The Man Behind The $100M Al Agent: The FASTEST WAY To Build Your First $1M App In The Era Of AI!

Amol Jain 0:00 We have people who've never written a line of code in their life build real businesses, make money off of that. I guarantee you it is very much possible. Just yesterday, I opened the Replit app and I go, "Hey, I need this app that lets me track 100 foods that I'm trying to feed my baby." 45 minutes later, when I'm done feeding her and cleaning up, the app is ready.

Calum Johnson 0:20 You can now deploy apps in minutes, not hours, not days. Can you pull up Replit and, like, let's get started.

Amol Jain 0:28 Let's do it.

Calum Johnson 0:33 Okay, so today's guest is Amol Jain, and let me start by giving some context. A few months ago, we got a vibe coding expert by the name of Matt Palmer onto the show. He's the development lead at Replit, and he broke down how you can build your first million-dollar app using only AI agents, no coding required. You guys loved that conversation. Then something interesting happened. A couple of months later, Replit actually released their new agent, Agent 3, which is able to build software almost entirely by itself. It's autonomous. And so when I heard about it, I reached back out to Replit, and we spoke with Amol, who's their head of product engineering and led the development of Agent 3, so that he could give you all of the tips and the information that you need to build your first successful app in minutes right from your phone. And so my intention with this episode is very simple. I believe that there are people watching at home, maybe it's even you, and you have all of these product ideas and business ideas, and the only reason you haven't brought them to life is because you tell yourself that you're not a software engineer, you're not a coder, you don't have the technical background to build things. Well, not today. After watching this episode, you'll be able to build software from your phone in minutes. And so my hope is that you listen to this conversation, you action some of the things that we talk about, and six to 12 months from now, you're DMing me to tell me that you've built your first successful app, and it all started with this video. That's my hope. But anyway, let's just get into the conversation. You're listening to "The Calum Johnson Show." What is gonna be the value to them of listening to the conversation that you and I are about to have?

Amol Jain 2:20 Yeah, um, in a lot of ways with, with, with Replit, our, you know, our, our mission is enabling the next billion developers. When we say developers, it's, it's folks being able to bring their ideas to life. Um, my hope is that in this conversation, we're able to, we're able to show them how easy it is with a platform like Replit to, in natural langua- language, describe what you're looking for, and Replit allows you to bring that vision to life, bring it to something that, that you can then, uh, have, you know, working and, and be able to give to your users, to your friends, what have you.

Calum Johnson 3:05 Mm-hmm. You know, like, what a big thing for me was, even in the preparation, and I, I know even from our conversations it was similar for you, there's this magic feeling of, like, building that first piece of software and that process of something going from, like, an idea to it's like a tangible thing, like, like it's right there. And so the interesting thing is, like I said in the beginning, you're the head of product engineering at Replit, but then you've also built your own startup in the past. And so I wanted to, I wanted to read this to you. You said this on your LinkedIn.

Calum Johnson 3:43 You said, "Coming off, coming off my own startup, I was planning to take a breather, but then Replit popped up. I'd always admired their mission of enabling a billion people to build software, but I was skeptical until I tried their new AI agent. It was pure magic. Prompt to a live app in five minutes." You then go on to say, "I built an app to find good House Hunter episodes for my wife." I built an app to find good House Hunter episodes for my wife. Can you just take us back for a second to, like, that first moment, that moment of magic, of building that app that you did with your wife? Like, take us into that.

Amol Jain 4:31 Yeah. Um, so I had spent six, seven years, um, building my own company, um, as, as you know, most of the listeners know, it's, uh, it's, it's hard. It, it takes a lot, lot of toll and, uh, um, I had just made the hard decision to, to leave that company, especially with my, with my second child. I had just had my second child and, and figured it was as good of a time to ever take a break as there ever could be. Um, and, uh, the idea was I'll just spend time with my family and, and, uh, you know, enjoy this special, special time. Um, uh, so both wife and I were on maternity and paternity leaves and were, uh, just spending time with the baby. One day, uh, my wife... As you would imagine, there was a lot of TV watching involved, uh-

Calum Johnson 5:24 Yeah

Amol Jain 5:24 ... because the baby sleeps and you watch TV. And my wife loves this show called House Hunters. It's a show about buyers in markets across America looking at multiple options, house options, and then deciding the one they wanna buy. She's obsessed

Amol Jain 5:42 with this show. She's obsessed with real estate. So we're sitting there binge-watching these episodes after episodes. Uh, she really... She wonders out loud of, like, "Hey, I, I, I, you know, aspirationally, I wanna see fancier homes," like, not in an elitist kind of way, but, but more just curiosity. She's like, "It'll be more fun if I could see fancy homes in Southern California or something." And A- at that moment, I was like, "I am an engineer. I can build this, but it'll take me days and days of work. I'm, I'm not, not gonna do that." But I had heard of Replit, and I had heard of, heard of this, this recent buzz around the company of this bold claim, idea to app in, in minutes. And at that point, I decided I'm just gonna pull this up. Um, I type up, "Help me build an app that lets me filter for 'House Hunters' episode by location and budget." And truly, within 10 minutes, I had this working prototype that let us filter, do exactly that, be able to filter down to, to an area we were interested in, set a budget limit, and, and be able to see those episodes. Um, and it was just a mind-blowing, mind-blowing experience 'cause it, it earned me so many brownie points with my wife right away too, 'cause she was... She got exactly what I wanted. We spent a lot of time from there watching a lot more episodes. Um, but it was just such a magical thing, uh, even for me, who's, who's an engineer, who's had this power to build software for the longest time. Uh, being able to do that without thinking about code, purely in natural language, purely just uttering those words out into the ether, and then voila, you get this app fully built. It was, it was, it was just such a magical thing. I think that, that was the moment I, I really thought that I, this is something I wanna work on. This is something I wanna contribute towards. Yeah.

Calum Johnson 7:33 Yeah. You, you know what's so interesting? I was, I was thinking about this as you were speaking, is like whenever... I think traditionally, whenever we think about technology and technology companies, you imagine, like, these huge problems, right? Like you think about, like, what Uber's doing or, um, like Airbnb, like these huge apps that are, like, very sophisticated, and there's teams of engineers. And it's kind of with, with AI and vibe coding and some of these capabilities, it's like you now have these, like, everyday problems that the consumer can basically just go and fix for themselves using apps like or agents like Replit Agent. And so I'm curious for you because you have a very interesting perspective of, like, someone that's actually building the software at Replit, but then obviously, like, you know, you have these, these moments with, like, your wife or people that you know where they're also just building it as a consumer. What do you think is, like, the impact on humanity, on all of us if Replit is actually successful in that mission of enabling, like, a billion people to build software? Like, what does that mean?

Amol Jain 8:44 Yeah. I think software is, brings leverage generally. With software, you can do more, you can automate things. You know, the world now runs on software. Uh, but that, but that power's been locked up with 40, 50 million software developers that we've had out there. What that means is a lot of, lot of ideas, a lot of, lot of problems go, uh, lot, lot of ideas go unexplored, a lot of problems go unsolved because it's not accessible to the rest of the world, to the, to the 7 and a half, 8 billion people. And if we make it possible for anyone to solve their, their problems, um, it, it, it should really help anyone be more productive at their jobs, um, be able to free up time in their daily lives for creative pursuits. Um, it should enable a lot more people to, to build businesses. They should be able to... You know, these, these ideas, they, they can create prosperity for themselves and for, for, uh, their communities. And w- with, with, with products like Replit, we're finally seeing that happen. We're seeing that happen in real time where we have, have, you know, people who've never written a line of code in their life, um, build real businesses, make money off of that. Um, so I, I, I think it's, it's, it's gonna really have a, uh, um, a monumental impact on how we operate as a society because now everyone has power of, of making their lives better and pursuing their passions.

Calum Johnson 10:24 Yeah. No, I love it. You know, w- we're, we're gonna get into it of, like, by the time that someone finishes listening to this conversation, they're able to build their, like, first app using Replit. Before we do, I know even from our prior conversations that, like, that ability to build things and entrepreneurship and building businesses and products, it's actually very, like, personal for you. Like, it's, it's very meaningful having that kind of ability. I'm just thinking for that person that's listening at home who's still, like... They're, like, a bit on the fence even. They're like, "Is this really possible?" kind of thing. Can you just speak on what has been the impact in your life of being able to build software?

Amol Jain 11:14 Absolutely. I think I attribute a lot of my, my success and, and, and, you know, even happiness to, to this ability that I was able, I was fortunate enough to, to pick up. Um, I grew up in India with, with access to education, but in a, in a, in a, you know, lower middle-class family. Um, and there was always this desire to, to do more, to, to, uh, leave a mark on this world. And, um, I was fortunate enough to get into a great programming school and, and get to learn programming. And, and it enabled me to make a move to the United States and, and, you know, to the, to the sort of, um, the heart of where all the software action is. It, it allowed me to build a, a great career, work on really successful products like, you know- ... Facebook or Airbnb or, or Replit now. Um, it's, it's, it's, it's just very, uh, it's been personally gratifying of, of, you know, building software that has an, has a direct impact on people's lives and their livelihoods. And, uh, uh, also, you know, personally, I've, I've, I've been able to accomplish so much and provide for my family and such. I think, uh, it wouldn't have been possible without me learning coding and, and, and, and being able to, to do all these things. Um, and I'm so, like, I... That's what gets me out of the bed every morning is, is the idea of bringing that exact power to everyone, that superpower of coding, 'cause I really, I really, really believe in that. I think, I think it's a, it's, it's superpower that everyone should possess.

Calum Johnson 12:56 Yeah. Yeah, I love that, the superpower of coding. Okay, you know what? Let's, let- let's bring it to people. And you know what's interesting, so, um, you know this, but a, a couple of months ago we filmed an episode with your colleague, uh, Matt Palmer, and I think, like, the thing that was kind of mind-blowing for me, it wasn't just what we were able to create with Replit. At the time it was Replit Agent 2. I know you guys have since come out with Replit Agent 3, which we're gonna talk about. But it wasn't just what we were able to create, it was the speed. And there was something that Matt actually said during that episode. He goes, "You can now deploy apps in minutes, not hours, not days. That changes the types of ideas you can even attempt."

Calum Johnson 13:41 Can you, can you just talk for a second about, like, being inside at Replit and seeing what people are actually building with the software. Are there any, like, success stories, examples that, like, you guys have seen internally of, like, this person was able to build something incredibly quickly and is now actually... They have, like, a full business that's built off the back of, uh, Agent?

Amol Jain 14:10 Yeah. We, we see it very frequently. Um, one of, one of my favorite examples, and, and maybe this is, this is not as widely a- applicable to, to starting a business, but I, I think it just showcases the power of, of what, what non-technical folks who don't know coding can, can achieve with Replit. Um, I talk about this example a lot. I, I personally love it. It, it, it matches to, you know, people I work with on the day-to-day as well. Um, but one of our, one of our, uh, Replit sells to large companies as well. Zillow is, uh, one of the real estate companies, is a client of ours. Um, uh, one of their, one of their, um, one of... Someone who works at, at, at Zillow, Mike Messenger, he's... He doesn't really know how to code. He's not on the engineering team, but he was able to... He discovered Replit. He had this idea in his head that he's always had for Zillow to be able to, um, connect real estate agents to, to, uh, to people who are buying properties. And, uh, he was able to, using Replit, vibe code this product, and it has since then delivered tens of millions of dollars of value for, for Zillow as a company. Um, it was, uh, so amazing that now Zillow, thousands of Zillow fo- Zillow, uh, workers actually use Replit, uh, to, to build software for, for themselves internally, but also for their customers. Um, I think, uh, it, it's a great story because in, in a lot of ways, Mike Messenger works at a company, a tech company, that has engineers a- at their disposal, but, but engineers, engineering time is limited, and it's... He usually doesn't get a, uh... Even with the, his ideas, he, he's not able to bring them to life because of that limitation on, on engineers' time. So the fact that he was able to do it by himself and, and create so much value for his company is, is pretty amazing to see. Um, and we see similar examples play out for, for folks that are just starting out businesses and side hustles, uh, and, and, and earning real money using

Amol Jain 16:26 Replit.

Calum Johnson 16:26 Yeah. You know what? I think that one of the, one of the things that I've seen and, like, speaking with people and even, like, listeners of the show as well, is you kind of have, like, two groups when it actually comes to building things. You have, um, examples of people that will have, like, a ton of ideas and they'll get inspired, and they never take, like, the, the step one to actually start building. And then on the, on the other side of things, you have these stories like what you just explained with Mike Messenger, where, like, he's working a job and now he's... He actually went out and built something that's creating, like, millions of dollars of value. And so before we get into, 'cause we're gonna get into the actual steps of how to build it on Replit, talk to me almost about the psychology, the way that someone should actually think about starting something on Replit so that this is actually the moment that they build their first kind of like successful piece of software. Okay, quick break. I had to make a change when it came to my diet. And so let me tell you guys, I have these days where I'm running around nonstop. I'm talking meetings to recordings to research to then my personal life, and by the time I even think about dinner, I've either ordered something greasy off the delivery apps or I've completely skipped it. And so a few weeks ago, I decided I had to end this unhealthy loop that I was in. And so based off my friend's recommendation, I started getting my meals from Cook Unity, the sponsor of today's episode. Cook Unity is this chef-powered marketplace that delivers fresh, ready-to-enjoy meals straight to your door. Literally, guys, when I tell you, I'm talking about restaurant quality dishes from award-winning chefs. And so to take you behind the scenes, this week I had jerk chicken with plantain. I also had a Singaporean chicken curry. Both of them really tasty. I'm not just saying that. And it's cool because they have over 300 different meal options on their website. And so honestly what my experience has been is that it saves me a lot of time. It takes two to three minutes to heat up the meals, and compared to buying a bunch of groceries and then cooking, and then you never end up using some of the groceries because you got busy, this is actually saving me money and time. And so if you wanna go and try it for yourself, go to the link in the description, try Cook Unity today. And I've actually hooked you guys up as well. So when you try Cook Unity using that link in the description, you'll get 50% off your first order. And so thank you to them for sponsoring this episode, and let's get back to the episode. Okay, quick pause. If you're looking to build something, this segment is for you. I remember when I started my first business. I spent more time figuring out how to start than actually doing the thing, and I was using all of these different website builders and plugins, payment tools. It felt like I needed five platforms just to sell one thing. Then Shopify, the sponsor of today's episode, changed the game. Shopify is a powerful set of tools that helps you start and manage your business, whether you're selling T-shirts, candles, coaching services, whatever the thing is. It gives you everything in one place. One cool feature they've recently added is Sidekick. It's your own personalized AI assistant, and you can actually ask it to help you set up your store, write product descriptions, tweak your homepage, even run marketing campaigns. And the cool thing is, is that it doesn't just follow commands. It reasons through multi-step tasks. It's literally like having an AI business partner, which is so helpful in the early stages of building something. And so if you've got an idea in your head right now and you're ready to make it real, go to shopify.com/calumjohnson to get started today. That's shopify.com/calumjohnson. Thank me later, and let's get back to the episode.

Amol Jain 20:26 Yeah. Uh, I think, I think you were j- y- you were just mentioning this. The i- it, it almost feels too good to be true. And I would, I would encourage folks to sort of start small. Think of, think of any problem that you've recently experienced just in your personal life and, and, and, and hoped that there was some app that solved for it and, and, and try to build that and, and experience the magic yourself. I'll, I'll give you an example. Um, my daughter, seven months old, starting to-- we're starting to introduce food to her. Um, we're a pescatarian family, so a, a lot of, lot of food charts out there don't quite apply to us. They, you know, we don't eat meat, so it's not a, not a... I'm not introducing beef to her. Um, and I just wished there was an app that, that let me track what she's eating, how many allergens has, has she checked off. Um, really something more bespoke that only I need but for my specific needs. And so, uh, just yesterday I opened the Replit app and typed this, as I'm feeding her some mushy carrots, I just pull the app up and I go, "Hey, I need this app that helps me track 100 foods that I'm trying to feed my baby as a way to expose her to all sorts of food." And 45 minutes later, when I'm done feeding her and cleaning up, the app is ready and, and we've been able-- we've-- last night we tracked all of her food on an ongoing basis, can see her progress towards that. I think it was, it was so magical 'cause I could have downloaded 50 apps that existed on the App Store, but none of them quite fit my, my exact use case, and I was able to create that bespoke software. So I think to me it's almost most folks have something or the other that they've always wished existed, a small something. Trying that just lets you see the, see the power of platform and, um, g- kick the tires a little bit, get used to what Replit can do, uh, get, get acquainted with it. And then from there you can really start to dial it up and, and, and with some of, with the Agent 3 changes, the agent is so powerful it can truly create large, viable production grade applications that you can sell or, or, or give to, give to a wider audience, uh, as, as a builder or a creator. Um, it, it, it feels intimidating at first. The idea feels a little far-fetched, but I think just trying the smallest of things can hopefully help people, um, quickly see the, see the value, see the potential and get, get acclimated to, to what's

Amol Jain 23:14 possible.

Calum Johnson 23:14 Yeah. You know, I, I really like that as a step one because I think so often when we-- And I've done this in the past, is when we try to like start companies or businesses, we come in with like all of this energy and like aspiration and ambition of like, "I'm gonna build this successful thing." And then that's kind of, I feel like, when the procrastination and like everything will start to build with that pressure. And the thing I love about what you described, it's like just start solving your own problems. And we've, we've all had that experience of like maybe you're, you're driving in the car or you're like trying to solve something or you're looking for something and you're like, "Isn't there an app for this?" Or someone-- "Hasn't someone already built a product? Like, why doesn't this already exist?" And now it's like the second that you have that pain and that frustration, you can just go build the thing. Um, so I, I love that as a, as a first point of like just start solving your own problems. Um, but you know what? Let's, let's get into it. Can you, uh, can you pull up Replit and like let's just, let's get started.

Amol Jain 24:27 Let's do it. Let's, um, let's dive into this, this, um, this app that I was just describing to, to track food eating for my, for my seven-month-old. This is, this is the app. Uh, I'll start here as I'm, I'm tracking 100 foods. She's tried three. Um, and the allergens and a breakdown of fruits versus vegetables versus proteins, et cetera. Um, and this took me about... Well, I, I spent about 45 minutes on this, if that, uh, while in the moment while I was experiencing this problem. Um, and so what I would love to do is just kind of, kind of show the journey of, of how this happened. I did all of this on mobile, but we can start right here and, and see what that, the-- how, how I went about it.

Calum Johnson 25:24 Okay, nice. Let's do it.

Amol Jain 25:26 Great. So, um, last night, as I was saying, I was feeding her, in the middle, middle of feeding her, and I go to my mobile app. Um, we're viewing this on, on desktop, but there is a Replit mobile app. I'm personally a big user of that because ideas come to you in the shower or as you're going about your life on a, or on a walk, and you just type them right in, let it do your-- let, let it do its thing while you are busy doing other things. Um, so basically my prompt was as simple as my daughter, here's her name, here's the app that I want to create, um, you know, tracking all the foods she's trying, laying out my, my specific scenario of she's pescatarian, we want to cross off all the allergens, um, and also help me recommend the next food she should try. I just type this away. I go back to feeding her. Replit Agent gets to work. It starts out by, by first making a plan. I think this is always helpful because it can very quickly from your, from your really vague attempt of an idea help you refine that idea. What exactly are you trying to build? So first thing Re-Replit Agent does is, is break it down into exactly the fol-following things it is going to build. Um, and at that point you can either decide to iterate on the plan. You can say, "Hey, actually I-- that's not what I was looking for. I would also love this. Maybe skip this." But once you've done that, you've really kicked it into gear. So at this point, for me it was, "This looks great. Start building," and it unleashes the, unleashes the agent.

Calum Johnson 27:09 Just one-

Amol Jain 27:09 As you can see, it's-

Calum Johnson 27:10 One

Calum Johnson 27:11 quick thing, Amol, sorry to interrupt, is when-

Amol Jain 27:14 Mm-hmm

Calum Johnson 27:14 ... so you go in and basically the first thing, so you get into Replit, and you're doing it on your phone, you're doing it on the mobile app, is you're just giving context on like the situation, like everything that's happening, what you're experiencing. When you kind of at the bottom of that first prompt, that first message that you put into Replit, are you asking it to build a plan or are you literally just saying like, "Here's the context of what I want to build," and then you hit enter and it's automatically just giving you how it's gonna go about it?

Amol Jain 27:47 Yeah. So it automatically does that. Um, the idea there is people have ideas, but they might not have thought a lot more about them quite yet. And if we, if the, if the Replit Agent understands exactly what you're looking for or does that clarification, then it's better set up for success, then it can truly meet your needs. And so we, we f- we found a lot of value in this planning phase, we call it the planning phase, where you can throw out a very vague idea, but Replit Agent is smart enough to go back and forth with you to, to come up with a plan and, and till, till both parties feel great and there's a clear path for, for AI to go do its job. Um, so every, every single time you give it a prompt, it will do some thinking, it will do planning, and it will try to first thing come up with a list of task list. It's almost it taking that vague prompt, clarifying, turning it into a, a, a clear task list that you can say, "Sounds good," or make changes to as, as, as you see fit.

Calum Johnson 28:59 I guess, I guess how-- So once it's shared the plan, how often do you find yourself actually going back and forth with Replit over the plan? Or is it more just you're taking a quick glance over it and you're actually like, "Yeah, this sounds pretty good to me," and you just carry on? What, what have you found in your experience?

Amol Jain 29:17 Yeah, it's been a, it's been a, a, um, it's been a learning experience of sorts if you will. At, at first my hunch or, or my ideas used to be very vague. I would just throw them, throw them out there, and in that case there was a quite amount of, quite a good amount of back and forth to, to clarify what exactly I needed. Over time, I've learned to give it more detail, um, and, and give it the specific type of detail it- Does well with. I think in a lot of ways, AI sometimes has personalities and its quirks, and you over time learn how to work well with it. Um, and so over time, I've really dialed and figured out what makes for a good prompt. And I think, I think, you know, laying out who you're building for, what's the core capability that, uh, or what's the job to be done, if you will, um, giving that, g- giving it those two things already helps you achieve much better results. So over time, I've gotten it dialed a lot more where my prompt, we are in sync, the Replit AI and I am in sync, where I give it a prompt, it comes up with a great task list. Most of the time it looks great, and I just go, go ahead, start building.

Calum Johnson 30:35 Yeah. That's actually, that's a... That's really cool the way that you explained it, 'cause it's like you actually become better at prompting the agent by just prompting the agent and, like, going through those reps. Um, no, that's, that's good stuff. Okay,

Calum Johnson 30:55 yeah, let's move on.

Amol Jain 30:55 Yeah. Yeah, absolutely. I think, I think, you know, to that point, we've all... The ChatGPT moment that happened two, three years ago, we all over time have learned how to work better, better with ChatGPT. We've gotten better at prompting. Uh, and I think similar, similar things are playing out across other AI products as well.

Calum Johnson 31:16 Yeah, 100%, 100%.

Amol Jain 31:19 Great. So the planning phase at this point, I feel good about it, so I go ahead and say, "Start building." At this point, what we like to do w- with Replit is, is, is, is free you up to go do, live your life and, and do the things that you were doing. And so at this point, Replit, especially Replit Agent 3 that we recently launched, is highly autonomous, where with that plan that we've agreed on, it's able to execute by itsel- by, uh, on it by itself. It's, it's able to go deep. It's able to really execute by itself and, and keep making progress. So once I did that, as you'll see, it starts to figure out I need to create a baby food tracking app. It starts building. One of the coolest things it does is every time it gets something working, it now knows to test its, uh, test the, test the functionality that it created. I think this is very important because it's AI checking its own work and making sure that it's tracking towards the tasks that we agreed on. So i- i- if it introduced a bug or if something broke by any chance, it is able to then, um, detect it itself, adapt its plan, introduce a new task item of, "I need to fix something that I didn't really go about doing." Um, I think this is, this has been incredibly powerful. We call it testing. But giving Replit agent the power of not just building, but also testing what it's built, really unleashes that power and gi- it makes it more autonomous, where it will get to the results of what, what you want, uh, much better and on its own.

Calum Johnson 33:03 Yeah. You know what? We, we hear, like, words, uh, like autonomous, and I, and I'm taking the perspective of someone, I'm someone that doesn't have, like, a coding background. And so this is even what you guys, um, I took this from your website. I took this from the Replit blog. It said, "Agent 3 runs on its own for up to 200 minutes, handling full tasks autonomously, building, testing, and fixing your app with minimal manual oversight, giving you hours of time back." We hear words like autonomous and it runs on its own. I just wanna make it so, like, real for people. What is... Can you talk about how big that is, the fact that it can go for up to 200 minutes by itself, and what is the impact in terms of what that person watching at home is gonna feel when they're actually using this?

Amol Jain 34:00 Yeah, it's, it truly, we, we, with, with Agent 3, we've truly pushed the frontier of what was possible with, with coding agents. Um, the idea behind agent running for a long time is it frees the human up to pursue other creative things, to pursue other things that are valuable, rather than watching an agent and babysitting an agent in its pursuit of completing the task you gave it. Um, in, in software development, a- and, and this is something engineers have, have felt for the longest time, but now as more and more people get the power of, of building with Replit noticed as well, it's very important to test every single functionality that you create. Um, but that's a very, very monotonous, time-consuming task that isn't really enjoyable 'cause you're going about clicking those same buttons. I think in, in a lot of ways, the power of and the premise of AI is that it can take away those mundane, monotonous tasks so humans are focused on more creative ones. And, and, and I think that's what this autonomy, the fact that it tests itself and corrects itself, unlocks, where you as a human aren't continuously being interrupted, or you have to, you don't have to sit here, babysit this agent, test every single thing it creates. It can do that by itself. Um, and keeps itself on track, keeps itself making forward progress, only interrupts you when you tr- when it truly needs your help, um, rather than interrupting you at every single task it completes, asking you to test it, asking, telling you that it's completed. Um, I think, I think this is, this is in some ways the future that we envision with AI. It, it is gonna Do a lot for the humans autonomously in the background while humans can go about living their life, spending time with their family, doing creative pursuits, as we talked about.

Calum Johnson 36:01 Yeah. It, it, it's interesting 'cause you're, you're exactly correct. It's like that feels like almost the, the utopia, like the, the thing that we heard about the future of AI and technology is that we would have these like AI agents in the background that are like performing tasks without us even really having to give them, uh, super detailed minute-by-minute oversight. Uh, and with this, it actually seems that we're, we're there. Um, yeah.

Amol Jain 36:30 Yeah. There's, uh... It, it was such a big jump from, from Agent 2 to Agent 3. The team just did an, uh, did a really amazing work pushing the frontiers here. And, and case in point, as I was, as I last night was trying to build this, this specific app, it ran for, I believe, 45 or so minutes, and by the time it sent me a notification, here we will see it right here, it had already accomplished all of these things. It had created the food logging system. It had created a food database and search. It had created that tracking mechanism, um, food suggestions. And at this point, it ran for 46 minutes without interrupting, so I could spend time with, with, with my child and, and, and actually feed her while this app was being created for me in the background without me having to test. So at this point, I get a push notification. She's gone to bed. I log in, and it's, it's already working really, really well. There are, there are a couple of rough edges that, uh, that I need to polish up, but I... But it was... It's always impressive. I do this often in, uh, as a part of my job, but also because it's just so... It, it, it still blows my mind even after having done it hundreds of times, uh, how powerful and capable Replit Agent is and, and, and how, how amazing, mind-blowing it feels every single time when you go from just that idea to a working application.

Calum Johnson 38:00 Yeah. You know what? It's almost like the, the four-step process that you follow is idea, and then you have like the plan and the spec kind of phase where you're like, which is what we showed in the beginning with the going over the plan. Then you kind of let the agent run, and that's the new thing with Agent 3 is that you can actually just let it run autonomously. That's step three. And then you kind of have the polish at the end. You know what? We, we're gonna get into the polish, but before we do, I wanna talk about, um, that, that phase just before where the agent is actually just running. The first thing that kind of went in my mind is that with Agent 3, with the new agent that you guys have built, the planning phase actually feels more important because if you're gonna let it go and run for... I'm assuming in the planning phase is when it's telling you what it will do once... when the agent is running. So you kind of wanna make sure that you're on the same page so that in that 45-minute span of time, it's building what you want. Is that what you've found, or are you actually actively kind of like checking on your phone even in that 45-minute space of time to make sure that what it's building is what you want?

Amol Jain 39:16 So our, our thinking on this is, is exactly what you said. That planning phase is now so, so important. Um, it's, it's a newer paradigm. You know, no, no coding agents have, have, have been able to run for so long. So we're really trying to create this, this comprehensive planning experience where you can, you and the agent can have an agreement on what exactly needs to be done, and then it has clear guidance and clear direction on, on what to go execute against. And so I think it, it... The, the planning step becomes incredibly important, and we are trying to encourage our users to spend more time there to really refine the task list. Almost think of it as a, as a to-do list that, that you are both aligning on that, that is really clear to the agent on what it needs to do, so you don't have to constantly be, be checking, uh, after you've set it in motion. Um, we still see users, and I myself finding, find doing this as well, the curiosity gets the better of you, and you wanna jump in and see what exactly is being built. Um, but that's, that's a really fun experience as well because in real time you see it cooking up different screens, cooking up different functionality. Um, uh, on this preview page, you, you truly see, uh, applet features and, and views appear, and that's a really magical thing too. So when... As users are, are, are trying out Replit for the first time, it certainly is a very, very, uh, rewarding experience to watch it work. But over time, once you start building a lot of apps, as long as y-you spend good amount of time on the planning phase, you're setting everyone up for success, yourself and the agent, and you can truly let it run and let it deliver on the, on the, on the tasks that you aligned on.

Calum Johnson 41:09 Yeah. You know what? So ju-just to go, to go on that planning phase, when you're looking at the plan, because... And, and I said this even in the beginning, y-you're the head of product engineering at Replit. Like, you're, you're a Replit expert, and so I think a lot of this stuff is probably pretty intuitive to you in terms of, "Okay, this plan looks good." I'm thinking about that, that person that's watching this for the first time and using Replit for the first time. Are there certain things that they should be looking for in the plan that it's like, "Okay, this is ready to run"?

Amol Jain 41:44 Great question, and maybe perhaps we can, we can try and, and do this live.

Calum Johnson 41:49 Let's do it.

Amol Jain 41:50 Um, so I'm, I'm gonna take exactly the idea that I've been working with, and I will hit- Start chat. Great. As it's working, we can, we can start to-- Uh, you, you can see it, it goes by default to the plan phase first, where it starts to... Where it's working, it's thinking, it's trying to come up with a plan. In terms of, of what to look out for here, um, in a lot of ways, an idea can be broken down into what are the specific user flows or, or, or what are some specific actions that the user needs for this idea that you, that you've, that, that you're thinking of. Um, it's almost features is a good way to think about it, is if I'm putting in an idea like this, the features I'm hoping for are log my food, uh, get recommendation for the next food item, um, be able to track overall progress. If I, if I truly s- the Agent tries to do exactly that. It tries to take that idea, that broad concept, and really break it down into specific features, specific actions that the user is going to take. Um, so in a lot of ways, what you as the, uh, as the builder are trying to do here is make sure that, that it, it matches your expectations, that it, it, it is indeed what you want your application that you're building here to do. So as I had put that, put that prompt in here, it comes back and, and lists out all those features, where it says, "I'm creating a food tracking dashboard with visual progress." Sounds good to me. Allergen checklist, that matters a lot in a lot of ways. The-that's the goal of introducing foods for a baby. Um, that food suggestion e-engine with age-appropriate pescatarian foods. I really like that it... Agent even took the liberty of, and, and it was smart enough to think about age-appropriate foods, 'cause that's so important. That's not something I had given to it, but it was smart enough to come up with that. Um, it was the, the styles of baby f-uh, feeding, a recipe database, that seems great. A food log, great. And then a food calendar. At this point, I could say, "This all looks good overall," or I could, uh, start by saying, "Change plan. I don't really care for a calendar. I don't n- I don't need to see dates of when my, my child consumed a certain food. I primarily care about all the foods that they've consumed." So at this point, this is what I will dial, dial in on and, and, and really, my recommendation for most folks that are trying Replit for the first time is try to build a smaller set to start. At this point, I will say, "Hey, I care about three things primarily. Let me change the plan to only include those," and then say, "All right, let's go ahead and build the entire app," and let Agent unleash from there.

Calum Johnson 44:48 Yeah. No, that's good stuff. I, I like the idea of like, of simplifying it, of starting... You're going over those features, 'cause you're so right in the sense that when we have an idea, we think about the big picture idea and what problem it's gonna solve. We don't think in the detail of what are the features that go into that. And so kind of in the beginning is actually getting to your three, like what are those three features? You know what? I, 'cause I see it at the bottom of the screen, the build the entire app and versus start with a design. You went with build the entire app. What is kind of the case for each one? Like why would someone go with build the entire app versus why would someone go with start with a design?

Amol Jain 45:35 Great question. So Replit gets used by a w-wide variety of people, and they're optimizing and looking for different types of outcomes. Um, one thing we often see is folks trying to create landing pages, for example. And in that case, the design, the visual look and feel matters a lot more. Folk-- Cer-certain users really want to see what my application's gonna look like, um, before they build out the broad functionality. Um, this is something, for example, in the tech world, product managers and designers, we often see them start with this. They, they're visual. They want to create the, the, the, the, the layout, the, the visuals first and be able to, um, feel good about that before saying, "All right, let's go build this entire app." Um, but if you're a person who-who's, who's more interested in I'm, I'm creating, I'm creating an application, this, this plan looks great to me, the feature set makes sense, it does make sense to just go ahead and build the entire application. 'Cause e- once, once Agent has created something, you can still iterate within on that design. You can still iterate on the colors, the branding, the, the layout, what have you. Um, it's just different use cases of what, what folks are, uh, are, are coming to Replit with. And start with design, we primarily see product managers and, and, and designers in the tech world use this as a way to create quick prototypes, where they're not looking for full-fledged applications. They're primarily looking to, to bring an, uh, to, to visualize an idea that they have for their product.

Calum Johnson 47:18 You know what? And, and I'm, I'm interested to get like your, your personal take on this, and, and maybe every person that we'd ask at Replit would say something different. But you, you kind of ex-explain, like there's people building apps on Replit, like software, building landing pages. Like they're building a variety of products. Is there a certain sort of almost like idea that you see that you just think that Replit, yes, it can build all of these things, but it excels at building this sort of either idea or product? I'm just c- I'm curious for you, like is there something that you've seen where you're like, "Okay, yeah, it actually does a really just great job of that"?

Amol Jain 48:01 I think Replit is ... extremely capable at building fully functional apps. It, it, it's really capable at building something that's production grade, that's end to end, something that you can indeed give out to real users, your friends, et cetera. Um, so my, my sort of leaning is, is, is for those things. Uh, y- I see Replit less of a... Or, or I would encourage folks to, for real i- if, if you're seeking to bring real ideas to life, this is, this is the place. Replit is the place for you. Um, if you're primarily trying to build a, a landing page or a, or a, or a website of sorts just to, just to, you know, a, a, a website, a front end website if you will, then there are, might be other places that are perhaps better suited for it. So I think that's the, the, the entire end-to-end production grade functional, um, performant application that you can, you're, you will feel proud of giving to real users, Replit really shines at that. It's, uh, it, it comes with all the bells and whistles and the building blocks that you need to accomplish that. Um, so that's my bias for it. That said, we see, uh, a ton of users also build landing pages and marketing surfaces and, and, and, you know, very visual only, only, only things with Replit.

Calum Johnson 49:33 Yeah. No, that's cool to know. I, I, I, I like that. It's good to have that clarity. Um, building apps. Okay. So you hit build the entire app, and then that's when you go into, in the, in the case of the baby food, that was that 45-minute window of time when it just went and kind of built it out for you.

Amol Jain 49:53 Exactly, yeah. So at that point, at-- This was the interactive part of the journey where I had to put in the idea, I had to look over the plan, I had to make sure it was good. Agent does a really great job at planning phase, and usually it is very good. And so for me, it was a one-click, this looks good, let's go ahead and build the entire app and set it in motion. Uh, that's the point where I would put my phone away and only look for notifications when the agent needs me for any certain inputs. Um, but at this point, we've unleashed the agent. It is going to do all... It is going to build every single one of these features. Every time it builds a feature, it'll test against it. It'll make sure that it works. Keep going, keep going, uh, uh, trying to build every single one of these, and then when it's done, it'll let me know.

Calum Johnson 50:46 You know, you know what's interesting? I was just thinking about this as you were speaking, is if we said, I don't know, it took you 46 minutes to build, uh, this app for your daughter, the vast majority of that time you weren't even really actually present. You-- It was, that was the, that was the agent running by itself building the thing out. It was actually more like 10 minutes, 10, 15 minutes of your time in terms of in the planning phase, and then I guess when we get into polish. It's actually pretty remarkable.

Amol Jain 51:18 It is. It's... It, it is, as we talked about, it, it, it's, it's truly bringing the promise of AI to fruition, where AI is doing the, the work for me while I get to live my life.

Calum Johnson 51:33 Hmm. You know what? Okay, so maybe we, we let this, uh, like run in the background. I'm curious, if we go to the one that you already built and, and we look visually at like even just the design of it and how it looks, how much of that... Like, I guess once you get out of the, uh, the autonomous stage, that final piece that we're seeing on screen, how much of that was like it was already done purely by the AI versus you were guiding the design and some of the specifics of that?

Amol Jain 52:07 In this case, I was very pleasantly surprised. Um, it's one of those things where we've invested a lot in making Replit Agent not just build, build functional apps, but also pretty visually appealing apps. And so, um, I, I really, as soon as I saw this, I was pleasantly surprised with the, with the playful tone it took to the, to the whole app. You know, it's a, it's an app for, for feeding babies, uh, for, for tracking baby food, so it- it's very playful. There's like light red and green colors. So I felt really good about this. However, if I wanted to change some of these things, I really like Replit gives me that, that, that, um, that power as well. So for example, I could just theme the entire thing where I could say, "Hey, I, I would actually prefer more of a, a black and white sort of theme," so I can just select this theme and, and truly control, um, what the look and feel is down to every single, every single button, every single element. It, it, it... There's, there's just a lot more control there. However, um, with our re-- with, with the recent advancements that we've made on our end on, on this product, it does a really great job at coming up with that great first theme, uh, and, and that, that I found often matches what exactly I'm looking for.

Calum Johnson 53:33 Yeah. You know what? So, so in the preparation of this episode, um, obviously I've watched, uh, I've watched back even the video that we did with Matt. I, like, was reading, um, even the, the press things that you had about Agent 3. I then actually came on, uh, Reddit, beautiful website of Reddit, and there's this like mega thread where people have left like feedback of their experience of using Replit Agent 3. And one of the things that, uh- One of the criticisms that people had of Agent 3 versus Agent 2 was, yes, there's this utility to the autonomous features, but it can take, um, a long time, like, on this autonomous portion, and then sometimes it actually doesn't fix the, the issues. Which I guess is why we come into, like, the polish phase. Can you speak to that? And then also in your personal experience, have you had it where it's like there's actually still problems even when it comes back in terms of the functionality, and how do you fix that?

Amol Jain 54:49 That thread was, was very, um, very, uh, sort of eye-opening to see certain, certain aspects that, that our users needed but that autonomy wasn't really providing. And, and to the, to the credit of our team and, and how fast they move, we took that feedback very quickly and we turned it that into this, this autonomy level, uh, mode where we user... We gave users more of that control where if they wanted to be a lot more in the driver's seat, then they could dial down the autonomy level to low or medium and have, be in more of the driver's seat where they're instructing the agent on smaller things. Um, and by giving, giving our power users this, uh, it, it, it hopefully meets their expectation or, or their ways of, of interacting with our product. Um, I think we are, we are continuously learning. We're, we're listening to our users, and the future is certainly where you as a human don't have to instruct the agent as much. Um, you know, I really like how, uh, Amjad, our CEO, describes it. This is in some ways the full self-driving versus the autopilot of self-driving-

Calum Johnson 56:11 Mm-hmm

Amol Jain 56:11 ... if you will. I think today there are, there are rough edges, and we are just at the, at the, at the forefront and frontier of this. What I, what I really love about Team Replit is its ability to take feedback and operate against that really quickly. So I wouldn't be surprised in a few weeks we've made the max autonomy make cont- with our continued investments in that full self-driving mode, made this much, much more useful where our power users also want to, want to switch over to this. But in the meantime, our, our sort of instinct was to meet where our users are and give them more control over exactly what they need.

Calum Johnson 56:53 Yeah. You know, I, I, I get what you're saying about the ideal is actually that most people, 99% of users, just actually use the max autonomy, and we kind of reach this reality where the AI can perform actions to, like, an elite level without human intervention. But in the time being, for the time being, have you found that actually the more experience that someone gets with Replit and vibe coding, and maybe even just, like, coding in general, they tend to want to have more... They tend to use, uh, kind of go in, like, the lower settings. Like, they want more control over how, uh, the agent runs. Is that what you found or seen?

Amol Jain 57:44 Not, not quite. So I'll, I'll share a couple of things. I think when we look at data of how folks are using this, it still leans heavily on max and high. Um, we've seen certain users move to low or, or medium, um, uh, in certain scenarios. The other, the other thing that I'm seeing a- as an engineer, we use a variety of, of, of, of, of products, uh, i- in the AI landscape, you know, other coding agents and such that are more about writing code than building apps. And more and more, I find myself, my colleagues, my friends leaning towards that autonomous mode more. There are, there are products that you truly, even the new ChatGPT with, with, with GPT-5, um, where you wanna give it a prompt and let it go on for a long time, do its deep thinking, and just come back with you a, a really good output rather than giving me its first thought. Um, so I think, I think a lot of the AI world is moving towards that. And as the AI capabilities keep increasing, we will, we will... The, the models and, and, and, and all the products that are built on top of these models will continue to get to that spot where they are, where, where humans start trusting them more and letting them run autonomously. Uh, in, in my personal experience, I definitely find myself leaning more and more there, of firing off tasks and trust- trusting the AI more to do, to do these tasks for me. But I think, uh, you know, change is hard. Um, humans have a tendency to want control, so it will take time for all of us, uh, to learn how to work better with, with this ever-evolving, incredibly fast evolving technology that AI is.

Calum Johnson 59:35 Yeah. You know what? So let, let's say for the beginner that's watching this, they're building their first, uh, piece of software. They're using the max. They're using, um, they're using AI Agent 3, and they're using it on the max setting. They've kind of gotten to this point similar to where we are now, where, like, they have this visual. Like they've, they've just had this magic moment of, like, it's before them, and now, like the A- the, the autonomous agent has produced the thing. Talk to me about that polish phase. Like, how do I go from Okay, it's 90% built and I can see it in front of me to this is actually the final, not the final, but it's like the final V1, like the final version one that I could actually go and deploy. Like take me through that polish phase, that final 10%.

Amol Jain 1:00:31 Absolutely. I think that, that 10% is often, often a, a very interactive small improvements type of journey where, where it's less asynchronous, it's more synchronous. You, you have to work closely with the agent on these small asks that you are, that you are, um, that you desire. For me personally, we can, we can look through some of this specific example as well, where when we first... After the first 46 minutes of running, it gets me to a, a state that I feel pretty good about. At this point, it's done 46, uh, minutes of work. I look at the product. My first thought, the visuals actually look really good. I am-- I, I, I like this and I do not want to change a lot of this. I will personalize this over time, you know, put my baby's face on there, uh, to, to make it more fun. Um, but first thought, I go ahead and I try to use the functionality. And what I see is it was, it was trying to show me progress day by day. So my first ask to it is, instead of that today's progress, I want you to show me headline metrics of how she's tracking, that how much percentage of the food journey is complete. And so that's one of the first things I ask it. As soon as I do that, in a minute later, it comes back and just show, gives me, "Hey, that's done." At this point I coulda, I could have batched my requests of I want, I want you to change the headline metrics, I want you to include this, that, and it would have gone on and done all of those tasks. But I was sitting in front of my computer testing this. As I was thinking of ideas, I'm just putting them in here and continuing to go about testing, testing the rest of the app and experiencing, uh, what, what the agent has created so far. And so even with, even with the max autonomous, max autonomy mode, if the task is small, the agent's smart enough where it just makes that task and comes back to you, "Hey, I'm done." So, so this was less than a minute of work. And then similarly, I realized as I was continuing to test of let me add a date for when she consumed a food, 'cause I might forget to log and I want to log something in the past. So I give it that. It is, again, able to just do that rather quickly. It realizes that adding a date is a significant action and things might break, so it figures out it needs to do some testing. It does all of that testing, and I believe a few minutes later it comes... It, it detects some things are wrong, makes all of those changes, and it comes back to me, "Hey, I've made your updates that you asked for." Um, yep, so the, the date picker is working, and this was a longer action that took about 20 minutes of work because it had to introduce the concept of date-based tracking rather than just tracking in general. So it's often, often certain things are small, certain things are large. It's hard to say just by looking at the app. Um, my personal preference is to just give, keep adding these messages in this chat box. There's even a queue here. Agent is smart enough to just queue them up and keep working on them one by one. Or oftentimes if I know it is multiple things, I'll just give it all to one. But, but this polish phase, this sort of like the last 10% phase tends to be a little bit more interactive, where you are experiencing the first version that it creates and trying to work your way to that, that production ready thing that you are happy with and want to, want to push out to the world.

Calum Johnson 1:04:12 Yeah. You know, it, it's kind of, um... It's cool 'cause listening to you talk about it almost feels like play in a way. Like you're kind of, you're, you're giving it certain things. You're like, "Okay, add this in," and then you're taking it away, and you're, you're kind of like playing around with like getting to that, that perfect thing for you. For that person that's actually looking to create an app that one day they have on the App Store on like iOS or on the, the Play Store on Android, are there any kind of secrets, tips that you have that's like, it's just gonna accelerate that polish phase and actually get you to something that's like, not only is it useful for you as an individual, that you could see similar to what Michael Messenger did with Zillow, like millions of, like a lot of people are actually using it. Like it's actually like a... It's reached a level of polish where a broad spectrum of people could actually be using it.

Amol Jain 1:05:18 Yeah. Um, I think few, few sort of things go into that, into that process of, of going from something that is, for the lack of a better word, a prototype to taking it to something that is truly production ready. And the way, the way sort of we see our users do it, the way we recommend is, uh, uh, Replit wants to take a lot of that burden upon itself. It wants to be able to give you all the tools that allow you to get to that spot quickly. Um, and so, you know, that's where, that's where things like testing come into play, because it is testing and it is discovering if there's any bugs. So before you push out, the bugs are, bugs are addressed. Um, that's where this full visual control comes into play, where you can truly polish to your heart's content of, "I believe that this needs to be a little bit, um, a little bit to the left or a little bit to the right." It allows me to do things like that. But even more importantly, it does things like security scan, where if you've built something, it will go ahead and, and, and do a scan of all your code and, and your functionality to make sure that it's actually secure, where you're not gonna leak your users' data. Um, so, uh, th- there's a lot of tools that we offer the users for visual control, functionality, um, you know, ensuring that the functionality's solid and stable, ensuring that it's safe and secure. And, and even so much so that we have, you know, our, our, our publishing flow lets you publish it in a, in a, in a scalable way where you might launch it to, you know, you, you might, you might launch it and, and one or two users use it, but then everyone suddenly discovers it and loves it. It will serve thousands, tens of thousands, hundreds of thousands of users no problem automatically without you having to worry about any of that. So our philosophy here is we've invested in a lot of tools, a lot of guidance, and a lot of help that you have available within the Replit platform to l- to actually enable you to take that idea to production. Um, so that's why we see users in fact publish these ideas.

Calum Johnson 1:07:38 Yeah. No, it's, uh, it's so cool. You know, on, on the design aspect, 'cause you, you mentioned earlier, you spoke about adding a picture of your daughter, and it kind of got this thought and idea, like, started in my mind, which is, I think for a lot of beginners and when you're just starting out with something, usually there's, like, an app or another product that has inspired you. You know, you're like, "Oh, I, I would love for my app to look similar to this app," or, "I, I'd love for my landing page to look like the Nike or the Air Jordan or the Apple, uh, like, website in terms of its design." Is that something that can actually be done with Replit, where, like, could I take a screenshot of, like, another app that I just like the way design-wise it looks on my phone, for example. Could I upload that screenshot into Replit and have it actually almost, like, mimic the design language and, like, look and feel of that app? I- is that possible or am I just living in, like, a, in a delusional land right now?

Amol Jain 1:08:44 That is absolutely possible. So we let you... If I wanted to change this to look more like, for lack of a better, better example, Instagram, I could take a screenshot of Instagram, just paste it right in here, and then let it, let it, um... Uh, the agent's smart enough to extract the, the visual look and feel, the theming, the styling, and apply it. Um, in fact, I, I have a great example for this. Um, there's this... I've always-- I wake up in the middle of the night quite a bit, can't fall back asleep, need a sleep story to fall asleep. I use a product that shall remain na- nameless, and it's fairly expensive, and, um, and, and with no real value. It's 20 static stories. I'm so tired of listening to the same stories. So I decided to, um, build an app that just does it. I want to publish this on, on, uh, one of the stores, take it live, make money off of it, in fact, hopefully someday. And, um, as a part of doing that, I took inspiration from certain, certain websites. I took a screenshot and just gave it, 'cause I wanted it to look calming like this, uh, 'cause in the middle of the night I wanna be looking at something calming like this. So I, I, I in fact took screenshot from another app and gave it for it to mimic that look and feel, and it did a, it did a pretty great job at something like that as well. Um, so within maybe a couple hours of work, I was able to create this, where I have full control over what story I wanna listen to, in what accent. British accent does really well for falling asleep. Um, uh, also, uh, you know, uh, what background sounds, and then in this very soothing theme that this app has, uh, and was able to build this. And now I'm really hoping to, to take this live and, uh, and let people create their own sleep stories and charge them money for it.

Calum Johnson 1:10:42 Yeah, we need to know, like, when, when, when is it... We need to know when it's gonna be live. That would be a good...

Amol Jain 1:10:46 Absolutely. I ne- I, I need the distribution, so I will let you know.

Calum Johnson 1:10:50 [laughs] I guess, because looking at this, this actually just looks like a professional product. Like, this just looks like something... You wouldn't be like, "Oh yeah, it took them a couple of hours to build this." You would just be like, "This is something that they're charging for." Like, this is just like a business. I guess we, we go through... So we go from idea to the planning, the agent running. We then polish it. At what point in your mind as, like, an engineer, as someone, like an expert in terms of Replit and then also coding, is it ready to deploy? So is it ready to deploy, and then how does someone go about doing that?

Amol Jain 1:11:31 There's this, there's this Silicon Valley axiom that you'll hear, hear people say, um, "If you're not embarrassed of, o- of something by the time you launch it, you, you're, you've l- launched it too late." Um, I believe in, I, I believe in that. I think a lot of people do. Because if you're trying to solve a problem as a business builder, you can always keep polishing, and polish can take a lot of time. But if the functionality works really well, my instinct is to put it out there, is to, is to let people have that functionality already. Like today, there aren't a lot of tools that let you create your, your sleep story like this. Um, so I'm not happy with the, with the exact look and feel here. I, there's, there's more work to, for me to do here. But I would rather be a little embarrassed by, by some of this as long as it can provide value. And, and you know, um- ... dri- users that, that ha- actually have this pain or have this desire will run through slightly buggy things or even unpolished things to actually use this app. So my hope is I would try to get this live and then in background continue to evolve on it. So there's a version one, a version two, a version three, that fast iteration. I think, I think building businesses, it's so important to, to test early, test often, validate that people do in fact need a sleep story app.

Calum Johnson 1:13:01 Yeah. You know what's interesting? Like in the, in the beginning we spoke about like the speed and actually figuring out like the fastest way to go live is actually really important. And so it's like you deploy actually the version one, you're getting feedback on that, and then in the, in the meantime, you're also like iterating rapidly to get to version two and version three. Like that's-- in, in your mind, that's how successful... Is that how successful AI businesses are gonna be built, you think, in the f- well, today and going into the future? It's gonna be about that speed.

Amol Jain 1:13:43 100%.

Amol Jain 1:13:44 I think, uh, the landscape is evolving so fast, the capabilities of what the foundational technology, the, the AI models is evolving so fast. And I think it, it's, it's, it's a crowded field with a lot of products being created. You know, with, with, with tools like Replit, we've truly unleashed that where, where anyone can be a builder, anyone can build something. So the, the, the field is crowded. There's lots happening. It's al- it's always better to s- uh, to, to, to anchor towards getting something out there quick and, and really getting in hands of your users and seeing that real feedback, that real giving the-- ensure that you are providing them the value that, that you set out to, um, rather than taking your sweet time. I, I think speed has always been a, a very critical thing for businesses, and it's even more so in today's AI world of fast-moving things where, where the barrier to entry of, of how long it takes for someone to build the, the, the, the technology or the product itself is so low that you want to, you want to anchor for speed.

Calum Johnson 1:14:54 Do you, do you think we get to the point where someone-- people are building multi-million dollar apps just like on their smartphone? Like on the-- they, like they're, they're experiencing a problem, they're typing about it into Replit agent or whatever the software is, and the first version of what that app looked like that went on to make millions of dollars was actually just fully built off their smartphone in that conversation.

Amol Jain 1:15:25 I absolutely see that happening. Uh, it's-- we are so early in this journey still. Um, y- what we're doing today is, is amazing. Six months from now, say Agent 4 comes out, I see this capability just exponentially increasing. And so as a builder, you, you are sitting down to dinner, you can just type up your idea and let the agent do things for you, and in a couple hours build you something that you can, with one click, publish to the world, um, even distribute it seamlessly from right from, from just that phone without having to spend a lot of time on it. Um, you know, at, at, at Replit, we truly believe that it's not just about helping people build apps, it's a, it's about helping them build real products, um, that, that we also help them distribute and monetize. How might we give you the capability to, to n- to talk to your users, to notify them, to send them emails, if you will? How do we create ways for you to grow your products? How do we create ways for you to easily monetize your products? So I can, I can easily add for every sleep story, charge a dollar to the user. So we are truly working on that, that end-to-end cycle. And as we do that, it should be very much possible for an entrepreneurial person sitting anywhere in the world to go to their smartphone, type up, type up their idea, iterate a couple of times, and set it live, uh, and see it catch fire.

Calum Johnson 1:17:02 Yeah. You, you know what? I'm, I'm curious to get your perspective. Like is there a downside to that? Because when I, when I-- And, and the reason I ask that is we have conversations like this, and it's like the opportunity, and honestly, I love sharing it with people, like what's possible. But then the general discourse around AI that I see for the most part on the internet is more of like a doomsday scenario. And so I'm a- I'm always thinking to myself, I'm like, "What's the, like what's the truth here?" Like w- is there gonna be a, a detriment and a negativity to humans being able to have that capab- that capability?

Amol Jain 1:17:45 I have a very optimistic view on this. I think AI as a technology is, is primarily going to take on the mundane, the, the annoying things that humans have to do today and free us up for the creative pursuits, free us up for family, for friends, for, uh, for all the things that a human might enjoy. Um, I almost think of this as, as instead of people having to do work for a living, they might work for their passions. Um, yeah, that's, uh, that, that, that, that's my view of how Us humans are gonna be able to build these technologies in, in, in, in, in a, in a safe way where it is, it is used for the betterment of the society. Um, you know, it, it, it takes us from the age of, in agriculture, people having to use, uh, their hands to sow seeds to, um, the combine that, that-- to creating the machine, the combine that does it for you, um, so you as the farmer can own more land, grow more crops, feed more people. Uh, uh, I, I truly think of it as that leap of just freeing us up to do other more creative pursuits and solve more problems.

Calum Johnson 1:19:16 Yeah. No, we hope so. We hope so. You know what? So the, the first iteration of Replit Agent, when was that? That was what... That, that's all kind of come in the last, like-

Amol Jain 1:19:29 Late last year.

Calum Johnson 1:19:30 Yeah. So in the last, like, 12 months. So if we went, and I'm not gonna ask you exactly what's in Replit Agent 4 or, like, what you guys are working on there. If we had to force... So 12 months ago, we didn't even have Replit Agent. Now it's, we have Replit Agent 3, and it's, like, running autonomously. 12 months into the future, where do you think we're at in terms of what AI and some of these things are capable of?

Amol Jain 1:20:00 I think that is, that is-- If I were to, if I were to really think of this in terms of, of, of, of, of Replit and, and what we are trying to do, I think that, that word that you brought up, capable of, the capabilities is the, is the exact thing. I think making the agent more capable across a variety of vectors is, is what we're going to see in the next six to 12 months. Um, you know, today it does a great job of building that app. How do we make it more capable so it does the entire software life cycle, where it's, it's, it does a better job at planning, it does a better job of writing and testing, it does better job of organizing the code so every subsequent action is easier. How does it do more than that? How does it help you distribute and monetize like we were talking about? Capabilities on the vector of, of what it can produce. Today, it produce we- it produces web apps primarily. Um, how do we h-help it create mobile apps that you can one click publish to the stores? How do we help you create automations which you can just set off in the background? Software is not forever going to be this UI that we interact with. How do we create software that does job in the background where you just assign tasks to it and, like, agent creating more agents, if you will, that sort of capability. Uh, capability of f- it, it connecting to more of your systems, so it has access to your, to your emails, it has access to your Slack, it has access to any app you use out there. Um, so it can build better software, it can tie them together, it can automate parts of your life, it can create-- it can let you create, you know, businesses to that end. So I truly see it in terms of capabilities. There is so much to do with all of those across all these vectors, and that's what a lot of investment we are making, um, so Replit Agent 4 should be a lot, lot, lot more powerful, allowing you to build businesses, allowing you to solve your own problems and across many dimensions.

Calum Johnson 1:22:08 Yeah. You know, before we get out of here, there, there was two things that I, that I thought about, and you mentioned them pretty early in the conversation. The first thing that you said is that, you know, you wanted to leave your mark. You mentioned that. You said, "I, I wanted to leave my mark. That's why I even wanted to build a business and be an entrepreneur in the first place." And then the second thing I think about, even some of the things that you're doing with Replit right now, where it's the mission is to enable a billion people to build software. I just think about that, those two things. Uh, like, you're working on this mission to enable a billion people. There's gonna be millions of people. There's already, like... I don't even know how many people you guys have seen have actually built things using Replit. On a personal level, Amol the human being, what does it mean to you, some of the things that you've been able to achieve, that your work and even, like, the impact of what you're seeing of, like, people are actually building their very first businesses, pieces of software through an AI agent that you had a hand in creating?

Amol Jain 1:23:22 Yeah. So, you know, there's... The, the, the, the whole idea of leaving a mark on the world is, sounds somewhat selfish as a pursuit. I think it's almost, you know, an innate desire that I have, other, other, other, you know, entrepreneurs have as well. But, but it's often, it often leads to great outcomes because you end up creating a product that solves a problem for, for people. For me, it's always been about helping people make a better living. Um, I joined Airbnb a long time ago, and i-initially when I joined, it was all about travel. Like, you know, travel is a great way of, of, of, of knowing the world and, and getting to understand other people. Um, I, I joined it from a traveler's perspective, but very quickly, what I really enjoyed at, at Airbnb was the host perspective, where hosts have these houses that, you know, they're paying giant mortgages on. How do we help them monetize those? How do they get to, uh, use this a- this big asset, which is usually a big chunk of their, of their net worth, they're able to monetize that. I think that was very powerful to see their stories and help them achieve that. Um- I got to do that for Airbnb Experiences, which was sort of the same idea of how do, how do we let people monetize their time, take pride in showing their city, but get paid in return for that, a new career, if you will. Um, from there I left to start a company which was again very much focused on, um, helping hourly workers. Hourly workers have had it really hard in America for the longest time. There's no flexibility, pay is bad, all those things. So really taking that angle of how do we help them earn a better living. Um, and then as I came across Replit, I really liked that part as well of, of by enabling people to create software, we are enabling them to perhaps become, you know, create businesses and create prosperity for themselves and others. Um, so I've, I've-- I, I think that is a problem of helping people achieve things, helping people make a living is something I get excited about, and I think this would be sort of the mark that I hope to leave on the world. And, and Replit especially has been very, very powerful at doing that. I think, you know, I'm so much closer to, to, uh, to the, to the, to the direct impact and, and seeing it unleash over the, you know, billions of people over the next few years.

Calum Johnson 1:25:53 Yeah. You know, it's awesome. B-before we get out of here, I'm so big on, like, these conversations are, like, actionable and practical, and as soon as someone, like, clicks off of this window or, like, ends this video, there's something that they can go and do. And so we're coming into the weekend. If you were gonna give someone a challenge of, like, on your Saturday morning or your Sunday afternoon when you have a couple of spare hours, like, go and do this and the time will not be wasted. Like, you will be infinitely grateful. Maybe you'll even experience that kind of magic that we spoke about earlier of, like, the first thing that you build, like being able to see that software built. What is a challenge that you would give someone of like, "Okay, you have some time this weekend, go and do this," and it's not something that you're gonna regret?

Amol Jain 1:26:53 I would say spend your Saturday being attuned with, with perhaps things that, that you're bothered by, things that you see that aren't efficient, some thoughts of, of you-- something you wish existed based on your personal life. I think it's a great exercise in itself of being attuned to, to what's going on and noticing problems. I think especially for business builders, that's the step one. And Sunday morning, open up the Replit app, type that, and in the next few hours see it come to life. Um, and, and if you are truly motivated, try to take it, try to take it to production. T-try to put it out there, try to send it to your friends, try to send to some potential users, and see if in a weekend you can go from discovering an idea to getting it in the hand of one real user. Um, I guarantee you it is very much possible.

Calum Johnson 1:27:47 Yeah. Amol, thank you so much, man. I really appreciate you making the time and, and being so open with us. I think it's gonna, it's gonna help a lot of people, so thank you.

Amol Jain 1:27:57 Thank you, Calum. Really thank you for having me. This was a great chat. Really loved sharing things about Replit, and I'm very excited to see what people build.

Calum Johnson 1:28:05 Okay, so if you enjoyed this episode, you're gonna absolutely love what we did with this one right here. Click here.