Behind the product: Replit | Amjad Masad (co-founder and CEO)

28:13So it created an admin account. We can ask it with these in the password and kind of go into it and manage it. But basically that's it. It's the app's complete in terms of what we ask for. We can send it out, I can give you your URL. Let's actually just deploy it really quickly. I'll show people how you can deploy it. Maybe in the show notes we'll link to the app. You could check it out. Sounds good. Okay, cool. That's amazing. So this is deploying it onto some like cloud provider. I don't know what you use, but we use Google Cloud. Okay, so it's, we abstract, we abstract all that away from you, but we use Google Cloud behind the scenes.

28:52Imagine a place where you can find all your potential customers and get your message in front of them in a cost -efficient way. If your B2B business, that place exists and it's called LinkedIn. LinkedIn ads allows you to build the right relationships, drive results, and reach your customers in a respectful environment. Two of my portfolio companies, Webflow and Census, are LinkedIn success stories. Census had a 10x increase in pipeline with a LinkedIn startup team for Webflow after ramping up on LinkedIn in Q4. They had the highest marketing source revenue quarter to date. With LinkedIn ads, you'll have direct access to and can build relationships with decision makers, including 950 million members, 180 million senior execs, and over 10 million sea level executives.

29:38You'll be able to drive results with targeting and measurement tools built specifically for B2B. In tech, LinkedIn generated a 2 -5x higher return on app spend than any other social media platforms. Audience is on LinkedIn, have two times the buying power of the average Web audience, and you'll work with a partner who respects the B2B world you operate in. Make B2B marketing everything it can be and get $100 credit on your next campaign. Just go to LinkedIn .com slash podlennie to claim your credit. That's LinkedIn .com slash podlennie. Terms and conditions apply. Let's go down this thread actually while this is happening.

30:16Just like what what allows for this to be possible technology wise, like what is the kind of a stack, whatever you can share that enables this to exist. Yeah, for sure. First of all, it's the all the abstractions that we built. So the way Repplet works is the very bottom layer, it's our runtime. So this is the operating system, this is the package manager, this is the language run times. We built a system that is able to install packages in any language, including native packages. So the AI anytime it needs a package, I can go here and show one of those. By the way, the AI can take screenshots as well, so that it checks the work.

31:06So here you can see it's taking screenshots to make sure that the home page is rendering. Here you can see, you know, I wanted to drag and drop library, and so it installed and so that. And so it has access to all the packages across all languages, including Linux and all of that. And then the layer on top of that is the editor and the infrastructure that runs the editor, including what I described as the multiplier editor. And then we expose all of that infrastructure to the AI. And there's like almost like a new discipline called AI computer interfaces. So sort of and turns out like LLMs need interfaces that are actually quite different than humans.

31:56They're trying to make them use human interfaces like anthropics computer use, but those are really expensive and you need to kind of process all this images and video. So instead we, you know, for the shell, for example, we give it a, you know, sort of a text representation of what the shell is doing at a certain increments. For package installation, we give it a certain tool for editing. We give it like an editor tool that like when it's writing the code, it's getting feedback on whether there are errors or not similar to what a human sees, but it's actually like old text, just to make it easier.

32:35So that's AI computer interface. And obviously all of that's sitting on foundation, foundation models. So the improvement in foundation models has allowed us to build this. The most important model that we use is the Sonnet model from Claude, from Anthropic. And it is the best model coding. So that's the model we use for coding. But we use models from OpenAI as well, because it's a multi agent system. And so we have models that are, you know, critiquing, we have manager editor model and we have like critique model. And different models will have different powers. We also train some of our models like the embedding model for search is something we trend internally.

33:30So, you know, I actually wrote about it back in like 22, I said, it's going to be a society of models like products, we've made of a lot of different models. And, you know, it's a quite a heavy engineering project. To say the least, we were talking offline and you said you've been working on this since 2009, when you first built the first idea of replete, is that right? Yes, yes. Oh my god. Here's the deployed app. I can send it to you and you can use it and you can see my requests even on the log that page. So I can register, I'm forwarded, log in as admin and move things around. We can see what's in progress, what's completed.

34:14Like this looks like a product I could see designer spending like days to, you know, designing, passing it to engineering, PMs, you know, having feedback, engineers taking a few days to build it. Yes. And here's just a prompt. Here's what I want. That's right. That's right. And we can already do it very easily. We can also iterate on the UI. We can, we can say, you know, I don't like this for that and it'll do a good job. So we can go here. We can start a new session on like a new session to create an entirely new feature here. And it'll, it'll just do the right thing. And it builds from that code base.

34:49It understands. Here's what you've built. I want to add this thing. Yeah. Okay. And that becomes, that becomes your, sort of, your history, right? Like this was the V1. And now I'm working on this new feature. And, you know, and, you know, it's almost like what engineers do and get commit messages. By the way, it generates get commit messages for every, for everything that it does. So you can roll back as well. And so we're trying to like make it so that yes, it's, it's for everyone. But we're trying not to abstract too much away. We want to build tools, right? For you to learn to use. And so we want power users to be able to understand the full power of, of, of, of Rapplet.

35:33And it's really deep product. I think you can spend, you can spend a couple of years to, to kind of master it. I want to talk about implications, but I want to come back to something you mentioned that I, is incredible that people may have missed. You basically built a computer specifically designed for the AI agent to use that is a different version of a computer specifically optimized for how AI wants to use a computer. Yeah. Yeah. So, you know, there's an entire discipline, uh, called like, HCI, right? Like a, like, computer interaction. Yeah. So, so now there are papers about AI computer interfaces and interactions.

36:16Um, and, and, and so, you know, large length of models are trained on large -stacks corpus from the internet. But there's still like, kind of alien creatures. So they're not like humans. So they have different behaviors. And like, it's unclear, like, what's the best way to give it an editor? So there's so many experimentation about like, what's the best way to give it a view on what's editing, how many files can you show it, like, you know, uh, before it starts to hallucinate and, and right now it's like more of an art than science, but it's becoming more, more like a science. That's insane. So it's a simple way to think about it.

36:59There's this foundational model. Here's what I want you to build. And here's a computer to use to build it. Yes. Oh my god. And here's a computer with a set of tools. Here's a tool to install a package. Here's a tool to, to edit the code. Here's a tool to run a SQL query. And also services. Here's a bunch of services you can graph from. Here's a database service. He's an object store service. He's an auth service. So you can think about it as a bunch of external services that computer with a bunch of tools and, and, and they're all interfacing with the, with a foundation model. It's funny listening to this how it starts to feel like, uh, the fact that we might be living in a simulation is not as far -fetched as it may feel like this is feels like the beginnings of what a simulation computer would be.

37:50Yes. Yes. I, you know, it's, uh, it's, pretty like, you know, you can, you can, uh, go really sci -fi on this and, and it's like, where, where has it had it, right? Like, um, you know, if we, if we give it, um, enough tools, like, let's say, you can, I can, I can drop it in, and, and slack. And instead of interfacing with it in this fashion, I want to interface with it in a totally autonomous way. So we actually have this feature coming up where instead of me testing it, we give it another agent. So here, you know, instead of me interfacing with it and, and saying, you know, the, uh, this is running or not running, we can give it another agent that is actually testing the application.

38:35Um, and so, and, and then let's say, I interfaced with an entire thrift lock and I'll say something like, um, get me, uh, uh, you know, give me Taylor Swift tickets that the moment they land. And so it'll build, uh, an app, uh, that continuously monitors the web for, for when Taylor Swift, uh, tickets land. And there's like an agent that's using the app, uh, to be able to get that. And then one, and, and you can imagine it has, has some kind of wallet or credit card. And then the moment it lands, it kind of gets it. I mean, what, what I'm trying to say is that software, like agents being able to do software is how AI gets more general because software runs our lives, runs the internet, runs our businesses.

39:28And so the more competent AI becomes at software, the more general they are in terms of what they can do. Okay. This can go in so many directions. I'm going to bring us back to the implications for people building products. They product managers, founders. How does this change that function that skill set? Like what skills do you see will matter more, matter less, which functions are maybe in some danger and they should start thinking about a different career path? What, what an interesting persona that we're seeing is the CEO, the CEO of, uh, startup, the CEO of, you know, uh, you know, uh, Andrew Wilkinson from, from tiny is, as a big user.

40:12Um, and so these people are, are typically, you know, creatives, right? They built a company, they hired people. A lot of them, like, can't code. A lot of them are, are designers or product managers or, or, or something else. And they, you can imagine a bottleneck. You can imagine a bunch of ideas in their head. Uh, and the ideas have to translate through them talking and then someone else listening to them and like assuming that someone else actually understands what they say and then that's someone else going and trying to build what they want to, what they want to build. And also assuming that person has, has time, right?

40:51Because a lot of times your engineers are kind of stuck building the current thing. They're not thinking about the future thing. And so, uh, what gets me excited is a lot of these CEOs are building the, the future concept, the next company, the next product they're going to build, the next, you know, say company they're going to build. And so, uh, it all locks, uh, the creativity. And again, sort of all blocks from that. And look, it's, you know, it's a V1 of the product, but it can push things forward. You can touch it, you can feel it, you can say, okay, this is, this really has lags and, and we should, we should work on it.

41:25You give it to your engineers and they can, they can improve on it from there. So, that's, uh, that's one persona, uh, but I'm really excited about it, the CEO slash, slash founder, um, in, uh, in sort of, uh, uh, companies, um, what, one of the things that, um, that I think is sort of hard about tech, tech companies is, uh, sort of these silos between, um, designers, product managers and, and, and engineers. And, you know, everyone feels that pain of kind of, we have low bandwidth communication, which is, which is language, which it then text on, on Slack and, and Zoom calls. And it leads to a lot of frustration because it's really easy to misinterpret, uh, people.

42:20And again, leads to sort of siloing where people working on, on, something and then you pass it on to the, to the next team and it's not really what they, what they expect that happens a lot between designers and engineers. Um, but, but like the common language that, uh, that, uh, that everyone shares is code, right? Like, ultimately, uh, and software tech companies, uh, everything that we're talking about need to eventually flush out in terms of code. And so, so what if the language becomes actually working prototypes and working applications? Um, you know, for example, we have the Figma extension that translate, uh, uh, you know, Figma mocks into, into React that runs on, on Replet.

43:08So instead of, instead of, um, you know, giving, giving the engineers, you know, just, just mocks or screenshots, whatever, you just say, oh, here's, here's a bunch of React code, you know, just make sure it runs on our infrastructure, but like don't mess with it, don't move the pixels around, right? Um, and, and so I think it, it just like opens, uh, opens up, uh, you know, silos of the companies make, make communication around, around product a lot more concrete because I can, I can give you a working, uh, prototype. And that'll change how, how people work. Like if you, if you can't imagine that everyone can, can make software, it's really kind of a radical, reimagining of not just what that companies are, but really what, what most companies are, because, because, you know, everyone can be more, more general.

44:04So say your, um, PM listening to this, an engineer, designer, what skills do you think, if you were one of these folks, if you were in building Replet right now, what kind of skills would you suggest folks focus on more and, which you think are just like, okay, this is going to be less valuable in the future. Don't worry about these sorts of things. And you can either pick one of those three functions or all three. I think a, a, a very important scale that, that's like perhaps harder to develop, but it's worth working on is being generative, being more generative, um, being able to generate new ideas, uh, quickly, because, you know, you can think about it as, as like a factory line, right?

44:49Like, so, so you have ideas, uh, you have the, the production of, of these, of these ideas or like the initial kind of production of, of these ideas. And then you have your other people that want to consume these ideas or work with you on these, on these ideas, and so typically your bottleneck to buy, by the middle kind of part where your ideas are kind of like, they're a lot of them and they're not fitting in because like, they need to be made and, didn't need to be made quickly. And so now you open up that bottleneck. So now, like actually making things, uh, is, is a lot easier. Actually, you become limited by how fast you can generate ideas.

45:31Um, and so, uh, and I find that true of, of myself as well. Like, you know, I, I consider myself, uh, quite generative, but, but now I have this tool and I can like build, build a lot more and explore a lot more. And I'm, I'm finding that, uh, well, actually I'm running out of ideas sometimes. And so, and so, so, uh, you know, training that, uh, that muscle, I think is, uh, is a good thing. Um, I think like learning a little bit of coding and like not the traditional way of learning coding. Like, when you go, it's like, if you go to like a coding bootcamp, they're going to start with like, what does get?

46:21Actually, my co -founder, high, uh, designer, well, when we're first building your raffle it together, uh, she went to, to, uh, web, web assembly, uh, to do like a coding course. Um, and the first day they were like, spend this whole time on Git. And she's like, what is that? Like, well, what does it do? Like, I'm like, I still don't know what to get exactly that. But, uh, it's, it's like, um, you're, you're inverting the process like you're giving the tool before the actual problem. And so I think all of that stuff you don't have to worry about. So things that you don't have to worry about, I think a lot of the, you know, as a PM, as a designer, as someone who's not like in your code editor every day, don't worry about all the tooling.

47:09And if you learn, uh, a little bit of coding, just by, you know, talking to an AI, doing a little bit of debugging, building something with the wrap -led, you know, running into a problem and trying to fix it, uh, just using AI, you learn a bit of coding. And, you know, I have this, um, I have this, that's been called, uh, not by me, dubbed as, I'm Jen's law, which is, um, their turn investment for learning code is doubling every six months. Um, and really, just learning a little bit of that skill, learning a bit of skill about how to, you know, prompt, uh, AI, how to read code and be able to debug it.

47:51Every, every six months, that's netting you a more and more power, uh, because you're gonna be able to create a lot more, you're gonna be able, it's gonna be easier to create, and you're gonna be able to create a lot, you know, a lot more complete, uh, things. Um, so, so that's, that's another, uh, skill that I think, uh, could be, could be necessary. This is super interesting. Okay. So this last point you made, I'm John's law, uh, it's interesting because when people, like, as someone's listening to this, like, it's see them being like engineers are in trouble. Why do you need engineers at this point?

48:28They're, these agents are building the code. Your point is specific engineering skills are going to be incredibly valuable and more and more about how often are they doubling when you say every year you said, no, every six months, every six months, these specific engineering skills are becoming more valuable. And the idea is this, you don't need to like, no, everything, you don't need to know the foundation, like to build the app as much. It's more to unblock the agent and understand the mental model of how this stuff is built so that you can move forward fast. That's right. That's right. Understanding the basic components of it.

49:02Yeah. So it's like, we need new engineering schools to teach you these very specific skills versus spending years on like algorithm algorithms and, and I think, I think no one has done that yet. Right. And I hate this is, this is like a, you know, uh, big business probably ready, like to get built. It's like, uh, AI native coding. It's totally different than, than like traditional coding. Yeah. That's why you know on Hacker News, there's so much skepticism about like AI native coding tools because they're like, yeah, it's a, it's a glorified auto complete. And I understand like if you're, you know, writing, operating system kernels, you know, it's not really doing that much for you.

49:52But if you building product, it's building it for you at this, at this point, right. And so, you know, if, if you're starting a school to teach AI native coding, you would skip so much of computer science and that, and the basic tools. And, uh, you would, you would teach the basic idea of how to structure an app. And then you would teach prompting, uh, and then you would teach, I think a little bit debugging. I think debugging is quite a, quite a good skill right now to learn. And interestingly, if you want to be a good at debugging, you like, there's a lot you need to understand, which is basically what you're saying is like, that's the subset of things to understand as things that break.

50:31And to do that, you have to understand how it works. What are servers, what are APIs, all these things. Okay. How far? So we've been talking about how this is very good right now, building a prototype, building a V1 MVP, people can use it, you can deploy, you deploy the zap, people can start using it. And there's like a scale it can reach. Do you see a future where you can build like a sales four size business fully, replicate or other tools that can scale to hundreds of billions of dollars of value, or is there just going to always be some limit of like, you need like actual engineers and designers sitting on this thing building it, you'd think, you'd awesome.

51:06If like my law is like, you know, directionally correct, even even if the months are not, uh, I don't know exactly right, the duration is correct. You're going to see compounding effect of the power. Like it's actually quite hard to convince yourself. But if you really convince yourself that we are on a massive scale of improvement and AI, then then the answer is yes, and it's like absurd to my engineering mind that I'm saying it is. But you know, Ray Kurzweil, this like, you know, futurist, you know, talks about how exponentials really hard for humans to grasp. And so actually when we started building the agent, you know, I told the team, it's easy and we fall in the strap before it's easy to build and optimize for today.

51:58You know, in in 22, we built like, you know, copied like thing and autocomplete, we trained our own models, we optimized the hell out of them. But at some point like that modality was kind of, you know, not the right modality, which is like the autocomplete modality. And the right modality is actually this, I think, for now, is being able to chat inside the programming environment and for the agent to create things for you. But in order for us to make that bet, you know, a year ago, the models were actually not there. Like the models could not do this. But we were like, okay, we're going to build for the models that are landing in six months.

52:35And it would truly like six months later, the models started to land that are capable of this of the reasoning that we need and whatever. And so that was like, you know, Son if you weren't, which is, oh, wow, like we switch right and the reasoning improved so much. And six months later, you have Son of you too. And so it's really almost like a six months cadence. And so if we're really on the on the stretch actuary, then, then, you know, I would say, you know, next year, you're able to scale maybe, maybe you get your thousands of users paying you the AI can do maintenance. You know, we're already showed that the AI doing like SQL queries and doing migrations.

53:14So they are able to do maintenance debugging things like that. I think where it gets really tough is that, you know, when you're hitting scale and you want to architect a system that is resilient. And so that means, you know, you would start, you know, shorting databases and we start like using different queue systems and components and things like that. And I think, you have the AI needs to have access to the entire suite of tools to be able to do this. And, and I think that that's going to be the next bottleneck. And I think the AI needs to get be a lot more reliable at doing that. But I could imagine like whatever, five years from now, someone running, you know, a billion dollar company with zero employees where it's like the support is handled by AI, the development is handled by AI.

54:14And, and, and, and you're just, you're just, building and, and creating this thing that is, you know, that people are finding valuable and are paying you for it. That being said, it's worth like thinking about the economics of it. Like, if the, you know, if the cost of software goes down a lot, like, then what is, you know, what is the price that you can charge on on software? So can you actually build the next Salesforce if anyone can generate Salesforce? And then, and then the question is like, what is the, you know, and this is why I emphasize being generative because I think then the, the thing that will make you better is like, by being able to iterate and improve the thing really quickly and generate new ideas.

55:00And stay ahead of all the other people building these tools so quickly. Yeah. Oh my god. An interesting other kind of mental model I'm seeing as you talk about this sort of thing, is not to offend religious folks, but there's this concept of God of the gaps. I imagine you've heard that. Yes. Where it's like God explains all the things that we don't yet understand and over time, that kind of space shrinks and God's like, all the things we don't get yet, those gaps, that was God. That's, that's what, that's, that's, that proves there's need, there to be a God. And it feels like right now humans are like the gaps in these tools, where these agents you talk about that you can hire within Replyt are like fixing these little gaps and over time AI will fix these things themselves.

55:44That's right. And these gaps will shrink. I mean, unless, unless we had some fundamental limit and the current regime of, of AI, which, you know, I'm not, I'm not an expert about like how far Transformers could could scale, but I, I feel like it is, you know, we found the thing that could, that could scale pretty far, but maybe there are limitations and data or, or, or other things like that that we could be, we could be surprised by, but if there isn't, then we are on a massive trajectory of removing these gaps quickly. Yeah, very true. We have no idea. We keep thinking it's just going to be, maybe it'll stop at some point.

56:29I could keep going and going, but I think we should also let people go play with these things and process all the things we've been talking about. Is there anything else they think might be helpful for folks to think about or learn or study? I'll give advice to, to, sort of, founders or leaders at companies. The way we work is, is going to change rapidly and it's important to, sort of, resell the end to that, that change. One thing that I think it's really difficult now is having roadmaps, especially if you're doing anything in AI, but really anything that AI could affect. You want to be able to react to it really quickly.

57:13And so, you know, when the, and Throbic dropped computer use, sort of capability, we slaughtered our road map because we don't really have an explicit road map. We, like, immediately jumped on it and started building things and we've launched some things around it. We're going to be doing more with it. But, like, there's going to be capabilities that are going to drop and you want to really, in some cases, if it really affects your business, you want to be able to jump on it really, really quickly. So, being agile, not being, you know, sort of stuck with, with roadmaps, being able to kind of just, just say, oh, we're just going to switch priorities right away.

57:57It's going to be super important. Not being, you know, like I said, with silos, I'd replicate there's so many people that are on the scale of, like, you know, designer to engineer, designer of product manager. Actually, I mentioned, I'm on earlier, he started as a designer at Replet and I was a product manager. We have people who start as designers become engineers. And we have people in the middle and we're comfortable with that, like design engineers. And that fit a different parts of the scale. And the design engineers go to the design, correct meetings and some designers go to the engineering meetings and you just got to be fluid, right?

58:39Because, you know, again, when designers can code and engineers can design, I mean, it's really becomes, you can't have a lot of structure around that. So, you want to build a culture and you want to build an environment or a milieu that is, like, really, really flexible, which is uncomfortable for a lot of people. Man, the future is wild. Everyone's a hybrid person now. Let me just actually double down on what you just said, which I think is really interesting. It's almost like if you're an engineer where your skill set will become most valuable is unblocking these AI tools and knowing debugging and figuring out an unblock, allow it to go further and further and further.

59:26Within PM and design land, based on what you're describing, where the skills will become more valuable is generating ideas, almost like finding opportunities, discovery, finding problems need to be solved. And then articulating that is clearly as possible to the AI tooling. That's right. Yeah, this is a very crisp sort of advice that people can follow today. Oh, man. What a world. Okay. I'm, John, this is incredible. My mind is racing. I've got to go build some apps immediately. Yeah, let's do it back. I will do that. So just to leave listeners with a couple things. One is just what should they know?

1:00:05Where do they find you? Where do they how do they try? Reply it? Anything else other than just going to repplet .com? Yeah, just go to repplet .com. It's an open beta right now. We're kind of quickly improving and going to exit beta, I think, in a few weeks. But if you're comfortable testing something that's not perfect, go to subscribe to our core plan. You should be able to access the agent and start using it. And we are, I think the place where we're most active is Twitter. So Twitter are like X, the handle, replet, REP, or my handle, a massage. Oh, yeah. One other thing I wanted to make sure we had a chance to touch on is you're working on something new, something that's coming in the very near future.

1:00:52Maybe the day this episode drops. Talk about that. All right. So depending on when the episode is coming out, this could be the first time people hear about it. But we have this product called agent. It is sort of high agency does everything from setting up the project and all of that, right? And so now we are working on assistant. So assistant is like, let's say the cousin of agent, it is a little less powerful, but a lot more controllable. So you can like focus on features or areas of the code that you want to change. And you still don't have to know how to code, but it is a lot more manageable and it is a lot faster.

1:01:38So you saw how it took some time to kind of create the project and code some of the things. Assistant is in the order of milliseconds and seconds to be able to respond to you. And so again, as I talk about the idea of tools, we want people to have as much power and autonomy as possible. And so there are certain instances where agent is the best. It's going to do the debugging for you. It's going to create the database for you. But if you want more control, assistant is going to give you that. Just so folks totally understand what this is going to do for them. What's like the mental model for what this is like if it's like a person who's helping you out.

1:02:17Agent is like having a developer, you give them that PRD, right? And they're going to go and build the thing. Assistant is like you're sitting next to them. So you, you know, they built the thing and now you walk over to their desk and you say, let me move this button three, three pixels to the left. Let me, you know, change this thing. So like small increments of changes that we want happen really quickly and you want it reliably, that will give you that. So it's just like much faster iteration on UI and things like that. Incredible. The future is wild. Final question I was asked everybody, how can listeners be useful to you?

1:03:04Come work at Repplet. We have a PM role. I think up if you're product manager, we're hiding engineers and product managers. So come work at Repplet or for some of Repplet. Especially if you're like our tools and you want them to get better, the best way to do that is to get us great people. We can hire. Well, you're about to get a full out of product managers applying. Amazing. I love that. Good luck. I'm John. Thank you so much for being here. This was incredible. Thank you. Thank you for your podcast in the community that you built and use letter and everything. It's been awesome to watch. Thanks, man.

1:03:40Appreciate that. Bye, everyone. Thank you so much for listening. If you found this valuable, you can subscribe to the show on Apple podcasts, Spotify or your favorite podcast app. Also, please consider giving us a rating or leaving a review as that really helps other listeners find the podcast. You can find all past episodes or learn more about the show at Lenny's podcast .com. See you in the next episode.