[0:00] interesting said software will eat the [0:01] world. I I feel like coding will eat all [0:03] knowledge work, right? And we're kind of [0:05] going that direction already. [0:06] The whole agent stack is emerging. [0:08] Identity, payments, marketing, even CLI [0:11] versus MCP. Like all of these are really [0:14] new things and I think a lot of the old [0:15] playbook goes away. [0:17] Yeah. It's a whole new world. I hope [0:19] more companies will stay small and I [0:21] think the founders of this generation [0:22] realize that like they want to stay as [0:23] small as possible. [0:24] Yeah. And instead of having like a 10 [0:26] person product team, you have like a two [0:27] or three person product team and you [0:28] just have a bunch of a agents help help [0:30] you. [0:30] Yeah. [0:30] Someone tweeted that like the the job [0:32] market is so bad that I can only pursue [0:33] my dreams now or something like that. So [0:35] like it's it's like you know it's like [0:37] Yeah. So maybe you lost your job but [0:38] like now you can actually do your own [0:39] thing% and have a shot at actually [0:41] achieving it, you know? [0:42] Yeah. [0:49] All right. Welcome everyone. I've got my [0:50] friend Peter Yang here. Peter, welcome. [0:52] Yeah. Great to be It's good to see you [0:54] again. [0:54] Yeah. Yeah. It's great to see you. Peter [0:55] and I worked together at Credit Karma [0:57] for a brief stint. Um and then we went [0:59] our separate ways and um you know, I [1:02] rediscovered um Peter from his prolific [1:04] posts on X and your YouTube and you [1:07] know, you've got a little bit of a a [1:08] Clark Ken Superman thing going because [1:10] you still got a day job, right? [1:11] That's right. I still have a day job. [1:12] Yes. [1:12] Yeah. Can you share where? [1:13] Uh yeah, I work on Roblox as a PM. [1:15] Amazing. Roblox and Dre Portfolio [1:17] Company. [1:17] Yes. [1:18] One of my favorites. Um well, [1:20] incredible, man. Let's get right into [1:21] it. Maybe I'll start with a softball fun [1:23] question and then you know we're going [1:24] to talk about everything in the claw [1:25] ecosystem. We're going to talk about [1:26] coding agents. We're talking about a [1:29] little bit about maybe what students [1:30] should study advice and some of the [1:32] things that you've talked about online. [1:33] Yeah, sure. [1:34] Um maybe to start, what is the name of [1:36] your Well, how many claws do you have? [1:38] And tell me their names. [1:38] Uh I only have one. I call her Zoe. [1:41] Zoe. [1:41] But I have like multiple conversations [1:42] going with her. [1:43] Okay. [1:44] Yeah. [1:44] And why Zoe? [1:46] Um I I I have two girls and I was going [1:48] to call my younger one Zoe and I did [1:50] not. So I'm I call my open claw Zoe [1:52] instead. [1:53] I see. I see. Yes. Yeah. This is your [1:55] fallback plan. Peter, tell me a little [1:57] bit about, you know, open claw, how you [1:59] discovered it, um how you're using it [2:01] today and and what you think the [2:03] implications are. [2:04] Yeah, I was lucky to interview Peter [2:06] Steinberger before he became super [2:07] famous and the whole thing blew up. [2:09] And um then right after I interview him, [2:11] I I like set up the thing. It took [2:12] forever to set up. It was super janky. [2:14] Um and and yeah, it does a lot of things [2:16] for me. It like pulls analytics for me [2:18] across um YouTube and like my Mercury [2:21] banking account. [2:22] It um can update Google documents for [2:24] me. It can build little web for me. But [2:26] if I was honest with you, dude, like I [2:28] mostly just talk to it through voice and [2:30] get voice replies and like every other [2:32] day I asked you to give me like a pep [2:33] pep talk like you know give me like like [2:37] look look through all your memory and [2:38] like give me some like deep insights [2:39] that I I don't know about. [2:40] Okay. and and like it gave me like like [2:43] I remember I was on a walk and it gave [2:44] me like a three-minute pep talk that was [2:46] like really amazing really amazing like [2:48] it was something about like uh like oh [2:50] you're like talking to me about your [2:51] career business and blah blah blah and [2:53] like your your job but like just [2:54] remember that your kids you know 74 are [2:57] going to grow up very soon and they're [2:58] not going to want to spend time with [2:59] you. [2:59] Wow. [3:00] So like you should really optimize for [3:01] you know them instead. [3:03] Yeah that's really cool and I mean very [3:05] cool but also something that all the [3:07] language models could have done prior. [3:08] Yeah. So what's the difference between [3:10] this and in a use case like that? [3:11] Yeah, that's a very good question. So I [3:13] don't know because I have it installed [3:14] on Telegram. It just feels like more [3:16] personal than using like cloud or chat [3:18] GPT [3:19] and and and it just feels like something [3:20] I can like text in bed. [3:23] It's probably not very healthy, but like [3:24] I text to it in bed. I I talk to it [3:26] during my commute [3:27] and it feels like it feels more like a [3:29] personal like like actual human. [3:31] Yeah. Yeah. So then, so how much for you [3:33] is OpenClaw about the kind of interface [3:36] like pushing it to messaging and you [3:37] know maybe helping to trick our brain [3:39] into feeling like hey this is a person [3:41] or personesque thing [3:42] um versus all the other components of [3:44] the stack the self modification skills [3:46] directory um every all the all the rest [3:49] I I think it's probably like 70 80% just [3:51] like the per personable part of it [3:53] because I mostly just talk to it and [3:54] like you know through voice but I also [3:56] think like is it's something for first [3:58] of all it is pretty janky it tends to [4:00] forget things a Yeah. [4:02] To keep reminding it. [4:03] But like any kind of zany idea that I [4:05] have, I just have to talk to it and it [4:06] can probably just do like like it's kind [4:08] of just like [4:09] um like the other day I was doing voice [4:11] replies with it. I was like, "Hey, can [4:13] we just have a live phone call instead?" [4:15] And and then I was like, "Okay, you you [4:16] got to connect Twilio. You got to do all [4:18] this stuff." And then okay, fine. I I [4:19] went off and did it. [4:20] Yeah. [4:20] And then we had a phone call. It called [4:22] my phone. [4:22] Oh, really? You have that set up? I've [4:23] been dying to set that up. Okay. [4:25] It's not very good though. Like the [4:26] latency is bad, but like the fact I was [4:28] able to get it going is like pretty [4:29] impressive. So it's kind of like any [4:30] kind of crazy idea I have it can kind of [4:33] kind of do. [4:33] And then in practice, how are you doing [4:35] that? Are you asking it to write a skill [4:37] on the fly? Are you discovering a skill? [4:39] How much of the code gen are you [4:40] actually using? [4:41] Um I mean I talk to in like a super [4:43] casual way with like just just like a [4:45] friend. So like hey you know hey Zoe can [4:47] can like can I have a phone call like [4:49] okay you got to do that. I said okay [4:50] fine I'll open my computer. I'll do all [4:52] this stuff [4:52] and then it's like give me a call and [4:54] then we troubleshoot a little bit and [4:55] then it works. So like I with with cloud [4:57] I have like very fancy prompts uh like [5:00] very long prompts but with open cloud I [5:02] just kind of text it. Yeah, it is [5:04] really interesting. So we sort of [5:05] touched on a couple things actually. So [5:07] one there's mobile messaging, [5:09] there's the memory system. Um there's [5:11] the sort of code generation component. [5:13] Um [5:14] how much do you think the memory system [5:16] like is it innovative because it's file [5:18] based? You said that it forgets things [5:20] but so do language models. Like do you [5:21] think the memory system is well done? [5:23] Does it hold it back or does it enable [5:24] it? I I I think the default memory [5:26] system is actually not that great. [5:27] Okay. [5:28] Like the way I understand it works is [5:29] like just like a memory MD text file. [5:31] Yes. And then every day per day. Per [5:32] day, right? And every day it updates and [5:34] it tends to forget things a lot. [5:36] Yeah. [5:36] So So I actually installed this like [5:38] three layer memory system that to be [5:39] honest I don't fully understand but it [5:42] has like [5:42] that's fancy. [5:43] It has like Toby's QMD search tool. [5:45] Okay. [5:46] Uh so I installed that and and then [5:48] install like a 2 GB thing and and then [5:50] it got a little bit better. [5:52] Okay. Um, but I I said to remind it like [5:54] I have to I had to put it into the [5:55] agents MD like hey like before you [5:57] answer any question from me like go [5:58] through all your memory and like check [6:00] everything. [6:00] Yeah. [6:00] And it also tends to forget that it can [6:02] do stuff like you know like can you [6:04] update my Google doc? It's like oh I [6:06] can't do that. Yes. Yes you can. It's [6:07] it's in your it's in your file. [6:08] Yes. Yeah. [6:09] So you have to remind it. Yeah. [6:10] Yeah. Really interesting. Well well [6:12] maybe let's get into a little bit of the [6:14] controversy. Um you know you'd said that [6:16] apps will die claw is going to be [6:18] everything and everywhere. I mean talk [6:19] us through that that point of view. [6:21] Yeah. Yeah. Well, for first of all, I I [6:22] I tweet all all kinds of random crap [6:24] that's like not super well thought out. [6:26] We take it all as fact. [6:27] Yeah. [6:27] Yes. [6:28] Um but I do think like um like ever [6:30] since I set up all these apps like [6:32] Mercury, MCP, and all this kind of crap [6:34] on my open call, like I I don't actually [6:36] open those apps much anymore, you know. [6:39] But I do agree with you like I I I think [6:40] the ones that are going to die first or [6:42] like maybe get less usage first is like [6:44] apps that you're just opening to try to [6:46] complete a task. like you actually are [6:48] trying to do something, you know, like [6:49] apps that you don't need to like get [6:50] entertainment can probably survive a [6:52] little bit longer, [6:53] but like apps to complete a task like [6:55] it's just way easier to text my agent to [6:57] do it for me. [6:58] Yeah. [6:58] It's like it's like you have a really [6:59] good admin just to do stuff for it for [7:00] you. [7:01] Yeah. [7:01] Yeah. [7:01] And so how much are you finding has this [7:03] like reduced your smartphone usage [7:04] outside of Modulo Open Claw? [7:06] Yeah. Um, no. Because like I'm I'm like [7:08] a Twitter addict, so I still use phone [7:10] way too much. But yeah, in terms of [7:12] using those apps has definitely reduced [7:14] it. [7:14] Yeah. [7:14] Yeah. Yeah. because you're not going to [7:16] ask Zoe like, "Hey, read my X for me and [7:18] tell me what's interesting." [7:19] I mean, it it sends me like a morning [7:21] briefing of like the top two tweets and [7:22] stuff that like trends, but but yeah, I [7:24] I I still open X. I look through it. [7:27] Yeah. Yeah. You know, it's interesting [7:28] because I've always had this theory that [7:29] people open apps on their phone because [7:31] they want to feel a feeling. [7:32] Yeah. [7:32] You know, and I think of course there's [7:34] some like functional set of needs which [7:35] is why you open calendar or something, [7:36] but I also think that, you know, [7:37] WhatsApp is you want to feel connected [7:40] and Slack is you want to feel productive [7:42] and of course, you know, Tik Tok is you [7:43] want to feel entertained. So I do wonder [7:46] with just one agent, how do you sort of [7:49] do the context switching of like when [7:51] are you flirting, when are you getting [7:52] done? I mean there you know in a [7:54] sense app gives you a nice it sort of [7:56] gives you a nice division of the [7:57] intents. [7:58] Yeah, you don't get with Zoe. [7:59] That's a good point. But I I do have [8:01] multiple channels set up with Zoe in [8:02] Telegram. Like one is just to random [8:04] voice replies and the other one is we're [8:06] actually working on our project [8:07] together. [8:08] Oh, [8:08] and another one I have like a public [8:09] channel where like I'm giving that [8:11] demos. I don't want to reveal private [8:12] information. Yes. So I have like [8:13] multiple channels [8:14] and is that um implemented as sub agents [8:16] or [8:17] No, it's just some janky setup I I found [8:19] online like you can set up a multiple [8:21] Telegram channels and then I'm not sure [8:23] if she actually remembers across context [8:25] across the channels but like you can [8:26] have separate conversations at least. [8:28] Got it. [8:28] Yeah. [8:29] And how how you know transparent are you [8:32] with your agent? Like does they do they [8:33] see your personal email or [8:35] uh I'm I'm like super transparent. Well [8:38] I I I I did buy the Mac Mini and set up [8:40] its own email. [8:41] Okay. And but I gave it like read access [8:43] to my email and like calendar and uh I [8:46] also gave it like right access to some [8:48] docs. [8:48] Yeah. [8:48] But it can like scroll my entire drive [8:50] or something you know. [8:51] Yeah. So [8:52] how do you imagine open claw which it's [8:54] sort of an architecture and a primitive. [8:56] How does it get productized packaged for [8:58] the world? [8:59] I mean I think that's what Peter [9:00] Steinber is working out at OpenAI right. [9:03] It's probably going to build something [9:04] to chat GBT which everybody uses uh so [9:08] that chat GP can actually get stuff done [9:09] for you and like maybe feels more human. [9:11] Yeah, [9:12] dude. Like let me rant about chat. [9:14] Yeah. Yeah. [9:15] For some reason for some reason they [9:16] they trained the model so that like at [9:18] the end of every conversation is always [9:20] like if you want I can also do X and Y. [9:22] Yeah. [9:23] And I I got so annoyed about it that [9:24] that kind of turned from Chad GPT. [9:26] Oh really? [9:26] Yeah. So so it probably increases their [9:28] matrix but like it's just like super [9:30] annoying. It's like why don't just do it [9:31] in the first place? Are you a cloud guy [9:33] now? [9:33] Yeah, I'm I'm a cloud guy now. But but I [9:35] I I do use codeex uh to code. [9:38] Yeah. [9:38] Yeah. [9:38] You like codeex, you prefer it to cloud [9:40] code or you use both? [9:41] Um codeex when I want to try to build [9:43] something real and cloud code is when [9:44] I'm just like vi vibing like very you [9:47] know well it's interesting. I think they [9:48] live at different points and you know [9:50] there's a sort of space of trade-offs. [9:52] Yeah. [9:52] Whereas I find um cloud code and opus 46 [9:55] it's a little more chatty. It makes more [9:56] assumptions but it can be more pleasant [9:58] for a synchronous experience. Whereas [10:00] codecs, it really thinks hard and it's [10:03] more often accurate. But sometimes it's [10:05] sort of like being in a conversation [10:06] where the other person pauses for like [10:08] three minutes to think. [10:09] Yeah. [10:10] You don't have to flow flow state, [10:11] right? It's hard to get a flow like [10:12] clock. Dude, I I tweeted the other day. [10:14] Clock is almost like uh like a slot [10:16] machine. It's like it's like has [10:18] different things each time. It's just [10:19] like [10:19] Oh, 100%. Look, I I do think that if you [10:22] think remember we were talking about in [10:23] the old social networking era, it was [10:25] variable scheduled rewards, right? That [10:26] was the whole magic of it. like you open [10:28] your Facebook feed and you know once in [10:30] a while it's like boring boring oh my [10:31] god this is so exciting and I the coding [10:33] agents have the exact same property also [10:36] the time is variable so sometimes you [10:38] get something in a second sometimes it [10:39] takes five minutes so up to a certain [10:42] point I actually think that both of [10:43] those things give it that casino like [10:45] feeling [10:45] yeah and and the other thing that's very [10:47] different about the product strategy or [10:48] maybe it's just the way it works is like [10:50] coding is kind of like self-explanatory [10:52] and clock hole you have all this crazy [10:53] you have like hooks and like skills [10:55] and like you have to you have to plug in [10:57] If if you if you're not following Yeah. [10:59] If you're not following X, you have no [11:00] idea how do you customize this thing. [11:01] Yeah. [11:01] But once you customize it, you kind of [11:03] feel like it's part of you. So it's it's [11:04] kind of hard to turn. [11:05] It's interesting with uh so I've [11:07] customized mine cuz also I read the long [11:09] thing that Boris put up. Yeah. [11:10] But I will say that I think that you [11:11] know cloud code a lot of the reasons [11:14] that I enjoy it are just harness [11:16] features. [11:17] You know like for example if you cut an [11:20] image you have to paste it into a file [11:22] before and then paste that file into [11:24] codeex. Okay. You can't just take a [11:26] subset of the screen, screenshot it, and [11:28] then paste it directly into codeex the [11:30] same way you can with cloud code. [11:31] Oh, really? Okay. [11:32] So, just like little things like that. [11:33] You know, cloud code added voice. It's a [11:35] little bit janky right now, but it's [11:36] going in the right direction. So, [11:38] they've just got a bunch of quality of [11:39] life things. [11:40] Yeah. [11:40] You know, Cloud Code speaks to Claude [11:42] and Chrome. [11:43] Okay. [11:43] And Codex doesn't speak to Atlas. [11:45] Got it. [11:45] So, I think these are all things that [11:46] OpenAI will fix. Yeah. I think Codeex is [11:48] actually a much better model. Um, but [11:51] they don't exist today. [11:52] Yeah. Yeah. They need to fix it. I mean, [11:53] they're going to go all all in on [11:55] cohortex, I'm sure. [11:56] Yeah. [11:57] Talk to me about coding agents. Like, [11:58] what's your general view? You know, do [11:59] you think it's the end of SAS? Do you [12:01] think these are just a toy? [12:03] Uh, well, first of all, I'm I'm I'm like [12:05] not an engineer, so I'm like a novice. [12:07] But I do hear that um [12:09] like I was talking to some folks uh the [12:11] other day and like like a AI native star [12:14] startup and they're basically trying [12:15] trying to they have a bunch of vibe [12:16] coders [12:18] and a lot of vibeers are just trying to [12:19] build internal tools that replace their [12:21] SAS that they're paying for. [12:22] Really? So it's an actual company that's [12:24] doing this. [12:24] It's an actual company. Yeah, [12:25] it's an AI native company. [12:26] Is it It's like one of the vi coding [12:27] companies, like one of the more popular [12:29] OS. [12:29] Interesting. [12:30] Yeah, [12:30] it's Oh, oh, I see. So they're actually [12:32] an appgen company. [12:33] They're appgen company and they they're [12:34] paying for a bunch of SAS like uh and [12:36] they want to get rid of the payment. [12:37] They want just buy by by coding internal [12:39] tools [12:39] by using Okay. So in that case that that [12:41] they might be the most extreme form of [12:43] adopter because their own product is [12:44] AppGen. So they should use Appgen for [12:46] everything. I guess is your prediction [12:48] though that the average company will [12:50] churn off of Slack or Deal or you know [12:54] um I don't think I I I feel like Slack [12:56] has a lot of legs because Slack can also [12:58] be the place where you talk to the [12:59] agents themselves. [13:01] Uh but some of the other ones they are [13:03] pretty complicated you know so like it's [13:04] kind of be hard to buy that kind of [13:06] stuff [13:06] but but I feel like if you have like an [13:07] app like maybe Calendarly or like [13:10] something more simple [13:11] Yeah. [13:11] then why why why should I why should I [13:13] pay for it? Like I just [13:14] why should I pay for it though? The [13:15] counter point is that it's not that [13:16] expensive. And do you really want to [13:18] maintain your own Calendarly thing? [13:21] Yeah. [13:21] You know, versus pay 20 bucks a month. [13:23] It always gets updated. It's always up, [13:25] you know, because there's just like a [13:26] fixed amount of capacity that anyone in [13:27] the organization is going to have for [13:28] all this stuff. [13:29] Yeah. That that that's true. Unless you [13:31] hire like dedicated viers like the [13:32] startup does, right? Voicable stuff. [13:34] But then it's like, you know, the cost [13:35] benefit versus just paying for [13:36] Calendarly. Yeah. Yeah. Yeah. Yeah. [13:38] Yeah. It's interesting about like for [13:40] example like a lot of people are [13:41] tweeting about Figma recently. [13:43] Yeah. uh like like the stock is down and [13:45] like you know are are you going to [13:46] survive? [13:47] Yeah. [13:47] And um [13:49] I feel like the jury is out there like [13:51] it's kind of hard to say. Yeah. Uh I I I [13:53] feel like all the designers are still on [13:54] fake Figma but like as a designer you [13:57] kind of need to learn how to vibe code [13:58] otherwise you're going to like if you [14:00] don't know how to do Figma. [14:01] Yeah. [14:01] Like you're probably going to be like [14:02] out of date [14:04] in a couple years. Yeah. [14:05] My counterpoint to that is that I think [14:07] that I've thought a lot about the sort [14:08] of thinking tools versus making tools, [14:09] right? The IDE was historically a making [14:12] tool. Mhm. [14:12] It's a place for execution. I think it's [14:14] migrating away from that. And now with [14:16] execution going to zero, I think these [14:18] sort of like multi- aent, you know, [14:20] nextg ides, a lot of them are about [14:22] trying things [14:23] and using the trial and error as a way [14:25] to inform your thinking. Like a lot of [14:27] times I'll just build a feature in a [14:29] really naive way and I'll hammer the [14:30] coding agent until it works. Then I'll [14:32] say, "Hey, write all the things that you [14:34] would have done differently and I'll go [14:36] back to the initial point and redo it." [14:38] So I wonder if and I think Figma [14:40] actually does both. I think it's a place [14:41] for design execution, but it's also an [14:43] important place for design thinking. And [14:44] I think that's their opportunity to be [14:46] highly relevant in the new stack. [14:48] Yeah, I totally agree. I totally agree. [14:49] Um, but I I I think A6Z has like uh you [14:53] guys investing pencil or something. [14:56] Pencil.dev. [14:56] Speedrun did. Yeah. [14:57] Yeah. Speedrun. And like um [14:59] yeah, Figma needs to like uh level up [15:01] his AI tooling because you know like [15:03] watching these agents collaborate with [15:04] you and like do stuff is like very very [15:07] interesting. [15:07] I know it's top of mind for them. Um [15:09] yeah, what do you think are the most [15:10] under discussed capabilities of coding [15:13] agents? You know, what's what's [15:14] underhyped and maybe what's overhyped as [15:16] well? [15:16] This is probably not underhyped, but [15:17] like but you know, like you know, I I I [15:19] feel like ent software will eat the [15:21] world. I I feel like coding will eat all [15:23] knowledge work, right? And we're kind of [15:25] going that direction already. Like I I [15:26] think lovable recently launched like [15:28] today. Yeah. [15:28] That they can support everything and [15:30] replic. [15:31] Yeah. [15:32] So, so yeah. So I and I feel like [15:34] everyone's cha chasing this like and is [15:37] probably in the lead. [15:38] Yeah. But like, you know, like I I don't [15:39] want to use PowerPoint anymore. I don't [15:40] want to like write a Google like I hate [15:42] writing Google Docs, dude. Like my [15:44] entire life. [15:44] Yeah. Yeah. So, so like but but but the [15:47] other day I was writing my blog post and [15:49] instead of just like typing it out, I [15:51] was like, "Hey, let's let me just use [15:52] clock code and like you know, let me [15:54] give you a bunch of feedback and you you [15:55] write it for me." [15:56] Yeah. [15:56] And then you just keep the [15:58] um [15:59] it it did the first 80%. The last 20% I [16:01] had to manually like go in there like [16:03] tweak tweak tweak stuff. [16:04] Yeah. [16:04] But like that that's the way I work now. [16:06] I I never start from zero. like I always [16:08] get the first 80% from AI, [16:10] right? [16:11] Yeah. [16:11] Yeah. It's interesting, you know, if you [16:12] look at there there are also like [16:13] historical analoges of this. I think [16:15] Satya said this um which is that Excel [16:18] is the most powerful or most popular [16:20] programming language in the world. [16:21] Yeah. [16:21] And that it's sort of a programming [16:23] language that millions and millions I [16:25] mean 100 million plus people must know [16:26] maybe even more. Um and yet we don't [16:29] think of it that way. It's a way to sort [16:30] of describe and solve problems. [16:33] Yeah. [16:33] And I think coding agents are going to [16:35] be that of course times a thousand. [16:37] Yeah. [16:37] Where even things that feel subjective [16:39] like writing Google docs can be [16:41] represented in the coding domain in such [16:43] a way that it's more satisfying [16:45] productive more to use agents to do it. [16:47] Yeah. Because Excel was like popular [16:49] because it's super appro approachable, [16:50] right? Yes. And like coding agents the [16:52] code is basically gone like it's like [16:53] appra just talking to some some agent [16:56] and getting to do do stuff. So [16:57] Yeah. Yeah. [16:58] Yeah. Exactly. [16:58] It's going to be hu huge. Yeah. [17:00] What do you think the future company [17:01] looks like? Is it just a bunch of agents [17:03] with a CEO? Is the CEO an agent? I mean, [17:06] what is the role for people in a company [17:07] in the future? [17:08] Okay. Well, I have some hot takes. So, [17:09] we we both worked at uh some companies [17:11] together and um [17:12] Uhhuh. [17:13] Uh let let me give you a hot take, man. [17:15] Maybe we cut this out, but like I feel [17:16] like as a company gets bigger, [17:17] it tends to get it tends to become [17:19] like a shitty shittier place to work, [17:20] dude. Yeah. Like like because there's [17:21] like a lot of people you have to align. [17:23] I think that's axiomatic. Yeah. [17:24] Right. And and I remember, you know, [17:26] maybe I should mention this company, but [17:27] I remember our company used to have all [17:29] these like OKR meetings and like I [17:30] remember sitting in a room for like [17:31] three hours talking about OKRs. I'm just [17:33] like, dude, this is like wasting my [17:34] life. [17:34] Yeah. So what where I'm going with this [17:36] is I hope more companies will stay small [17:38] and and I think the founders of this [17:40] generation realize that like they want [17:41] to stay as small as possible. [17:42] Yeah. [17:43] Um and instead of having like a 10 [17:45] person product team, you have like a two [17:46] or three person product team and you [17:47] just have a bunch of agents help help [17:49] you. [17:49] Yeah. [17:49] You know, because I I I think it's way [17:51] easier to cross launch a line with a [17:52] agents than with with humans. [17:54] Yeah. Yeah. Well, actually in a in a [17:56] sense the agents actually because it [17:57] takes the emotion out of it too. Like [17:59] you can imagine if I sent my agent, you [18:01] sent your agent to go negotiate [18:02] something. Yeah. [18:03] And they came out with some conclusion. [18:05] It's not emotional. [18:06] It's not [18:07] for either of us. You know, [18:08] it's very objective. Yeah. [18:08] Yeah. Yeah. Exactly. Yeah. It's funny. [18:10] You know, one of the things that we've [18:11] been talking a bunch about is like [18:13] what is the procase for AI at work in [18:16] terms of employee experience? And I [18:17] think it's what you're describing, [18:18] right? Like how do you increase the NPS [18:20] of work? Yeah. So, if we like go all the [18:22] way back or even broadly the NPS of the [18:24] human experience, right? Like think of [18:25] the NPS of the day-to-day human in [18:27] 10,000 BC when it's like just don't get [18:29] eaten by the lion and that's like a good [18:31] day, right? Or you know maybe a 100 [18:33] years ago it's like okay don't get [18:34] killed at the factory [18:36] um crushed by the the steam press or [18:38] whatever else. And um and now a lot of [18:40] it is like h like just don't get sucked [18:42] into some high emotion you know sort of [18:45] negotiation with another VP's [18:48] subordinate. Um [18:49] yeah like a 50 me stack thread going [18:50] back and forth. [18:51] Yeah. Exactly. And then eventually [18:52] everyone's like, I don't want to tell [18:53] the CEO. And eventually it goes there [18:55] and it's just terrible. Um, so maybe the [18:57] future of this is that a lot of that [18:59] emotional subjective work gets handled. [19:01] Yeah. [19:02] And we're sort of guiding the process [19:04] but not in the middle of it in a way [19:05] that just doesn't suit us as humans. [19:07] Yeah. Like you know, I I I lead a double [19:09] life as a PM creator and like I feel [19:10] like all the PMs actually just want to [19:12] create products. They want to create [19:14] products and and [19:15] well that's why we all got into it. You [19:16] know, it's it's so interesting. I mean, [19:17] Nicole talks about this all the time, [19:19] but like every PM's sort of view view of [19:22] the ideal PM is the innovator, you know, [19:24] like I came up with the new thing and [19:26] it's like I sort of like had the big [19:28] insight and it unlocked the product. [19:29] Yeah. [19:30] I think the black pill is I don't think [19:32] most PMs know how to do that. In fact, [19:34] many companies have zero people that [19:36] know how to do that at all in any [19:37] function. [19:38] Yeah. So, nonetheless, I think PMs [19:40] aspire to be able to do that and they [19:42] should either do it and either be [19:44] successful or maybe not successful and [19:46] move to a different function. I I also [19:47] feel like my hot take is like uh like [19:49] basically all the PMs I know are trying [19:51] to vibe code at night on weekends. [19:53] Yeah. [19:53] And I feel like my hot take is that like [19:55] I feel like if you're actually [19:55] unemployed like you probably have more [19:57] time to be a builder and like to be [19:58] innovative because you can actually like [20:00] play all this stuff and like learn all [20:01] this stuff. [20:02] A lot of PMs are trying to [20:03] or maybe be an engineer in the team, you [20:05] know? I used to be an engineer and I got [20:07] sort of I don't know if I got forced to [20:08] be a PM. Maybe I also perceived PM as [20:10] like being a little more high status [20:12] when I joined Google. But then [20:14] eventually you like this is terrible, [20:16] you know, like you never really get the [20:18] satisfaction of actually shipping other [20:20] than like you know once a quarter when [20:22] you ship. [20:22] I mean the PM skills of like talking to [20:24] users and like trying to figure out what [20:25] to do like what problem to solve like [20:26] those are very important still. [20:28] Yeah. [20:28] And and like but yeah, you got to wear [20:30] multiple hats, dude. You got you got to [20:31] like go go go build a thing yourself, go [20:32] prototype it and get some feedback and [20:34] then maybe bring engineer along. [20:36] How much do you think that everyone has [20:38] to go as fast as you know I mean like [20:40] Gary was talking about stemmies and [20:42] skipping sleep and 10 like you know [20:44] GStack I mean is hey I mean is that like [20:47] the default way that we all need to work [20:49] or do you think there's a trade-off for [20:50] thoughtfulness? [20:52] I I I think it's very easy now with all [20:54] these AI tools just going like 10 [20:55] different directions at once. [20:56] Yeah. So sometimes you do have to slow [20:58] down and try to figure out where you [21:00] want to go. [21:00] Yeah. [21:00] But I also believe that like u the [21:02] traditional process where you like do [21:03] annual planning and like do all this [21:04] like I just feel like that [21:06] doesn't really work anymore, [21:08] you know? [21:08] Yeah. Yeah. And for the record, I love [21:10] Gary and I like think the world of him. [21:12] So here's my view on that because I [21:13] thought a bunch about that. You know who [21:14] asked me this the other day? Hithan um [21:16] who's really really impressed. Yeah. Ch. [21:18] Yeah. [21:19] So we're talking about this like sort of [21:20] productivity porn and everybody's got 20 [21:22] agents running and 20 monitors and blah [21:24] blah blah. And like I do think when it [21:26] comes to [21:28] um fully realizing a local a sort of [21:32] local maxima, you should go very fast, [21:34] right? So let's say you kind of hill [21:36] climb, you get to the bottom of a new [21:37] local maxima. I think with agents, you [21:39] should be able to get to the top of that [21:41] hill extremely fast, right? You have a [21:43] new insight, build everything around the [21:44] insight so it's fully expressed. But [21:47] then I think to get to the next, you [21:48] know, the next sort of hill, [21:50] you've got to probably slow down and [21:51] almost stop and go touch grass and do [21:53] whatever. Yeah. [21:54] So I think there's this combination of [21:56] like fast and slow. That's probably the [21:58] future way. [21:58] Yeah, I think so. And and like you got [22:00] to go on that random walk trying to find [22:01] hard market fit which take takes us a [22:03] while, right? So it's not like [22:04] Yeah. [22:04] So we were talking before we started [22:06] recording about some of the business in [22:08] a box platforms. [22:10] Have you looked at them? Do you have a [22:11] view? [22:12] I've looked at post yet that we talked [22:13] about like I don't know if the guy like [22:16] intentionally made it the opposite of AI [22:17] slab or or is it kind of you [22:19] I think so. Yes. Yes. [22:22] That's funny. Well, I mean, I have a [22:23] pretty big public presence, right? So, I [22:25] connect all my to to it. And then, [22:26] um, I mean, it's it's definitely gives a [22:29] good peak into what's possible, but like [22:31] right now it's probably still pretty [22:32] like early stage. Like, it's telling me [22:34] to run like fa Facebook ads. Like, [22:36] why am I running Facebook ads? [22:38] Yeah. [22:38] You know, so [22:39] I don't know. Yeah. Yeah. Yeah. I I [22:40] mean, I'm very excited about it because [22:42] it does feel like it's a path for more [22:43] people to build companies. Yeah. Even if [22:46] they're single oneperson companies, you [22:48] know, like if you think about how [22:49] competitive it is to build a billion [22:50] dollar business, like the markets that [22:52] support it, the number of people trying [22:53] versus hundred million versus 10 million [22:55] versus $100,000 TAM. Yeah. [22:58] Like maybe there are these pockets all [23:00] over the country, all over the world [23:01] where there are opportunities for [23:03] $100,000 TAM products. Yeah. [23:05] And that would change somebody's life. [23:07] Now, that's not an enterprise [23:08] ventureback company, but that's okay. [23:10] Yeah. So, I hope that whole thesis works [23:12] because I do think it's a way to get [23:14] more people to participate, you know. [23:15] Yeah. That that that's my plan for my [23:16] kids, dude. Like I want them to just [23:18] build like um bootstrap businesses in [23:20] high school. [23:21] Yeah. [23:21] And they can skip the whole college and [23:23] like uh core corporate life. [23:24] Yeah. [23:25] Well, dude, I think this is like, you [23:26] know, for 10 years there's this moral [23:28] panic about the kids want to be [23:29] YouTubers. Yeah. You know, you're a [23:30] YouTuber. [23:31] Um you know, you know, in the in the [23:33] vein of Mr. And and I think the like pro [23:36] case for that actually is that the kids [23:38] wanted to be entrepreneurs or have [23:40] agency. And the only channel for people [23:42] if they weren't programmers was creating [23:44] YouTube videos at least online. [23:45] Yeah. [23:45] So if you're like an online native [23:47] generation, you want to create something [23:48] you're not a programmer, you make a [23:49] YouTube show. Now you can make a lot [23:51] more than that. [23:52] Yeah. You you can build whatever you [23:53] want. Yeah. [23:53] Exactly. [23:53] So Exactly. [23:54] It'd be very exciting. Yeah. [23:55] Yeah. Any other hot takes for us? [23:57] I'm curious about your thoughts about [23:58] this actually. Like so I feel like uh a [24:01] lot of people are saying like agents [24:02] will interact with your product first, [24:04] right? And and then you see all these [24:06] great companies like uh building like [24:07] APIs and MCPS. [24:09] Mh. [24:10] But like how do how do you think about [24:11] like you know you've been consumer for a [24:12] while. So like consumer is like you got [24:15] to get the user to come back and use [24:16] your product, right? [24:18] Yeah. [24:18] But but now like the user is like hey go [24:20] send the agent to use So, so how do you [24:21] think about retention and all this like [24:22] basic stuff like how do you you know or [24:24] even like brand equity because the agent [24:26] just like point some API like you know [24:28] how do you [24:28] yeah I don't okay so I think one of I [24:31] don't know is the is the truth but um [24:33] there's I have a few thoughts so one I [24:35] think that a lot of the sophistication [24:38] sophistication that happened in consumer [24:40] happened because we had to have indirect [24:43] monetization [24:44] okay like we just were never charging [24:46] consumers directly for these products [24:47] which is why you got [24:48] got ads and stuff [24:49] ads and large scale networks and we all [24:52] obsessed with retention and engagement [24:53] and whales and all of these things [24:55] really really mattered because we didn't [24:57] simply charge people for products. [24:59] So I think one big thing that's actually [25:01] really helped in the AI era with that is [25:02] that consumers are now excited to try [25:04] new things. They're willing to pay [25:05] they're willing to pay a really high [25:06] price point. [25:07] Um there's also consumption revenue in [25:09] consumer for the first time [25:10] like token and stuff. [25:11] Yeah. Yeah. For like tokens. You have [25:12] your subscription plus your token. So [25:14] and then the actual like the sort of [25:16] blessing in disguise is that there are [25:18] real costs as well. you have these [25:19] inference costs. So you're like, "Wow, [25:21] we have to charge our customer on day [25:23] one." [25:24] So I think one thing is that like the [25:25] business model simplification I think [25:27] will really help with a lot of what [25:29] you're describing. [25:30] Two, I think that a lot of the products [25:31] will have a sort of, you know, it'll [25:33] have an API interface for your agents to [25:35] interact with or for, you know, for [25:37] transactional sort of wrote things. [25:39] Yeah. [25:39] And then it'll have like a consumption [25:41] based interface as well. [25:42] So you can also imagine a like a mobile [25:44] app where there's like the feed, but [25:46] then you can kind of turn it over to [25:47] where the wires are. [25:48] Mhm. And you can just ask for things to [25:50] get done or you can just see the log of [25:52] the things that got done. [25:53] Yeah, maybe people would do both, right? [25:54] I mean, you can imagine Credit Karma [25:55] where we worked, you know, like once in [25:56] a while you want to just take a look at [25:58] your score history and a few other maybe [26:00] credit card offers. I don't know. I [26:01] mean, [26:01] yeah. Yeah. If I get my score with all [26:03] kind of credit card offers, I'll [26:04] definitely do that. [26:04] Yeah, 100%. Exactly. [26:06] On the other hand, like sometimes you [26:07] want to just be like, yo, like can you [26:08] just fix all my stuff or like what stuff [26:10] did you fix this week? How much money [26:11] did I save? You know, [26:12] got it. Got it. [26:13] Yeah. [26:14] It's definitely interesting. Yeah. [26:15] But look, I also just think the whole [26:17] agent stack is emerging. [26:18] identity, payments, marketing, we don't [26:22] even even CLI versus MCP. Like all of [26:24] these are really new things and I think [26:26] a lot of the old playbook goes away. [26:28] Yeah, it's a whole new world and like in [26:30] 2025 I thought agents was overhyped but [26:32] now I think it's really coming like [26:34] me too. I know it's just the word is [26:36] frustrating because it gets so [26:37] overloaded. [26:38] Yeah, there's like workflows like all [26:40] this kind of [26:40] Totally. I've been trying to just say [26:41] like can we just say like model in a [26:43] loop? [26:43] Yeah, exactly. Model that use tools in a [26:45] loop. That's the best definition. Yeah. [26:46] Yeah. [26:46] But nobody likes to hear that. It's [26:48] Asians is much flashier, you know. [26:49] Yeah, it's flashier. [26:50] Yeah. [26:51] My my hope is that um [26:53] all all this stuff's like a lot of [26:54] people think like we're going to lose [26:55] our jobs, which probably what will [26:56] happen at some point, but like I hope [26:58] all this stuff [26:59] makes just makes like uh human work more [27:01] fun like our jobs more fun, [27:03] you know? [27:03] Dude, I don't think we're all going to [27:04] lose our jobs. Like I really think the [27:06] and we see this a lot of companies, you [27:07] know. So we look at a ton of companies [27:08] and we've seen two different buckets. So [27:11] one bucket is hey, we dramatically [27:12] increase productivity for a person or a [27:15] team. We see this in like recruiting, [27:17] but we couldn't do 100% of the job. So, [27:19] we could do the phone screen, but we [27:21] couldn't obviously, you know, show the [27:22] candidate around the office or we could [27:24] do the phone screen and we could like [27:26] answer all the questions about the [27:27] company and we could even do the like [27:28] comp negotiation, but we couldn't do the [27:30] onboarding. [27:30] Yeah. [27:31] The other style of company which we see [27:33] which is maybe a decagon, right, or a [27:35] happy robot is hey, we did 100% of a job [27:37] like customer support. [27:39] You know, the the customer called in, [27:40] they had a question, we hopefully [27:42] resolved their query and then that's it. [27:43] M and that is 100% automated. I'd say [27:46] that that second group where you have [27:47] 100% automation of a job function is [27:49] really rare. Almost every AI product AI [27:52] native X or Y we see is able to provide [27:55] dramatic lift but it's not able to do [27:57] 100%. [27:58] So last 10% use humans. [27:59] Yeah, it's still it's today anyway it's [28:01] still humans that do that stuff and it's [28:03] interesting too because the buyer looks [28:04] at that as software as expensive [28:06] software whereas in the case of [28:08] something like a happy robot docking [28:10] Sierra they look at it as like cheap [28:11] labor. M [28:12] so I do think there's a different buyer [28:14] mindset but because there's been this [28:15] difficulty of getting to 100% automation [28:17] I think a lot of the efficiency gain [28:19] shows up in just a different way [28:21] probably not less jobs maybe we get like [28:23] the European style 4 day work week maybe [28:25] companies get like twice as productive I [28:27] have no idea [28:28] yeah but but you don't think that like [28:29] uh I I I feel like there's going to be a [28:31] transition from like these like 10,000 [28:34] plus people companies laying a lot of [28:35] people off to hopefully like more [28:37] smaller companies like solarreneurs and [28:39] stuff like that [28:40] I I think yes I think that they'll the [28:42] the sort of shape of the economy is [28:44] going to change like the amount of [28:45] concentration but I just don't think [28:46] there's going to be less jobs. I think [28:48] human ambition has no ceiling. [28:49] You know, human desire has no ceiling [28:51] and just read any mildly interesting [28:53] science fiction book [28:54] like there's no way this is the peak [28:55] expression of all the stuff that we want [28:57] and we need and we're going to convince [28:58] ourselves and you know all the new [29:00] things that you read about every day is [29:01] these luxuries peptides and you know [29:03] everybody is going to have all of that [29:05] stuff and want even more. You know, [29:07] dude, I I saw a really good tweet about [29:09] this. Like someone tweeted that like um [29:11] the the job market is so bad that I can [29:13] only pursue my dreams now or something [29:15] like that. So like it's it's like you [29:16] know, it's like Yeah. So maybe you lost [29:18] your job, but like now you can actually [29:19] do your own thing. [29:20] And have a shot at actually achieving [29:21] it, you know? [29:22] Yeah. Yeah. [29:23] Cool. [29:23] Well, awesome, man. Maybe that's a good [29:25] uh positive note to end on. [29:26] Yeah, that's good to know. Yeah. Cool. [29:27] Good thing you do. Yeah. [29:28] Thanks, Peter. Yeah. Thank [0:00] interesting said software will eat the [0:01] world. I I feel like coding will eat all [0:03] knowledge work, right? And we're kind of [0:05] going that direction already. [0:06] The whole agent stack is emerging. [0:08] Identity, payments, marketing, even CLI [0:11] versus MCP. Like all of these are really [0:14] new things and I think a lot of the old [0:15] playbook goes away. [0:17] Yeah. It's a whole new world. I hope [0:19] more companies will stay small and I [0:21] think the founders of this generation [0:22] realize that like they want to stay as [0:23] small as possible. [0:24] Yeah. And instead of having like a 10 [0:26] person product team, you have like a two [0:27] or three person product team and you [0:28] just have a bunch of a agents help help [0:30] you. [0:30] Yeah. [0:30] Someone tweeted that like the the job [0:32] market is so bad that I can only pursue [0:33] my dreams now or something like that. So [0:35] like it's it's like you know it's like [0:37] Yeah. So maybe you lost your job but [0:38] like now you can actually do your own [0:39] thing% and have a shot at actually [0:41] achieving it, you know? [0:42] Yeah. [0:49] All right. Welcome everyone. I've got my [0:50] friend Peter Yang here. Peter, welcome. [0:52] Yeah. Great to be It's good to see you [0:54] again. [0:54] Yeah. Yeah. It's great to see you. Peter [0:55] and I worked together at Credit Karma [0:57] for a brief stint. Um and then we went [0:59] our separate ways and um you know, I [1:02] rediscovered um Peter from his prolific [1:04] posts on X and your YouTube and you [1:07] know, you've got a little bit of a a [1:08] Clark Ken Superman thing going because [1:10] you still got a day job, right? [1:11] That's right. I still have a day job. [1:12] Yes. [1:12] Yeah. Can you share where? [1:13] Uh yeah, I work on Roblox as a PM. [1:15] Amazing. Roblox and Dre Portfolio [1:17] Company. [1:17] Yes. [1:18] One of my favorites. Um well, [1:20] incredible, man. Let's get right into [1:21] it. Maybe I'll start with a softball fun [1:23] question and then you know we're going [1:24] to talk about everything in the claw [1:25] ecosystem. We're going to talk about [1:26] coding agents. We're talking about a [1:29] little bit about maybe what students [1:30] should study advice and some of the [1:32] things that you've talked about online. [1:33] Yeah, sure. [1:34] Um maybe to start, what is the name of [1:36] your Well, how many claws do you have? [1:38] And tell me their names. [1:38] Uh I only have one. I call her Zoe. [1:41] Zoe. [1:41] But I have like multiple conversations [1:42] going with her. [1:43] Okay. [1:44] Yeah. [1:44] And why Zoe? [1:46] Um I I I have two girls and I was going [1:48] to call my younger one Zoe and I did [1:50] not. So I'm I call my open claw Zoe [1:52] instead. [1:53] I see. I see. Yes. Yeah. This is your [1:55] fallback plan. Peter, tell me a little [1:57] bit about, you know, open claw, how you [1:59] discovered it, um how you're using it [2:01] today and and what you think the [2:03] implications are. [2:04] Yeah, I was lucky to interview Peter [2:06] Steinberger before he became super [2:07] famous and the whole thing blew up. [2:09] And um then right after I interview him, [2:11] I I like set up the thing. It took [2:12] forever to set up. It was super janky. [2:14] Um and and yeah, it does a lot of things [2:16] for me. It like pulls analytics for me [2:18] across um YouTube and like my Mercury [2:21] banking account. [2:22] It um can update Google documents for [2:24] me. It can build little web for me. But [2:26] if I was honest with you, dude, like I [2:28] mostly just talk to it through voice and [2:30] get voice replies and like every other [2:32] day I asked you to give me like a pep [2:33] pep talk like you know give me like like [2:37] look look through all your memory and [2:38] like give me some like deep insights [2:39] that I I don't know about. [2:40] Okay. and and like it gave me like like [2:43] I remember I was on a walk and it gave [2:44] me like a three-minute pep talk that was [2:46] like really amazing really amazing like [2:48] it was something about like uh like oh [2:50] you're like talking to me about your [2:51] career business and blah blah blah and [2:53] like your your job but like just [2:54] remember that your kids you know 74 are [2:57] going to grow up very soon and they're [2:58] not going to want to spend time with [2:59] you. [2:59] Wow. [3:00] So like you should really optimize for [3:01] you know them instead. [3:03] Yeah that's really cool and I mean very [3:05] cool but also something that all the [3:07] language models could have done prior. [3:08] Yeah. So what's the difference between [3:10] this and in a use case like that? [3:11] Yeah, that's a very good question. So I [3:13] don't know because I have it installed [3:14] on Telegram. It just feels like more [3:16] personal than using like cloud or chat [3:18] GPT [3:19] and and and it just feels like something [3:20] I can like text in bed. [3:23] It's probably not very healthy, but like [3:24] I text to it in bed. I I talk to it [3:26] during my commute [3:27] and it feels like it feels more like a [3:29] personal like like actual human. [3:31] Yeah. Yeah. So then, so how much for you [3:33] is OpenClaw about the kind of interface [3:36] like pushing it to messaging and you [3:37] know maybe helping to trick our brain [3:39] into feeling like hey this is a person [3:41] or personesque thing [3:42] um versus all the other components of [3:44] the stack the self modification skills [3:46] directory um every all the all the rest [3:49] I I think it's probably like 70 80% just [3:51] like the per personable part of it [3:53] because I mostly just talk to it and [3:54] like you know through voice but I also [3:56] think like is it's something for first [3:58] of all it is pretty janky it tends to [4:00] forget things a Yeah. [4:02] To keep reminding it. [4:03] But like any kind of zany idea that I [4:05] have, I just have to talk to it and it [4:06] can probably just do like like it's kind [4:08] of just like [4:09] um like the other day I was doing voice [4:11] replies with it. I was like, "Hey, can [4:13] we just have a live phone call instead?" [4:15] And and then I was like, "Okay, you you [4:16] got to connect Twilio. You got to do all [4:18] this stuff." And then okay, fine. I I [4:19] went off and did it. [4:20] Yeah. [4:20] And then we had a phone call. It called [4:22] my phone. [4:22] Oh, really? You have that set up? I've [4:23] been dying to set that up. Okay. [4:25] It's not very good though. Like the [4:26] latency is bad, but like the fact I was [4:28] able to get it going is like pretty [4:29] impressive. So it's kind of like any [4:30] kind of crazy idea I have it can kind of [4:33] kind of do. [4:33] And then in practice, how are you doing [4:35] that? Are you asking it to write a skill [4:37] on the fly? Are you discovering a skill? [4:39] How much of the code gen are you [4:40] actually using? [4:41] Um I mean I talk to in like a super [4:43] casual way with like just just like a [4:45] friend. So like hey you know hey Zoe can [4:47] can like can I have a phone call like [4:49] okay you got to do that. I said okay [4:50] fine I'll open my computer. I'll do all [4:52] this stuff [4:52] and then it's like give me a call and [4:54] then we troubleshoot a little bit and [4:55] then it works. So like I with with cloud [4:57] I have like very fancy prompts uh like [5:00] very long prompts but with open cloud I [5:02] just kind of text it. Yeah, it is [5:04] really interesting. So we sort of [5:05] touched on a couple things actually. So [5:07] one there's mobile messaging, [5:09] there's the memory system. Um there's [5:11] the sort of code generation component. [5:13] Um [5:14] how much do you think the memory system [5:16] like is it innovative because it's file [5:18] based? You said that it forgets things [5:20] but so do language models. Like do you [5:21] think the memory system is well done? [5:23] Does it hold it back or does it enable [5:24] it? I I I think the default memory [5:26] system is actually not that great. [5:27] Okay. [5:28] Like the way I understand it works is [5:29] like just like a memory MD text file. [5:31] Yes. And then every day per day. Per [5:32] day, right? And every day it updates and [5:34] it tends to forget things a lot. [5:36] Yeah. [5:36] So So I actually installed this like [5:38] three layer memory system that to be [5:39] honest I don't fully understand but it [5:42] has like [5:42] that's fancy. [5:43] It has like Toby's QMD search tool. [5:45] Okay. [5:46] Uh so I installed that and and then [5:48] install like a 2 GB thing and and then [5:50] it got a little bit better. [5:52] Okay. Um, but I I said to remind it like [5:54] I have to I had to put it into the [5:55] agents MD like hey like before you [5:57] answer any question from me like go [5:58] through all your memory and like check [6:00] everything. [6:00] Yeah. [6:00] And it also tends to forget that it can [6:02] do stuff like you know like can you [6:04] update my Google doc? It's like oh I [6:06] can't do that. Yes. Yes you can. It's [6:07] it's in your it's in your file. [6:08] Yes. Yeah. [6:09] So you have to remind it. Yeah. [6:10] Yeah. Really interesting. Well well [6:12] maybe let's get into a little bit of the [6:14] controversy. Um you know you'd said that [6:16] apps will die claw is going to be [6:18] everything and everywhere. I mean talk [6:19] us through that that point of view. [6:21] Yeah. Yeah. Well, for first of all, I I [6:22] I tweet all all kinds of random crap [6:24] that's like not super well thought out. [6:26] We take it all as fact. [6:27] Yeah. [6:27] Yes. [6:28] Um but I do think like um like ever [6:30] since I set up all these apps like [6:32] Mercury, MCP, and all this kind of crap [6:34] on my open call, like I I don't actually [6:36] open those apps much anymore, you know. [6:39] But I do agree with you like I I I think [6:40] the ones that are going to die first or [6:42] like maybe get less usage first is like [6:44] apps that you're just opening to try to [6:46] complete a task. like you actually are [6:48] trying to do something, you know, like [6:49] apps that you don't need to like get [6:50] entertainment can probably survive a [6:52] little bit longer, [6:53] but like apps to complete a task like [6:55] it's just way easier to text my agent to [6:57] do it for me. [6:58] Yeah. [6:58] It's like it's like you have a really [6:59] good admin just to do stuff for it for [7:00] you. [7:01] Yeah. [7:01] Yeah. [7:01] And so how much are you finding has this [7:03] like reduced your smartphone usage [7:04] outside of Modulo Open Claw? [7:06] Yeah. Um, no. Because like I'm I'm like [7:08] a Twitter addict, so I still use phone [7:10] way too much. But yeah, in terms of [7:12] using those apps has definitely reduced [7:14] it. [7:14] Yeah. [7:14] Yeah. Yeah. because you're not going to [7:16] ask Zoe like, "Hey, read my X for me and [7:18] tell me what's interesting." [7:19] I mean, it it sends me like a morning [7:21] briefing of like the top two tweets and [7:22] stuff that like trends, but but yeah, I [7:24] I I still open X. I look through it. [7:27] Yeah. Yeah. You know, it's interesting [7:28] because I've always had this theory that [7:29] people open apps on their phone because [7:31] they want to feel a feeling. [7:32] Yeah. [7:32] You know, and I think of course there's [7:34] some like functional set of needs which [7:35] is why you open calendar or something, [7:36] but I also think that, you know, [7:37] WhatsApp is you want to feel connected [7:40] and Slack is you want to feel productive [7:42] and of course, you know, Tik Tok is you [7:43] want to feel entertained. So I do wonder [7:46] with just one agent, how do you sort of [7:49] do the context switching of like when [7:51] are you flirting, when are you getting [7:52] done? I mean there you know in a [7:54] sense app gives you a nice it sort of [7:56] gives you a nice division of the [7:57] intents. [7:58] Yeah, you don't get with Zoe. [7:59] That's a good point. But I I do have [8:01] multiple channels set up with Zoe in [8:02] Telegram. Like one is just to random [8:04] voice replies and the other one is we're [8:06] actually working on our project [8:07] together. [8:08] Oh, [8:08] and another one I have like a public [8:09] channel where like I'm giving that [8:11] demos. I don't want to reveal private [8:12] information. Yes. So I have like [8:13] multiple channels [8:14] and is that um implemented as sub agents [8:16] or [8:17] No, it's just some janky setup I I found [8:19] online like you can set up a multiple [8:21] Telegram channels and then I'm not sure [8:23] if she actually remembers across context [8:25] across the channels but like you can [8:26] have separate conversations at least. [8:28] Got it. [8:28] Yeah. [8:29] And how how you know transparent are you [8:32] with your agent? Like does they do they [8:33] see your personal email or [8:35] uh I'm I'm like super transparent. Well [8:38] I I I I did buy the Mac Mini and set up [8:40] its own email. [8:41] Okay. And but I gave it like read access [8:43] to my email and like calendar and uh I [8:46] also gave it like right access to some [8:48] docs. [8:48] Yeah. [8:48] But it can like scroll my entire drive [8:50] or something you know. [8:51] Yeah. So [8:52] how do you imagine open claw which it's [8:54] sort of an architecture and a primitive. [8:56] How does it get productized packaged for [8:58] the world? [8:59] I mean I think that's what Peter [9:00] Steinber is working out at OpenAI right. [9:03] It's probably going to build something [9:04] to chat GBT which everybody uses uh so [9:08] that chat GP can actually get stuff done [9:09] for you and like maybe feels more human. [9:11] Yeah, [9:12] dude. Like let me rant about chat. [9:14] Yeah. Yeah. [9:15] For some reason for some reason they [9:16] they trained the model so that like at [9:18] the end of every conversation is always [9:20] like if you want I can also do X and Y. [9:22] Yeah. [9:23] And I I got so annoyed about it that [9:24] that kind of turned from Chad GPT. [9:26] Oh really? [9:26] Yeah. So so it probably increases their [9:28] matrix but like it's just like super [9:30] annoying. It's like why don't just do it [9:31] in the first place? Are you a cloud guy [9:33] now? [9:33] Yeah, I'm I'm a cloud guy now. But but I [9:35] I I do use codeex uh to code. [9:38] Yeah. [9:38] Yeah. [9:38] You like codeex, you prefer it to cloud [9:40] code or you use both? [9:41] Um codeex when I want to try to build [9:43] something real and cloud code is when [9:44] I'm just like vi vibing like very you [9:47] know well it's interesting. I think they [9:48] live at different points and you know [9:50] there's a sort of space of trade-offs. [9:52] Yeah. [9:52] Whereas I find um cloud code and opus 46 [9:55] it's a little more chatty. It makes more [9:56] assumptions but it can be more pleasant [9:58] for a synchronous experience. Whereas [10:00] codecs, it really thinks hard and it's [10:03] more often accurate. But sometimes it's [10:05] sort of like being in a conversation [10:06] where the other person pauses for like [10:08] three minutes to think. [10:09] Yeah. [10:10] You don't have to flow flow state, [10:11] right? It's hard to get a flow like [10:12] clock. Dude, I I tweeted the other day. [10:14] Clock is almost like uh like a slot [10:16] machine. It's like it's like has [10:18] different things each time. It's just [10:19] like [10:19] Oh, 100%. Look, I I do think that if you [10:22] think remember we were talking about in [10:23] the old social networking era, it was [10:25] variable scheduled rewards, right? That [10:26] was the whole magic of it. like you open [10:28] your Facebook feed and you know once in [10:30] a while it's like boring boring oh my [10:31] god this is so exciting and I the coding [10:33] agents have the exact same property also [10:36] the time is variable so sometimes you [10:38] get something in a second sometimes it [10:39] takes five minutes so up to a certain [10:42] point I actually think that both of [10:43] those things give it that casino like [10:45] feeling [10:45] yeah and and the other thing that's very [10:47] different about the product strategy or [10:48] maybe it's just the way it works is like [10:50] coding is kind of like self-explanatory [10:52] and clock hole you have all this crazy [10:53] you have like hooks and like skills [10:55] and like you have to you have to plug in [10:57] If if you if you're not following Yeah. [10:59] If you're not following X, you have no [11:00] idea how do you customize this thing. [11:01] Yeah. [11:01] But once you customize it, you kind of [11:03] feel like it's part of you. So it's it's [11:04] kind of hard to turn. [11:05] It's interesting with uh so I've [11:07] customized mine cuz also I read the long [11:09] thing that Boris put up. Yeah. [11:10] But I will say that I think that you [11:11] know cloud code a lot of the reasons [11:14] that I enjoy it are just harness [11:16] features. [11:17] You know like for example if you cut an [11:20] image you have to paste it into a file [11:22] before and then paste that file into [11:24] codeex. Okay. You can't just take a [11:26] subset of the screen, screenshot it, and [11:28] then paste it directly into codeex the [11:30] same way you can with cloud code. [11:31] Oh, really? Okay. [11:32] So, just like little things like that. [11:33] You know, cloud code added voice. It's a [11:35] little bit janky right now, but it's [11:36] going in the right direction. So, [11:38] they've just got a bunch of quality of [11:39] life things. [11:40] Yeah. [11:40] You know, Cloud Code speaks to Claude [11:42] and Chrome. [11:43] Okay. [11:43] And Codex doesn't speak to Atlas. [11:45] Got it. [11:45] So, I think these are all things that [11:46] OpenAI will fix. Yeah. I think Codeex is [11:48] actually a much better model. Um, but [11:51] they don't exist today. [11:52] Yeah. Yeah. They need to fix it. I mean, [11:53] they're going to go all all in on [11:55] cohortex, I'm sure. [11:56] Yeah. [11:57] Talk to me about coding agents. Like, [11:58] what's your general view? You know, do [11:59] you think it's the end of SAS? Do you [12:01] think these are just a toy? [12:03] Uh, well, first of all, I'm I'm I'm like [12:05] not an engineer, so I'm like a novice. [12:07] But I do hear that um [12:09] like I was talking to some folks uh the [12:11] other day and like like a AI native star [12:14] startup and they're basically trying [12:15] trying to they have a bunch of vibe [12:16] coders [12:18] and a lot of vibeers are just trying to [12:19] build internal tools that replace their [12:21] SAS that they're paying for. [12:22] Really? So it's an actual company that's [12:24] doing this. [12:24] It's an actual company. Yeah, [12:25] it's an AI native company. [12:26] Is it It's like one of the vi coding [12:27] companies, like one of the more popular [12:29] OS. [12:29] Interesting. [12:30] Yeah, [12:30] it's Oh, oh, I see. So they're actually [12:32] an appgen company. [12:33] They're appgen company and they they're [12:34] paying for a bunch of SAS like uh and [12:36] they want to get rid of the payment. [12:37] They want just buy by by coding internal [12:39] tools [12:39] by using Okay. So in that case that that [12:41] they might be the most extreme form of [12:43] adopter because their own product is [12:44] AppGen. So they should use Appgen for [12:46] everything. I guess is your prediction [12:48] though that the average company will [12:50] churn off of Slack or Deal or you know [12:54] um I don't think I I I feel like Slack [12:56] has a lot of legs because Slack can also [12:58] be the place where you talk to the [12:59] agents themselves. [13:01] Uh but some of the other ones they are [13:03] pretty complicated you know so like it's [13:04] kind of be hard to buy that kind of [13:06] stuff [13:06] but but I feel like if you have like an [13:07] app like maybe Calendarly or like [13:10] something more simple [13:11] Yeah. [13:11] then why why why should I why should I [13:13] pay for it? Like I just [13:14] why should I pay for it though? The [13:15] counter point is that it's not that [13:16] expensive. And do you really want to [13:18] maintain your own Calendarly thing? [13:21] Yeah. [13:21] You know, versus pay 20 bucks a month. [13:23] It always gets updated. It's always up, [13:25] you know, because there's just like a [13:26] fixed amount of capacity that anyone in [13:27] the organization is going to have for [13:28] all this stuff. [13:29] Yeah. That that that's true. Unless you [13:31] hire like dedicated viers like the [13:32] startup does, right? Voicable stuff. [13:34] But then it's like, you know, the cost [13:35] benefit versus just paying for [13:36] Calendarly. Yeah. Yeah. Yeah. Yeah. [13:38] Yeah. It's interesting about like for [13:40] example like a lot of people are [13:41] tweeting about Figma recently. [13:43] Yeah. uh like like the stock is down and [13:45] like you know are are you going to [13:46] survive? [13:47] Yeah. [13:47] And um [13:49] I feel like the jury is out there like [13:51] it's kind of hard to say. Yeah. Uh I I I [13:53] feel like all the designers are still on [13:54] fake Figma but like as a designer you [13:57] kind of need to learn how to vibe code [13:58] otherwise you're going to like if you [14:00] don't know how to do Figma. [14:01] Yeah. [14:01] Like you're probably going to be like [14:02] out of date [14:04] in a couple years. Yeah. [14:05] My counterpoint to that is that I think [14:07] that I've thought a lot about the sort [14:08] of thinking tools versus making tools, [14:09] right? The IDE was historically a making [14:12] tool. Mhm. [14:12] It's a place for execution. I think it's [14:14] migrating away from that. And now with [14:16] execution going to zero, I think these [14:18] sort of like multi- aent, you know, [14:20] nextg ides, a lot of them are about [14:22] trying things [14:23] and using the trial and error as a way [14:25] to inform your thinking. Like a lot of [14:27] times I'll just build a feature in a [14:29] really naive way and I'll hammer the [14:30] coding agent until it works. Then I'll [14:32] say, "Hey, write all the things that you [14:34] would have done differently and I'll go [14:36] back to the initial point and redo it." [14:38] So I wonder if and I think Figma [14:40] actually does both. I think it's a place [14:41] for design execution, but it's also an [14:43] important place for design thinking. And [14:44] I think that's their opportunity to be [14:46] highly relevant in the new stack. [14:48] Yeah, I totally agree. I totally agree. [14:49] Um, but I I I think A6Z has like uh you [14:53] guys investing pencil or something. [14:56] Pencil.dev. [14:56] Speedrun did. Yeah. [14:57] Yeah. Speedrun. And like um [14:59] yeah, Figma needs to like uh level up [15:01] his AI tooling because you know like [15:03] watching these agents collaborate with [15:04] you and like do stuff is like very very [15:07] interesting. [15:07] I know it's top of mind for them. Um [15:09] yeah, what do you think are the most [15:10] under discussed capabilities of coding [15:13] agents? You know, what's what's [15:14] underhyped and maybe what's overhyped as [15:16] well? [15:16] This is probably not underhyped, but [15:17] like but you know, like you know, I I I [15:19] feel like ent software will eat the [15:21] world. I I feel like coding will eat all [15:23] knowledge work, right? And we're kind of [15:25] going that direction already. Like I I [15:26] think lovable recently launched like [15:28] today. Yeah. [15:28] That they can support everything and [15:30] replic. [15:31] Yeah. [15:32] So, so yeah. So I and I feel like [15:34] everyone's cha chasing this like and is [15:37] probably in the lead. [15:38] Yeah. But like, you know, like I I don't [15:39] want to use PowerPoint anymore. I don't [15:40] want to like write a Google like I hate [15:42] writing Google Docs, dude. Like my [15:44] entire life. [15:44] Yeah. Yeah. So, so like but but but the [15:47] other day I was writing my blog post and [15:49] instead of just like typing it out, I [15:51] was like, "Hey, let's let me just use [15:52] clock code and like you know, let me [15:54] give you a bunch of feedback and you you [15:55] write it for me." [15:56] Yeah. [15:56] And then you just keep the [15:58] um [15:59] it it did the first 80%. The last 20% I [16:01] had to manually like go in there like [16:03] tweak tweak tweak stuff. [16:04] Yeah. [16:04] But like that that's the way I work now. [16:06] I I never start from zero. like I always [16:08] get the first 80% from AI, [16:10] right? [16:11] Yeah. [16:11] Yeah. It's interesting, you know, if you [16:12] look at there there are also like [16:13] historical analoges of this. I think [16:15] Satya said this um which is that Excel [16:18] is the most powerful or most popular [16:20] programming language in the world. [16:21] Yeah. [16:21] And that it's sort of a programming [16:23] language that millions and millions I [16:25] mean 100 million plus people must know [16:26] maybe even more. Um and yet we don't [16:29] think of it that way. It's a way to sort [16:30] of describe and solve problems. [16:33] Yeah. [16:33] And I think coding agents are going to [16:35] be that of course times a thousand. [16:37] Yeah. [16:37] Where even things that feel subjective [16:39] like writing Google docs can be [16:41] represented in the coding domain in such [16:43] a way that it's more satisfying [16:45] productive more to use agents to do it. [16:47] Yeah. Because Excel was like popular [16:49] because it's super appro approachable, [16:50] right? Yes. And like coding agents the [16:52] code is basically gone like it's like [16:53] appra just talking to some some agent [16:56] and getting to do do stuff. So [16:57] Yeah. Yeah. [16:58] Yeah. Exactly. [16:58] It's going to be hu huge. Yeah. [17:00] What do you think the future company [17:01] looks like? Is it just a bunch of agents [17:03] with a CEO? Is the CEO an agent? I mean, [17:06] what is the role for people in a company [17:07] in the future? [17:08] Okay. Well, I have some hot takes. So, [17:09] we we both worked at uh some companies [17:11] together and um [17:12] Uhhuh. [17:13] Uh let let me give you a hot take, man. [17:15] Maybe we cut this out, but like I feel [17:16] like as a company gets bigger, [17:17] it tends to get it tends to become [17:19] like a shitty shittier place to work, [17:20] dude. Yeah. Like like because there's [17:21] like a lot of people you have to align. [17:23] I think that's axiomatic. Yeah. [17:24] Right. And and I remember, you know, [17:26] maybe I should mention this company, but [17:27] I remember our company used to have all [17:29] these like OKR meetings and like I [17:30] remember sitting in a room for like [17:31] three hours talking about OKRs. I'm just [17:33] like, dude, this is like wasting my [17:34] life. [17:34] Yeah. So what where I'm going with this [17:36] is I hope more companies will stay small [17:38] and and I think the founders of this [17:40] generation realize that like they want [17:41] to stay as small as possible. [17:42] Yeah. [17:43] Um and instead of having like a 10 [17:45] person product team, you have like a two [17:46] or three person product team and you [17:47] just have a bunch of agents help help [17:49] you. [17:49] Yeah. [17:49] You know, because I I I think it's way [17:51] easier to cross launch a line with a [17:52] agents than with with humans. [17:54] Yeah. Yeah. Well, actually in a in a [17:56] sense the agents actually because it [17:57] takes the emotion out of it too. Like [17:59] you can imagine if I sent my agent, you [18:01] sent your agent to go negotiate [18:02] something. Yeah. [18:03] And they came out with some conclusion. [18:05] It's not emotional. [18:06] It's not [18:07] for either of us. You know, [18:08] it's very objective. Yeah. [18:08] Yeah. Yeah. Exactly. Yeah. It's funny. [18:10] You know, one of the things that we've [18:11] been talking a bunch about is like [18:13] what is the procase for AI at work in [18:16] terms of employee experience? And I [18:17] think it's what you're describing, [18:18] right? Like how do you increase the NPS [18:20] of work? Yeah. So, if we like go all the [18:22] way back or even broadly the NPS of the [18:24] human experience, right? Like think of [18:25] the NPS of the day-to-day human in [18:27] 10,000 BC when it's like just don't get [18:29] eaten by the lion and that's like a good [18:31] day, right? Or you know maybe a 100 [18:33] years ago it's like okay don't get [18:34] killed at the factory [18:36] um crushed by the the steam press or [18:38] whatever else. And um and now a lot of [18:40] it is like h like just don't get sucked [18:42] into some high emotion you know sort of [18:45] negotiation with another VP's [18:48] subordinate. Um [18:49] yeah like a 50 me stack thread going [18:50] back and forth. [18:51] Yeah. Exactly. And then eventually [18:52] everyone's like, I don't want to tell [18:53] the CEO. And eventually it goes there [18:55] and it's just terrible. Um, so maybe the [18:57] future of this is that a lot of that [18:59] emotional subjective work gets handled. [19:01] Yeah. [19:02] And we're sort of guiding the process [19:04] but not in the middle of it in a way [19:05] that just doesn't suit us as humans. [19:07] Yeah. Like you know, I I I lead a double [19:09] life as a PM creator and like I feel [19:10] like all the PMs actually just want to [19:12] create products. They want to create [19:14] products and and [19:15] well that's why we all got into it. You [19:16] know, it's it's so interesting. I mean, [19:17] Nicole talks about this all the time, [19:19] but like every PM's sort of view view of [19:22] the ideal PM is the innovator, you know, [19:24] like I came up with the new thing and [19:26] it's like I sort of like had the big [19:28] insight and it unlocked the product. [19:29] Yeah. [19:30] I think the black pill is I don't think [19:32] most PMs know how to do that. In fact, [19:34] many companies have zero people that [19:36] know how to do that at all in any [19:37] function. [19:38] Yeah. So, nonetheless, I think PMs [19:40] aspire to be able to do that and they [19:42] should either do it and either be [19:44] successful or maybe not successful and [19:46] move to a different function. I I also [19:47] feel like my hot take is like uh like [19:49] basically all the PMs I know are trying [19:51] to vibe code at night on weekends. [19:53] Yeah. [19:53] And I feel like my hot take is that like [19:55] I feel like if you're actually [19:55] unemployed like you probably have more [19:57] time to be a builder and like to be [19:58] innovative because you can actually like [20:00] play all this stuff and like learn all [20:01] this stuff. [20:02] A lot of PMs are trying to [20:03] or maybe be an engineer in the team, you [20:05] know? I used to be an engineer and I got [20:07] sort of I don't know if I got forced to [20:08] be a PM. Maybe I also perceived PM as [20:10] like being a little more high status [20:12] when I joined Google. But then [20:14] eventually you like this is terrible, [20:16] you know, like you never really get the [20:18] satisfaction of actually shipping other [20:20] than like you know once a quarter when [20:22] you ship. [20:22] I mean the PM skills of like talking to [20:24] users and like trying to figure out what [20:25] to do like what problem to solve like [20:26] those are very important still. [20:28] Yeah. [20:28] And and like but yeah, you got to wear [20:30] multiple hats, dude. You got you got to [20:31] like go go go build a thing yourself, go [20:32] prototype it and get some feedback and [20:34] then maybe bring engineer along. [20:36] How much do you think that everyone has [20:38] to go as fast as you know I mean like [20:40] Gary was talking about stemmies and [20:42] skipping sleep and 10 like you know [20:44] GStack I mean is hey I mean is that like [20:47] the default way that we all need to work [20:49] or do you think there's a trade-off for [20:50] thoughtfulness? [20:52] I I I think it's very easy now with all [20:54] these AI tools just going like 10 [20:55] different directions at once. [20:56] Yeah. So sometimes you do have to slow [20:58] down and try to figure out where you [21:00] want to go. [21:00] Yeah. [21:00] But I also believe that like u the [21:02] traditional process where you like do [21:03] annual planning and like do all this [21:04] like I just feel like that [21:06] doesn't really work anymore, [21:08] you know? [21:08] Yeah. Yeah. And for the record, I love [21:10] Gary and I like think the world of him. [21:12] So here's my view on that because I [21:13] thought a bunch about that. You know who [21:14] asked me this the other day? Hithan um [21:16] who's really really impressed. Yeah. Ch. [21:18] Yeah. [21:19] So we're talking about this like sort of [21:20] productivity porn and everybody's got 20 [21:22] agents running and 20 monitors and blah [21:24] blah blah. And like I do think when it [21:26] comes to [21:28] um fully realizing a local a sort of [21:32] local maxima, you should go very fast, [21:34] right? So let's say you kind of hill [21:36] climb, you get to the bottom of a new [21:37] local maxima. I think with agents, you [21:39] should be able to get to the top of that [21:41] hill extremely fast, right? You have a [21:43] new insight, build everything around the [21:44] insight so it's fully expressed. But [21:47] then I think to get to the next, you [21:48] know, the next sort of hill, [21:50] you've got to probably slow down and [21:51] almost stop and go touch grass and do [21:53] whatever. Yeah. [21:54] So I think there's this combination of [21:56] like fast and slow. That's probably the [21:58] future way. [21:58] Yeah, I think so. And and like you got [22:00] to go on that random walk trying to find [22:01] hard market fit which take takes us a [22:03] while, right? So it's not like [22:04] Yeah. [22:04] So we were talking before we started [22:06] recording about some of the business in [22:08] a box platforms. [22:10] Have you looked at them? Do you have a [22:11] view? [22:12] I've looked at post yet that we talked [22:13] about like I don't know if the guy like [22:16] intentionally made it the opposite of AI [22:17] slab or or is it kind of you [22:19] I think so. Yes. Yes. [22:22] That's funny. Well, I mean, I have a [22:23] pretty big public presence, right? So, I [22:25] connect all my to to it. And then, [22:26] um, I mean, it's it's definitely gives a [22:29] good peak into what's possible, but like [22:31] right now it's probably still pretty [22:32] like early stage. Like, it's telling me [22:34] to run like fa Facebook ads. Like, [22:36] why am I running Facebook ads? [22:38] Yeah. [22:38] You know, so [22:39] I don't know. Yeah. Yeah. Yeah. I I [22:40] mean, I'm very excited about it because [22:42] it does feel like it's a path for more [22:43] people to build companies. Yeah. Even if [22:46] they're single oneperson companies, you [22:48] know, like if you think about how [22:49] competitive it is to build a billion [22:50] dollar business, like the markets that [22:52] support it, the number of people trying [22:53] versus hundred million versus 10 million [22:55] versus $100,000 TAM. Yeah. [22:58] Like maybe there are these pockets all [23:00] over the country, all over the world [23:01] where there are opportunities for [23:03] $100,000 TAM products. Yeah. [23:05] And that would change somebody's life. [23:07] Now, that's not an enterprise [23:08] ventureback company, but that's okay. [23:10] Yeah. So, I hope that whole thesis works [23:12] because I do think it's a way to get [23:14] more people to participate, you know. [23:15] Yeah. That that that's my plan for my [23:16] kids, dude. Like I want them to just [23:18] build like um bootstrap businesses in [23:20] high school. [23:21] Yeah. [23:21] And they can skip the whole college and [23:23] like uh core corporate life. [23:24] Yeah. [23:25] Well, dude, I think this is like, you [23:26] know, for 10 years there's this moral [23:28] panic about the kids want to be [23:29] YouTubers. Yeah. You know, you're a [23:30] YouTuber. [23:31] Um you know, you know, in the in the [23:33] vein of Mr. And and I think the like pro [23:36] case for that actually is that the kids [23:38] wanted to be entrepreneurs or have [23:40] agency. And the only channel for people [23:42] if they weren't programmers was creating [23:44] YouTube videos at least online. [23:45] Yeah. [23:45] So if you're like an online native [23:47] generation, you want to create something [23:48] you're not a programmer, you make a [23:49] YouTube show. Now you can make a lot [23:51] more than that. [23:52] Yeah. You you can build whatever you [23:53] want. Yeah. [23:53] Exactly. [23:53] So Exactly. [23:54] It'd be very exciting. Yeah. [23:55] Yeah. Any other hot takes for us? [23:57] I'm curious about your thoughts about [23:58] this actually. Like so I feel like uh a [24:01] lot of people are saying like agents [24:02] will interact with your product first, [24:04] right? And and then you see all these [24:06] great companies like uh building like [24:07] APIs and MCPS. [24:09] Mh. [24:10] But like how do how do you think about [24:11] like you know you've been consumer for a [24:12] while. So like consumer is like you got [24:15] to get the user to come back and use [24:16] your product, right? [24:18] Yeah. [24:18] But but now like the user is like hey go [24:20] send the agent to use So, so how do you [24:21] think about retention and all this like [24:22] basic stuff like how do you you know or [24:24] even like brand equity because the agent [24:26] just like point some API like you know [24:28] how do you [24:28] yeah I don't okay so I think one of I [24:31] don't know is the is the truth but um [24:33] there's I have a few thoughts so one I [24:35] think that a lot of the sophistication [24:38] sophistication that happened in consumer [24:40] happened because we had to have indirect [24:43] monetization [24:44] okay like we just were never charging [24:46] consumers directly for these products [24:47] which is why you got [24:48] got ads and stuff [24:49] ads and large scale networks and we all [24:52] obsessed with retention and engagement [24:53] and whales and all of these things [24:55] really really mattered because we didn't [24:57] simply charge people for products. [24:59] So I think one big thing that's actually [25:01] really helped in the AI era with that is [25:02] that consumers are now excited to try [25:04] new things. They're willing to pay [25:05] they're willing to pay a really high [25:06] price point. [25:07] Um there's also consumption revenue in [25:09] consumer for the first time [25:10] like token and stuff. [25:11] Yeah. Yeah. For like tokens. You have [25:12] your subscription plus your token. So [25:14] and then the actual like the sort of [25:16] blessing in disguise is that there are [25:18] real costs as well. you have these [25:19] inference costs. So you're like, "Wow, [25:21] we have to charge our customer on day [25:23] one." [25:24] So I think one thing is that like the [25:25] business model simplification I think [25:27] will really help with a lot of what [25:29] you're describing. [25:30] Two, I think that a lot of the products [25:31] will have a sort of, you know, it'll [25:33] have an API interface for your agents to [25:35] interact with or for, you know, for [25:37] transactional sort of wrote things. [25:39] Yeah. [25:39] And then it'll have like a consumption [25:41] based interface as well. [25:42] So you can also imagine a like a mobile [25:44] app where there's like the feed, but [25:46] then you can kind of turn it over to [25:47] where the wires are. [25:48] Mhm. And you can just ask for things to [25:50] get done or you can just see the log of [25:52] the things that got done. [25:53] Yeah, maybe people would do both, right? [25:54] I mean, you can imagine Credit Karma [25:55] where we worked, you know, like once in [25:56] a while you want to just take a look at [25:58] your score history and a few other maybe [26:00] credit card offers. I don't know. I [26:01] mean, [26:01] yeah. Yeah. If I get my score with all [26:03] kind of credit card offers, I'll [26:04] definitely do that. [26:04] Yeah, 100%. Exactly. [26:06] On the other hand, like sometimes you [26:07] want to just be like, yo, like can you [26:08] just fix all my stuff or like what stuff [26:10] did you fix this week? How much money [26:11] did I save? You know, [26:12] got it. Got it. [26:13] Yeah. [26:14] It's definitely interesting. Yeah. [26:15] But look, I also just think the whole [26:17] agent stack is emerging. [26:18] identity, payments, marketing, we don't [26:22] even even CLI versus MCP. Like all of [26:24] these are really new things and I think [26:26] a lot of the old playbook goes away. [26:28] Yeah, it's a whole new world and like in [26:30] 2025 I thought agents was overhyped but [26:32] now I think it's really coming like [26:34] me too. I know it's just the word is [26:36] frustrating because it gets so [26:37] overloaded. [26:38] Yeah, there's like workflows like all [26:40] this kind of [26:40] Totally. I've been trying to just say [26:41] like can we just say like model in a [26:43] loop? [26:43] Yeah, exactly. Model that use tools in a [26:45] loop. That's the best definition. Yeah. [26:46] Yeah. [26:46] But nobody likes to hear that. It's [26:48] Asians is much flashier, you know. [26:49] Yeah, it's flashier. [26:50] Yeah. [26:51] My my hope is that um [26:53] all all this stuff's like a lot of [26:54] people think like we're going to lose [26:55] our jobs, which probably what will [26:56] happen at some point, but like I hope [26:58] all this stuff [26:59] makes just makes like uh human work more [27:01] fun like our jobs more fun, [27:03] you know? [27:03] Dude, I don't think we're all going to [27:04] lose our jobs. Like I really think the [27:06] and we see this a lot of companies, you [27:07] know. So we look at a ton of companies [27:08] and we've seen two different buckets. So [27:11] one bucket is hey, we dramatically [27:12] increase productivity for a person or a [27:15] team. We see this in like recruiting, [27:17] but we couldn't do 100% of the job. So, [27:19] we could do the phone screen, but we [27:21] couldn't obviously, you know, show the [27:22] candidate around the office or we could [27:24] do the phone screen and we could like [27:26] answer all the questions about the [27:27] company and we could even do the like [27:28] comp negotiation, but we couldn't do the [27:30] onboarding. [27:30] Yeah. [27:31] The other style of company which we see [27:33] which is maybe a decagon, right, or a [27:35] happy robot is hey, we did 100% of a job [27:37] like customer support. [27:39] You know, the the customer called in, [27:40] they had a question, we hopefully [27:42] resolved their query and then that's it. [27:43] M and that is 100% automated. I'd say [27:46] that that second group where you have [27:47] 100% automation of a job function is [27:49] really rare. Almost every AI product AI [27:52] native X or Y we see is able to provide [27:55] dramatic lift but it's not able to do [27:57] 100%. [27:58] So last 10% use humans. [27:59] Yeah, it's still it's today anyway it's [28:01] still humans that do that stuff and it's [28:03] interesting too because the buyer looks [28:04] at that as software as expensive [28:06] software whereas in the case of [28:08] something like a happy robot docking [28:10] Sierra they look at it as like cheap [28:11] labor. M [28:12] so I do think there's a different buyer [28:14] mindset but because there's been this [28:15] difficulty of getting to 100% automation [28:17] I think a lot of the efficiency gain [28:19] shows up in just a different way [28:21] probably not less jobs maybe we get like [28:23] the European style 4 day work week maybe [28:25] companies get like twice as productive I [28:27] have no idea [28:28] yeah but but you don't think that like [28:29] uh I I I feel like there's going to be a [28:31] transition from like these like 10,000 [28:34] plus people companies laying a lot of [28:35] people off to hopefully like more [28:37] smaller companies like solarreneurs and [28:39] stuff like that [28:40] I I think yes I think that they'll the [28:42] the sort of shape of the economy is [28:44] going to change like the amount of [28:45] concentration but I just don't think [28:46] there's going to be less jobs. I think [28:48] human ambition has no ceiling. [28:49] You know, human desire has no ceiling [28:51] and just read any mildly interesting [28:53] science fiction book [28:54] like there's no way this is the peak [28:55] expression of all the stuff that we want [28:57] and we need and we're going to convince [28:58] ourselves and you know all the new [29:00] things that you read about every day is [29:01] these luxuries peptides and you know [29:03] everybody is going to have all of that [29:05] stuff and want even more. You know, [29:07] dude, I I saw a really good tweet about [29:09] this. Like someone tweeted that like um [29:11] the the job market is so bad that I can [29:13] only pursue my dreams now or something [29:15] like that. So like it's it's like you [29:16] know, it's like Yeah. So maybe you lost [29:18] your job, but like now you can actually [29:19] do your own thing. [29:20] And have a shot at actually achieving [29:21] it, you know? [29:22] Yeah. Yeah. [29:23] Cool. [29:23] Well, awesome, man. Maybe that's a good [29:25] uh positive note to end on. [29:26] Yeah, that's good to know. Yeah. Cool. [29:27] Good thing you do. Yeah. [29:28] Thanks, Peter. Yeah. Thank