0:00Jordan Metzner says he wanted to build something like an algorithmic DJ where every press of the AI button generates a different version of a beat and changes the sounds it uses.
2:03Jordan Metzner says Strudel is a JavaScript live coding library for music composition he found via TikTok; its code is easy to read and edit, LLMs write it well, and its community has remade songs like Charli XCX's 360.
5:08Jordan Metzner says his tool has two decks with multiple channels, master effects and an AI button that writes new Strudel beats. He built it mostly with OpenAI Codex and says it would otherwise have been a heavy, expensive build.
8:25Sam Nadler says AI lets people get hands-on with crafts that took years to master, and he could build a basic mixer in a couple of hours. Jordan Metzner says LLMs are now affecting music after video and image design.
9:47Jordan Metzner says he gave Claude Google's 57-page 10-K and asked for a CFO dashboard, getting an executive summary plus charts on revenue, net income and free cash flow. Sam Nadler says OpenAI's output on the same document was weaker.
13:55Jordan Metzner says Codex was underdeveloped when he tried it months ago, but over Labor Day it built most of his DJ tool and the team is moving to it as the best model available, though that may change.
15:54Jordan Metzner says OpenAI paid $1.1 billion for Statsig, an A/B testing and feature flag tool OpenAI already used, and expects more deals as OpenAI pushes into enterprise and coding, where Sam Nadler says Anthropic leads.
18:05Jordan Metzner says Figma shares fell about 15% even though revenue grew 40% year over year and it slightly beat expectations; he thinks Figma is expanding in enterprise and has capital to acquire complementary AI tools.
19:33Jordan Metzner says Google settled the antitrust case without having to sell Chrome, good news for Google but mixed for Apple. He adds that You.com raised $100 million while focusing almost entirely on enterprise.
20:41Jordan Metzner says back office work suits AI well and Ryz Labs automated almost its entire invoicing with AI. Sam Nadler says executives want AI everywhere but people in the trenches show some resistance until the culture changes.
Strudel AI DJ Mixer: Live Coding Music Built With Codex: the key questions from this episode, answered with quotes and timestamps.
Jordan Metzner [0:00]: So I wanted to make something that was kind of like, Algoriddim djay. So every time I hit AI, I'm getting a different version. And I kinda change the sounds that it's using.
Sam Nadler [0:26]: Hey, everyone, and welcome to built this week, the podcast where we share what we're building, how we're building it, and what it means for the world of AI and startups. I'm Sam Nadler, cofounder at Ryz Labs, and I'm joined each and every week by my friend, cohost, and business partner, Jordan Metzner. Jordan, how are doing today?
Jordan Metzner [0:49]: Hey, Sam. How's going? Happy to be back. Episode 11. Lots going on in the AI world.
Sam Nadler [0:54]: Episode 11, and we have a full docket today. But before we jump into the agenda, please remember to like and subscribe. We have a new episode out every Friday, so like and subscribe on your favorite podcast platform, Spotify, Apple Podcasts, YouTube, wherever. We'd love to get in your feed. So, with that, I'm gonna jump right into the agenda. Is that okay?
Jordan Metzner [1:17]: Yeah. Let's go for it.
Sam Nadler [1:18]: Okay. So like every week, we're gonna cover a recent tool that we've built. This week, we're gonna cover somewhat of a passion project of yours. Then we're gonna jump into a use case of Claude, which maybe people are familiar with, but it's something we use all the time and I think is quite impressive. And if people aren't familiar with, they just, hopefully will unlock something that has been really beneficial to you and I on almost a daily basis. And lastly, there's, every week, some hot AI news, market moving news this week, but seems like there is every week. So with that, I'd love to see what you built this week, Jordan.
Jordan Metzner [2:02]: Yes. So busy week. I was scrolling and perusing through TikTok, one of my favorite pastimes, and I found some TikTokers making music using code. And what they were using was a library called Strudel, which is a JavaScript library for developing music, and I'd never heard of it before. So I took a quick screenshot of what they were working on, and then started to explore it a little bit. And Strudel is this library built off of another library that is essentially meant for music composition. Cool thing about Strudel, it's got a really easy language to be able to read. It's pretty editable. And as you know, LLMs are great at writing code. So, it's a live coding platform for music and allows you to make edits in real time and do a bunch of other stuff. It has a pretty cool online community. The community has done all different cool things. So I think before I show my demo, let me just show this one. So this is some people using Strudel to make a Charli XCX song, a remix cover of the song 360. You can you know, it might be hard to see here, but, if you look in here
Sam Nadler [3:09]: Through, like, kinda how this works, because, like, a lot of the secret sauce is is right here at the top.
Jordan Metzner [3:15]: Yeah. So, I mean, this looks like, almost any other programming language, let, like, lead synth, arrange and, here's the arrangement. So it's a three in one arrangement, and it's e three to b three with c four, and then, kind of runs through this and then it runs through four times. Starts with like a VOC sample up here at the beginning, and the note is a sawtooth, and then, it kinda goes down into each section. Here's the bass, etcetera. If I just press play here, you should be able to hear it. Hopefully, we can sync it all up in post, but should sound like Charli XCX. Let's see how this goes. Alright. So what you can see here is, like, it's, like, running through this three, four times, then it hits this one, then it runs through it again. Right? So you can actually see it happening in real time, and I think I did this like kind of as an example, but like, if we just changed, the entire song or at least the intro to be e three instead of these different notes like e three to as f three, etcetera. Yeah. I think, hopefully, this should work. What we'll be able to see is that, like, the song should come out totally flat.
Jordan Metzner [4:34]: I mean, we can keep those e fours in here, but it's just one note. And so if that plays, let's see if this works. So there you go. And so you can kinda see how that works a little And so I had never really heard of it. I guess maybe it's starting to take some ways online, but I decided to take this and take it up a notch, and maybe I can show you what I built.
Sam Nadler [5:03]: Yeah. So yeah. How did this inspire what you decided to build next?
Jordan Metzner [5:08]: Yeah. So I love using Ableton Live and Ableton to produce music, and I've been DJing since I was a little kid. I think I first got turntables when I was, like, 10 or 12 years old. So I love to make music and to produce music. And so I wanted to make something that was kind of like Algoriddim djay Pro 2, and a combination of Ableton, and this is when I built. So I don't have a good name. I just called it Strudel. It's like the name of the library. But what you can see here is I have a two deck mixer, and you can actually mix between kind of deck one and deck two. Each deck has different multiple channels on it. On each deck, there's actually like a sub channel mixer between a and b that you can kind of manipulate here down below and selecting the channels. It's got a bunch of additional master effects and features, synchronization. The coolest feature, it has AI. So, all of these channels have some AI code behind them or have regular code behind them, Strudel code. And the AI button will help us, like, write new beats.
Jordan Metzner [6:09]: So let's see if this works right. It might be a little buggy, but hopefully, it'll all work. So we should be able to get some techno music here. Let's see. Alright. And I didn't promise it'd be like amazing techno music. So, here's the drums and I could, increase the delay. Immediately, you should start to hear, you kinda have the drums picking it up there a little bit. And since the drums are on channel a, I can kinda mix it between channel a and drive it all the way down channel b where there's no drums. Got it. Right? So there's the mix. But now, know, what we can do is hopefully hit these drums and change the b on them, like, in the AI button. So every time I hit AI, I'm getting a different version. And I kinda change the sounds that it's using. I guess every time I hit AI, it resets it again. But you could see here it's creating different beats. And then, we could basically apply those and you can do this across the whole stack. So you can start to make the whole song sound different.
Jordan Metzner [7:38]: And you can keep going across the stack. So, it's just the beginning early days. I think we've talked about in the past that we've used, you know oh, there we go. We've talked about in the past that we've used things like Suno to produce music, but this is another type of, like, music production. Just another way to leverage LLMs, JavaScript, React. I think I used mostly Claude. Actually, I think I used mostly OpenAI Codex to build this. I think, I could have never built something so dynamic, so progressive without the use of LLMs. This would have been a very, very heavy and expensive development cycle to build something so complex. And, yeah, while it's just the beginning, I think this type of libraries, these Strudel libraries and others like it will allow, the LLMs to continue to help people make cool stuff, like, in this case, music. You know?
Sam Nadler [8:24]: Yeah. That's what I love about AI. I mean, it really brings access to previously complex, I guess, trades or something that would take, decades or years to really master, and you can really get your hands in and, so, have an impact pretty quickly. Not saying there's not a learning curve. Like, there's definitely a learning curve. I'm like, I'm not familiar with mixing music, and this looks a little complicated to me. But, I think I could probably build myself a pretty basic dumbed down version and within, like, a couple hours. And, like, previously, that would have been impossible. It'd also help me learn the technology and how to learn to mix music. So that's what I absolutely love about it. Just, the accessibility of almost anything. If you're willing to spend a few hours and work with the tools that are available, it's pretty amazing.
Jordan Metzner [9:16]: Yeah. I mean, I think we've talked in the past of kind impact of, the Veo 3 models from Google and the impact on video and what that has on Hollywood. And, I know last week we spoke about Nano Banana and the impact on Photoshop and image design, know, and here's LLMs having impact on music production. So, I think, we're gonna continue to see LLMs have an impact whether they like it or not on the creative arts. And, hopefully, that generates a whole new wave of, creative artists, actually.
Sam Nadler [9:47]: Perfect. And that leads us to our kinda next segment. We're gonna talk about a use case for Claude. I'm assuming most people are familiar with Claude and have used it for many different tasks. However, this specific use case, we use a lot, and I think it's great to show our network and our listeners about, this amazing feature from Claude. And, we've also we kind of run this test with other LLMs, and it's not nearly as good. So why don't you walk us through what we're doing and, a, this example, and then how we use it?
Jordan Metzner [10:24]: Yeah. So I think, we don't really need to do this, but in our prep for today's today's call, we ran this through some other LLMs and we found, that continually, Claude continues to outperform. And what I mean by outperform is you can give Claude a CSV, a spreadsheet, or a document, a PDF document, and it'll build you a pretty beautiful dashboard that is pretty cohesive. And so in this example, I gave it Google's latest financial filings. I think it was a 10 k. I asked something super simple. Make me a dashboard as if I'm a CFO after reading this document. And yeah. So I just had Claude put that all together. It gave me an executive summary here that we can read in the text, but it also gives me this beautiful dashboard of, Google's revenue for the quarter, including its, 96,000,000,000 in revenue and 28,000,000,000 in net income. And it even shows kind of a breakdown of where that revenue is coming from, operating expenses by segment, free cash flow, employees, all the way down to the bottom of the balance sheet. So, if I had additional questions or want to ask certain things, I could certainly do that. It looks like they have a settlement liability charge here for some privacy concerns and some other key considerations of risk. But, the document that I used to generate this was a 57 page document in PDF format with, financials and all different types of documentation. So, it's a great tool for investors, but it's a great tool for kind of almost any small business.
Jordan Metzner [11:52]: You can export, your financials from, Google spreadsheet, QuickBooks, Excel, kind of anywhere you want, and just have it make a dashboard for you. And, one of the things I like about this is that, it's almost like a throwaway tool. I'll use it once, I'll check out the dashboard, I'll look at it, and, next time I wanna make a dashboard, I'll just make a new dashboard. I don't need to save this, I don't need to it for anything. So it's almost like a one time use site. I find that really valuable and really helpful and, incredibly fast.
Sam Nadler [12:18]: Exactly. I think in this use case, we're using Google. But in our day to day use case, we're using, QuickBooks or other data that we have from our own businesses. And instead of maintaining a dashboard, we can quickly throw it up in Claude, and it's so incredibly digestible and beautiful. It's, it's it's hard to argue that there could be something better in the amount of time this takes. And we did, just for context, put it in, the same document in OpenAI. It wasn't nearly as good. It didn't provide the immediate visual. The summary was pretty good, but I would argue that it was too much to read. And then we had to prompt it a couple more times just to get the visuals, and the visuals, were there, but not nearly as I think it made four graphs.
Jordan Metzner [13:05]: You know? I made four graphs. And, here, we don't have any graphs, but we have really good charts that actually help you really understand what's happening in the financials. And, I mean, I didn't read the document, but I know pretty well how Google's doing just by, looking at this dashboard. So it definitely hits the spot as far as being effective and efficient.
Sam Nadler [13:24]: And you can do this with really any data, whether it's financial data. I was playing with some operational metrics today and all very, very useful in Claude. And it's, it's kinda funny how these LLMs excel or accelerate, and then who's better at what continues. Seems like it's constantly changing, which actually leads me to kind of our next topic, which is Codex. We've talked about a lot about using LLMs to write code. A week ago, two weeks ago, I can't remember, we were talking about Claude Code and how powerful it was with, the multiple agents, and I went through an ATS that I built in a couple hours. But you, since Codex, these since the most recent launch of Codex, you're you've told me that Codex is even better, and our engineers are switching to Codex.
Jordan Metzner [14:16]: We keep switching. And I think that's gonna be the game until, there was there's some type of supremacy, and it seems like it's gonna be a winner take most model. I think since starting to code with AI, we've gone through a bunch of different iterations from using ChatGPT to using tools to moving over to cursor, and then, Claude Code and then Codex, you know. I tried Codex a few months ago, and I thought the product was highly underdeveloped. It was a pretty poor experience. And, yeah, over the Labor Day weekend, I saw a lot of people talking about it, so I decided to give it a test run. It helped me build, like, the majority of what you saw inside my DJ tool. It's incredibly well developed. You can definitely see that they're spending a lot of effort in improving the models and improving the Codex experience. Yeah. I think we're gonna continue to see a change in workflows. I think, yeah, the team is moving over because it's the best model we found available on the market today. But that may change, based on something new coming out tomorrow. So it is exciting. It's a exciting space.
Jordan Metzner [15:16]: It is showing that, these things are moving very, very quickly. But, yeah, I think to bet on, what our workflow is gonna be like by the end of the year would be crazy, because I just think it's gonna change a bunch more times. And speaking of the Codex, just, some news came in that Alex AI or Alex Codes is a cursor for Xcode. So basically, build Swift based apps is joining OpenAI to join the Codex team as well. So, they're definitely they, being OpenAI, they're definitely investing in Codex as a tool, trying to beat Anthropic in this space. Team is is adapting to it quite
Sam Nadler [15:54]: And they made another acquisition. Correct? Statsig?
Jordan Metzner [15:57]: Yeah. Statsig acquisition by OpenAI. I had the link up here. Let me try to find it. But, yeah, OpenAI paid $1,100,000,000 for Statsig. I don't know if you know, but Statsig is a tool that allows you to do a b testing and other types of optimizations, launch flags, and things like that. And I know that OpenAI was using it as well as others in order to test their product and make, marginal improvements and things like that, AB testing. So but this just shows, like, huge acquisition. I mean, billion dollars is a massive acquisition for a startup. Super exciting. I think some of those people are gonna roll into product roles at OpenAI, and I think this is, again, this is just like inning one, inning two. I think we're gonna see a lot more acquisitions, a lot more deals like this.
Sam Nadler [16:45]: Super exciting. It seems like OpenAI has captured, the consumers and is trailing a little bit in enterprise adoption. So with Anthropic maybe leading in the thus far in the code race, So, I think they're gonna probably prioritize this effort, and it makes sense. Like, enterprises are gonna be spending the most amount of money.
Jordan Metzner [17:06]: Yeah. And, I mean, I think it was Dario from Anthropic who kinda predicted that, like, know, by the end of the year or something like that, 90% of code will be written by bots. And, I don't want know what percent it is here for Ryz Labs. I know my code is like almost a 100% written by bots. I know companies are getting to that percentage, or close to a 100% really, really quickly. It's gonna have an impact, but, I think everyone is gonna continue to float to the best tool du jour tool of the week or whatnot or, maybe we'll see somebody take a lead where no one can catch up. But, it is surprising that, Microsoft with owning GitHub and its cloud is not the best. Amazon with its cloud is not the best coding tools, and even Google who, obviously, one of the best data companies in the world, plus its cloud is as hasn't produced the best coding tools yet. But, I wouldn't count any of these big players out, and I think we're gonna continue to see more change in the market as it's so critical to kind of own that developer relationship.
Sam Nadler [18:04]: I know you wanted to talk about Figma, and a couple weeks ago, we covered the IPO, which I think, opened at somewhere around $30.35 dollars. I think eventually got up to the $1.15, $1.20, but, has, come down a little bit. I don't know what the price is as of today. Is it in the seventies? Is that accurate?
Jordan Metzner [18:30]: Yeah. I mean, I know today shares plunged, like, 15% or something like that. I think revenue increased 40% year over year, so, company is definitely growing. I think they just slightly beat expectations and earnings are breakeven. But, yeah, I think overall, it just shows kind of this company is still growing. You and I were talking that, we in general use less Figma today than we did before, especially due to AI. But my guess is that, Figma is probably expanding heavily in the enterprise space where, they were not using tools like Figma and, they're still slow to adopt AI inside their enterprises. And so, still a great opportunity. And, yeah, I mean, Figma, probably with all this new capital they raised, probably has a war chest to go, acquire some AI tools that probably will be complementary to their business today, and are probably some of the tools that you and I are talking about on a weekly basis. So it's really early.
Jordan Metzner [19:21]: I would not count out Figma or Dylan or the rest of the team. Since there's really only competition is Adobe, it seems like, they're still in a pretty good spot until we see something different.
Sam Nadler [19:30]: Cool. Anything else to cover today?
Jordan Metzner [19:32]: Yeah. Just one or two more last things. Google settled the antitrust case, so they do not have to sell Chrome, and that will have some impact on their deals with companies like Apple. It was overall really good news for Google. Stock was soaring.
Sam Nadler [19:47]: Good news for Apple as well, isn't it? And the stock buckle a bit.
Jordan Metzner [19:51]: Mixed. I think mixed for Apple. Yeah. I think the stock did perform higher, but it's unclear if this is going to be a great deal because, as far as I understand, Google can't charge Apple for the exclusivity of its search engine, which may or may not be a good thing for Apple. I think it's it's too early to say. Just a few more random quick stories. You.com, which started out as a kind of smarter Google search type tool, just raised a $100,000,000, but focused almost exclusively on enterprise. And I think this is kind of like, know, the reason I wanted to bring up the story was like, this is the trend we're gonna see a lot is, know, the application of AI is happening best and is most effective at the enterprise level. While there are some AI tools for consumers, the application of AI inside of enterprise is really where the big money is today. And so I think this is just, like, kinda one more story that shows that.
Sam Nadler [20:40]: Did you see I think it was a week ago, there was a Wall Street Journal article about the kind of, impact AI has had, at, I don't know, 500 enterprises, and kind of the response seemed like it was lackluster. And where it had the biggest impact was in back office operations, if I remember correctly. Did you see that
Jordan Metzner [21:04]: There's a lot of startups, obviously, that were promising a lot of things. And, it's always a question about kind of too early you're being in the arena. And, I mean, you've seen Eddie, the amount of imp improvement we've seen from, ChatGPT 3.5 to four o to, o three to five is is significant. And it's sometimes hard to remember how kind of poor the 3.5 was compared to how good something like five is today. But, you gotta be building on those stones, on that foundation in order to be able to get there. So, it is possible that a lot of these companies kind of overpromised with the hopes that kind of these models would improve what they delivered. But, yeah, I think, as you've mentioned, back office seems to be a place where AI works incredibly well. I know for us, like, we've automated almost our entire invoicing platform leveraging AI. Yeah. Like we mentioned, obviously, coding is is probably number one.
Jordan Metzner [22:00]: But, companies are gonna continue to see the implementation of AI inside their organization in all different fields. I think from HR to marketing to, like we mentioned, obviously, recruiting with Ntrvsta, and then sales Finance? And pretty much, yeah, finance. Like, every function of the business is is gonna get impacted by AI. And if it's not, you're gonna be left behind. And that's just the reality of how fast this thing is moving.
Sam Nadler [22:25]: Well, one thing I heard, and I didn't read this, I just heard it from somewhere else, so don't quote me on it. But was there just actual you know, executive team really wants AI everywhere. They see the potential. They see the cost savings. But the people who are, like, in the trenches, there's some degree of resistance. And, I think that will probably exist until, like, can the mentality and the culture change around using AI and, working with it to solve the problems. But, I can see how teammates not in the c suite are, apprehensive about these AI tools.
Jordan Metzner [23:04]: Yeah. All type of change is scary. And, I mean, AI is probably the biggest cultural change we've seen in a long time that's gonna impact organizations. So, again, like, hard to see where it's all gonna shake out. It's definitely gonna be highly impactful. And, like we mentioned earlier, it seems to be just the early innings right now. So, anyway, Sam, this was an awesome episode. So great to chat with you and talk about AI and all the products that we built just this week. So I'll love to catch up again next week. And to all of our fans and friends, don't forget to like and subscribe on YouTube, Spotify, and Apple Podcasts, and all your other favorite podcasting platforms.
Sam Nadler [23:41]: Thanks, Jordan. Great episode, and see everyone next week. Cool. See you later, Sam.
