Peter Yang on Small Teams, Coding Agents, and Why Human Ambition Has No Ceiling

0:00The interest in said software will eat the world.

0:02Anish Acharya:I feel like coding will eat all knowledge work, right? And we're kind of going that direction already. The whole agent stack is emerging. Identity, payments, marketing, even CLI versus MCP. All of these are really new things. And I think a lot of the old playbook goes away. Yeah, it's a whole new world. I hope more companies will stay small. And I think the founders of this generation realize that. They want to stay as small as possible. And instead of having a 10 % product team, you have a two or three person product team. And you have a bunch of agents to help you. Someone tweeted that the job market is so bad that I can only pursue my dreams now.

0:35So maybe you lost your job, but now you can actually do your own thing.

0:39Anish Acharya:Yeah, 100 % and have a shot at actually achieving it. Yeah.

0:45Most people open their phone to feel something. Connected, productive, entertained. Each app is a door to a different emotional state. In 2007, the iPhone gave us the app grid. 19 years later, billions of people still tap the same colored squares dozens of times a day. The interface became so familiar, it stopped feeling like technology. It became reflex. Now a generation of builders is collapsing that entire grid into a single conversation. One agent that checks your analytics, updates your documents, runs your errands, and gives you a pep talk on your morning walk. Not because the apps failed, but because talking is faster than tapping.

1:30The question is, what happens to products, companies, and careers when building software costs almost nothing? A16Z general partner Anish Acharya speaks with Peter Yang, creator and product lead at Roblox.

1:48Anish Acharya:All right, welcome everyone. I've got my friend Peter Yang here. Peter, welcome. Yeah, great to be here. It's good to see you again. Yeah, it's great to see you. Peter and I worked together at Credit Karma for a brief stint, and then we went our separate ways. and I rediscovered Peter from his prolific posts on X and your YouTube. And you've got a little bit of a Clark Kent Superman thing going because you've still got a day job, right? That's what I said with a day job, yes. Yeah, can you share where? Yeah, I work on Roblox as a PM. Amazing, Roblox. And Drees and Portfolio Company. Yes. One of my favorites.

2:16Anish Acharya:Well, incredible, man. Let's get right into it. Maybe I'll start with a softball fun question and then we're going to talk about everything in the claw ecosystem. We're going to talk about coding agents. We're talking about a little bit about maybe what students should study, advice and some of the things that you've talked about online. Yeah, sure. Maybe to start, what is the name of your, how many claws do you have? And tell me their names. I only have one. I call her Zoe. Zoe. But I have like multiple conversations going with her. Okay. Yeah. And why Zoe? I have two girls and I was going to call my younger one Zoe and I did not.

2:45So I'm like, I call my open claw Zoe instead.

2:47Anish Acharya:I see. Yeah. This is your fallback plan. Peter, tell me a little bit about open claw, how you discovered it, how you're using it today and what you think the implications are yeah i was lucky to interview peter steinberger before he became super famous and the whole thing blew up and then right after i interviewed him like set up the thing it took forever to set up it was super janky and yeah it does a lot of things for me it like pulls analytics for me across youtube and like my mercury banking account it can update google documents for me it can be a little web websites for me but if i was honest with you dude like i mostly just talk to it through voice and get voice replies and like every other day i asked to give me like a pep talk.

3:25Like look through all your memory and like give me some like deep insights that I don't know about. Okay. And it gave me like, like I remember I was on a walk and it gave me like a three minute pep talk that was like really amazing. Really amazing. It was something about, oh, you're like talking to me about your creator business and blah, blah, blah, and like your job. But just remember that your kids, seven and four are going to grow up very soon and they're going to want to spend time with you. Wow. So you should re-optimize for them instead. Yeah.

3:49Anish Acharya:Yeah. That's really cool. Yeah. And I mean, very cool, but also something that all the language models could have done prior. Yeah. So what's the difference between this and a use case like that? Yeah, that's a very good question. So I don't know, because I think it's still on Telegram, it just feels like more personal than using like Cloud or ChatGPT. And it just feels like something I can text in bed. It's probably not very healthy, but like I text to it in bed. I've talked to it during my commute and it feels like it feels more like a personal, like actual human. Yeah. Yeah. So how much for you is OpenClaw about the kind of interface, like pushing it to messaging and maybe helping to trick our brain into feeling like, hey, this is a person or a person-esque thing versus all the other components of the stack, the self-modification, the skills directory, all the rest?

4:32I think it's probably 80 % just the personal part of it because I mostly just talk to it and like, you know, sort of voice. But I also think like it's something, first of all, it is pretty janky. It tends to forget things a lot. Yeah. To keep reminding it. but like any kind of zany idea that i have i just have to talk to it and it can probably just do it's kind of like the other day i was doing voice replies with it i was like hey can we just have a live phone call instead and then it's like okay you gotta connect to it you gotta do all this stuff and okay fine i went off and did it yeah and then we had a phone call it called my phone really you have that set up i'm dying to set that up okay it's not very good though like the latency is bad but the fact that i was able to get it going is like pretty impressive so it's kind of like any kind of crazy idea i have it can kind of kind of do and then in practice how are you doing

5:14Anish Acharya:that? Are you asking it to write a skill on the fly? Are you discovering a skill? How much of the code gen are you actually using? I mean, I talk to it in a super casual way with like, just like a friend. So I'm like, Hey, Hey, Hey, Zoe, can you have a phone call? Okay. You got to do that. I was like, okay, fine. I'll open my computer. I'll do all this stuff. And then give me a call and it will troubleshoot a little bit and then it works. So I, with cloud, I have like very fancy prompts, like very long prompts, but with open cloud, I just kind of text it. Yeah. It is really interesting. So we sort of touched on a couple of things, actually.

5:43Anish Acharya:So one, there's mobile messaging, there's the memory system, there's the sort of code generation component. How much do you think the memory system, like, is it innovative because it's file-based? You said that it forgets things, but so do language models. Do you think the memory system is well done? Does it hold it back or does it enable it? I think the default memory system is actually not that great. Okay. Like the way I understand it works is just like a memory.md text file. Yes. And then every day, per day, right? and there they updates and it tends to forget things a lot. Yeah. So I actually installed this like three layer memory system that to me I don't fully understand but it has like fancy.

6:17It has Toby's QMD search tool. Okay. So I installed that and then it installed like a two gigabyte thing and then it got a little bit better. Okay. But I still have to remind it I have to put it into the agents like MD. Hey like before you answer any question from me go through all your memory and check everything. Yeah. And it also tends to forget that it can do stuff like can you update my Google Doc? It's like oh I can do that. yeah yes you can it's in your it's in your file yes so you have to remind it yeah yeah really

6:41Anish Acharya:interesting well well maybe let's get into a little bit of the controversy you'd said that apps will die claw is going to be everything and everywhere i mean talk us through that point of view yeah well for first of all i i tweet all kinds of random crap that's not super well thought out we take it all as fact yeah yes but i do think like ever since i set up all these apps like mercury mcp and all this kind of crap on my open claw like i don't actually open those apps much anymore but i do agree with you like i i think the ones that are gonna die first or like maybe get less usage first it's like apps that you're just opening to try to complete a task like you actually are trying to do something you're not like apps that you're opening to get entertainment can probably survive a little bit longer but like apps that i want you to complete a task like it's just way you should text my agent to do it for me yeah it's like you have a really good admins it's a do stuff for it for you yeah yeah and so how much are you finding has this reduced your smartphone usage outside of modulo open call yeah no because i'm like a twitter addict so i see use my phone way too much.

7:35But yeah, in terms of using those apps, it's definitely reduced it. Yeah. Yeah. Because you're not going to ask Zoe, hey, read my apps for me and tell me what's interesting. I mean, it sends me like a morning briefing with the top two tweets and stuff, the like trends, but yeah, I still open X and look through it.

7:49Anish Acharya:Yeah. It's interesting because I've always had this theory that people open apps on their phone because they want to feel a feeling. Yeah. And I think, of course, there's some like functional set of needs, which is why you open calendar or something, but also think that WhatsApp is you want to feel connected and Slack is you want to feel productive and of course TikTok is you want to feel entertained so I do wonder with just one agent how do you sort of do the context switching of like when are you flirting when are you getting shit done I mean in a sense app gives you a nice it sort of gives you a nice division of the intents yeah that's what you don't get with Zoe that's a good point but I do have multiple channels set up with Zoe in Telegram like one is just to random voice replies and otherwise we're actually working on our project together and then I want to have a public channel where like I'm giving demos.

8:32I don't want to reveal private information. Yes. So I have like multiple channels. And is that implemented as sub-agents or? No, it's just some janky setup I found online. Like you can set up multiple Telegram channels and then I'm not sure if it actually remembers across contexts across the channels, but like you can have separate conversations at least. Got it. Yeah.

8:49Anish Acharya:And how transparent are you with your agent? Do they see your personal email? I'm like super transparent. Well, I did buy the Mac Mania and set up its own email. Okay. But I gave it like read access to my email and like calendar. And I also gave it like write access to some docs. Yeah. But it can't screw my entire drive or something. Yeah. So how do you imagine OpenClaw, which it's sort of an architecture and a primitive. Yeah. How does it get productized, packaged for the world? I mean, I think that's what Peter Steinberg is working out at OpenAI, right? Yeah. He's probably going to build something to ChatGPT, which everybody uses so that ChatGPT can actually get stuff done for you and like maybe feels more human.

9:28Yeah. Dude, let me rant about ChatGPT. Please. Yeah, yeah, yeah. For some reason, they trained the model so that at the end of every conversation, it's always like, if you want, I can also do X and Y. Yeah, yeah. And dude, I got so annoyed about it that I kind of churned from ChatGPT. Oh, really? Yeah. So it probably increases their metrics, but it's just like super annoying. It's like, why don't you just do it in the first place? It's like, are you a quad guy now? Yeah, I'm a quad guy now. But I do use Codex to code. Yeah, yeah. You like Codex, you prefer it to Code code or you use both? Codex, when I want to try to do something real and Cloud Code is when I'm just like vibing.

9:58Anish Acharya:Yeah. Well, it's interesting. I think they live at different points and there's a sort of space of trade-offs. Yeah. Whereas I find Quad Code and Opus 4.6, it's a little more chatty. It makes more assumptions, but it can be more pleasant for a synchronous experience. Yeah. Whereas Codex, it really thinks hard and it's more often accurate, but sometimes it's sort of like being in a conversation where the other person pauses for three minutes to think. Yeah. So you don't have to flow state, right? It's hard to get a flow. Like Cloud Code, dude, I tweeted the other day, The clock almost like a slot machine.

10:25It has different things each time. Oh, 100%. It's like a slot machine.

10:28Anish Acharya:Look, I do think that if you think, remember we were talking about in the old social networking era, it was variable scheduled rewards, right? That was the whole magic of it. Like you open your Facebook feed and once in a while it's like boring, boring. Oh my God, this is so exciting. And the coding agents have the exact same property. Also, the time is variable. So sometimes you get something in a second. Sometimes it takes five minutes. So up to a certain point, I actually think that both of those things give it that casino-like feeling. Yeah. And the other thing that's very different about the product strategy or maybe just the way it works is like, coding is kind of like self-explanatory.

10:59Like Cloud Code, you have all this crazy shit. You have like hooks and like skills and - Plugins. If you're not following Twitter. Yeah, if you're not following X, you have no idea how to customize this thing. Yeah. But once you customize it, you kind of feel like it's part of you. So it's kind of hard to churn.

11:11Anish Acharya:It's interesting with, so I've customized mine because also I read the long thing that Boris put up. Yeah. But I will say that I think that Cloud Code, a lot of the reasons that I enjoy it are just harness features. Like, for example, if you cut an image, you have to paste it into a file and then paste that file into Codex. Okay. You can't just take a subset of the screen, screenshot it, and then paste it directly into Codex the same way you can with Quadcode. Oh, right. Okay. Okay. So just like little things like that. Yeah. Quadcode added voice. It's a little bit janky right now, but it's going in the right direction.

11:42Anish Acharya:So they've just got a bunch of quality of life things. Yeah. Quadcode speaks to Cloud in Chrome. Okay. And Codex doesn't speak to Atlas. Got it. So I think these are all things that OpenAI will fix. Yeah. I think Codex is actually a much better model, but they don't exist today. Yeah, yeah. They need to fix it. I mean, they're going to go all in on Codex, I'm sure. Yeah. Talk to me about coding agents. What's your general view? Do you think it's the end of SaaS? Do you think these are just a toy? Well, first of all, I'm not an engineer, so I'm a novice. But I do hear that, like I was talking to some folks the other day, an AI native startup, and they're basically trying to have a bunch of Vibe coders.

12:17And all the Vibe coders are just trying to build internal tools that replace their SaaS that they're paying for. Really? So it's an actual company that's doing this? It's an actual company. It's an AI-native company. It's like one of the bi-coding companies. It's one of the more popular-wise.

12:28Anish Acharya:Interesting. Yeah. Oh, I see. So they're actually an AppGen company. They're an AppGen company, and they're paying for a bunch of SaaS, and they want to get rid of the payment. They want to just bi-coding internal tools. Okay. So in that case, they might be the most extreme form of adopter because their own product is AppGen, so they should use AppGen for everything. I guess, is your prediction, though, that the average company will churn off of Slack or Deal or... Or I don't think, I feel like Slack has a lot of legs because Slack can also be the place where you talk to the agents themselves.

12:57But some of the other ones, they are pretty complicated. So it's kind of hard to buy or that kind of stuff. But I feel like if you have an app like maybe Calendly or something more simple, then why should I pay for it? I just... Why should I pay for it?

13:08Anish Acharya:Though the counterpoint is that it's not that expensive. And do you really want to maintain your own Calendly thing? Yeah. Versus pay 20 bucks a month. It always gets updated. It's always up. Yeah, yeah. Because there's just like a fixed amount of capacity that anyone in the organization is going to have for all this stuff. Yeah, that's true. Unless you hire like dedicated VibeCore like the startup does, just VibeCore stuff. But then it's the cost benefit versus just paying for Calendly. Yeah, yeah. It's interesting about, for example, like a lot of people are tweeting about Figma recently. Yeah.

13:35It'll stock us down, like, you know, are you going to survive? Yeah. And I feel like the jury's out there. It's kind of hard to say. Yeah. I feel like all the designers are still on Figma, but as a designer, you kind of need to learn how to VibeCore. Otherwise, you're going to, if you want to know how to do Figma, Yeah. You're probably going to be like out of date in a couple of years. Yeah.

13:53Anish Acharya:My counterpoint to that is that I think that I've thought a lot about the sort of thinking tools versus making tools, right? The IDE was historically a making tool. It's a place for execution. I think it's migrating away from that. And now with execution going to zero, I think these sort of like multi-agent next-gen IDEs, a lot of them are about trying things and using the trial and error as a way to inform your thinking. Like a lot of times I'll just build a feature in a really naive way and I'll hammer the coding agent until it works, then I'll say, hey, write all the things that you would have done differently.

14:22Anish Acharya:And I'll go back to the initial point and redo it. So I wonder if, and I think Figma actually does both. I think it's a place for design execution, but it's also an important place for design thinking. And I think that's their opportunity to be highly relevant in the new stack. Yeah, I totally agree. I totally agree. But I think A16Z has like, you guys investing Pencil or something? Pencil.dev? Speedrun did, yeah. Yeah, Speedrun. And yeah, Figma needs to like level up this AI tooling because like watching these agents collaborate with you and do stuff is like very interesting. I know it's top of mind for them.

14:52Anish Acharya:Yeah. What do you think are the most under discussed capabilities of coding agents? What's under hyped and maybe what's over hyped as well? This is probably not under hyped but you know I feel like and just as software will eat the world I feel like coding will eat all knowledge work right? And we're kind of going that direction already like I think Lovable recently launched like today that they can support everything and can make decks. Yeah so so yeah so I and I feel like everyone's chasing this and anthropic is probably in the lead yeah i i don't want to use powerpoint anymore i don't want to write a google i hate writing google docs dude plus my entire life so like the other day i was writing my blog post and instead of just like typing it out i was like hey let's let me just use clock code and let me give you a bunch of feedback and you write it for me yeah and then you just keep it did the first 80 the last 20 i still had to manually go in there like to take stuff yeah but like that's the way i work now i i never start from zero like i always get the first 80 % from AI, right?

15:44Yeah.

15:45Anish Acharya:Yeah, it's interesting. If you look at it, there are also like historical analogs of this. I think Satya said this, which is that Excel is the most powerful or most popular programming language in the world. Yeah. And that it's sort of a programming language that millions and millions, I mean, 100 million plus people must know, maybe even more. And yet we don't think of it that way. It's a way to sort of describe and solve problems. Yeah. And I think coding agents are going to be that, of course, times a thousand. Yeah. Where even things that feel subjective, like writing Google Docs, can be represented in the coding domain in such a way that it's more satisfying, more productive, more high leverage to use agents to do it.

16:20Yeah, because Excel was popular because it's super approachable, right? Yeah. And coding agents, the code is basically gone. It's like apps shut away. You're just talking to some agent and getting to do stuff. Yeah. Yeah. Exactly. It's going to be huge. Yeah.

16:31Anish Acharya:What do you think the future company looks like? Is it just a bunch of agents with a CEO? Is the CEO an agent? I mean, what is the role for people in a company in the future? Okay, well, I have some hot takes. So we both worked at some companies together. And let me give you a hot take, man. Maybe we cut this out. But I feel like as a company gets bigger, it tends to become like a shittier place to work, dude. Yeah. Because there's a lot of people you have to align. I think that's axiomatic, yeah. Right? And I remember at, maybe we should mention this comment, but I remember our company together, we used to have all these like OKR meetings.