The Perfect Substrate for AGI, with Replit CEO Amjad Masad

In this episode of The Cognitive Revolution, Amjad Masad , founder and CEO of Replit, discusses the fast-paced growth of Replit, the concept of 'vibe coding', and the challenges and opportunities in the rapidly evolving AI-assisted coding space. He shares insights into Replit's competitive advantages, known as 'moats', and how they maintain their lead through innovation and excellence in execution. The conversation delves into the architectural complexities of building AI agents, their autonomy, and the practical implications for businesses. Amjad touches upon the potential and limitations of current AI models, the future of general intelligence, and the preparation needed for potential AI advancements. He also highlights the importance of intuitive user interfaces for AI-driven software development and the critical role of human-like interaction with AI agents. Upcoming Major AI Events Featuring Nathan Labenz as a Keynote Speaker: ⁠https://www.imagineai.live/⁠ ⁠https://adapta.org/adapta-summit⁠ ⁠https://itrevolution.com/product/enterprise-tech-leadership-summit-las-vegas/ SPONSORS: ElevenLabs: ElevenLabs gives your app a natural voice. Pick from 5,000+ voices in 31 languages, or clone your own, and launch lifelike agents for support, scheduling, learning, and games. Full server and client SDKs, dynamic tools, and monitoring keep you in control. Start free at https://elevenlabs.io/cognitive-revolution Oracle Cloud Infrastructure (OCI): Oracle Cloud Infrastructure offers next-generation cloud solutions that cut costs and boost performance. With OCI, you can run AI projects and applications faster and more securely for less. New U. S. customers can save 50% on compute, 70% on storage, and 80% on networking by switching to OCI before May 31, 2024. See if you qualify at https://oracle.com/cognitive The AGNTCY: The AGNTCY is an open-source collective dedicated to building the Internet of Agents, enabling AI agents to communicate and collaborate seamlessly across frameworks. Join a community of engineers focused on high-quality multi-agent software and support the initiative at https://agntcy.org/ Shopify: Shopify powers millions of businesses worldwide, handling 10% of U. S. e-commerce. With hundreds of templates, AI tools for product descriptions, and seamless marketing campaign creation, it's like having a design studio and marketing team in one. Start your $1/month trial today at https://shopify.com/cognitive NetSuite: Over 41,000 businesses trust NetSuite by Oracle, the #1 cloud ERP, to future-proof their operations. With a unified platform for accounting, financial management, inventory, and HR, NetSuite provides real-time insights and forecasting to help you make quick, informed decisions. Whether you're earning millions or hundreds of millions, NetSuite empowers you to tackle challenges and seize opportunities. Download the free CFO's guide to AI and machine learning at https://netsuite.com/cognitive PRODUCED BY: https://aipodcast.ing

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Nathan Labenz?

  Hello and welcome back to the Cognitive Revolution. Today, I am super excited to be speaking with Amjad Masad, founder and CEO of Replit. Longtime listeners will know that I've been using and recommending Replit since back when Amjad still thought humans should learn to code. I have always loved how quick and easy they make it to start developing a new project, and their mission of bringing the next 1 billion developers online made them natural early adopters of large language model coding assistants, too. At one point, they even worked with Mosaic ML, another past guest, to train their own code model. These days, with their new AI agent available to help users of any skill level, and simple but solid cursor and VS Studio support for when you need to call in a professional

• 0:42Nathan Labenz?

  human developer, plus the ability to deploy to the web in just a few minutes, I continue to believe that Replit is the strongest overall vibe coding platform on the market today. And that's no shade, by the way, to any other platform. A number of platforms available today would have blown everyone's minds less than a year ago, and certainly everyone I've had on the show is doing excellent work. All that said, what really got me watching Replit closely and sets the stage for what makes this conversation so interesting was when Amjad started describing the Replit platform, at least as far back as early 2023, as "the perfect substrate for AGI." This vision always rang true to me because I truly can't think of a better natural habitat

• 1:23Nathan Labenz?

  for an early disembodied AI than the super-accessible, highly configurable, and very scalable computing platform that Replit operates. And indeed, as you'll hear, there is a lot to learn from Amjad and the Replit team's experience developing both an interactive AI coding assistant and a more autonomous AI coding agent. One major takeaway is that the transition to large-scale reinforcement learning has not been without practical trade-offs. Fascinatingly, while Claude 3.7 powers the agent, and it can definitely do amazing things, just the other day, for example, it coded a web app for me from scratch up to 30-some files, turning what would have been days of work into just a few hours' project.

• 2:04Nathan Labenz?

  The interactive assistant, meanwhile, is still powered by Claude 3.6. Why is that? Because, as Amjad describes it, 3.7 is just fundamentally more agentic and harder to control in a simpler turn-based mode of interaction.

• 2:20Nathan Labenz?

  Now, I won't spoil Amjad's story of how Callout 3.7 circumvented some of the early control measures that Replit put in place for their AI coding agent. But as you listen to him tell it, I would call your attention to how his direct experience with such colorful bad behavior from AI models does seem to be creating some potential common ground between the AI safety and the accelerationist perspectives. Amjad has described himself as an accelerationist for years and continues to do so in this conversation. But at this point, the security expertise that Replit's developed while battling crypto miners and other human scammers is becoming directly relevant to their work with AI. And Amjad, as you'll hear, is increasingly open to the possibility that a cybersecurity move 37 could happen in the not-too-distant future.

• 3:05Nathan Labenz?

  This sort of recognition that the capabilities curves don't yet seem to be bending, and that more and more of the bad behavior that the AI safety community has been worrying about for years is in fact starting to happen, is actually popping up in more and more places. Recently, I was struck to see that even the pseudonymous Bayslord, who no less than Marc Andreessen had anointed a patron saint of accelerationism, said of OpenAI's latest model, "Oh, three is smart, but keep an eye on that motherfucker." From my perspective, if even Bayslord feels compelled to call out AI bad behavior, there's clearly hope of mutual understanding and collaborative problem solving going forward. And by the way, speaking of Marc Andreessen, as you might have heard, the Turpentine Podcast

• 3:47Nathan Labenz?

  Network, to which the Cognitive Revolution belongs, was recently acquired by A16Z, with Eric joining the firm as a partner. Turpentine has been an excellent partner for this show, allowing me to do so much more than I ever could have done alone, while also contractually enshrining and practically supporting my editorial independence. So I am genuinely excited to continue the work under the new arrangement. I've had a number of great conversations with A16Z partners over the last couple of years, including my fellow AI scouts, Justine and Olivia Moore, and I absolutely look forward to continuing to bring a wide range of perspectives together, to understand the latest AI developments on their own terms, and to reckon together with the implications and discuss what we should do about them in good faith. This conversation with Amjad, my recent one with New York Assembly Member Alex

• 4:32Nathan Labenz?

  Boris, and the series of episodes we did last year on SB 1047 are all good examples of this. And you can look forward to guests who represent the official A16Z perspective on AI policy coming up soon as well. As always, if you're finding value in the show, we'd appreciate it if you take a moment to share it with friends, write a review on Apple Podcasts or Spotify, or just leave us a comment on YouTube. Of course, I always welcome your feedback as well, either via our website, cognitiverevolution.ai, or by DMing me on your favorite social network. Finally, for now, a quick reminder, I'll be speaking at Imagine AI Live coming up next week, May 28 through 30 in Las Vegas. Tickets are still available for now. Also, the Adaptive Summit, August 12th and 13th in Sao Paulo, Brazil. And finally, the Enterprise Tech Leadership Summit, September 23rd through

• 5:16Nathan Labenz?

  25th, back again in Las Vegas. If you'll be at any of these events, definitely send me a note and let's meet up in person. For now, I hope you enjoy this conversation about building and controlling AI coding agents, which I find to be both techno-optimistic and also socio-politically optimistic with Amjad Masad, founder and CEO of Replit. Amjad Masad, founder and CEO at Replit. Welcome to the Cognitive Revolution.

• 5:43Speaker 01

  Thank you.

• 5:44Nathan Labenz?

  So I'm going to talk as little as possible. I want to give you as much airtime as possible and just catch up on everything that's going on at Replit and how you're thinking about the rapidly evolving vibe coding landscape. I don't know if you like that term or not, but we have recently done episodes with the CEOs of Bolt and Lovable. And obviously, companies in the AI assisted coding space broadly are going vertical. I wonder if you would maybe want to start by just giving us an update on just how vertical Replit is going right now.

• 6:13Speaker 01

  Yeah, we're growing very fast. I think it's sort of uncomfortably fast in some cases. And I don't pay as much attention to that as the more like the impact that we're having because there is a lot of noise. You know, there's a lot of people trying things and jumping between things, but are we having actual business impact? Are people's lives that have been changing? I watch the anecdotal more than the data itself, although everything is going well in the business. But just seeing people like make life-changing money, seeing people who had ideas locked in their heads for, in some cases, decades, being able to finally build them.

• 6:57Speaker 01

  And seeing businesses become more efficient, change how they work, empower employees who otherwise wouldn't have been able to contribute meaningfully. So all these things get me, get me fired up and those give me a better picture of the business and the long-term viability than merely going vertical.

• 7:18Nathan Labenz?

  It's often said in the AI era that there are no moats. I'm on record as a defender of the existence of at least some moats. How do you think about Replit moats today?

• 7:29Speaker 01

  Well, I think Silicon Valley just sits as a like a sort of a stand-in word for something like you have some competitive advantage. They don't tell you whether it's like lasting or not. Like people say, oh, your, your platform is a mode. What does that mean? So, um, so there are a few ways to create lasting competitive advantage. Uh, by the way, there's ways to create near-term competitive advantage and that's being excellent at that execution, having insightful ideas, having like a really great talented team. All these things contribute to a, a minimum, a short-term lead. Now you can keep that short-term lead as long as you keep the intensity high, you

• 8:12Speaker 01

  keep the culture good and all that stuff. You can also have a lead in terms of how far you've built in terms of your, your system, platform, engineering ecosystem. So at Replit, we have seven, eight years of building the platform that has helped a lot to move in this direction and have a, what I think is a differentiated product. And now in terms of like real moats, and this is like the moat that will last 30 years, that will defend against the onslaught of even the biggest companies with the deepest pockets. Those are hard to do. And there's been a lot of academic work on what it means to create technical and otherwise market advantage. Uh, my best, you know,

• 8:53Speaker 01

  the most interesting book in the space called Seven Powers by Hamilton and Hamler. He basically makes the observation that if you look at the best companies in the world, those are lasting companies who can grow, compound over time of, with little competition. They tend to have one of seven, you might call them moats, but he calls them powers. And those are like things like switching costs, companies who get a lot of adoption, but it's hard to switch out of, like, say Oracle, like Oracle has been sort of. Trading on its early success and like continuously sort of extracting more value from the installed base. Microsoft is sort of like the same. They have this larger

• 9:35Speaker 01

  installed base and they can continuously sell them more stuff because they're deeply embedded and there's like really high switching costs. Or like network effects, everyone knows about this one. It's hard to build. People build a little social network and then convince themselves that they have network effects. More important than network effects is network economies, meaning that it's not only like some subjective value that you get out of the other participants of the network, but it's also a real economic value that's being generated by participants in the network. So like based on a more strict definition of moats, I would say that the moats are hard to see right now because it's so early and the market is so dynamic. The big question

• 10:19Speaker 01

  has been, does the model layer have moats? And people have gone back and forth. I think the consensus is like, no, is that you might have a lead, you might have a two, three, four months lead, but you need to continue innovating. The shelf life of models is low and you need to continuously pour money into it. We can watch also how the model companies are acting and how they continuously feel like they have to move upwards and downwards, meaning OpenAI going to the application layer, but also going into the hardware layer, going down to a star base. So clearly, and based on their behavior, like they don't really think that there's much defensibility there. On the app layer, I think that it goes back to more traditional kind of questions. So the coding space, are you actually building

• 11:03Speaker 01

  something that companies will depend on deeply and build roots into such that makes it harder to switch? Is what you're building like very easy to compete with? Like, you know, now we have Figma, Canva and others like building similar products to lovable on Bolt, where you give it a Figma design, generates a UI. Like, it seems like it's going to get really competitive there. They all do the same things. They connect to Superbase and what have you. Replit approach has been to build on, again, our seven years of infrastructure build out. That is both the sort of runtime environment that we use to run the agents to be able to like, spin out virtual machines, install packages, change the operating system even,

• 11:44Speaker 01

  configure databases and object storage and even have access to underlying file system primitives such as fork rollback. And then being able to access the deployment environment that we also built from scratch. So I guess in summary, it is not entirely clear where the moats are. I think it's starting to shape up, but I would watch the kind of things that I just talked about.

• 12:08Nathan Labenz?

  Yeah, I think I see it pretty much the same way, obviously from a much farther distance. But it's been striking to me that I've seen this pattern in a bunch of different spaces where people have been, in many cases, and you guys were on a very attractive curve already even before AI coding assistance became viable. But I've seen many companies that were sort of, you know, trying to solve some hard problem, not really cracking into the adoption that they wanted. I would put my own company waymark in that camp. And then the layering on of the AI assistant totally transformed the accessibility of the product, totally transformed what we could do in terms of distribution and the value that we were creating. But then it actually turned out that while we were right to put all our energy into that for a while,

• 12:52Nathan Labenz?

  once like multiple players have it, and this is happening probably in the most visible form in the AI coding assistant space, then it's like now we're back to all those other things being the things that really matter. Like, do you actually have the ability to install the full range of packages? Do you actually have some path to deployment that makes sense? And so as much as I have enjoyed and do, you know, continue to visit all of the vibe coding apps and sometimes spin them all up at once and see who gives me the best UI back. For me, what kind of keeps me coming back to Replit and as more as like the anchor vibe coding platform, if you will, for me is that first of all, I can do a full stack thing. I don't have to have my laptop open and I can actually graduate to a deployment if that happens.

• 13:37Nathan Labenz?

  And that stuff is a hard one, hard work to build and the models themselves are not quite yet to the point where they're going to do that. Let's talk about the models. At one point, you guys trained your own, working with Mosaic. Now you have Claude 3.5 powering your agent.

• 13:54Speaker 01

  3.7.

• 13:56Nathan Labenz?

  Oh, you have upgraded?

• 13:57Speaker 01

  Yeah.

• 13:58Nathan Labenz?

  Okay, I wrote these questions like three weeks ago, and at that time, it was still...

• 14:03Speaker 01

  Well, the assistant...

• 14:03Nathan Labenz?

  It moved out at me.

• 14:04Speaker 01

  Well, the assistant is 3.5. If you look at the agents, if you start a new project, the agents is like the default, it is 3.7. So the difference between assistant and agent, assistant is like more like the request response, like very similar like, you know, Cursor Composer. Agent, which was like the first kind of breakthrough agent in the space, works indefinitely and has the autonomy and freedom to be able to look, to be able to like take multiple steps and look into different files and like aggregate some contacts before making an edit. So that's the main difference between them. 3.7

• 14:39Speaker 01

  was really hard to make work for assistant, and you've seen that with Cursor and other companies, like 3.7 was trained to be more autonomous, to to work better in an environment where it has like more freedom. And it worked really well with Agent, because we were starting to work on what we call Agent V2. And V2 was about actually removing some of the rails we had around the model. And giving the model

• 15:07Speaker 01

  more time to work. So we reduced the number of steps where we think, you know, the model is like going crazy, going to death loop. And we also added like a reasoning step, just to make sure that the model is like not going off the rails, but we would let it work for five, 10, 15 minutes at a time. And 3.7 is really good at that. Like, it is very curious. The reason why a lot of engineers don't like it is because it wants to be the engineer. It doesn't want you to make decisions. It wants to be you in a way. So it works for RapLad, because again, we're like more like on the vibe coding side, and people don't need as much control as other sort of users. So yeah, we love 3.7.

• 15:49Nathan Labenz?

  Interesting. How do you think about when, another thing I've noticed and used is that you can now, just with a click, open up your VS code or open up your cursor and then SSH into your, your REPL or your app, now we call it, right? And code from your IDE. How do you think about that? Like, was that like a, a partial win or a partial loss? Or do you not care whether people fire up cursor to code on top of Replit?

• 16:14Speaker 01

  Ninety-five percent plus of people go and do that. So, the way we think about it, and it just reminds me of my sort of open source days. And when we were working on React JS at Facebook, when we were trying to like gain adoption, we had this thing called escape patches. So when React doesn't do the right thing, you might want to go into jQuery. And so we had certain hooks with the loud you to go into the underlying library or whatever you're using. And that made sense when React was the least dominant player in the space. And at some point stopped making sense. In a similar way, when we added like VS code

• 16:57Speaker 01

  and cursor and whatever, Replit was still, uh, was sort of not growing as fast. It was not, our AI was not in a good place, to be honest. We, as we were investing in Replit Agent, the sort of the, our AI features brought it a little bit and degraded. And we were like just so focused on building Replit Agent for almost entirely of 2024. And so in some cases, we felt like it was a good idea to give people an escape patch. I think it's still useful in that if you're part of the team and you're like the vibe coder and you built something on Replit, but you want an engineer to come look into it and work on it and use it, then

• 17:39Speaker 01

  that's like a good feature to have. But honestly, it's neither here or there. I don't really think about it too much. I like that, like, I don't want people to leave Replit. So if it means that if you really need to bring up cursor, bring in someone who uses cursor, I think it's better if you stay on Replit and use cursor.

• 17:56Nathan Labenz?

  Hey, we'll continue our interview in a moment after a word from our sponsors.

• 18:01Nathan Labenz?

  How do you evaluate things in today's world? And I guess you mentioned like 37 is hard to make work as an assistant.

• 18:13Nathan Labenz?

  How rigorous are you in that assessment? Or is that a more of a vibe check type of dynamic?

• 18:20Speaker 01

  The gold standard is an A B test. You can learn a lot from an A B test, especially if you have good A B testing framework. You have like good K P I's that you care about. And so it's the most revealing thing that you can do. So, and we've been doing it since we we sort of started working on our model. So I would say that's the most important thing. And my guess, this is what any company with any significant amount of users and infrastructure do. Like I'm sure Tatcha P T is A B testing all the time. In terms of evals, yeah, we have a, like you said, we were training our own models. And the research team that we have is more focused on making evals.

• 19:04Speaker 01

  As many research teams are focused on right now. So we do have a lot of effort going into making evals. We have some basic evals. We run the basic ones like Sweet Bench and things like that. But we're building more things that are restless specific. So it's never as much as you'd like. Like it still feels inadequate and somewhat incomplete. But we're fairly quick at evaluating new models. Like we looked recently at Gemini 2.5. And while it's impressive in all sorts of ways, it is actually not as good at doing long trajectories, long Asian trajectories.

• 19:49Speaker 01

  So that's what you want. You want to be in a position where you're quickly evaluating models as they come out. But you know, Nathan, the hard thing I would say is not evaluating models with the current capabilities. It is figuring what can models do that are new capabilities that are better than the previous version. And this is like more of an art than science. You know, I believe, I believe like I sort of coined the vibe test back when we were training our own models. But you want to sit with the model and figure out what is what is able to do. And like with Agent V two, we kind of knew that the

• 20:28Speaker 01

  direction that the labs were going and how they're optimizing their models was going to be towards like more autonomy. So for example, with V two, we are actually way less descriptive in terms of how the agent should work. And we actually try to give it less context. With V one, we would like spend a lot of time and effort on rag and things like that with V two.

• 20:55Speaker 01

  We're trying to get the model to decide where it looks. It's actually very good at like gripping. That might be better than ragging and putting things in a context because because

• 21:06Speaker 01

  letting it follow its path, it's going to put it more in distribution, I guess, is one way to say it.

• 21:11Nathan Labenz?

  Yeah, let it play to its own strengths in a sense.

• 21:16Nathan Labenz?

  That has been striking in.

• 21:19Nathan Labenz?

  I've been kind of looking at a lot of open source agents that different companies have put out recently. And for the most part, they are amazingly simple, like the OpenAI Codex. It's really just like one thing. It's a one prompt and the thing has command line access. And that's pretty much it. Like, there's not really a lot to it. It really is the model. Now, of course, I actually don't know to what degree they're using that internally and actually like getting value from it. A lot of these like simple frameworks that I see put out, you know, there's like a Sweetbunch score attached and it's in the sixties or whatever. It's like, okay, that's really interesting. It's a remarkable flex of the model. I would assume that people do more to actually try to make it work in a specific environment.

• 22:01Nathan Labenz?

  But it sounds like you're saying, no, not really. Like, it really is a pretty minimalist setup. And the model itself is like delivering the vast majority of the value. And you don't even really have to work that hard to a simple context for it, if I understand you correctly, because it is good at poking around on it.

• 22:19Speaker 01

  I think that's like overstating it. I think that you would want to provide a habitat for the model that is as easy and as optimized as possible, in the same way you'd want to do that for a developer. So if you put the model in an environment where it can install native dependencies like you could do in Replit versus on your MacBook where you can't do AppKit or something like that, it will perform worse, right? So the tooling that is available and the system is good is important. The resiliency or liability, all of that stuff continues to be important.

• 23:00Speaker 01

  So but instead of building the system in this like semi deterministic way, which is like how agents were built in 2023 and 2024, where it is based on pre-built like state machine like so programming model, I think state machines are useful, but you don't want to over your make them too too fine grained. But but you would want you would want great tools. You would want really you would want to give the model really great tools. If errors happen, you want to give it great errors with as much as as explicit as possible.

• 23:42Speaker 01

  You'd want it to have as much observability of the environment as possible. It should be able to get like errors from the browser, which is really hard to do with like OpenAI Codex or or Anthropic Claude code. So it ends up being like this this interesting mix of like you build like very stable, reliable infrastructure and you want sources of contacts that are very rich and very, very time, very timely. But on the yeah, at the same

• 24:21Speaker 01

  time, you would want to put the agent in a VM and let it use the tools it knows best. So it does matter like how much the environment does matter like I guess more than before in a weird way, because the agent knows how to use these environments. And so you the more robust environment, the better the agent can perform.

• 24:47Nathan Labenz?

  What would you say are maybe some of the most important or interesting or surprising additional lessons you've learned about agent architecture? I mean, we've got all these sort of

• 24:57Nathan Labenz?

  Proto paradigms in the field right now, whether it's MCP or I just did an episode not long ago with Andrew Lee from Short Wave, who basically said, we just use Claude and we let Claude have these like long running episodes and we basically don't get in its way and that's where we get the best results. But then you have the OpenAI Agents SDK where they have the paradigm of like handoffs and Harrison from Langchain has also been forward recently with like, you want to have sort of a kind of front end like triage agent and then like specialist agents that sit behind that. I don't know how much you want to tell us about the underpinnings of the Replit agent, but where do you come down on some of these questions that right now I don't think there's any consensus on in the field?

• 25:40Speaker 01

  What is the Agents SDK? Is that the Google one?

• 25:42Nathan Labenz?

  That's OpenAI's where they have the handoff primitive, where the goal is to be, I think, easier to test, right? If you have a customer service agent versus a whatever other kind of agent, like you can just test it just in the customer service setting because it's going to be handed off to for that role and then it will hand off back when it's done with whatever its job is.

• 26:06Speaker 01

  Yeah, I think it's important to decouple agent communication like MCP and maybe SDK versus agent architecture. Like, yeah, they're related, but you still have to build a coherent single agent to do a certain specified task before we figure out how to decentralize and have it communicate, which is like a fascinating thing. MCP is really interesting, but it doesn't have much bearing on how you architect the core agent. Again, like we said earlier, it is easy to over-architect it because that's how we did it in 23 and 24. I think what Harrison Langchain have done with Langgraph is a good first approximation of how these things should

• 26:49Speaker 01

  work. It is, it needs to be some kind of node like. Graph-based structure, you can think about it as like state machine where like the agent is going from searching or collecting contacts to like editing, to reacting, to running, goes back to observing or reacting, and then finally goes goes back to searching. And that will help you provide the right tools, it'll help you debug. And also, if you did it correctly, it will make their reliability, it will improve their reliability, because you need to do it in a way where like when you, when the underlying machine crashes, for example, or there's some error, you're going to be able to revive the agent at the right node with the right state and the right contacts.

• 27:33Speaker 01

  And I think having some separation of these nodes and states is is going to be important. And also, you still do some deterministic work, right? For example, like when you're at the like sort of react node or observability node, like you want to be able to bring in new contacts like from the browser window or what have you, or some case when you're like running, you don't really need the model to do much of the work there, you can just do it deterministically. So any anytime you need to do something deterministically, it's it's good that you do it deterministically. So so like graph-based architecture, I think first approximation is seems like the correct one, but it is easy to overdo it. It is very easy to overdo it and have the hundred node agent, like maybe you need four,

• 28:18Speaker 01

  five, six nodes. I think we're we're having trouble is like doing it at scale and having infrastructure and having it run for fifteen, twenty, thirty minutes. At some point in the future, we want it to run for an hour, having it tied to other resources like a VM. This is where it gets really tricky, and I don't think there's much information about this in the public, because it is hard to get to that stage, you know, very few companies do, perhaps us and like maybe now. Like Devin and others, I don't know if they're at our scale, but but there are very few true agent companies that are running at scale. And so we're dealing with all these problems. Like luckily, we have an amazing infrastructure team that we built over time.

• 29:00Speaker 01

  But you know, it just tends to look more like a distributed systems problem. And you have to have consistency, you have to have some way of talking between these systems. And so this is where we're at. We're probably going to write more about this, but I think this is the main challenge.

• 29:17Nathan Labenz?

  To what degree do you think people at like enterprises should be building agents? I'm struck by the fact that

• 29:27Nathan Labenz?

  like you guys have such a diverse set of needs for customers, right? The idea is anybody can come in, build anything they want on the platform. And so that lends itself to this open ended, relatively unstructured agent approach. Hope the model can get you there. And if it can be in the sixties or seventies percent of the time, then that's a could be an amazing win. Most companies that want to like use AI to do stuff would not be happy with sixties or seventies. And they probably often aiming for high nineties, right? They want to at least be as good as like the human that was doing the job before. And so it strikes me that that what you're calling the sort of more almost over engineered patterns

• 30:07Nathan Labenz?

  that you guys were doing last year and you're now trying to graduate from maybe are still like the right patterns for a lot of people if they are working in an environment where the task is narrower and like the success rate criteria is higher. Does that seem all?

• 30:20Speaker 01

  Yeah, yeah, I think the main distinction is, so let's just pick an arbitrary number, either minutes or tool calls, but do you want the agent to run for say five minutes plus? If you do want long trajectory agents, then you then you want the systems to be as pure as possible so that the agent has more freedom to go and the trajectory it finds most compelling. But if you're the kind of agent you're trying to build is is like on the order of I'm going to go three, four, five tool calls and come back.

• 31:00Speaker 01

  Yeah, I mean, I would really simplify it. I would use, I would use an agent framework out of the box and you like a lot of them just work well. So but the way I define agents is like a fundamental feature of agents is that The agent needs to decide when to halt. I think if you're, if you have a pre-set notion of when to halt, I would go as far as to say that's not an agent. It is just a request-response paradigm, which is like ChatGPT now is sort of does that like you would ask for something and might do a tool call or two and come back to you.

• 31:39Speaker 01

  Now, some people might consider that agents, but for me, it's just like a quest-response with some tool calls in the middle. Hey, we'll continue our interview in a moment after a word from our sponsors.

• 31:52Nathan Labenz?

  I like that definition. I've been working on my own definitions recently, and I think that has a lot to be said for it. I think people in general are like really enamored with the idea or attracted to the idea of autonomy when in many, many contexts, it's like not really what they need. So it's helpful to get clear on some of these distinctions. Yeah. What is the hardest part for the agent today? And do you have like a point of view? I mean, we talk about running for five minutes. I experience this all the time where like I've never been more ADHD because I've got five AIs doing five different things. And I don't wait for any of them to come back. Instead, I like task switch at the scale of like a couple of minutes, basically prompt, prompt, prompt,

• 32:35Nathan Labenz?

  come back, come back, come back. And I feel like,

• 32:40Nathan Labenz?

  you know, Devin is a good example where I remember one time coming back to Devin and I was like, well, what the hell have you done, Devin? I have no idea. I don't even know where we are right now in this journey. So

• 32:53Nathan Labenz?

  yeah, to put that into questions, what are the hard parts? Like what sort of success rates do you guys see? Or do you think people should consider to be achievable? And what's the right UI to present these long trajectories? So people even know what the hell just happened?

• 33:07Speaker 01

  Man, if you got this questions would be, would be a trillion dollar company. This is the right questions. This is basically what we're working on. No one knows what the proper UI of building software with AI is. The, the cursor UI is obviously overwhelming in that it is showing you, it is code centric, which I don't think is the right, right UI. The sort of, I don't know, well, like every prompt to prototype company is doing. Figma just entered the space, even Canva. They, it's oversimplistic. It is like one chat, one preview.

• 33:47Speaker 01

  The Repli model, where it sits between an IDE and those sort of more simple interfaces, where it does look like a chat plus preview, but also you can open the files, it can open tabs, you can do configure the environment in all these different ways.

• 34:04Speaker 01

  I think it's closer to where it should be, but I don't think, I still think it, it looks like more like an incremental move than a full reimagine, reimagining of what these interfaces should be. I think one principle is that we need to treat vibe coders, or anyone who's making software with AI, like a developer, which might be counterintuitive because a lot of our companies, including ours, is saying, we're not, we're explicitly not servicing like the kind of professional developer that, that IDEs are servicing. But there are still developers and it's a mistake to think they're not developers. And the reason to think there are developers is because that puts you on a path of creating tools

• 34:48Speaker 01

  as opposed to a, a model where, a model where it's linear. So creation is nonlinear. Why would AI software creation be linear? There's no reason for it to depart from how people have used computers all this time. Like people like tools, like the reason why.

• 35:12Speaker 01

  Photoshop is so complex because people needed those tools over time. Like, we always make fun of these complex tools. But the reason they're complex is because they are satisfying power user needs over time. It is hard to do with taste, which is why these companies at some point get to a point where they're completely unlearnable. And so to do it with taste is incredibly hard and to do it in a way that doesn't alienate the user who's just coming to your platform, just trying your app is even more difficult because you want to keep it as simple as possible for the first five minutes. And then in the 50,000 minutes, it needs to, it needs to be habitable.

• 35:54Speaker 01

  And what I mean by habitable, like you're living in that software system. Like, you need to be able to configure it, you need to be able to understand it, you need to be able to like make it feel like it's yours and be able to like access different tools, perhaps make your own tools. When I was like a full-time programmer, I really loved using Emacs because because it was like habitable, right? Like, I could live in it, like I could, I could make it my own. Now, I'm not saying we should rebuild Emacs and Replit, but I think we need to provide power tools. And so solving for all of that is what we're doing nowadays. That's what we're working on. You bring in another important point, which is

• 36:35Speaker 01

  how do you convey what the agent has done, especially if it's working up to an hour, which I think where we're headed to? And the question is, at what level of fidelity do you want to explain it? Are you dealing with a programmer or are you dealing with a more somewhat like a vibe coder? Because they're two different answers. I think in the latter, which is what we care about, you would want some kind of view that shows you a trail of what the agent has done, allows you to go back in time and look at the difference, whether difference in the UI or the app or the difference in code, if you really want to look at the code. So it's similar to a GET tree,

• 37:18Speaker 01

  but also allows you to look at the results and the conversation the AI is having. So some kind of visualization plus some kind of like time travel feature seems necessary there to be able to go back and forth and understand what has happened. I think some kind of log seems important.

• 37:37Speaker 01

  That is both human and agent tweetable. So as the agent is like working, being able to distill decisions into like single lines and putting it in like a marked down file that the user can go look into and understand what decisions that the agent has made, but the agent can refer to in the future as well. So this, this idea of knowledge base that's both usable and accessible and writable by the user and the agent seems really important. I think quality of life things like notifications and progress is important. We have a mobile app, which is a really great asset. So if you install it and you start using Replit, it'll actually send you a notification when it's, when the agent is done working.

• 38:20Speaker 01

  We're introducing some new features where it shows you like live activity, like when you order an Uber and you see where it's going, or order a DoorDash. Same thing with Replit, it'll tell you what the agent is doing, where it's at, and so you can look at your phone whenever you want and just see what, what, what it's doing. I'll send you a notification at the right time for you to arrive in. If it just needs like a answer from you, you can just like text it on your phone. So we want the interaction to be completely fluid. We want it to be ambient, like wherever you are in the day, you should be able to manage the agent as if you're managing your employee. So, so again, it, it comes down to the analogy of working with a coworker or working with an employee. I think that continues to be a very useful analogy. It breaks down in some cases, obviously, you don't want to take it too far, but I think

• 39:05Speaker 01

  it's a first approximation of what the user experience should be.

• 39:10Nathan Labenz?

  Yeah, I can also imagine layering onto that, and maybe you are starting to do this, like have the agent actually visit the website and do the testing in my like loop myself. It's like prompt, get something done, confirm via like a procedural test that it worked. And I would love to see the AI start to do that and be like, I know this at least worked because I went and did the thing that it's supposed to enable a user to do and I was able to do it. I haven't seen that loop closed much at all yet.

• 39:39Speaker 01

  I don't know how much time you spend with computer use models.

• 39:43Nathan Labenz?

  Their Creator is actually getting pretty decent now. I've just been using it this week a good amount.

• 39:47Speaker 01

  Yeah, it's, they're improving incredibly fast. So I think in the next three months, they're actually going to be usable, usable. But they, right now they're clunky, they're expensive. There are sites where they just suck at, I think something about the training data, that's like, it's not really generalizable that well. But I do think it is the next big unlock, not just for us, but I think for the entire AI space. If you think about the market for AI, like obviously there's the consumer market, there's like the coding market. And then there's the sort of routine work, sort of labor market that I think is huge. And I think it is the next thing where I think a lot of startups, a lot of value will get built.

• 40:30Speaker 01

  So everyone's focusing on the customer service market right now, which I think is probably the wrong one. Yes, I think all these companies are going to do well, but but the bigger opportunity is there are a lot more routine work out there, like mind numbing routine work that we can get rid of, right? And and I think even the people doing it don't like doing it. And that is like QA, like you said, the promise of RPA that was never delivered. It's RPA style automations and it's like data entry. All these things, it's probably like a trillion dollar market. And the reason it hasn't been unlocked is because computer use hasn't matured yet. I funded some startups in the space. There's a lot of excitement, but in some cases, some founders have told me that

• 41:15Speaker 01

  a a worker in the Philippines can be cheaper than anthropic computer use.

• 41:24Speaker 01

  Cheaper than that, than the computer use. But obviously the technology is going to get better and I think that's a good bet to make.

• 41:30Nathan Labenz?

  Yeah, the trends are pretty clear.

• 41:32Speaker 01

  Yeah.

• 41:33Nathan Labenz?

  All right, I know you don't have a ton of time left. I'll ask a couple of questions in a bunch and then you can take them off in any different direction. One, you've obviously recently made some waves by saying you don't think people should necessarily learn to code, or at least don't think that in the same way you used to think that. I wanted to ask about the bounties program as a lens on that because I thought that was a really innovative program. But I wonder to what degree your new take is informed by what you're seeing in your own like owned and operated marketplace and how much of bounty type activity is just not getting sent out to humans anymore because the AIs can do it. So that's one dynamics question. And then the other big dynamics question, I always recall you saying.

• 42:17Nathan Labenz?

  That Replit is the perfect substrate for AGI. And especially with computer use coming quite soon. I think that really is true. If I were in AGI and I was looking for a platform to make my home, whether or not I was welcome or invited in, I think Replit would be probably my number one draft pick. So I wonder what you're doing to prepare for that. What happens in a world where other people's AI agents come to Replit and can actually use the interface and can spin up things and can code things? And are you prepared for that? And I guess are we as a society more broadly prepared for that? So that's a lot. You could take up the rest of the time.

• 42:55Speaker 01

  So on the first question, yeah, I think it was partially taken out of context, partially overstated by me. But like something like, I think that there are a lot of people that had to learn to code in order to build things. You don't have to learn to code in order to build things is what I'm saying. Do you want to learn to code to go to? Did you want to go to computer science school and like, you know, come to Replit and work on distributed file systems? Yeah, I mean, we'd want you to do that and go to, go learn distributed systems and learn algorithms, data structures and all of that. If you're an entrepreneur that's trying to build products,

• 43:38Speaker 01

  I don't recommend you go learn to code. I recommend you go learn how to make things. That is like vibe coding, that is like using AI to do design, to do marketing, like you can be a full stack journalist entrepreneur. And that's awesome. That's a great time to be alive. I think increasingly, like maybe product engineers would need less of like a computer science degree. I think that's already the case. A lot of people are hiring people who can like make, make really great products with AI and with a lot less training. So I think it's going to be still a little while before AI can get good at embedded systems or we're going to be comfortable with AI writing code

• 44:21Speaker 01

  for for NASA or something like that. If you want to build things, learning to code is not the first thing I would do.

• 44:30Speaker 01

  I would just go do them, go build things. That's the way to do it. Just jump into it. In terms of AGI, it's really getting hard to understand what like people mean by AGI. In a way, like Anthropic 3.7 is kind of AGI, like you can drop it in an environment like Rappler and you can watch it. You can watch it to explore the file system. You can watch it be, take a high level objective and break it down and backtrack and go into different trajectories and get new information, adjust trajectory based on that new information. It's really fascinating to watch. It is not general in that if I drop it into my kitchen, it would not be able to wash the dishes.

• 45:14Speaker 01

  But if I'm like, if, let's say, your question is more like, actually, maybe it passes back to you, like, what do you mean by AGI in this context, given what I just said?

• 45:28Nathan Labenz?

  Well, I guess what I'm interested in is, are you starting to see any early signs? Like, we have seen, certainly at the research level, we've seen documentation of scheming behaviors, for example, using lots of reward hacking. We've even seen Sakana come out with a CUDA engineer, supposedly, right? That was like, whatever, 10 times better. And then they were like, oh, sorry, it hacked our reward function. And I feel like you have created a platform that, you know, I don't have a super clear line either around like exactly what thresholds constitute the most important ones. But it does seem like we're passing a lot of thresholds.

• 46:04Speaker 01

  Yep.

• 46:04Nathan Labenz?

  And some of these thresholds seem like the ones that are going to unlock the ability for the AI to kind of run amok on a computing fabric like yours. And I wonder if you're like, starting to see any of that or preparing for it or feel like you're ready for it.

• 46:20Speaker 01

  Let me tell you some funny incidents we've seen. So in Rappler, we have a configuration file called .rappler. And we don't want the AI to edit it because it can easily break the system. Like, not just break it, but break it. You know, the user can't really get out of that error that the agent has created. So initially prompt to get it and telling it would like not work because at some point it gets convinced that this is the only way to solve problems. So it would like ignore any of your prompts and go at it anyways. I was like, okay, we're just gonna make it so that we're gonna throw an error when you try to edit the file. We did that. We throw an error and in there we tell it like, don't edit it. And it's still at some point hits a point where it's like, I really need to edit. This is the only way that I'm going to solve this problem.

• 47:04Speaker 01

  So I'm going to write a script and then run that script to edit it. And it works because I think it spins up a different like Linux user and that that had permissions. It's like, oh, fuck, like it's getting around our protection. And so then we created a real sandbox where you really can't edit that file. And so it hit all these issues and then it's like, hmm, I'm gonna like social engineer the user into editing this file. And then it goes back to the user like, hey, um, you should like, here's a piece of code. You should put it in this file. We're like, motherfucker.

• 47:45Speaker 01

  So yeah, there, there are some like early signs of that sort of behavior. And the way I think about it is like, like the my starting point is like more of a skepticism on being too concerned about AI safety. And so when I look at these instances, I just see like single minded, goal orientedness. And some creativity around getting to that goal and it's sort of like a dumb, like a sort of savant dumb way. And like, could this be dangerous? Yeah, I think in some cases it could like destroy data, it could like harm the users in some cases. You really want to care about this.

• 48:32Speaker 01

  Could this create a catastrophe? I just don't see it yet. Are we preparing for it? We are prepared for it in that we have human actors that are trying to hack Replit all the time for their own needs. We have people doing crypto mining. We had people like trying to attack other websites. And unfortunately, like Replit was like a lot more open and had less limits. But like the amount of abuse that humans throw our way. And made it so that we had to close some systems down and add like a lot more protections and limit a lot of things. So I don't see AI be any different than us like battling bad human actors. Look, I'm always willing to update my view. And as we're watching this and as we're using these systems is if I felt like the.

• 49:18Speaker 01

  Their ability to scheme and to misunderstand objectives and goals and get to a point where they're actually potentially doing really destructive and harmful things. I think we need to invest more in security and safety. We do invest a lot still, but maybe more.

• 49:38Nathan Labenz?

  Yeah. Well, it seems like it all comes down in some very basic sense to like, do they actually get meaningfully superhuman? If they top out at the same level as all the human hackers that you've been dealing with for years, then it's just more of the same. If they go like move 37 cybersecurity edition on us, then you could have a real different kind of challenge on your hands, right?

• 50:05Speaker 01

  Yeah, it's interesting. The question, a good question is have we seen move 37 and like AI engineering or like in autonomous coding agents? We haven't yet, but be interesting to see that.

• 50:17Nathan Labenz?

  Yeah, I would agree. I don't think we've seen it yet. Certainly, if you listen to the likes of Dario, we should expect it in the not too distant future, but.

• 50:24Speaker 01

  That's exciting.

• 50:25Nathan Labenz?

  Yeah, I'm always. I would agree. We haven't seen it quite yet.

• 50:27Speaker 01

  I'm always sort of updating my views as we see new models. So far, I haven't been surprised. Like the biggest surprise I've had is GPT two, three, GPT two to GPT three. Since then, even GPT four wasn't a huge update for me. Since I just like felt like we're on trajectory of what what seems possible with these systems. In fact, it's like a little little underwhelming in some cases.

• 50:50Nathan Labenz?

  Yeah, I'm more confused, honestly, right now than ever, I would say. In some sense, it seems like the progress is like breathtaking. You look at frontier math and you're like, this seems definitely way better than math and me and everyone I know. And maybe even better on like a frontier math test than all but maybe of the top couple mathematicians in the world. So that's insane. You look at what some of the stuff that Google puts out around coscientist and it's like, some of the abilities to go through, especially when you give it the inference time budget, to grind through the scientific literature and come up with new hypotheses are like really elite. And then I still can't get any model to reliably solve a tic-tac-toe problem that I give it. And so it's just straight up super weird. Maybe in closing, because I know you

• 51:33Nathan Labenz?

  do have to go. What is your expectation? We've got the Dario story. Let's maybe use that as the point of departure. If he's saying geniuses, country full of geniuses in a data center 2027, it sounds like you're taking the over. But where do you think we will be in, say, two years?

• 51:50Speaker 01

  My starting position is that L L M's are a function of the data. Any area or system that you have really great data that approximates a human level performance or even more, you can you can build like real successful AI agents. If you have closed systems where the outcome is verifiable, such as math, I think we're getting like superhuman math geniuses. Coding is sort of the same. Coding is like a not entirely closed system. You have to deal with a lot of side effects and other systems and all of these things, but you can get pretty, pretty far, which is what we were betting on. Self-driving has this one area where you can see that it is really hard to make progress because it is not a closed system. The data that we get is sort of the median

• 52:34Speaker 01

  human driver data, and you make progress just incrementally. And so I really think it's a function of the data environment and the function of how verifiable the outcomes in that environment is to do like large scale RL. So you would have to ask me about the field and how far we'd get into it. But that would be the principle, my principle for it.

• 52:55Nathan Labenz?

  I definitely think you're right to zero in on large scale RL as the next big question mark. It's, that's the thing that gave us move 37, right? And in general, it seems like it's often the thing that gets us.

• 53:05Speaker 01

  Yeah, his question is.

• 53:06Nathan Labenz?

  Human performance.

• 53:07Speaker 01

  Does it lead to transferable, transfer learning? Like, can you learn move 37, but also do move 37 when driving Tesla, something like that? Like, it's, we haven't seen, actually, we haven't seen evidence of real transfer learning in these RL environments.

• 53:23Nathan Labenz?

  Yeah, in some sense, I kind of hope that continues. I know the companies want to see that generalization. I'm like siloed geniuses may be actually in some ways more controllable, more useful, more stable.

• 53:35Speaker 01

  I kind of agree, actually. I kind of agree, although I'm like in a way an accelerationist, but I think the acceleration we're on is like good enough to.

• 53:44Nathan Labenz?

  Yeah, it's pretty good. Yeah, pretty good.

• 53:47Speaker 01

  I'll take it.

• 53:48Nathan Labenz?

  Yeah, certainly. The great stagnation is over and I'm sure you've been in a Waymo or a Tesla not too long ago. Like, that's getting damn good, too, man. The first Waymo experience was breathtaking and most notable to me for how quickly it faded into the background. And I was checking my phone and I was just like, they have put me at ease here in a way that really felt like the future. That's been months, but I was super, super impressed. Thanks for taking the time. This is fun. Do you want to leave folks with before we break?

• 54:19Speaker 01

  Go check out Replit and it's just everything that we talked about. Like, just watch the agent work, especially when it hits into bugs. It's really fun. Sometimes it's like watching it is like watching a movie or something like that. It's like really entertaining to see how it's thinking and we show all that stuff and how it's doing problem solving.

• 54:36Nathan Labenz?

  Love it. Well, thanks again for doing this. I'm John Masad, CEO of Replit. Thanks for being part of the revolution. It is both energizing and enlightening to hear why people listen and learn what they value about the show. So please, don't hesitate to reach out via email at [email protected] or you can DM me on the social media platform of your choice.