It Ships While You Sleep. The AI Agent OS a Former Google Engineer Built After Leaving Google.

David Petrou: Token costs are continuing to go down, but what's not improving is the availability of, you know, the human attention span. Organizations, whether big or small, the more they adopt agentic coding, the more they struggle with problems of communication.Jordan Metzner: Built this week, breaking it down. Built this week, we show you how. A fresh idea, a clever tweak you locked in. You builtSam Nadler: Hey, everyone, and welcome to Built This Week, the podcast where we share what we're building, how we're building it, and what it means for the world of AI and startups. I'm Sam Nadler, cofounder here at Rise Labs, and each and every week, I'm joined by my friend, cohost, and business partner, Jordan Metzner. How are doing today, Jordan?Jordan Metzner: Hey, Sam. How's it going? Happy to be back, along summertime, and happy to have a first episode back in the New Year or the the new fall. Lots going on in the AI world. I think, like, three new models dropped today.So I think everyone's kinda back to business as usual again. But, yeah, looking forward to this episode and and today's guest.Sam Nadler: Yeah. Absolutely. Super excited about our guest who I'm about to introduce. If you haven't yet, please like and subscribe. We had a short hiatus for some vacation time.We're back to episodes every week. So if you don't mind like and subscribe, I think we've crossed over the 35,000 subscriber mark on YouTube. And with that, I'd like to introduce David, the CEO of Continua, who is gonna walk us through, a new product called Weely. David, if you don't mind, please introduce yourself, give us a little bit of background, and then we'll we'll just jumpDavid Petrou: into it. Sounds great. Good to meet you guys, Jordan and Sam, and happy to be helping you kick off your fall season. Yeah. My name is David Petru.I run Continua. We're working right now on building an agentic development operating system, and I'll talk through all about that stuff. But just briefly, my life's been around computers. I, got my PhD from Carnegie Mellon, and then I went to Google. I was there for almost eighteen years and got to see Google through a number of different phases and sort of had a front row seat to technology.Decided, maybe about three or three and a half years ago that, boy, this is just the perfect time to leave big tech and start my own company. We can go through the reasons why on the timing later if you like. But, in any case, we went through several iterations of different products, and we really landed on something that it I I just can't tell you how excited we are every day to be working on this problem and what an incredible time it is right now in computing history. And that's all around how can we get this incredible power of agentic coding to work better for you, to work more cheaply, to get things done in a more measurable fashion. And we decided to call that product Wheelie.Sort of a play on words. Everybody talks about agentic loops, and so we think about the bicycle, and we think about the wheels turning, and also having a lot of fun in the process.Sam Nadler: Very cool. I would just love to hear about, you know, one, you know, as founders ourselves, we love the founder journey. You mentioned, you know, leaving Big Tech to starting something. What was the inspiring moment you had, to jump all in? And then let's I would love to hear about your product.David Petrou: Yeah. Sure. Yeah. So, you know, my my whole life and I I was really inspired by by my my dad in this. You know, he he was an inventor, and he had, many patents.And he always, taught me the importance of coming up with new ideas and combining things in new ways and doing things more efficiently. And so my career at Google was always one of serial entrepreneurship. I would come up with an idea. I would create the prototype. I'd evangelize it.I'd get a team together. We'd prototype it. We'd take it to production. We'd launch and then and then repeat. And I had a lot of fortune in in meeting a lot of great engineers at Google and having a lot of great managers.I was on the founding team of Google Glass, the wearable computer with trackpad on the side. I was also on the founding team of Google Goggles, which was the world's first universal mobile visual search engine. You now think of that as a Google Lens, you know, in the latest incarnation, and and a bunch of other projects. And around, I guess, it was, you know, in the early twenty twenties, you know, around the COVID time, it was just amazing what we were starting to see internally at Google. And then the whole world saw externally from OpenAI, I think, in around November 2022 when Chat came out.And, boy, it felt to me like the timing would be right to see how fast I could run on my own. You know, coming from Google, a company with near infinite resources to a place where, boy, you're just, like, staring down the abyss of, like, oh my god. You know, there's so many things that can go wrong. But at the same time, you have a completely clear calendar. You can partner with whomever you want.You can make decisions immediately. There's no overhead. And and that's really what the last few years for me have been.Jordan Metzner: Yeah. Tell us about Weely.David Petrou: Yeah. Okay. So Weely came about while we were working on our previous product. So I'm gonna spend just a few minutes on the previous one because it actually is important to not only how came to be, but also the opportunity that we see that's still unsolved with things like agentic coding. So before Wealy, we were working very hard on something that we called social AI, meaning that everybody uses these chatbots and it's changed the world, but it's always in the context of what you might call single player mode.You type something, the agent responds, it's calling a response, but people are social. People work in groups. People collaborate. People are inspired by each other. So we thought, why don't we take an AI chatbot and put it into the context where people are already communicating, whether it's WhatsApp, iMessage, Slack, Telegram, whatever.And then the AI can be really smart and have this kind of social awareness or social intelligence or etiquette on should it respond, should it hang back, how should it add value, should it take notes, Should it bring in information from the world that's happening right now in ways that can really help the group chat experience? And we saw tons of magical moments with that. We saw a lot of really interesting use cases. Around September of last year, so September 2025, we finally saw coding agents get good enough where they can operate somewhat autonomously. Now, obviously, we've been in this area and following this area for a very long time, so it's not like, you know, we tried coding agents for the first time then, but it was the first time that it was really starting to click.I would say that our experience, we used chatbots, just from, like, web interfaces to do some coding. We pretty much skipped over the whole cursor tab tab tab, revolution and kind of jumped straight into using agentic coding harnesses such as the one from pi, so pi.dev, a very, very nice one. And we saw that, boy, we could completely revolutionize our business by changing every single dashboard, every single piece of business intelligence at Continua instead of it being something that is just for a human to look at. You know? What do our costs look like?How many users? How many monthly actives? What's the retention? What if instead we threw an API on top of everything? Because once you do that, you can send off a hoard of agents to do all kinds of business intelligence analyses overnight, and you can have agents looking at logs and making sure that everything is working well.So this is the classic story of in Silicon Valley, you set off to build a. In order to build a, you build b. And then when you build b, you realize, oh my god. This is actually the bigger opportunity because every business out there is going to want to have access to better business intelligence, ways of making decisions faster, not to mention a better system for creating their software itself, which is what we're doing in Wheatley.Jordan Metzner: Obviously, you know, like you mentioned, you know, late last year, the LLMs and kindSam Nadler: of AI in general changed in its ability to kinda do a step function quality improvement. And then, obviously, we'veJordan Metzner: seen just insane velocity since then. But, yeah, tell us now, you know, obviously, there's a lot of coding tools and agents and harnesses and all different things on the market even by different phrases or names. But tell us a little bit about Wheelie, who's using it, or, like, kind of who's the the archetype of of who wouldDavid Petrou: be using it. So so, I mean, the archetype would be solopreneur, if that's the term, also prosumer, CTO of a small company, up to medium size. Really, you can think of it like, okay. If you've done agentic coding in a harness, you kinda know what the experience is like. You have a terminal open.You're doing the work. You have various degrees of autonomy. It started where you type something. You'd wait thirty seconds. You get an answer, you type something back.Now it's going on for, like, longer, you know, goal oriented stretches. And then you get to a point where you realize, wow, I can open up two of these windows. Now you have two agents functioning at the same time, and then you can go to four and so on. And then what you realize as a developer is, you're starting to get very exhausted. You're starting to have a lot of trouble holding in your mind all the state that's going on.A lot of companies now are experimenting with sub agents and plan mode and stuff like this, but they all suffer a number of problems. And just to tie in our experience with social AI into the story of Wheelie, human attention is the most precious resource that's out there. Token costs are continuing to go down, but what's not improving is the availability of, you know, the human attention span. And so what we've seen is that organizations, whether big or small, the more they adopt agentic coding, the more they struggle with problems of communication, say amongst a team. We're seeing situations where, there could be a production failure, and people on the team, they just don't even bother to talk to others on the team to give, like, a proper postmortem, like, what happened?How did the fix go? Because it's faster just to make the fix and roll forward. Right? So what Wheelie is trying to solve is figure out how to get the right information to the right developers at the right time so that you're not exhausted and so that the money that you're spending on these tokens is being maximally used. So it's a very opinionated platform, and I think that you're right, Jordan, that this space there's a lot of players in this space, and different companies are take taking different views on how to do things.By opinionated, the approach that we're taking at Wheelie is that a lot of the systems that have been built up to now have been built for the shape of human attention. Think about Git. Think about GitHub. If you were to design source control from scratch in an agent first world, you wouldn't do it that way. We see scaling issues all over the place in the stack.Okay. You have a problem. You state the problem. Wheelie decomposes it, figures out the right number of agents, figures out where they should run, figures out when to interrupt you if it has a question. It knows your habit.It knows what time zone you're in. It knows what time zone your coworkers are in. So if there is a production issue and you're off for the night, it could contact someone in a different time zone that might still be awake. So it's this type of optimization of communication flow that Wheelie provides. We are early in this, adventure.So at the moment, we are a wait list that you can sign up, and we take people off the wait list, and we hold your hand through different parts, and we ask exactly what's most important to you. We refine our product based on that. But at the end of the day, what we're trying to do is provide maximum value on a per token basis for doing agentic coding, and that's why we call it an agentic development operating system.Jordan Metzner: Okay. That's a that's a cool overview. I think if I'm to understand it here, this isn't a replacement for individual developers kind of coding tools or harnesses, but more of an umbrella that sits over an entire system to help developers.David Petrou: We use foundation models underneath. We also have some fine tunes for for various purposes, but but you would not be using Cloud Code directly. You would not be using Codecs directly. We are aiming for think of it like the space between a developer that wants to use the low level tools and someone who would use Lovable. So we're in the middle space.That's that's what we're targeting.Jordan Metzner: Okay. I gotcha. And I think, yeah, as you mentioned, like the prosumer market. Is there any kind of verticals where you think kind of these or like types of businesses where it's kind of an optimal use case?David Petrou: Yeah. I mean, the there's a lot of things that seem to come up. One is just modernizing existing software. So things that just need to be upgraded and brought into cloud or or clunky pieces of code that you want a better front end. What's very interesting about the space that we're in right now, again, getting back to my point before about how software development is going to change dramatically with these tools, is that the the types of SDKs and libraries that we've had in the past don't necessarily make sense today.Take something like front end development, like React programming. React's great if you're a person and a human. It also is incredibly bloated and can be slow. Now an agent that cuts through all of these layers can modernize, front ends and modernize different experiences in a very bespoke personalized way that is faster, has a better user experience, less bloat, and requires less third party code. So it's this type of rethinking the entire software development life cycle and the software development stack that we're doing at Wheelie.So I would say if you are a company willing to experiment with either modernizing some old code or you have a greenfield project and you want to experience how can you maximally use agentic coding without the baggage of any prior dependencies that you have, then then Wheelie, I would say, would be a great place to start.Speaker 4: Every company knows they need AI. Almost none know where to start. We do. Rise Labs works inside your org to identify real use cases and ship them end to end. Not innovation theater.Real deployment. This is Rise Labs.Sam Nadler: David, thanks so much. I'm gonna queue us up for two news articles. Would love your thoughts. Both of you come, you know, with your experience at at Google with Google Glass and Jordan. Jordan had a big hardware project earlier in his past.Would love both your thoughts on the Apple foldable phone. So apparently, the first foldable iPhone is going to be released q three this year. Is this gonna be a huge hit? Is it gonna be a bust? David, if you don't mind, any high level thoughts?David Petrou: Yeah. I mean, so first of all, I'm like, I'm a I'm a pixel guy. I I shall just show you something funny here. I have like, we might have to edit this, but I have, like, a whole rack of of Pixel phones for for various hacking. And so, I mean, I do have a couple iPhones laying around here, but I haven't even gotten into the foldable Pixel yet.High level, here's my thought because, and I'm sure Jordan can talk more about the foldable aspect. I think that Apple is in an incredibly strong position with regards to, bringing AI to as many consumers as possible given that they have this unified memory architecture. So we're seeing a lot of people experiment with iMac minis and getting large language models running on them, and we're gonna see more and more of that push to the edge, is great for privacy. It's great for latency. There is a problem though.It is now possible to do much more radical, innovative user experiences on an Android device because you can unlock it. You can have access to all the sensors. You can do disintermediation of the applications by, you know, remixing different parts of different apps. You know, say you want to get a car ride somewhere. You just want the lowest price.Why couldn't the OS run the Lyft app and the Uber app and, like, render off screen, do screen understanding, and then present you, like, the choices, and then you select. That's something that is possible on Android. It's not gonna be possible anytime soon on Apple. So I guess my point here is that we're in interesting times because hardware wise, Apple, with this acceleration and and all of all the the unified memory and so forth, is in a fantastic position. But from a closed ecosystem perspective, I can see I can see things shifting toward Android in the near future.Jordan Metzner: That's a interesting perspective. I think you're right. Like, controlling the OS at an agentic level sounds like a new opportunity that hasn't kind of been presented on either OS yet. I think what's interesting about Apple foldable phone is it's a new interface screen size, kind of like a double wide screen size. And I know Samsung obviously has, like, kind of the foldable there as well.David Petrou: Yeah. Pixel Pixel does as well.Jordan Metzner: Yeah. And Pixel as well. But, you know, does that become kind of a new computing interface? Because, like, kind of, you know, a phone historically was too small, tablet became kind of more powerful, and so does this come become, like, you know, an interface that becomes significantly more popular.David Petrou: But but the the question is, why, though? So if you think about I mean, I'm seeing a lot of developers who are now comfortable on laptops because agentic coding gives them the ability to just use natural language. And so if we think about a deeper embedding of AI across the stack, I'm not so sure that we need the extra real estate. I mean, maybe once in a while when you're on a plane and you wanna watch a movie.Jordan Metzner: Yeah. I I don't know. I mean, I think, you know, it is hard to, like, view a web browser and view your terminal. And I don't know, again, you know, or whether we'll move, you know, continue to manage development in this type of state.David Petrou: Maybe that maybe the laptop disappearing will just have foldable phones.Jordan Metzner: Well, I mean, obviously, we've seen the proliferation of voice as like a commanding center of a way to control your device. So, you know, yeah, maybe we'll see the keyboard disappear as well. So I don't know.Sam Nadler: Alright. One more, guys. I think this released yesterday, Fable 5.1. I haven't used it yet. I've heard it's better at writing than Fable five.Obviously, I think the the top line is it's cheaper. But, yeah, any experience thus far? I'll start with you, Jordan.Jordan Metzner: I just been playing with it for a little bit, but there's so many models that came out today. There's the updated Google stuff, and I think there were some new stuff from OpenAI on Astra. But, yeah, I updated and I'm not sure I can tell you I can see a difference specifically just, like, from, like,Sam Nadler: the small coding task I gave it so far.Jordan Metzner: But, yeah, I think everybody's pushing the frontier pretty hard. Elon said there's a new Grok model coming out any day now. So it seems like the competition is heating up and from a developer perspective, that that feels good. Give us kind of more optionality and choice. And then also been playing with some of the open source models.How about yourself, David?David Petrou: Yeah. I think there's a few things here. I'll just start from the developer experience. First of Anthropic is a fantastic company. I have a number of friends there.OpenAI has been a lot friendlier in in certain ways. So, Anthropic, does not let you use, the subscription option with a different harness. So you have to go through Cloud Code. You can't use PIE or or anything else. OpenAI has just been much more welcoming, I would say.I believe Anthropic has fixed some of the problems that, were in a Fable five regarding the type of language it would use, and it would be quite loquacious and sometimes use terms that are quite oblique. And so 5.1 is an improvement on that. Still, though, when you look at cost and trying to be, you know, Pareto optimal on on cost versus quality, things like Gemini 3.8 Flash, which just came out today, it looks really good. But I also think that moving further, it's just look at deep sea. Look at, you know, when look at what's what's happening with with these Chinese models, and we're starting to see the market really be really moving or or, you know, spending their money in places where it might not be the absolute frontier, but you can get much more tokens per second, much more tokens per dollar.And in that case, it really comes down to the harness. Right? How are you managing memory? How are you decomposing large problems into subtasks? And so I think it's gonna be quite interesting, the coevolution of large language models along with the harness.And what we might see from certain labs is actual development of new programming languages that are intended for agents. And that could open up a number of new efficiencies, especially when you think about program verification.Jordan Metzner: Okay. I don't wanna get too deep into it, but new programming languages for agents to speak with. Just one quick, I guess, question there. Why why would the English language or even maybe a semantic language that already exists for programming not be good enough for what we have today?David Petrou: We should not assume any, any of the ergonomics, that currently exist are is the optimal set of things for agents. I mean, if if you think about, say, Kubernetes, you know, people talk about, okay. Was Kubernetes invented or discovered? Right? So it it works well for humans.It's not necessarily the right abstraction or an agent to manage, you know, processes that that run-in the cloud. When it comes to programming, what I've found what we found at Continua is that using formal verification like TLA plus and a bunch of other ones, they're really useful if you do so at particular seams. For example, places where there could be a number of there could be a higher chance of concurrency bugs, API seams, when clients talk to servers. So in these various places where you have a bunch of state machines, it's really great to talk about what's the preconditions, what's the post conditions. Right now, something like a Go program and your formal statement of what it's supposed to do, these are separate things.So you could imagine programming languages that bring those two closer together, which would be kind of difficult for a human to deal with, but not so hard for an agent. Now whether you can single shot it or you need to fine tune the LLM, that's a whole separate story. But I would say, just to get back to my my primary point, the coevolution of the model along with the harness and potentially programming languages, I think, is really where this is going, and we can't just look at some of these existing benchmarks that have already been around and are kind of showing their age right now.Jordan Metzner: Oh, that's a an amazing and unique perspective and obviously comes from probably some of years of experience. Okay. Well, this was a pretty awesome episode. David, thanks for joining us. How can people get ahold of you?David Petrou: You can find me on Twitter, d p e t r o u. I'm also on LinkedIn, and you could just go to wheelie.dev and get yourself on the wait list.Jordan Metzner: Awesome. Thanks so much. Thanks, Sam. Great episode.Sam Nadler: Thanks, Joy, and thanks, David. See everyone next week.Jordan Metzner: See you.