation.

Bryce

Yeah.

Conor

And anyways, when you when you go to the Python implementation, it's range, aka iota for length, because that's what they call you know size, and Python is length. Dot keep self. And so it's it's just it's it's this kind of stuff where you're building a summary of like implementations of two different libraries to see what are the missing pieces, where do they differ.

Bryce

I can't believe it got that like it did this on like the first try.

Conor

One-shotted, one-shot in. And it's it's just so nice. It's so nice.

Bryce

I can write your paper for you.

Conor

Well, I mean, I'm gonna write the paper myself, but this table, this table is just like a deterministically, that's just like a waste of my time if you can get but like the idea of this kind of stuff, right? Like having some another dashboard or some kind of thing where you're able to have an LLM, you know, generate this thing and check, you know, where the changes are. You know, like I said, is I I feel like unit tests and like lint tests or static tests, this is just like the small, like it's the beginning. Like now, yeah, I f I feel like in the future at NVIDIA we're gonna have a bunch of like performance, you know, because we're we're getting we're gonna get these AI produced diffs on our libraries.

Bryce

And we're gonna have to be able to do performance validation on it. Yeah, totally. Like, like, like automating automating the like code optimization process. Yeah, that's like a very interesting topic. Like, you know, you can it's not because you need to not just do performance testing, but you also have to hook in all of the various profiling tools to the uh to the LLM, and you have to make sure that it's it's reward mechanism, that the way that it's measuring is is you know gonna be accurate, that it's it it like you have to have your functional checks correct to have your performance checks correct. Because if you've if you don't have really good functional checking, then you can't even start performance checking, because then if you start doing performance optimization without the good functional checks, then you'll regress functional behavior. So you have to have really solid functional testing, and then you have to set up good performance validation that is not going to be accidentally cheated, and then you have to like hook up the tools to all the various different ways that we identify and fix performance problems.

Conor

Yeah, I mean it's such it's such an interesting like problem to think about solving of like you can imagine a world where we build systems by launching these sophisticated LLMs to generate code, but there's admittedly going to be a bunch of issues, right? Like, but how do you and and that's already what these Frontier labs are doing, you know, how do you build context or add context to their prompts so that they give you better code and better whatnot? But you can imagine like some like some static analysis tools in C and like Clang, I know they have like a complexity metric where if you have like a certain amount of nesting, you know, it f it has some flag, right? So like you can imagine, and one of the one of the issues that I run into with ArrayBox is it duplicating code instead of having like a single source of truth for something that shouldn't be duplicated, like the set of primitives. And if you can build that into like some kind of unit testing framework that it it is able to track when you duplicate something that shouldn't be duplicated, if you like if you can build enough like guardrails, whether that's for unit tests or for performance or for complexity analysis or like I don't know what you call it, like the code quality, we're just off to the races then. Do we do we you know we are just gonna be people that kick these things off and and then that's it? Uh you know? Right? Yeah, ship it. Yeah, we retire.

Bryce

I don't know. Are we gonna I I don't know about that, but yes, yes.

Conor

I mean my I already have my retirement plan. I gotta I'm just gonna do code I've decided. Array languages, worst case, they're just fun. They're just fun. And I like thinking about tacit programming models. I like thinking about how awful mathematics notation is and how Iverson tried to reinvent mathematical notation and failed. And there's no reason we can't, you know, we can't change mathematical notation. That's what we you know. I feel I think I'm gonna take the Bartaj Meluski path. You know, we talked about the four horsemen of the Pacific Northwest C meetup. There was Andre Alexandrescu, Eric Niebler, Walter Bright, and Bartaj Meluski. And all were hanging out while they were falling in love with functional programming and Haskell and whatnot. Eric took the pragmatic path, and he went and tried to bring the the power of functional programming to a very popular language like C via ranges. That's not gonna be you. Walter went and tried to do a whole language better than C called D. Andre joined that fight at a certain point and was working on the D language for like a decade, but then came back to C. And Bartaj Meluski, he just went and tried to evangelize functional programming and category theory for his whole career, basically.

Bryce

Yeah, I think that's have you seen him recently? I haven't seen him in so many years. He's a great guy.

Conor

I met him whenever I was at Zuri Hack, which was definitely like three or four years ago. I actually why did I go to I went to Switzerland for a couple weeks for vacation, and the Zuri Hack ended up being there, and so I popped in and then I met that I met him in person there. And then I've seen him on I think a couple online talks since then, but not in person. Yeah. Have you had you met him in person before? I I assume I guess C now.

Bryce

The original BoostCon or the last BoostCon, 2011, I think I met him. Maybe maybe he was at like one C now or two C now.

Conor

Yeah. He's given a bunch of talks to C now that I've seen. I haven't been. Although they were they were before the recording quality was like peak. Yeah. Good old days. Yeah. Good old days.

Bryce

Oh man. So yeah, we should talk about Parrot, but maybe we maybe we're gonna save that for next time.

Conor

Yeah, it's funny too, because someone was asking Discord if I had released the YouTube version of the Parrot Talk, because the C under the sea one still isn't out. And I had said that I was gonna record it and put it on YouTube back in like November, December, and I still have not done that. Mostly because I feel like this work that I'm doing right now, I want to give like an updated version of that talk where I talk a little bit more about the implementation. Because the talk that I gave at C under the sea was mostly about usage, not about the implementation. And admittedly, I think the usage is more important, but the implementation might be more interesting. But anyways, be sure to check these show notes either in your podcast app or at adspthepodcast.com for links to anything we mentioned in today's episode, as well as a link to a GitHub discussion where you can leave thoughts, comments, and questions. Thanks for listening. We hope you enjoyed and have a great day.

Bryce

Low quality, high quality. That is the tagline of our podcast.

Conor

That's not the tagline. Our tagline is chaos with sprinkles of information. So when it when when's your what's your due date? Due date is September 15th.

Bryce

Okay.

Conor

Yep. Pretty pretty exciting, I guess.

Bryce

Yeah, it's pretty exciting. I'm excited for you. I'm sure you're gonna be a great dad. We're gonna have lots of we'll have lots of baby pictures. That's awesome. Yeah.

Conor

I will not take any pictures of my baby.

Bryce

You won't take any pictures of your baby.

Conor

No, just kidding.

Bryce

So you you're gonna take you're gonna take some time off? Yes, at some point. But like not wait when the kid's born.

Conor

You don't know what your plan is yet? We haven't really figured out exactly what we're doing there yet, but because we still got six months or something. Yeah. Cool.

Bryce

Exciting.

Conor

Well tell sh tell Shima, congratulations. Yeah, we gotta get I gotta get the pod guide editor done so I can just hit the button. Otherwise, we might be taking a a break. But uh I I I we've made it what 280 episodes without taking a break yet. A child shan't stop me.

Bryce

He says now. He says now.

Conor

Yeah. We'll see though. Maybe, maybe, uh, maybe we should record a bunch of like a month or two worth of uh we'll we'll like uh we'll buffer them up, right? Yeah, yeah. And just say, listen, folks, you're gonna get 20-minute episodes for the next two months, and then I'm gonna resurface in in November.

Bryce

Alright, I should uh I should probably call it because I gotta go make all these slides and uh Yeah. Are you what you you're doing training or just giving a talk next week? I'm doing both, and I've just been oh man, you know, I feel like the AI makes things more intense. Like I feel like I'm working more intensely than I ever have in the past. And I don't know, I just I'm almost like I almost feel a little drained. It'll be nice to I need to find a time to take a little bit of a break after GTC.

Conor

Uh yeah. Yeah, no, it's true. I mean, I said that whenever the last time we recorded is that like I work way more now.

Bryce

Yeah.

Conor

Because definitely think that's true. Yeah. I I just feel like there's like a window here too. Like if I'm gonna do like everyone's gonna be trying to do this in a year because it's gonna be you're not gonna need to know what like ADB is and how to host a website and how how to whatever. Like there is there's gonna be like a tsunami of like everyone can do everything for a fee. All right, buddy. Good luck. Talk to you later.

Bryce

Congratulations again.

Conor

Yeah, yeah, thanks. It's very exciting. Later. Later.