1 00:00:00,080 --> 00:00:01,880 We have something like 2000. 2 00:00:01,920 --> 00:00:04,190 Open source GitHub repository is a usable product. 3 00:00:04,290 --> 00:00:06,760 1.4 million monthly downloads on npm. 4 00:00:06,800 --> 00:00:10,400 It really blew up on Twitter, and it was probably the first time that I pulled 5 00:00:10,400 --> 00:00:12,160 something that went viral. 6 00:00:12,200 --> 00:00:14,640 Everyone around me was saying, Chad is dead. 7 00:00:14,840 --> 00:00:19,560 Yeah. I think people are much more willing to accept chats now than in 2024 when I 8 00:00:19,560 --> 00:00:23,360 saw it. Building chat interfaces seems very easy. 9 00:00:23,400 --> 00:00:28,800 The hard part is state management, ChatGPT Claude and all these other AI that 10 00:00:28,800 --> 00:00:31,320 consumers are familiar with set a high bar. 11 00:00:31,360 --> 00:00:35,520 So consumers are expecting the same level of polish in every app. 12 00:00:35,560 --> 00:00:40,120 If you want to roll yarn in AI agents, that's where CI comes in and gives you all 13 00:00:40,120 --> 00:00:42,080 the basic features built in. 14 00:00:42,120 --> 00:00:46,480 I think the classic role of you have a company where someone specifies the 15 00:00:46,480 --> 00:00:51,120 features completely and then gives it to engineering to build doesn't work anymore 16 00:00:51,480 --> 00:00:56,040 because the building part, if you have a really good spec, is just fully automated 17 00:00:56,040 --> 00:00:57,590 by coding agent. 18 00:00:57,600 --> 00:01:03,840 And so the bottleneck is really writing good arms and having that sort of taste. 19 00:01:03,880 --> 00:01:08,200 What was the early signal when you said, okay, let's make this into a product? 20 00:01:08,320 --> 00:01:11,280 I think what drove me was. 21 00:01:19,960 --> 00:01:23,640 Hello and welcome everyone to another episode of the merge, our podcast brought 22 00:01:23,640 --> 00:01:26,560 to you here from our San Francisco office at Code Rabbit. 23 00:01:26,800 --> 00:01:33,280 Today I'm super excited because I'm joined by Simon Freshly, the founder and CEO of 24 00:01:33,280 --> 00:01:37,080 assistant UI and Open Source Project that you will be learning a lot more about in 25 00:01:37,080 --> 00:01:39,770 just a minute. Simon, how are you doing today? 26 00:01:39,800 --> 00:01:41,040 Great. And thanks for having me. 27 00:01:41,040 --> 00:01:42,880 I'm excited to be here. Awesome. 28 00:01:42,920 --> 00:01:44,900 Well, we're excited to have you, too. 29 00:01:44,920 --> 00:01:47,840 I think the project has seen great success. 30 00:01:47,840 --> 00:01:50,880 Certainly. I've heard about it actually a while back already. 31 00:01:50,880 --> 00:01:54,880 But some people might not be so familiar already with what you're doing right now. 32 00:01:54,880 --> 00:01:56,840 So could you give us a little bit of a brief? 33 00:01:56,880 --> 00:02:00,560 Of course. So assistant UI gives you the UX strategy between your own app. 34 00:02:01,720 --> 00:02:04,240 We give you react front end components. 35 00:02:04,520 --> 00:02:06,120 You can customize them the way you want. 36 00:02:06,160 --> 00:02:12,480 And we handle the state streaming and connections to OpenAI, as well as the 37 00:02:12,480 --> 00:02:14,480 agent frameworks like Master Lang chain. 38 00:02:14,600 --> 00:02:15,800 And the AI is 39 00:02:16,840 --> 00:02:18,950 right on a very important work. 40 00:02:18,960 --> 00:02:23,960 I think one of the quotes that I read is I think it was from Sast. 41 00:02:24,280 --> 00:02:29,320 That's the idea. But they were claiming that you're revolutionizing the way UI is, 42 00:02:29,360 --> 00:02:34,480 is done for the new era of of of chat, which I think is certainly the interface 43 00:02:34,480 --> 00:02:36,960 that we've been all using the most lately. 44 00:02:37,680 --> 00:02:40,630 Can you tell me what was wrong with the way it was done before? 45 00:02:40,640 --> 00:02:43,400 Why was why did we need assistant UI? 46 00:02:44,000 --> 00:02:46,480 The project started as a hackathons project. 47 00:02:46,480 --> 00:02:48,560 I didn't plan to make it into a startup. 48 00:02:48,600 --> 00:02:53,840 It was something that I needed myself when I started the project in 2024. 49 00:02:53,880 --> 00:02:57,920 Everyone was making fun of AI wrappers, and 50 00:02:58,920 --> 00:03:04,080 because AI chat was very popular, but at the same time, everyone felt a little bit 51 00:03:04,320 --> 00:03:09,040 like it seemed so repetitive that nobody thought this was going to become a huge 52 00:03:09,040 --> 00:03:14,080 thing. I think since then, we've adopted chat as the default interface for AI 53 00:03:14,120 --> 00:03:17,840 agents. This was not very obvious around the time that I started it. 54 00:03:18,840 --> 00:03:23,160 I think I just saw that I needed this so much, so often in my own personal 55 00:03:23,160 --> 00:03:27,480 projects, and also when I was trying to start other startup ideas. 56 00:03:27,560 --> 00:03:32,520 So I built it for myself and it's caught on a little bit. 57 00:03:32,680 --> 00:03:34,960 Yeah. What is what is the traction right now? 58 00:03:35,000 --> 00:03:36,440 How many people are using it? 59 00:03:37,120 --> 00:03:41,630 We have something like 2000 open source GitHub repository. 60 00:03:41,680 --> 00:03:45,560 I use our product 1.4 million monthly downloads on npm. 61 00:03:45,600 --> 00:03:47,880 This includes npm updates as well. 62 00:03:47,920 --> 00:03:49,660 That's. Yeah, that's very impressive. 63 00:03:49,680 --> 00:03:51,040 Congratulations. Thank you. 64 00:03:51,080 --> 00:03:54,070 It's about like 1.5% of the market. 65 00:03:54,080 --> 00:03:55,760 So we have long ways to go. 66 00:03:55,800 --> 00:03:57,480 I mean, it's a big market. 67 00:03:57,960 --> 00:04:01,240 Yeah I saw a first demo of you. 68 00:04:01,240 --> 00:04:02,800 I think it was at the AWS. 69 00:04:03,040 --> 00:04:05,720 I built a lot of it was 2024. 70 00:04:06,640 --> 00:04:11,400 Can you tell a bit about the story where you started back then and then where are 71 00:04:11,400 --> 00:04:13,250 you going? How did you see the shift? 72 00:04:13,360 --> 00:04:16,200 What was the what were the new challenges that came up? 73 00:04:16,600 --> 00:04:22,080 Yeah, that demo, I think I made it like about two weeks and after I decided that's 74 00:04:22,080 --> 00:04:25,360 the thing, I was going to become the thing that I'll really focus on. 75 00:04:27,200 --> 00:04:29,160 I randomly signed up to this event. 76 00:04:29,160 --> 00:04:30,560 I didn't even prepare much for it. 77 00:04:30,560 --> 00:04:34,960 I started preparing for that event like an hour before, and I pulled up one of the 78 00:04:34,960 --> 00:04:41,320 examples that I had built trying to sell to another company and e-commerce 79 00:04:41,360 --> 00:04:46,640 store. And it really blew up on Twitter, and it was probably the first time that I 80 00:04:46,640 --> 00:04:49,400 built something that that went viral. 81 00:04:49,440 --> 00:04:50,120 And 82 00:04:51,520 --> 00:04:57,160 I think it only has views, but it seems like it's highly concentrated in San 83 00:04:57,160 --> 00:04:59,710 Francisco and the exact right community. 84 00:04:59,720 --> 00:05:00,280 So 85 00:05:01,240 --> 00:05:05,480 there was this sort of impression that a lot of people around me suddenly knew what 86 00:05:05,480 --> 00:05:06,720 the system UI was. 87 00:05:06,760 --> 00:05:10,400 I remember I learned about it and then I yeah, I also used it in one of my 88 00:05:10,400 --> 00:05:15,920 projects. So very excited to to hear that you were coming and that I get a chance to 89 00:05:15,960 --> 00:05:19,680 chat to you, because I feel like it was really one of those early first pivotal 90 00:05:19,680 --> 00:05:23,920 moments. And I saw a lot of people building in the city on assistant UI, 91 00:05:24,200 --> 00:05:28,040 maybe for people who are have not dived into it yet. 92 00:05:28,520 --> 00:05:31,280 Can we explain what can you explain a bit more deeply? 93 00:05:31,280 --> 00:05:37,520 Why did we need it, a new interface, why was like whatever Streamlit was providing, 94 00:05:37,520 --> 00:05:39,760 for example, not sufficient. 95 00:05:40,040 --> 00:05:42,480 Building traffic interfaces seems very easy. 96 00:05:42,800 --> 00:05:45,320 The hard part is state management. 97 00:05:46,480 --> 00:05:52,800 You've got the state that streams in token by token, and you want to stream this 98 00:05:52,800 --> 00:05:57,040 efficiently. There's all sorts of post-processing that you want to do, like 99 00:05:57,080 --> 00:05:58,400 markdown parsing, 100 00:06:00,080 --> 00:06:05,080 things like file attachments, queuing messages, voice mode. 101 00:06:05,120 --> 00:06:10,240 These are all things that you need to add on top of the original chat interface. 102 00:06:10,240 --> 00:06:16,600 Message editing is another one that's sometimes challenging, and I just 103 00:06:16,600 --> 00:06:19,600 bundled all of these standard features into the 104 00:06:21,080 --> 00:06:22,600 same component library. 105 00:06:22,760 --> 00:06:27,480 The bar is pretty high because ChatGPT Claude and all these other AI chats that 106 00:06:27,520 --> 00:06:30,040 consumers are familiar with set a high bar. 107 00:06:30,080 --> 00:06:34,360 So consumers are expecting the same level of polish in every app. 108 00:06:34,360 --> 00:06:40,520 If you want to roll yarn in app AI agents, and that's where CI comes in and gives 109 00:06:40,520 --> 00:06:43,480 you all the basic features built in. 110 00:06:43,680 --> 00:06:47,480 Great. Can you tell us some names, like do you have some prominent companies that 111 00:06:47,480 --> 00:06:50,960 are using it right now that are relying on your your project? 112 00:06:51,480 --> 00:06:56,560 Yeah. So I went through by Combinator and they're using it internally. 113 00:06:57,690 --> 00:07:01,600 Lang chain uses us master users us. 114 00:07:03,000 --> 00:07:05,790 We got a few fortune 500 enterprises. 115 00:07:05,800 --> 00:07:11,840 I'm not so sure if I can name them, but they use us mostly internally, sometimes 116 00:07:11,840 --> 00:07:15,640 also as part of their developer kits. 117 00:07:16,810 --> 00:07:21,400 Well, how do you balance, I guess, you know, feature requests coming in from 118 00:07:21,400 --> 00:07:23,320 these companies and these bigger names? 119 00:07:23,320 --> 00:07:27,120 Or is this maybe something that the community seems to focus because we have 120 00:07:27,120 --> 00:07:29,600 so much traction, so many downloads a month? 121 00:07:29,600 --> 00:07:34,440 I can only imagine there must be a lot of people pulling on on your project, trying 122 00:07:34,440 --> 00:07:36,720 to navigate you in one direction or another. 123 00:07:37,440 --> 00:07:42,080 One of the cool design decisions that I made was to make everything so modular. 124 00:07:42,600 --> 00:07:46,440 That's every customer can build our next feature themselves. 125 00:07:47,720 --> 00:07:54,040 So when customers that are paying us ask for new features, we go ask them for 126 00:07:54,040 --> 00:07:56,760 repository permissions and build it in a code base. 127 00:07:56,920 --> 00:07:59,280 And then we try it out. 128 00:07:59,320 --> 00:08:01,000 Do this again for another customer. 129 00:08:01,000 --> 00:08:04,920 And this way we figure out the patterns and the design space. 130 00:08:04,920 --> 00:08:07,120 And then we put it into the main library. 131 00:08:07,940 --> 00:08:12,440 Okay, okay. Everyone can build our next feature themselves. 132 00:08:12,600 --> 00:08:18,920 And this doesn't apply to everything, but 95% of new things we ship works 133 00:08:18,920 --> 00:08:23,000 like this okay. Do you get a lot of quality pull requests from externally as 134 00:08:23,000 --> 00:08:23,520 well? 135 00:08:24,520 --> 00:08:28,120 I'm small. Scopes are usually very high quality. 136 00:08:28,360 --> 00:08:33,480 I've actually hired like three people from the open source contributors, and it's 137 00:08:33,480 --> 00:08:39,280 been great because they understand the product direction and have 138 00:08:39,480 --> 00:08:43,520 similar taste to me, and that's really how to find other ways. 139 00:08:43,920 --> 00:08:47,920 Yeah, I think taste in general is an interesting conversation topic. 140 00:08:49,040 --> 00:08:51,200 Seems like now with the new era of Gen. 141 00:08:51,240 --> 00:08:56,880 I this is what matters the most for for programmers, especially as we become full 142 00:08:56,880 --> 00:09:00,040 stack and more of a product management role as well. 143 00:09:00,040 --> 00:09:03,400 Somehow I like that term product engineer to a certain sense. 144 00:09:04,160 --> 00:09:04,560 Sure. 145 00:09:05,720 --> 00:09:07,040 How do you decide for that? 146 00:09:07,080 --> 00:09:08,480 How do you hire for that? 147 00:09:08,920 --> 00:09:13,640 I think the classic role of you have a company where someone specifies the 148 00:09:13,640 --> 00:09:18,520 features completely and then gives it to engineering to build doesn't work anymore 149 00:09:18,920 --> 00:09:23,440 because the building part, if you have a really good spec, is just fully automated 150 00:09:23,440 --> 00:09:25,120 by coding agent. 151 00:09:25,880 --> 00:09:31,800 And so the bottleneck is really writing good prompts and having that sort of 152 00:09:32,200 --> 00:09:33,040 taste. 153 00:09:34,200 --> 00:09:39,440 I think we feel this a lot ourselves as well, because assistant UIs product is 154 00:09:39,440 --> 00:09:43,120 mostly about design and less about engineering feat. 155 00:09:43,160 --> 00:09:48,040 It's like we give you a well-designed interface, but we also give you a good 156 00:09:48,040 --> 00:09:50,160 developer experience. 157 00:09:50,800 --> 00:09:55,480 APIs that make sure as you grow, your code doesn't become spaghetti and remains 158 00:09:55,480 --> 00:09:59,720 maintainable. And these are like design decisions on how we make you structure 159 00:09:59,720 --> 00:10:03,000 your code that lead to a more maintainable code base. 160 00:10:03,200 --> 00:10:09,320 And then for the agents that run in your app, we try to give you a few APIs that 161 00:10:09,480 --> 00:10:12,480 for the LM to use so that you can interact with your app. 162 00:10:12,480 --> 00:10:17,160 And there the economics matter as well for the genetic experience, so that the agent 163 00:10:17,160 --> 00:10:20,200 is able to be productive in the UI. 164 00:10:20,240 --> 00:10:25,040 So if the user just interacted with something in those, the context of that, I 165 00:10:25,040 --> 00:10:29,000 saw that I mean assistant UI as many built in TypeScript, right. 166 00:10:29,320 --> 00:10:31,520 But then you also a lot of your 167 00:10:32,800 --> 00:10:34,560 customers, I guess they build on long chain. 168 00:10:34,560 --> 00:10:36,480 At least that's what I saw in your demo as well. 169 00:10:37,320 --> 00:10:43,040 Do you how do you what else is there to how do you see people building? 170 00:10:43,080 --> 00:10:46,080 Chet Epstein's days, I guess is my question. 171 00:10:46,760 --> 00:10:49,720 Market is all react developers and 172 00:10:50,800 --> 00:10:56,000 I would say 60% use TypeScript, 40% use Python backend. 173 00:10:57,600 --> 00:10:59,520 The best agents I've seen. 174 00:11:02,680 --> 00:11:07,440 Use Python. But then, like, I think the average quality of the TypeScript projects 175 00:11:07,440 --> 00:11:08,840 is higher. So it's like interesting. 176 00:11:08,840 --> 00:11:14,440 It's like you have a few teams that are very data driven and actually 177 00:11:14,960 --> 00:11:19,720 obsess over every word in the system prompt and the tool design and all that. 178 00:11:20,440 --> 00:11:25,240 The kind of engineer who obsesses over that, I think on average is a Python 179 00:11:25,240 --> 00:11:31,120 developer. And so that's I see a few really good agents being built in 180 00:11:31,120 --> 00:11:31,880 Python. 181 00:11:33,400 --> 00:11:36,960 I would say it's much easier to build good agents with TypeScript though. 182 00:11:39,880 --> 00:11:44,680 We support both. We have adapters and for assistant UI in both Python we have a 183 00:11:44,680 --> 00:11:48,480 small Python library that we maintain as well as TypeScript. 184 00:11:49,600 --> 00:11:54,800 There are a few users who use esoteric languages, but 99% of the market is 185 00:11:55,120 --> 00:12:01,200 TypeScript and Python. Yeah, I had of master here as well. 186 00:12:01,200 --> 00:12:07,360 He also had some some opinions on that, which is interesting, I feel, and many 187 00:12:07,360 --> 00:12:11,600 people agree with me, that mainly the main market used to be in Python. 188 00:12:11,600 --> 00:12:17,600 I mean, the entire length chain train came and took, I think, a major part. 189 00:12:17,640 --> 00:12:22,760 Do you feel like that is shifting in a certain sense that more people are moving 190 00:12:22,760 --> 00:12:24,520 into into TypeScript? 191 00:12:24,520 --> 00:12:28,520 Or do you do you think that that will stay in Python, just like you mentioned 192 00:12:28,520 --> 00:12:32,680 before? I'm more bullish on TypeScript and it's mostly where my heart is at, I 193 00:12:32,680 --> 00:12:38,160 think. I think TypeScript as a language, as a developer community culture, the 194 00:12:38,160 --> 00:12:43,320 TypeScript developer community is more obsessed about design, and I think that 195 00:12:43,320 --> 00:12:46,110 matters a lot, especially also thanks to master. 196 00:12:46,160 --> 00:12:52,360 They're bringing a ton of evals and the culture around being data driven to 197 00:12:52,400 --> 00:12:58,720 TypeScript. So I think what they're doing there with their evals, platform and 198 00:12:59,400 --> 00:13:01,000 tests and stuff is really cool. 199 00:13:01,720 --> 00:13:03,920 Yeah, I agree, I think they've done a great job. 200 00:13:04,080 --> 00:13:10,000 I'm building most of my agents now on Lastra because it's also it's so easy to 201 00:13:10,040 --> 00:13:14,920 use. I think they have done a great job of making it accessible for, for an agent as 202 00:13:14,920 --> 00:13:18,880 well, to just dive into a project and formulate that out. 203 00:13:19,460 --> 00:13:23,200 That's right. I realized that the programing language is probably matter 204 00:13:23,200 --> 00:13:24,560 less going forward. 205 00:13:24,800 --> 00:13:26,120 Thanks for coding agents. 206 00:13:26,680 --> 00:13:32,480 It's you'll have a lot of data teams that will start feeling comfortable using 207 00:13:32,520 --> 00:13:38,240 TypeScript. Given that we barely write any code by hand anymore and it's all 208 00:13:38,640 --> 00:13:39,760 AI. Anyway, 209 00:13:40,680 --> 00:13:43,160 I realized this in hiring as well as like 210 00:13:44,280 --> 00:13:50,720 you can hire a front end engineer or a data scientist and 211 00:13:50,760 --> 00:13:52,280 have them work full stack. 212 00:13:52,880 --> 00:13:53,760 And 213 00:13:54,720 --> 00:13:58,880 this sort of like ownership across the whole stack has been is really productive, 214 00:13:58,880 --> 00:14:04,480 I think. So that's the way we hire is like, we hire folks who can own and 215 00:14:04,480 --> 00:14:10,640 maintain features and to end so they they come up with the idea of the feature, do a 216 00:14:10,640 --> 00:14:14,760 little bit of research in the design space and then go and build it. 217 00:14:15,560 --> 00:14:20,520 That's interesting because personally, I feel like certain domain expertise and 218 00:14:20,520 --> 00:14:23,560 programing even still matters even with coding agents. 219 00:14:23,600 --> 00:14:29,920 Like for example, I recently built an app and I'm not a database expert. 220 00:14:30,240 --> 00:14:35,840 Actually, my background is in embedded C, so a little bit different and I build it 221 00:14:35,840 --> 00:14:41,120 with some other folks around that, you know, probably also went database experts. 222 00:14:41,120 --> 00:14:45,880 And then when we onboarded a lot of users at the same time, the app crashed because 223 00:14:45,880 --> 00:14:48,960 we there was some inefficient querying as well. 224 00:14:48,960 --> 00:14:52,920 And then some of the, the scaling of the database just didn't work. 225 00:14:53,600 --> 00:14:54,280 And 226 00:14:55,520 --> 00:14:59,680 I feel like if we would have had somebody with more expertise in our team because we 227 00:14:59,680 --> 00:15:02,600 were using AI, that probably would have not happened to us. 228 00:15:02,600 --> 00:15:04,720 So how do you how do you see that? 229 00:15:05,160 --> 00:15:10,840 I try to hire a diverse team in terms of like interests. 230 00:15:11,240 --> 00:15:14,520 So some people who are more data driven, some that are 231 00:15:15,440 --> 00:15:20,760 upset, that obsessed over infrastructure, some that obsess over design, some that 232 00:15:20,760 --> 00:15:22,840 obsess over efficiency. 233 00:15:23,440 --> 00:15:24,280 And. 234 00:15:26,960 --> 00:15:33,120 We operate in a way where engineers and my team pick up tasks that they enjoy doing. 235 00:15:33,120 --> 00:15:35,920 So there's not a fixed roadmap for there. 236 00:15:35,920 --> 00:15:41,480 So the order to which we do we implement features depends on like the interests of 237 00:15:41,480 --> 00:15:45,920 the engineers and by hiring a more diverse team there, 238 00:15:47,320 --> 00:15:49,480 I aim to like even the product out. 239 00:15:49,480 --> 00:15:53,760 So hopefully when something like this happens where one of our careers is 240 00:15:53,760 --> 00:15:56,000 inefficient, the engineer who 241 00:15:57,360 --> 00:16:00,040 looks at our dashboard sees at the. 242 00:16:00,560 --> 00:16:06,080 There are a few spikes in the average latency, and they get super curious and go 243 00:16:06,080 --> 00:16:08,400 and figure out why that's the case. 244 00:16:08,440 --> 00:16:09,960 Okay, that makes sense. Yeah. 245 00:16:10,280 --> 00:16:14,210 Let's switch gears a little bit and talk a bit about open source. 246 00:16:14,240 --> 00:16:19,000 I think one thing that is very interesting about assistant UI is that you're making 247 00:16:19,000 --> 00:16:23,160 a business out of something that started as a hobby project, and you're making it 248 00:16:23,160 --> 00:16:24,720 all open source and available. 249 00:16:25,160 --> 00:16:30,120 What is your business model and what is the trick behind making it sustainable? 250 00:16:30,840 --> 00:16:34,080 Yeah, it's trickier than your usual business. 251 00:16:34,920 --> 00:16:39,520 It's hard to make money with UI component libraries, which is what the idea of the 252 00:16:39,680 --> 00:16:44,040 day is. I saw Clerc and was inspired by them. 253 00:16:44,080 --> 00:16:50,080 They give you authentication and also react components as the API for 254 00:16:50,200 --> 00:16:52,440 interacting with the service. 255 00:16:52,720 --> 00:16:58,480 And so we have a system cloud and it's our backend and it powers it's a database. 256 00:16:58,480 --> 00:17:01,360 Basically it powers a chat product. 257 00:17:01,840 --> 00:17:06,360 When you upload files we can store them in Assistant Cloud. 258 00:17:06,360 --> 00:17:11,960 And then we give you an analytic so you understand what sort of questions people 259 00:17:11,960 --> 00:17:17,720 ask in your product, how many people use your product, the AI chat in your product. 260 00:17:17,720 --> 00:17:24,000 And we're expanding to add a few more features that like 261 00:17:24,040 --> 00:17:26,720 an AI gateway. Basically, that's so you can 262 00:17:27,640 --> 00:17:33,560 use all the different AI models in there, a billing service, so that you can let 263 00:17:33,560 --> 00:17:37,640 your users add tokens and a wallet and money in a wallet and use it for their 264 00:17:37,640 --> 00:17:41,280 token usage. This is for the AI app builders that. 265 00:17:43,520 --> 00:17:45,280 Have like longer running agents. 266 00:17:45,280 --> 00:17:50,360 And yeah, hopefully users make money is like one of the ways we're going to make 267 00:17:50,360 --> 00:17:54,600 money. Okay, so you're basically as soon as somebody comes becomes more successful 268 00:17:54,600 --> 00:17:58,960 with their project and you can see that you can yeah, charge a little subset of 269 00:17:59,000 --> 00:18:01,160 that. And that's been working out well. 270 00:18:01,480 --> 00:18:03,800 We're kind of like charging permanently active user 271 00:18:05,240 --> 00:18:08,200 for the storage service and analytics. 272 00:18:08,400 --> 00:18:10,960 So if a user interacts with your AI chat 273 00:18:12,560 --> 00:18:14,400 we charge a small fee for that. 274 00:18:14,840 --> 00:18:16,560 How did you come up with that? 275 00:18:16,560 --> 00:18:20,120 I think that's a very difficult barrier to find. 276 00:18:20,120 --> 00:18:22,520 And also the concept itself is hard to develop. 277 00:18:22,880 --> 00:18:24,840 Maybe many people will be listening to this. 278 00:18:24,880 --> 00:18:28,400 We're also thinking, oh, I've got an open source project that somehow seen some 279 00:18:28,400 --> 00:18:33,280 traction. Where did you start drawing the line between this is going to be freely 280 00:18:33,280 --> 00:18:36,680 available. This is what we're going to use as a service to sustain ourselves. 281 00:18:37,760 --> 00:18:40,200 Yeah. It has pros and cons to do open source. 282 00:18:40,200 --> 00:18:46,400 I think it's just in UI because you cannot directly really monetize 283 00:18:46,520 --> 00:18:48,280 UI libraries on a venture scale. 284 00:18:48,280 --> 00:18:50,120 There's not that many examples of it. 285 00:18:50,160 --> 00:18:54,560 It doesn't see that much competition from other venture backed companies. 286 00:18:54,880 --> 00:18:56,880 And that's a good thing. 287 00:18:56,880 --> 00:18:59,200 That makes our lives a little bit easier. 288 00:19:00,360 --> 00:19:03,840 The bounce side is that our business model is more convoluted. 289 00:19:04,120 --> 00:19:06,360 We have to get traction on the open source product. 290 00:19:06,360 --> 00:19:10,080 And then there's this whole separate product, which is not just assessing UI as 291 00:19:10,080 --> 00:19:14,920 a hosted service. It's supplemental add ons that plug into assisting UI. 292 00:19:15,280 --> 00:19:17,440 And for that part we charge. 293 00:19:17,480 --> 00:19:21,240 And so you need to get product market fit twice. 294 00:19:21,280 --> 00:19:26,400 You need to make the open source product successful, and then you need to upsell 295 00:19:26,440 --> 00:19:28,920 the open source customers on the pay product. 296 00:19:28,920 --> 00:19:32,680 And that's a whole different product that needs its own product market fit. 297 00:19:32,960 --> 00:19:35,280 Well. How big is your team right now? 298 00:19:35,320 --> 00:19:37,400 Every for four engineers full time. 299 00:19:37,440 --> 00:19:39,120 Wow. So no marketing. 300 00:19:40,120 --> 00:19:46,400 It's it's all inbound. The nice thing is that under cloud product we get a 301 00:19:46,400 --> 00:19:49,680 ton of enterprise inbound requests. 302 00:19:49,720 --> 00:19:54,400 Just because we have champions in the companies that use the open source 303 00:19:54,400 --> 00:20:00,640 product. And through our documentation, we do a good, 304 00:20:01,160 --> 00:20:04,600 decent job at selling the cloud product and the need for it. 305 00:20:04,600 --> 00:20:07,520 And when the customer is ready, they come to us. 306 00:20:07,840 --> 00:20:11,820 Do you plan on expanding the team or expanding the footprint further? 307 00:20:11,840 --> 00:20:15,640 Is there maybe another product you're thinking about working on? 308 00:20:15,680 --> 00:20:20,000 We're actively hiring, but I we tried. 309 00:20:20,080 --> 00:20:26,440 I try to keep our product scope more on the smaller and and just going really deep 310 00:20:26,440 --> 00:20:32,440 on a few design spaces and getting really good in that area, because 311 00:20:33,440 --> 00:20:38,080 I think building a very basic AI chat is already doable with cloud code. 312 00:20:38,080 --> 00:20:43,440 And the benefit that we give you is that we've just gone so deep in the design 313 00:20:43,440 --> 00:20:44,600 space that 314 00:20:45,640 --> 00:20:50,320 it won't be the intuition for your AI coding agent to do this on the first try. 315 00:20:50,520 --> 00:20:55,040 Yeah, that's an interesting question to where do you see the interface going? 316 00:20:55,080 --> 00:21:00,120 Because some people might argue that maybe chat is not going to be the box. 317 00:21:00,120 --> 00:21:04,600 I think I've read a couple ex posts on, you know, chatters dead and we need 318 00:21:04,600 --> 00:21:08,760 something new. What what do you think will be the next interface for us? 319 00:21:08,840 --> 00:21:11,280 I think dialog is here to stay. 320 00:21:12,080 --> 00:21:16,000 So it's going to be either chat or voice, 321 00:21:17,600 --> 00:21:23,120 where experimenting with a few multi-modal interfaces, like you point your mouse at 322 00:21:23,120 --> 00:21:27,280 an element and then use your voice to ask about it. 323 00:21:28,520 --> 00:21:32,960 At the end of the day, I think chat and dialog is going to stay and it's just 324 00:21:32,960 --> 00:21:34,520 going to get more interactive. 325 00:21:35,720 --> 00:21:38,480 What does that interactive element look like? 326 00:21:38,560 --> 00:21:43,840 So the AI can show you not just text, but generate components on the fly. 327 00:21:44,560 --> 00:21:49,760 A can interact with other services and apps and do things for you. 328 00:21:50,840 --> 00:21:55,640 I do think that the interface is going to overall shrink, just because language is 329 00:21:55,640 --> 00:21:59,760 more expressive than many of the dashboards and UI interfaces that we have 330 00:21:59,760 --> 00:22:05,400 today. So unless you're showing rich information like charts, maps, diagrams, 331 00:22:05,560 --> 00:22:07,360 you might get away with just text. 332 00:22:08,040 --> 00:22:14,320 You just made me think for a little while after I remember the first Streamlit app 333 00:22:14,320 --> 00:22:19,320 that I built, and I felt like it was feeling so clunky, but it somehow worked. 334 00:22:19,320 --> 00:22:25,440 So I assume that you kind of went on the same journey when you when you started 335 00:22:25,440 --> 00:22:31,240 with assistant. I what was that first adoption curve moment like? 336 00:22:31,240 --> 00:22:36,960 Because I remember you writing, you know, we talked about that AWS Genii 337 00:22:37,000 --> 00:22:40,960 moment where you give that presentation and then that LinkedIn post. 338 00:22:40,960 --> 00:22:45,840 I think there was also something that I found about an opinionated piece on the 339 00:22:45,840 --> 00:22:50,520 internet, where you talked as one of the earlier ones about the limitations of for 340 00:22:50,640 --> 00:22:53,440 writing JSON structure, JSON output. 341 00:22:53,440 --> 00:22:56,080 So it seems like you've been a fort leader in the space. 342 00:22:56,120 --> 00:23:00,880 Did you find it easy to to find traction for that project, or was that something 343 00:23:00,880 --> 00:23:02,960 that you also had challenges with? 344 00:23:02,960 --> 00:23:07,160 What was the early signal when you said, okay, let's make this into a product? 345 00:23:07,200 --> 00:23:09,200 I think I resisted the 346 00:23:10,520 --> 00:23:14,760 signals that I was seeing, the pulled out I was seeing for the longest time because 347 00:23:14,760 --> 00:23:17,480 I was like, this is a UI component library. 348 00:23:17,560 --> 00:23:19,600 Not sure how to monetize this. 349 00:23:19,840 --> 00:23:22,440 Everyone around me was saying, chat is dead. 350 00:23:22,600 --> 00:23:27,000 Yeah, I think people are much more willing to accept chat now than in 2024. 351 00:23:27,040 --> 00:23:29,040 When I started, I 352 00:23:29,960 --> 00:23:36,160 think what drove me was some sort of fear that it might become like a 353 00:23:36,160 --> 00:23:41,960 super app. And I didn't also like that sort of proposal that sites like ChatGPT 354 00:23:42,360 --> 00:23:47,400 would replace the web as the primary interface that we interact with the world. 355 00:23:47,920 --> 00:23:52,560 And this has also been something that guides us is like, how do we create an 356 00:23:52,560 --> 00:23:54,240 open ecosystem for 357 00:23:55,200 --> 00:23:56,650 the interfaces of the future? 358 00:23:56,650 --> 00:23:59,880 And that's why I see is open source and the open source. 359 00:23:59,880 --> 00:24:01,120 As much as we can 360 00:24:02,480 --> 00:24:08,320 make sure that the interface and just like the the share access to 361 00:24:08,760 --> 00:24:14,880 high quality UX is not what drives consumers to use ChatGPT or other 362 00:24:14,880 --> 00:24:21,000 products. And then we have a few things brewing over distributed memory for 363 00:24:21,040 --> 00:24:24,480 agents, because personalization is another potential. 364 00:24:24,480 --> 00:24:27,280 Centralizing modes for platforms like. 365 00:24:30,080 --> 00:24:33,320 The other one is like token subscriptions. 366 00:24:34,480 --> 00:24:39,920 That one's interesting. There's a window proposal that Chrome has implemented which 367 00:24:39,960 --> 00:24:43,160 allows the browser to supply about sites with tokens. 368 00:24:43,160 --> 00:24:44,320 I think that's very cool. 369 00:24:44,380 --> 00:24:48,920 Oh, okay. So that's why maybe you just subscribe to tokens through your browser, 370 00:24:48,920 --> 00:24:52,120 and then that supplies all your websites and all the assistants. 371 00:24:52,120 --> 00:24:52,680 But 372 00:24:53,600 --> 00:24:55,280 tokens. Yeah, there could be. 373 00:24:55,320 --> 00:24:57,800 Yeah. Another another interesting surface right. 374 00:24:58,000 --> 00:25:01,720 Tell me a little bit what where do you see this going. 375 00:25:01,760 --> 00:25:05,960 Because there's obviously there's the open source angle for you as a business to 376 00:25:06,000 --> 00:25:11,600 sustain that to also with the new wave of maybe we can talk about AI slop PR and all 377 00:25:11,600 --> 00:25:16,680 of that part and then the the overall space of course of 378 00:25:17,760 --> 00:25:19,920 chat as an interface. 379 00:25:20,200 --> 00:25:24,280 You just mentioned ChatGPT becoming the omni product that everybody uses for 380 00:25:24,280 --> 00:25:27,640 everything. How do you want to position yourself? 381 00:25:27,760 --> 00:25:33,840 And as an open source company and as a business in order to see success in 26, 382 00:25:33,880 --> 00:25:38,920 27? Yeah, I think for us, if you look at the Apple ecosystem, which is very 383 00:25:38,920 --> 00:25:45,190 centralized, Apple puts users first and developers second, like the I, 384 00:25:45,920 --> 00:25:49,480 I predict that is going to be similar. 385 00:25:50,600 --> 00:25:55,760 And we're trying we're going to be on the side of the developers first and see what 386 00:25:55,760 --> 00:25:58,840 they want, prioritize their business interest first. 387 00:25:58,840 --> 00:26:00,200 What do they want right now? 388 00:26:00,360 --> 00:26:06,560 I think the industry still very early, and it's just like getting things to work 389 00:26:07,000 --> 00:26:08,440 is such a hard problem. 390 00:26:10,760 --> 00:26:12,240 Making the agents better, 391 00:26:13,400 --> 00:26:18,920 giving them better context is like, is that something you do with the on the 392 00:26:18,920 --> 00:26:23,760 front end side? So what did the user do in the last five seconds before interacting 393 00:26:23,760 --> 00:26:25,240 with the AI chat? 394 00:26:25,480 --> 00:26:30,600 What is the user looking at in the app where assistant guys, I'm better than who 395 00:26:30,640 --> 00:26:36,800 the user is. A few hooks for these signals from the browser is what we give 396 00:26:36,800 --> 00:26:37,200 you. 397 00:26:39,920 --> 00:26:41,160 I want to do more there. 398 00:26:41,160 --> 00:26:41,640 I think 399 00:26:43,160 --> 00:26:45,520 things are still moving too slow and 400 00:26:46,920 --> 00:26:48,480 we could build so many. Cool. 401 00:26:48,520 --> 00:26:55,000 We have so many cool demos and prototypes, and the real world apps that I 402 00:26:55,000 --> 00:26:57,720 use today still have very few of that. 403 00:26:58,880 --> 00:27:05,240 I think the main benefit of, like adding your own AI chat is that you get to 404 00:27:05,280 --> 00:27:10,080 integrate it deeply in your product, and it understands the primitives of your 405 00:27:10,080 --> 00:27:16,280 product, can help you use a product faster and smoother, and the click around the 406 00:27:16,280 --> 00:27:18,360 product for you, you logged in. 407 00:27:18,360 --> 00:27:23,160 It has all the data that you're that you have in a way that's much harder for like 408 00:27:23,200 --> 00:27:24,640 a super app to replicate. 409 00:27:24,840 --> 00:27:26,560 If you're building your own application. 410 00:27:26,600 --> 00:27:31,960 Obviously you cannot just rely on on a baked out interface that you just pull. 411 00:27:32,880 --> 00:27:36,760 So you will be the pioneer of sort of say, the AI interface. 412 00:27:36,760 --> 00:27:38,560 That's what you see working on it. 413 00:27:38,600 --> 00:27:42,920 Yeah, constantly bring in bring out new primitives for this. 414 00:27:43,120 --> 00:27:46,080 What's one primitive you're thinking about doing next? 415 00:27:46,120 --> 00:27:51,080 One thing we're seeing very often is that customers are building generative 416 00:27:51,080 --> 00:27:57,560 dashboards. And this is like you ask the AI agent, how many new customers 417 00:27:57,560 --> 00:27:59,600 did I close last week? 418 00:28:00,080 --> 00:28:05,760 And this is like in a CRM and it gives you a graph and you're able to from the chat, 419 00:28:05,760 --> 00:28:07,200 drag and drop this 420 00:28:08,920 --> 00:28:13,640 graph into a dashboard that constantly updates and stays up to date. 421 00:28:13,640 --> 00:28:15,400 So it's like a dashboard builder. 422 00:28:16,800 --> 00:28:21,840 I've had to build this like ten times already for customers in their code bases. 423 00:28:21,880 --> 00:28:27,920 Yeah, and I want to build this into a primitive that that sounds like a very 424 00:28:27,920 --> 00:28:33,520 exciting for me. If I think the you see applications like base 44 and 425 00:28:34,080 --> 00:28:40,080 other lovable coding tools that constantly are flooded with 426 00:28:40,080 --> 00:28:43,560 requests like this, I think, and it would be super useful. 427 00:28:43,560 --> 00:28:46,480 I saw my my girlfriend actually building an application. 428 00:28:46,480 --> 00:28:50,720 She's I wouldn't say non-technical because she's been really diving into it, but 429 00:28:50,720 --> 00:28:52,840 certainly he does not have an engineering background. 430 00:28:52,840 --> 00:28:58,560 And the amount of dashboards that I've seen, well, not very well working a little 431 00:28:58,600 --> 00:29:01,920 bit clunky, that could benefit from something like that. 432 00:29:01,920 --> 00:29:05,000 Then maybe some some closing thoughts. 433 00:29:06,000 --> 00:29:10,510 We can talk a little bit about your ideas on what you've seen. 434 00:29:10,520 --> 00:29:14,920 You already mentioned some primitives for your product, but more be more for the 435 00:29:14,920 --> 00:29:16,160 industry itself. 436 00:29:16,360 --> 00:29:21,000 I think 26 is is a very interesting year, because 25 was kind of seeing that a huge 437 00:29:21,000 --> 00:29:24,240 shift of adoption everywhere outside of Silicon Valley. 438 00:29:24,240 --> 00:29:28,280 I think we've been a little bit ahead to to speed. 439 00:29:28,760 --> 00:29:32,560 Yeah, be frank. But there are more people building UIs. 440 00:29:32,560 --> 00:29:36,440 They are building people building with application with AI. 441 00:29:36,960 --> 00:29:38,320 How do you see 442 00:29:39,400 --> 00:29:43,360 the role of the developer shifting in the next year to come? 443 00:29:43,840 --> 00:29:46,640 And how do you see the influx of new 444 00:29:47,640 --> 00:29:53,000 AI enabled developers, or by coders playing into your business, as well as our 445 00:29:53,000 --> 00:29:56,920 business as dev tool companies think there's going to be a shift to more full 446 00:29:56,920 --> 00:30:01,400 stack developers, and everyone will need to become, 447 00:30:03,640 --> 00:30:07,920 full and on both the front end, the back end, the data structures 448 00:30:09,400 --> 00:30:13,880 obviously are going to be different interests, but I think that's a good 449 00:30:13,880 --> 00:30:17,680 place. You can grow as an engineer today is like if you've been doing a lot of 450 00:30:17,680 --> 00:30:23,880 backend work. You can now go dive deep into frontend very easily, something 451 00:30:24,160 --> 00:30:27,000 you had to learn over weeks worth of experimentation. 452 00:30:27,000 --> 00:30:31,720 You can learn in like a few hours by asking the AI to answer all your questions 453 00:30:31,720 --> 00:30:33,880 and just pairing with it. 454 00:30:36,520 --> 00:30:41,520 Similarly, I think the other thing that I've been doing a lot more is running 455 00:30:41,520 --> 00:30:43,280 agents in a loop. 456 00:30:45,440 --> 00:30:46,640 I was one of. 457 00:30:50,400 --> 00:30:56,320 I was pretty early in using quadruped and the Tabata complete of GitHub 458 00:30:56,320 --> 00:30:57,160 copilot 459 00:30:58,520 --> 00:30:59,880 when it was better. 460 00:30:59,880 --> 00:31:00,520 And 461 00:31:01,440 --> 00:31:07,640 I think one really cool loop is having an agent look at the code rabbit 462 00:31:07,960 --> 00:31:11,480 errors and then fixing them and pushing updates. 463 00:31:11,680 --> 00:31:16,160 And this is something that's constantly running in the background on my machine. 464 00:31:16,440 --> 00:31:20,920 Okay. There's not that many that much good tooling for this though yet that's like 465 00:31:20,960 --> 00:31:23,200 these automated loops. 466 00:31:23,240 --> 00:31:29,720 One of them is look at the code review bots suggestions and the easy to pick 467 00:31:29,720 --> 00:31:32,240 ones. You can already merge the other ones. 468 00:31:32,640 --> 00:31:37,880 Give it to me in a summary I can like I go in cloud code and like it gives me bullet 469 00:31:37,880 --> 00:31:39,880 point a numbered bullet points of. 470 00:31:42,320 --> 00:31:44,240 This idea seems good. I'm not sure about it. 471 00:31:44,240 --> 00:31:46,040 This idea seems good. Not sure about it. 472 00:31:46,040 --> 00:31:49,600 I just implemented the rest and I can say one, two, three are good. 473 00:31:49,800 --> 00:31:52,400 Skip four, skip 5 to 6. 474 00:31:52,560 --> 00:31:55,360 Then it goes and does that and then the PR is ready to merge. 475 00:31:57,040 --> 00:32:02,720 CI is failing because there's a lint error and my agent running in the background 476 00:32:03,120 --> 00:32:05,040 automatically sees this and pushes an update. 477 00:32:05,040 --> 00:32:05,600 Their. 478 00:32:07,120 --> 00:32:12,360 The other thing we do is run a coding agent in a while loop to 479 00:32:13,440 --> 00:32:19,360 check our docs, and if there's anything out of date, it goes on, fixes it, and 480 00:32:19,360 --> 00:32:22,920 it's like a script that goes through every single file and it runs overnight, 481 00:32:24,280 --> 00:32:25,320 and that works pretty well. 482 00:32:25,320 --> 00:32:30,760 And I'm pretty sure you can think of like 20 or 30 other patterns like this that are 483 00:32:30,760 --> 00:32:32,760 generalizable across projects. 484 00:32:33,560 --> 00:32:39,920 And I'm missing the platform or primitive to run this end to. 485 00:32:40,200 --> 00:32:42,280 I think that's coming pretty soon. 486 00:32:42,800 --> 00:32:44,480 That's a very interesting take. 487 00:32:44,520 --> 00:32:45,760 I love it. 488 00:32:46,840 --> 00:32:51,400 So you're basically running Ralph Loops constantly on your entire production code 489 00:32:51,440 --> 00:32:54,040 base. I mean, not pushing to production immediately, obviously. 490 00:32:54,040 --> 00:32:58,600 But yeah, Dexter showed me the rough loop 491 00:32:59,520 --> 00:33:01,840 paradigm and I 492 00:33:03,680 --> 00:33:09,080 dragged him into my hackathons to port a few projects to a different coding. 493 00:33:09,080 --> 00:33:14,800 I encourage and report a browser use and a few others that browser use the 494 00:33:14,840 --> 00:33:17,160 TypeScript and call it better use. 495 00:33:17,160 --> 00:33:18,640 And that blew up on. 496 00:33:21,160 --> 00:33:27,000 Hacker News as well. And since that's caught on a lot, and just recently 497 00:33:27,640 --> 00:33:32,520 some imported cloud code to Python, and that's very interesting to see as well. 498 00:33:32,560 --> 00:33:37,680 And porting software to a different language is definitely one of these things 499 00:33:37,680 --> 00:33:39,240 that you can run in a rough loop. 500 00:33:39,560 --> 00:33:44,560 And like I said, checking your docs is another one, and I'm pretty sure there's 501 00:33:44,600 --> 00:33:46,880 like a few others out there that work really well. 502 00:33:47,560 --> 00:33:49,960 Ralph soups. Great. 503 00:33:49,960 --> 00:33:50,400 Yeah. 504 00:33:51,440 --> 00:33:55,600 I've been seeing a little bit of problems with running some of them on, on some of 505 00:33:55,600 --> 00:33:59,680 my projects, but maybe I have to go back and reiterate. 506 00:33:59,720 --> 00:34:01,400 It really depends on what you're trying to do. 507 00:34:01,400 --> 00:34:03,000 For some things it works well. 508 00:34:03,000 --> 00:34:05,080 For many others it doesn't. 509 00:34:05,320 --> 00:34:10,440 What will be your top three tips for doing it successful besides reset docs already? 510 00:34:10,760 --> 00:34:14,920 We we said having bullet points so giving some kind of decision narrative. 511 00:34:15,440 --> 00:34:18,200 Is there a third one a recommendation? 512 00:34:18,720 --> 00:34:23,840 I have a blog post coming up and it's called Lean Forward and it basically says 513 00:34:23,840 --> 00:34:26,960 as an engineer in 2026, you have two options. 514 00:34:27,360 --> 00:34:32,880 You can lean back and just let the AI do the vibing. 515 00:34:32,920 --> 00:34:38,840 Yeah that the I cook and it you get to an 80% good result. 516 00:34:39,440 --> 00:34:45,760 Or you can lean forward and it's like you're working like ten times harder. 517 00:34:46,040 --> 00:34:48,720 You're working as much as you did in 2018. 518 00:34:48,840 --> 00:34:55,200 But in return you get like the remaining 20% of what you're 519 00:34:55,200 --> 00:34:59,440 capable of. And psychologically, that seems like a really bad trade off because 520 00:34:59,440 --> 00:35:03,840 you get 80% of the work done with 10% of the effort. 521 00:35:05,000 --> 00:35:09,520 I'm arguing that you shouldn't do that, and all the alphas in the remaining 20%, 522 00:35:09,560 --> 00:35:13,480 and you want to go and really, really focus. 523 00:35:14,760 --> 00:35:20,280 Because what else does differentiate you from any other software program or non 524 00:35:20,280 --> 00:35:21,680 programmer, even at the end. 525 00:35:21,720 --> 00:35:23,040 Right. That's right. Yeah. 526 00:35:23,080 --> 00:35:26,320 All the alpha is going to be in the remaining 20%. 527 00:35:26,480 --> 00:35:30,280 And you working as hard as you had to in 2018. 528 00:35:30,280 --> 00:35:34,160 In 2018 you then have the choice to lean back and chill. 529 00:35:34,200 --> 00:35:35,520 No code got rid of them. 530 00:35:36,080 --> 00:35:37,960 Wouldn't compile if you miss that semicolon. 531 00:35:38,000 --> 00:35:44,400 But yeah yeah I think that's a that's a great, great point to start wrapping 532 00:35:44,400 --> 00:35:48,400 up this conversation. I have a little tradition that I do with every amount of 533 00:35:48,400 --> 00:35:52,280 my guests here. I have a small round of rapid fire questions. 534 00:35:52,640 --> 00:35:55,040 Yeah. If you're ready I'll ask them to. 535 00:35:55,080 --> 00:35:57,840 You can try and answer them shortly. 536 00:35:57,840 --> 00:36:01,000 If it's something that you feel strongly about, please elaborate more. 537 00:36:01,040 --> 00:36:02,800 Okay? Okay. All right. 538 00:36:03,160 --> 00:36:06,240 First question. Your favorite programing language TypeScript. 539 00:36:06,960 --> 00:36:08,800 All right. No doubt there. 540 00:36:09,160 --> 00:36:12,240 What do you think about the rest of the web? 541 00:36:12,280 --> 00:36:18,420 We just talked about the fact that you can now basically use a roof loop to write, 542 00:36:18,480 --> 00:36:23,080 rewrite everything. You think it would be beneficial to have a rust version of a 543 00:36:23,080 --> 00:36:27,400 sustained UI? Not in the near future, at some point, maybe. 544 00:36:27,520 --> 00:36:31,360 One thing I've learned from TypeScript is that you can get unreasonably effective at 545 00:36:31,400 --> 00:36:34,880 optimizing the language that has some built for optimization. 546 00:36:35,200 --> 00:36:36,800 That's really interesting to see. 547 00:36:36,840 --> 00:36:38,640 So we're betting more on that. 548 00:36:38,760 --> 00:36:40,640 Okay. Yeah. Cool take. 549 00:36:41,160 --> 00:36:44,400 What's your favorite coding model still cloud. 550 00:36:44,440 --> 00:36:48,930 Just because I've worked so much with it and I know it's quirks okay. 551 00:36:49,000 --> 00:36:51,440 Sticking to cloud code as an as a harness. 552 00:36:51,760 --> 00:36:55,200 All right. What's the coolest thing you built with assistant UI? 553 00:36:55,520 --> 00:37:00,200 What? I built myself as a language learning app that just lets you converse 554 00:37:00,200 --> 00:37:03,200 in a different language, and it corrects you if you make a mistake. 555 00:37:03,280 --> 00:37:05,080 That's pretty. Yeah. Pretty handy. 556 00:37:05,120 --> 00:37:07,480 Yeah. What language are you learning? 557 00:37:07,560 --> 00:37:09,440 French. Oh, okay. 558 00:37:09,480 --> 00:37:11,080 Cool. Very cool. 559 00:37:11,160 --> 00:37:15,000 Circling back into our conversation before, do you think there's too many AI 560 00:37:15,080 --> 00:37:18,480 tools right there in the market right now for developers, or do you think we still 561 00:37:18,480 --> 00:37:23,400 need more? I think the tools out there cluster around a few things that already 562 00:37:23,400 --> 00:37:27,800 work. There's just so much that I do by hand, and I don't understand why this 563 00:37:27,800 --> 00:37:29,840 isn't automated yet. 564 00:37:30,160 --> 00:37:32,000 Yeah, okay. Makes sense. 565 00:37:32,040 --> 00:37:37,040 What's the most underrated feature of system UI that people should know about? 566 00:37:37,080 --> 00:37:42,040 You can customize every pixel, and most people stick to the default chat style. 567 00:37:42,100 --> 00:37:42,360 So 568 00:37:43,760 --> 00:37:46,160 okay, okay, we give you zero CSS. 569 00:37:46,680 --> 00:37:50,040 We give, you know, course and it all lives in your code base. 570 00:37:50,080 --> 00:37:51,960 You can go and customize it like crazy. 571 00:37:52,200 --> 00:37:53,640 Awesome. Yeah. 572 00:37:53,800 --> 00:37:56,920 Is there one last thing you would like to to highlight? 573 00:37:56,960 --> 00:38:00,200 Maybe that people should know about you or about the team assistant UI? 574 00:38:00,440 --> 00:38:04,310 We're hiring and we mainly care about design. 575 00:38:04,320 --> 00:38:10,320 So if design is something that interests you, regardless of how much 576 00:38:10,560 --> 00:38:14,680 coding experience you have and how much design experience you have, if that's 577 00:38:14,680 --> 00:38:16,480 where your heart says, I want to talk to you. 578 00:38:16,640 --> 00:38:19,000 Okay, great. We'll close without that. 579 00:38:19,040 --> 00:38:20,760 Thank you so much for coming through. 580 00:38:20,800 --> 00:38:22,720 Thank you as well. Awesome.