A conversation with Replit's President and Head of AI Michele Catasta

SPEAKER_00

Michele, it is so fantastic to have you here today.

SPEAKER_01

Thanks for having me.

SPEAKER_00

So you started working with Replit in 2022.

SPEAKER_01

Correct.

SPEAKER_00

A lot has happened since then. Can you walk me through what the product was in 2022? Who was it for? What did the product look like? And then I want to get to 2024, but what was it like four years ago? [SPEAKER_01] So Replit started in 2016 up to 2023. It was a cloud development environment. We made it as easy as possible for developers to kickstart their code on our platform. In reality, though, we were building an amazing platform that eventually allowed us to build Replit Agent.

SPEAKER_01

[SPEAKER_00] And tell me about Replit Agent. Where did that idea come from? When did you guys start working on Replit Agent?

SPEAKER_01

So I would say there was always the vision in a sense. If you go back and check the slide decks that Amjad, the CEO and co-founder of Replit was using to raise money, he always had in mind something that looked like Replit Agent, where AI would have generated the code for our users, and in turn, our users would have focused on business ideas, and literally what is crucial about running a business, rather than the code required to create products. It took a while to get there, and not only because it took a while to build Replit, but because the entire field of AI was growing really fast at the time, and maybe we were not expecting that to happen so quickly. And that buildup, I would say, started around 2020, when we got access to the first token source models, like GPT-2 from OpenAI. Back in those days, I was a researcher at Stanford, so I was very early on LLMs applied to source code, and it was exactly the same bet that Replit was making. I did it from a research point of view. Replit was doing that from a product point of view. And then we met halfway. So I met Amjad in 2021. I showed him a demo of something that looked like Replit agent, but way more primitive compared to what we launched. And we realized we had exactly the same life mission.

SPEAKER_01

[SPEAKER_00] Wow. Okay. So you were doing research in this space, and Amjad had this very similar vision of what the future should look like. You guys came together, and wow, this is just a match made in heaven. Yeah. [SPEAKER_00] So and then you joined immediately as president and head of AI, or what was the journey there?

SPEAKER_01

The journey was, I joined to build the AI team zero to one. And my only fixation was to create something like Replit agent. I recall, I think, maybe two weeks after I officially joined, I wrote a manifesto, an AI manifesto for the company. And if you go back and read it today, it basically describes what not only we launched with Replit agent B1 back in September 2024. The reason why I wrote it is because I realized, and also the reason why I joined the company, to be honest, because I realized they built an amazing platform where all the tools necessary to actually build an agent were already in place. If you think about it, what an AI agent does, especially for coding, is to use the same tools that we built for developers in the past. And it just learns how to orchestrate them correctly, as if it was a human developer. And everything was there. And I told myself, I'd rather join a team that knows exactly how to build that. It's already very advanced in terms of engineering culture, and loves exactly the same problem I want to work on, rather than trying to raise money once again and build the right team, and make mistakes, because early stage it's very probable to make mistakes in creating the right team.

SPEAKER_00

I want to talk about what it was like to launch Replit agent, because I'm sure that was a magical moment. But before we do that, I'm so curious. I think it's always very interesting how companies think about when to invest in a bet and structure their teams around a bet, or something that they don't know is going to work, versus when to continue investing in the thing that's maybe not at full product market fit, but it seems to be working in some sense. So what was the structure of the team? How many people were working on Replit agent? How did you guys make that happen?

SPEAKER_00

[SPEAKER_01] I would love to tell you that it was just one of the many bets, and if it didn't pan out, everything would have been all right. The truth is the opposite. I think we were almost at the finish line of the company, not because, you know, we had money in the bank, but the product we built up to there was not easy to monetize. It was amazing for the educational markets. So schools across the globe used us, that's why our user base was growing so fast. But it was hard for us to justify the growth of the company, our valuation and basically becoming what we are today. So Replit agent was both a bet that matched our company mission, which is empowering the next billion software creators, but also a necessity. We had to find strong product market fit that allowed us to grow to where we are today. And it started as a small group of misfits. So I created a very small AI team and I told them, regardless of where this goes, let's try to build this. Let's try to show the world how this new technology is going to look like. Let's build something for non-developers, because it's going to be highly differentiated. And it's a much harder technical problem, to be honest. And I think the last two months before launching, more than half of the company was working on Replit agent full time.

SPEAKER_00

So how many people was that, about 60 people, 30 people?

SPEAKER_01

It was 30 people, roughly. [SPEAKER_00] We were down to a skeleton crew. I would say more than 80% of the company was technical. All the other functions were run part-time by some of us. Let's try to show the world how this new technology is going to look like. Let's build something for non-developers, because it's going to be highly differentiated. And it's a much harder technical problem, to be honest. And I think the last two months before launching, more than half of the company was working on Replit agent full time. So how many people, was it about 60 people, 30 people? It was 30 people, roughly. [SPEAKER_00] We were down to a skeleton team.

SPEAKER_01

I would say more than 80% of the company was technical. All the other functions were run part-time by some of us. And then once we had product market fit, and we sat on that for a while when we realized, okay, this is actually working well, then we started to grow the account again. Wow. So right before launch, it seems like it was probably a hard time for the company then. [SPEAKER_00] You scaled down. [SPEAKER_00] Darkest possible times. [SPEAKER_00] Yes. [SPEAKER_00] Describe that for me. [SPEAKER_00] What did that feel like coming into work every day? So early 2024, we attempted to sell the product to the enterprise. [SPEAKER_00] The previous one, not Replit agent.

SPEAKER_01

[SPEAKER_00] And we found a lot of interest in terms of the philosophy of our product. Everything integrated, one single place where it can go from idea to creating your software. And it was very hard to actually have companies embrace it because it would have displaced a lot of different tools.

SPEAKER_00

[SPEAKER_01] And as you know, every developer has a lot of their own specific tools. [SPEAKER_01] And it would have been really hard to tell them, drop everything, just embrace Replit. [SPEAKER_01] So we hit that dead end. [SPEAKER_01] And then we went in quick order. [SPEAKER_01] We relocated away from San Francisco down to Foster City. [SPEAKER_01] So brand new, much bigger office. [SPEAKER_01] We had to go through a layoff because our sales and marketing team was not, first of all, not required at that point in time because we didn't have anything to sell. [SPEAKER_01] And also they were there with a different product in mind.

SPEAKER_00

[SPEAKER_01] And six months ahead of us where we just knew we had a deadline that I dropped on everyone's calendar.

SPEAKER_01

I said, this is the day where we're going to be launching Replit Agent V1. Everything else was just us showing up at the office every single day, heads down, knowing that we had this bet that must have worked by then. It required a lot of grit and determination to actually go through all of that. But I think it gave us an opportunity to be extremely focused. And I never experienced that again because since the launch, when we actually hit product market fit, my life changed completely. I don't spend every single second of my life thinking about the technical problem. Now I have way more duties. And I don't regret that. I think it's a lot of fun.

SPEAKER_01

I'm learning a lot of new skills with that. But it will never come back again. Those six months were both peak productivity and also peak determination for us because we had to make it. Oh, man. That sounds both an incredible time and an incredibly hard time. Incredibly stressful for sure. [SPEAKER_00] So you set this deadline the night before you're ready to launch Replit Agent. [SPEAKER_00] What are you thinking? What are you feeling? [SPEAKER_00] Are you confident it's going to work? [SPEAKER_00] What's going through your head? I'll have to give you all the shameful details about what happened in the couple of days before the launch.

SPEAKER_01

I planned two different dogfooding sessions with the entire company. 36 hours before launch, literally nothing was working.

SPEAKER_00

[SPEAKER_01] We had prototypes in the previous days where the product was pretty okay and it was still very rough on the edges but was working most of the time. [SPEAKER_01] And then we hit the worst possible planet alignment where we had a few bugs that we couldn't understand where they were. [SPEAKER_01] And it felt very disheartening because I brought in the few people who were not working on the project, including non-technical employees at the company. [SPEAKER_01] And I told them just give this a try, it's going to be amazing. [SPEAKER_01] And the experience was awful. [SPEAKER_01] I was this close to deciding not to launch.

SPEAKER_00

[SPEAKER_01] And then I told the core team that was working on Agent and we said guys, we're going to pull another all-nighter. [SPEAKER_01] We're going to get to the bottom of this. [SPEAKER_01] We're going to find out what happened. [SPEAKER_01] And we have another 36 hours to do it. [SPEAKER_01] Fast forward, I think 16 hours from that moment, we did another dogfooding session.

SPEAKER_01

The product was finally working to the surprise of everyone at the company. And we said, okay, tomorrow morning. [SPEAKER_00] What does finally working mean? So I think especially for the non-technical employees at Replit, it was the first time that they went from a prompting natural language to a very basic application showing up in front of their eyes without doing anything different. And those are folks that never wrote a single line of code in their lives, even though they were working at Replit. Because the truth is, no matter how much we ask them, please you should dogfood the product. Learning how to write code is not an easy endeavor.

SPEAKER_01

And a lot of them have very busy lives. They work. They have families. They're not going to learn how to write code as a side activity. And then we realized, imagine what's going to happen if this is across the entire world. [SPEAKER_00] What did you guys see that even the most AGI-pilled, deepest research teams had not seen yet? [SPEAKER_00] What was it that you guys got to this point sooner than anyone else? So personally, when I decided to break from my research career and focus on product, I felt I removed shackles from my wrists on how to think of technical problems.

SPEAKER_01

And I think I can put myself in the shoes of researchers that were at OpenAI and Anthropic in those days where you always have certain expectations in terms of how good your model has to be, how good the product experience that you want to launch. There was way more skepticism about AI back in those days compared to now. Suddenly I felt free. I said, we're going to create a new product experience from the ground up. Even if it's something that I'm almost ashamed of, it's totally fine to bring it out there. I want to learn from users what's their feedback, what's their reaction, what do they love, what are they disappointed about.

SPEAKER_01

And it was literally a mandate from me, and it would be the first on the market.

SPEAKER_00

[SPEAKER_01] Every single day that we, before launch, I was panicking that someone else wouldn't have done it.

SPEAKER_01

How good is the product experience that you want to launch?

SPEAKER_01

There was way more skepticism about AI back in those days compared to now. Suddenly I felt free. I said we're going to create a new product experience from the ground up. Even if it's something that I'm almost ashamed of, it's totally fine to bring it out there. I want to learn from users what's their feedback, what's their reaction, what do they love, what are they disappointed about. And it was literally a mandate from me and it will be the first on the market. Every single day that we, before lunch, I was panicking that someone else wouldn't have done it. And I think if we were not first on the market, we probably wouldn't be where we are today as a company.

SPEAKER_01

So what you're describing strikes me as you were treating it as an applied research project. [SPEAKER_00] We were thinking it's okay if it's not great. It's okay if it's not commercializable. I'm just curious about how it's going to go. Yeah, let's see what happens. Is that how you run product development now?

SPEAKER_01

Pretty much. So that specific launch we internally said is going to be a low-key launch. We're just going to do a very basic video. We're going to put it out there. Let's see what happens. And then it went insanely viral. So the same formula that we use for Agent 1, I keep applying for our engineering and product teams where, first of all, I tell everyone, let's not get emotionally attached to anything that we built, both on the backend as well as on the product surface. There are assumptions we made back in the days that were already successful from day one. And I think they really influence what people expect from our AI-coding product today. Because many of them got inspired by Replit. And some assumptions that were wrong. Many of them likely came from me, and I admitted in front of the company, I think it's time for us to change. I think it's time for us to do something different.

SPEAKER_01

What was a wrong assumption? Let me think. I can give you one about Agent 3 where we put a lot of emphasis on autonomy.

SPEAKER_00

Mm-hmm.

SPEAKER_00

[SPEAKER_01] I wouldn't call it a wrong assumption per se, but we thought most of our users want to go from point A to point B with just a single prompt. And they just care about the agent doing the entire work themselves. In principle, that is true. In practice, you end up hurting engagement in the product. Because we are the first one to market, Replit Agent 3 can run for 200 minutes in a row. That's amazing. But what are you going to do in those three and a half hours of your life? You need some feedback, some interaction. You need to keep them on the platform and engage. And it's something that we fixed with Agent 4 and now we have way more engagement than we had before.

SPEAKER_00

So we make some bets and then we have to correct course depending on how our users react to our launches. [SPEAKER_01] But I always try to think about what we build as partially a very structured engineering project, but at the same time also as a research project where we're trying to find out exactly what is the right product to build. Because the truth is no one knows the playbook for how to build agents. Not only coding agents, but in general agents are something that we are shaping as the field improves. And if you have a very prescriptive approach, you're just lagging behind compared to anyone else. So it's not worth it to play that way.

SPEAKER_00

So Replit strikes me as particularly good at not getting emotionally attached to the last thing that you built. You're just reinventing, evolving, and willing to burn down the last thing in favor of the next new thing. Is that a cultural value? Is that something that comes from you and Amjad? Do you look for that in the people you hire? How do you create that?

SPEAKER_01

Everything you said. Definitely the fact that I spent several years in my career in research means that I'm used to thinking about the state of the art as something that is ephemeral. And eventually it will be replaced by something much better. The truth is it doesn't only happen in AI research. It happens as well in engineering. There are new abstractions, new libraries, new frameworks. And if you want to stay up to date, you don't fossilize on something that lasts for a decade. You're always looking to improve yourself, to improve your product. So Amjad and I both really embraced that kind of philosophy. Myself now by running the entire engineering and product roadmap. Everyone knows how I like to work. Everyone has seen what made us successful. I think at this point it permeates the entire company. I don't even have to remind it anymore.

SPEAKER_00

I recall, for instance, when I changed my role and I tried to address the entire company, telling them what I wanted us to accomplish. I said this explicitly. I talk about not being emotionally attached. I talk about accepting the fact that some projects will fail. Now I don't even have to say it anymore. I think everyone embraces it and everyone loves it. It's way more exciting. Every day you know they are going to be shaping something that will really determine our future as a company. It's also hard though, right? Because you have users using your old products.

SPEAKER_01

Yeah. [SPEAKER_00] How do you manage the now many millions of users that are using your existing product with the next version of agent? And where do you think about that balance?

SPEAKER_01

Likely we have one common denominator with agents, which is the fact that you interact in natural language. So regardless how much you revolutionize the product surface or how many new features you add, the transition from one version to another is never completely something that gets you completely lost in the process. So the primitives aren't completely changing anymore. Primitives are there. The expectation of you creating software is there. The fact that agent will reply back to you in more or less advanced ways has just kept evolving gradually over time. So we're not going from an interface based on natural language to something that's just point and click. But absolutely, it requires sometimes to do these transitions.

SPEAKER_01

[SPEAKER_00] So regardless of how much you revolutionize the product surface or how many new features you add, the transition from one version to another is never completely something that gets you completely lost in the process. So the primitives aren't completely changing anymore. Primitives are there. The expectation of you creating software is there. The fact that an agent will reply back to you in more or less advanced ways has just kept evolving gradually over time.

SPEAKER_01

So we're not going from an interface based on natural language to something we're just pointing and clicking. But absolutely, it requires sometimes to do these transitions. And every single time we go through a big launch, I have to ask myself, shall we keep around both experiences and then allow for a smooth transition? Or shall we just have a clear cut? It's a muscle that you have to train in a company and you always have to balance between innovation and smooth transitions. I tend to be more of a fan of innovation.

SPEAKER_01

You have an impossible product job, which is that you're building a product for now 50 million users. 500,000 of them are professional users, but your median user is the vibe coder type user. You have enterprises that are using your product. At any other company, this kind of range would be product sabotage. [SPEAKER_00] You're trying to build a product for basically everyone. How do you do that? Just walk me through your product development process.

SPEAKER_01

[SPEAKER_00] First of all, you need to grow a very thick skin. Because the amount of feedback you will be getting is honestly overwhelming. In a good way. I keep telling everyone at the company, it's great that people care about it. Even if, of course, we are biased on reading just the negative feedback. Because the positive feedback doesn't last long. You feel supported, you feel loved, and then you go back to the things that are actually broken.

SPEAKER_01

But similar to changing the mindset of the company, being very innovation focused and not getting emotionally attached, I don't think we get offended anymore by feedback because we have seen oftentimes users literally flipping in a few hours between hide this product, it has become trash because they've done these wrong choices. And then we actually got them through the different user journey they had in mind and then suddenly they love it.

SPEAKER_01

So it's necessary to be able to digest feedback correctly and react in the right way. It's also a blessing because we are building a product for humans. We're not building a product that tries to replace humans with AI, which I think is a very dystopian view of the future. And if we didn't have people who care so deeply about it, our product would be strictly worse today.

SPEAKER_01

So yes, it's hard, but we are in a place where I never feel lost about what we should be doing next. If anything, if I could 10x the size of the company overnight and make it run as smooth as it does today, I will absolutely do it. That's as much work as we have in our hands today that we could potentially accomplish. Do you find that your different types of users all love the same things about Replit or is it different things?

SPEAKER_01

Different things, especially now that we launched Replit Agent 4 back in March this year. We put a lot of emphasis on the fact that it's a general purpose agent. Yes, building applications, building automations is the reason why most of the users know and love Replit. [SPEAKER_00] But the truth is you also come to Replit to interact with datasets and extract insights, to generate slides, to generate animation videos, and this list of additional features and skills will keep growing over time.

SPEAKER_01

And so I think it's a really important thing. It forces us though to really have as minimalistic as possible an approach to how the product should look like. We don't try to tailor it towards a specific type of persona. And one of the principles we follow internally is that I don't want to hear the word persona during a meeting basically. I want to say we should optimize this for PMs or we should optimize this for designers. Not because I don't allow them as a user base. I think I actually mentioned two early adopters among all the other types of employees in a company that really embrace Replit as early as possible. But rather because everyone out there, every single knowledge worker has a need to actually create software. So why would we only make it bespoke just for PMs or designers?

SPEAKER_01

And it's served us well until now because that's the reason why we keep growing so fast. So in your product development process, are you looking for the things that every knowledge worker needs across every persona versus, hey, here's the thing specifically that a PM needs or specifically that a designer needs? Are you trying to look for those foundational primitives? I think many would be curious to understand how you even follow that process. What does your UXR look like?

SPEAKER_01

[SPEAKER_00] So our UXR, we go straight to both our power users and the new adopters, early adopters, especially seeing how they evolved on the platform. And we almost do open-ended interviews. So we ask them, why do you love Replit? Why have you been building for several months?

SPEAKER_01

Why did you leave Replit? What made you move away somewhere else? Why did you leave a lot of clients? And based on all the signals that we collect, then we simply try to make their life easier on the product. Scrubbing away what they do for a living. We talk about them by first name. I don't say Louis PM. I say this person told us X, Y, and Z. And their concerns are valid. And we should be fixing our publishing pain. The tone of the agent should be different. This interface is too overwhelming. And we keep making steady progress based on all the feedback that we believe to be useful.

SPEAKER_01

And you'd be surprised. There are always universal themes across the feedback that we receive. A good product is good across a very large user base. In our case, a good product is good across a large user base. I love that. I heard you had one sales rep when you shipped Replit Agent? Yeah, for quite a while. [SPEAKER_00] I'm sure you didn't. And then we had two and then both of them had amazing performance during that year. What is the profile of the sales reps that you hired today? Are they traditional enterprise reps? Are they product-minded folks? The vast majority are not.

SPEAKER_00

And many of them have never done sales in their lives. We try to find people that love the product, can resonate with it, used it, and have fallen in love with it. [SPEAKER_01] I love that.

SPEAKER_01

I heard you had one sales rep when you shipped Replit Agent?

SPEAKER_00

[SPEAKER_01] Yeah, for quite a while. I'm sure you're not. And then we had two and then both of them had amazing performance during that year. What is the profile of the sales reps that you hired today? [SPEAKER_01] Are they traditional enterprise reps?

SPEAKER_01

Are they product minded folks? The vast majority are not. [SPEAKER_00] And many of them have never done sales in their lives. [SPEAKER_00] We try to find people that love the product, can resonate with it, abused it, have fallen in love with it. [SPEAKER_00] Perhaps have built something useful in their past job. Or some of them are reaching out to us because they realize I want to sell Replit. It's so amazing that I want to be part of the journey. Those folks end up being our top performers. And I think what a buyer wants to hear on the other side, especially today where testing AI products is not that hard. So all of them already got their hands dirty with Replit.

SPEAKER_01

They don't want to hear the basics. They want to hear, why should I buy it? And there's no better person to tell you why than someone who's actually been impacted positively by it. So those are our best sales reps. And I think we're going to keep using this philosophy because they need to have a good combination of storytelling and being technical to an extent. They need to really know the nitty gritty details of why Replit is so powerful. The nice thing about sales is that it's such a meritocratic thing. You can literally measure whether someone's good or not.

SPEAKER_01

Even if that goes against traditional sales knowledge of hiring your traditional sales folks, the proof is in the pudding. [SPEAKER_00] So everyone's trying to figure out pricing in the time of AI.

SPEAKER_00

I talked to a ton of users who are actually thinking of turning off their free trials, for example, because it costs too much compute to serve these users who may not ever convert. You all have a free tier.

SPEAKER_01

[SPEAKER_00] You have your somewhat standard SaaS tiers right now, SaaS pricing tiers. [SPEAKER_00] How well is that pricing model working for you? [SPEAKER_00] The tiers that we have are fairly standard, as you said. [SPEAKER_00] We have our core seat, which is running like $20 a month. [SPEAKER_00] And then we have a pro seat that has a few more features that are very important, including premium support and better SLAs of the infrastructure. [SPEAKER_00] And then from there, you step over to the enterprise plan. I would say that only describes part of our pricing philosophy because the truth is agents fundamentally have to be charged by usage.

SPEAKER_01

So by something proportional to the amount of compute that they are using or, in the AI jargon, the amount of tokens that they are burning.

SPEAKER_01

So we embrace usage-based billing very early. And I think by being trailblazers in this space, also when we launched the first version of the product, we were also trailblazers on when we changed the pricing model. And as you can expect, the backlash from the community during those couple of months was fairly substantial. But I think over time they realized why.

SPEAKER_00

[SPEAKER_01] Now you look around the industry, everyone is doing exactly the same. [SPEAKER_01] And we had to do it because of how agents work, it is relatively impossible to predict how much they're going to be running for a specific task.

SPEAKER_01

So any level of fixed pricing or even purely subscription-based wouldn't scale for a company like ours. And the fact that agents are more and more widespread means there is better acceptance of us having to charge in that way. And that said, if I had a magic wand, I would love to figure out outcome-based pricing for our product. The reason why it's going to be practically a research project for us is because the variety of tasks that you can accomplish on Replit is such that how can you even come up with a rate list of the different outcomes that you can have on Replit? So, yes, I know that we want to get there.

SPEAKER_01

[SPEAKER_01] I do know that it will come with a certain amount of goodwill from the community when we actually accomplish it. But I also know that it's not a short-term goal that we can crack very easily. And I think we need to see some step function improvement on how the entire AI models are working today in order to make that happen. Outcome-based pricing has somewhat similar issues with normal tier pricing because you wouldn't be pricing it based off of usage, right? You'd be basing it off of outcome.

SPEAKER_01

It's based on outcomes, but at least it doesn't have that perception of non-determinism where you just write a prompt and you don't know how much you're going to be charged. [SPEAKER_00] Now, we try to put some more guardrails on our product experience today. We have three different agent modes: light, economy, and power. And respectively, each one comes with certain expectations on how autonomously the agent will be running. In light mode, we do our best to have a very narrowly scoped task. Even if you write way too much, we carve out the core of your request and we don't allow the agent to run for long.

SPEAKER_01

If you're in power mode and you write a very long PRD, off to the races, the agent will run literally for several hours and put in front of you the entire work being done. Expert users are amazed by that. And a lot of them are using our agent exactly in that way. But if that's one of your first experiences in the product, of course, it can be confusing why you spend so much money and why the agent did more than you expected. So the alignment between user expectations and what the agent accomplishes is one of the hardest AI problems that not only us, but the entire field faces today.

SPEAKER_00

[SPEAKER_01] And you guys are thinking not only obviously about monetizing your own product, but you also want to help users monetize their applications that they build on Replit. [SPEAKER_01] When did that become a part of the product? How do you guys think about building that infrastructure? [SPEAKER_01] We always wanted to do it because the goal of this is to have our users coming to Replit to get work done and then being able to create products that they can monetize, create new sources of income, start a new company.

SPEAKER_00

[SPEAKER_01] That is what we are really passionate about. Those are the success stories for us. Now, in order to make payments easier, of course, the first area that we explored was Stripe integration that we work on together in the last few months. [SPEAKER_01] We launched that in Q4 last year and it has been growing at an exceptional rate, which makes me both proud for how easy we made it together in our product, as well as proud of our users that are really crushing it.

SPEAKER_01

Those numbers going up really make me understand that the vision we always had for our product is going to be something I'm going to be passionate about for my entire lifetime.

SPEAKER_00

[SPEAKER_01] That is what we are really passionate about. Those are the success stories for us. Now, in order to make payments easier, of course, the first area that we explored was Stripe integration that we work on together in the last few months. We launched that in Q4 last year and it has been growing at an exceptional rate, which makes me both proud for how easy we made it together in our product, as well as proud of our users that are really crushing it. Those numbers going up really make me understand that the vision we always had for our product is going to be something I'm going to be passionate about for my entire lifetime. It's not just building our company. We are actually going to become enablers for an ecosystem of companies.

SPEAKER_01

[SPEAKER_00] You're a big proponent of the one person billion dollar company, the one person unicorn. Yes. [SPEAKER_00] I would love to talk to you about this because I've always found this framing confusing because wouldn't you assume for as long as adding a human incrementally improves your company, competitive forces will always require companies to want to add more humans. It seems hard to stay with the one person billion dollar company if everyone can start a one person billion dollar company, but a 10 person, 10 billion dollar company. So where does this sit with you?

SPEAKER_01

I see your point. I don't think there should be a reward for someone who carries the burden of creating a massive company just on their shoulders. I think our point is more about how you can scale a business really fast on your own orchestrating a lot of agents way better than you could even imagine six months ago. Because the truth is, even if you are a researcher on the bleeding edge and you really understand exponentials and you're predicting what the future is going to be, mid last year, I don't think many people would have thought a one billion dollar single person company will actually exist. And here we are, we're starting to see a few entrepreneurs that are on track to make that happen.

SPEAKER_01

[SPEAKER_00] On Replit?

SPEAKER_01

On Replit, yes. So it's going to happen. And of course, there will be maybe hiring more people and maybe it's going to stop at say 250 million ARR instead of one billion. But it's still very exciting to know that if you have the entire product vision in your mind and you want to be as effective as possible with little resources at the beginning, you can actually do it. What matters is that we're removing the friction and the inertia to make that happen. And then, of course, more people will be hired and these businesses will become bigger. But what I think will generate wealth in the era that we're creating with AI is the fact that way more people will become entrepreneurs. Way more companies will be born for that reason.

SPEAKER_01

[SPEAKER_00] What is going to be the scarce good in that new era? If there's an abundance of software and coding and entrepreneurs, where are the bottlenecks going to be?

SPEAKER_01

I think it's real good business ideas. What should we be working on? It's easy to test them. If anything, it's never been as easy to test ideas. But if you go and see through the long tail of unsuccessful business ideas tried on Replit and other similar coding agents, most of them, you wouldn't be shocked why they don't work out. Maybe they target a very small niche. Maybe they were just focusing on attacking a problem rather than something that the users need. And it's very similar to what we're facing before launching the first version of our agent. You're left wondering, do people really care about this? Should we build it or is it a waste of time? So by far, that's always going to be the hardest question that you're going to be asking yourself. And the fact that way more ideas can be tested, on one hand, will perhaps lower the value of the average good business idea. On the other hand, if you strike gold, you're going to get there much faster than you ever did in your life. So even more than in the past, I think I will encourage people to not stop at the first failure attempt. It's something very common people say about startups. You shouldn't wear it as a badge of honor, but it's fine to fail as long as you stand up again and try. Now, the frequency which you have to do that is going to be even more than before. A lot of folks growing thick skin.

SPEAKER_01

[SPEAKER_00] Will agents be able to tell us whether a business idea is a bad idea?

SPEAKER_01

That's a great question. I think to a first approximation, yes, why not? Especially if it's a business that has some data aspects where you can go out and do some level of marketing analysis and then you can run user studies on test labs and you can completely automate that. So I do believe that at least the first steps can be completely automated with agents. Then I love to think that humans and taste will still play an important role. But one of the moonshot projects that we have at Replit that Amjad and I talk about all the time is, can we go from a prompt to an agent that basically on every single day tries to craft a business idea at zero to one and then puts it in front of an MVP that is already running, is already looking for customers, is already shooting emails and running ads on social media and see, okay, how fast is this idea taking off? Is it worth pursuing or not? So the zero to MVP and the basic steps required to run a business, like including incorporating an LLC with Stripe Atlas, all of that can be easily automated. From there to understanding where should I bet, where should I be putting managers, I do think that we as humans will still play a key role because after all, a lot of products will still be tailored towards being sold to other humans. And I love to believe that we understand each other much better than machines do.

SPEAKER_01

[SPEAKER_00] In your manifesto that you wrote, where you sort of wrote out Replit Agent V0 or V1 and how it looks now, what does it look like in three, four or five years?

SPEAKER_01

We have been creating basically applications since the first time we launched our agent. So websites, web apps, automations, always software that often requires a human on the other end to interact with it. And if you think about it, user interfaces existed in the last decades just because we haven't found a better way to have humans interacting with machines until a few months ago, where agents became powerful enough where perhaps even yourself in your daily job today, you probably do quite a lot of interactions that are not point and click based on your keyboard. You're dictating to your phone and that prompts end up somewhere in an agent that runs some job in the background and then gives you back some results that you care about. I believe that most of the interactions we're going to be having with computers in the near future

SPEAKER_01

with it. And if you think about it, user interfaces existed in the last decades, just because we haven't found a better way to have humans interacting with machines until literally a few months ago, where agents became powerful enough, where perhaps even yourself in your daily job today, you probably do quite a lot of interactions that are not point and click based on your keyboard. You're dictating your phone and that prompts end up somewhere in an agent that runs some job in the background and then gives you back some results that you care about.

SPEAKER_01

I believe that most of the interactions we're going to be having with computers in the near future will be based on that pattern, not anymore interacting with UI. So what does it mean? In the short term, we're probably going to see even more applications being created. Good for us, we're amazing app builders. So more of them will be born, especially because Replit exists. And in parallel, though, I think we are already experiencing an exponential growth of how many agents are being built on the market. I even see today inside Replit, a lot of the work that we do is not anymore about creating dashboards or creating workflows. We have agents that scour our entire company and get the job done. And then there is a human assessing the results that they obtain. I expect more and more people wanting to build that. So the future that I see for us as a product is not just building applications, but it's allowing everyone to create very powerful and advanced agents with the same ease of use that we got everyone used to with Replit Agent when it comes to creating software. Amazing. I'm excited for that future. Likewise. So Replit is known as a very high intensity mission driven culture. How do you hire for intensity? Wow, that's a good one. I've been asked by my Italian team several times how I would do that. On one end, I think I learned to recognize some patterns from the people that we have internally that are amazing performers. And then I go and look for some pattern matching with the new candidates that I get to meet during the interview process. Definitely something that helps is to find out if they were former founders. And I would say maybe today in our engineering team, more than 40% of the members are actually former founders. So including myself, that always tells you something about that person. You know, they've been crazy enough, at least for a window of time in their career to believe so much in themselves and have a lot of agency and ownership to try to do zero to one on something. It doesn't really matter if they've been extremely successful or not, but that trait, the level of intensity is there. And then a lot of other folks that are part especially of the team—I want to see that they work on projects that they care about. And you'd be surprised. If during an interview, you ask someone to tell you in depth about a technical project that you worked on, what are the hard choices that you made, what kind of confrontation you have to have with your colleagues—you immediately realize if they cared about it or not. And I want to see more of that than how good they are technically today.

SPEAKER_01

On one end, because interviews are always faulty in terms of the signal you collect. Not everyone is amazing at doing interviews. And then they show up and they're absolutely incredible engineers. So I only partially rely on the technical interviews, but also because I want to see exactly that behavior once they come and wrap it. We give an incredible amount of agency scope to every single IC from day one. So they need to love that level of responsibility, not be overwhelmed about it. Going back to the 2024 launch, I think 2024 revenue at the beginning of the year was what, two and a half million? [SPEAKER_00] Yeah.

SPEAKER_00

[SPEAKER_01] And what was that at the end of the year?

SPEAKER_01

At the end of the year, I think we were roughly 10, if I recall correctly. Okay. So we're in the ballpark. Yes. And if you can share, what are you guys at now? What's your run rate for 2026? Let's say we're on track for 1 billion at the end of this year. And I'm starting to tell the company I'd be disappointed if we hit that in December or not earlier. [SPEAKER_00] Congrats. [SPEAKER_00] Thank you. Yeah. What are the metrics that you obsess over at this point with this kind of exponential growth?

SPEAKER_01

I care a lot about engagement. Engagement tells me if users are actually finding value in our product or not. Yes, of course, we care about how many subscribers we have. If they keep using the product on a monthly basis, I care about how many applications they build, how many of them they also publish online. But after knowing that they find Replit more and more useful and they come back several times a week, that's what tells me that we built something magic for them and something that makes them more productive. So you're looking at daily active users, weekly active users? Correct. D7, yeah, these kinds of metrics.

SPEAKER_00

[SPEAKER_01] Are you looking at any sort of surprising engagement metrics or metrics that other companies might not look at, like what parts of the product they're spending time on?

SPEAKER_01

I think something is very peculiar to Replit because we care about the software creation lifecycle end to end—is the fact that we also have a deployment product. So you can take whatever you create on Replit and then publish it. And it could be within your company, so an internal tool. It could be a link that you share only with a few friends. It could be something that you put online for everyone and it goes viral. And the beauty is you don't even realize because everything scales magically for the users. For me, that's one of the most exciting metrics to look at on a daily basis because it tells me, did they build something that they care so much about that they're willing to spend more money to actually publish it? Thumbs up, thumbs down, feedback—everything is valuable—but nothing speaks as loud as money. If you're willing to invest more of your money because you're proud of something that you built, then I know that we actually made something right. And if I need to dig deeper on what our agent is doing correctly or some debug that we have, I always put more emphasis on the agent traces that brought us to a deployment rather than those who didn't, because I know that's what a user considers end-to-end work. It's a unit of valuable work for them to be done. Yeah. Amazing. Favorite Replit app?

SPEAKER_01

that they're willing to spend more money to actually publish it? Thumbs up, thumbs down, feedback, everything is valuable, but nothing speaks as loud as money. If you're willing to invest more of your money because you're proud of something that you built, then I know that we actually made something right. And if I need to dig deeper on what our agent is doing correctly or some debug that we have, I always put more emphasis on the agent traces that brought us to a deployment rather than those who didn't, because I know that that's what a user considers end-to-end work. It's a unit of valuable work for them to be done. Yeah. Amazing. Favorite Replit app? I have to mention things that we build internally because they're absolutely amazing. I hired a very small team of full-time AI coders. And I called the lead of the team, my AI chief of staff. And the reason is, it's someone who has to embed himself across every single team of the company, spending a couple of weeks with them, understanding what they need to build. And they don't have bandwidth because they're already working full steam on their current project.

SPEAKER_01

[SPEAKER_00] goes back to his desk and builds with rapid, exactly the internal tool that makes a difference for them. So my support team has a very standard ticketing system that we always use for many years. And then on top of that, we built a very futuristic dashboard that tells us everything about trends in terms of sentiment of our user base and what our response rate is, if we're doing well on enterprise versus our pro plan versus our core plan. And we have all these dashboards scattered across the company. So exactly for doing well or not, among all the different teams.

SPEAKER_01

We have another tool for our HR team where we have our entire internal data index, our org chart, our desk positioning. The moment you onboard, you just go on this tool and you can get everything done in one place. I love to talk about them because oftentimes people think that AI coding is just creating a small side project. You can run a company with these tools, literally. And we do it. We believe so much in it that first we build them and then realize we have to sell this to enterprise. It actually works extremely well. Do you buy any external software at this point or is it? Very little. I don't want to be part of this, maybe. The SaaS apocalypse. Yes. Doomsdayers.

SPEAKER_01

I think it's psychosis. Also, we have seen the public market readjusting after the drop and I'm not surprised that it happened. Of course, there are very large SaaS vendors that do have a reason to exist. And I will dare to say most of their value is not in the software that they built. It's in the business processes that they refine over many years.

SPEAKER_00

[SPEAKER_01] The domain expertise.

SPEAKER_00

[SPEAKER_01] The enterprise domain expertise. The system of records that they own. There is a lot of value that is not in the lines of code. But conversely, there is a very long tail of small SaaS vendors where by definition they can't really customize what you need to your needs because they need to build a generic tool. And all of them we have really built ourselves or we didn't even buy. As the company grew and we realized that we needed something, I didn't even have to ask most of the times. The specific team that needed something to be built already created that on Replit. We never bought a single software to do order forms. It was built on Replit from day one with a much older version of the agent. If you see today, it's a very professional tool. Sometimes we wonder maybe we should start to sell it as a side project, you know.

SPEAKER_00

[SPEAKER_01] Your next vertical. [SPEAKER_01] Exactly. Who knows. [SPEAKER_01] How much of Replit, the user facing product, are you building with Replit?

SPEAKER_00

[SPEAKER_01] At this point, it's quite a lot. Our designers probably spend 80-90% of their time building Replit on Replit. So they do all the different design variations. We have the entire design guidelines of our product, the branding, everything in one place. And they just iterate there. Which is amazing because when I show up at the other product sync or a design sync, I see them updating based on the feedback you give them real time. You know, they write their prompts and we see the interface changing and then we make decisions on the spot. The cycle that it takes to make product improvements probably went down by an order of magnitude easily.

SPEAKER_01

So much more satisfying, right? It's just like the software is fun again. Yeah, it's malleable, you know. You can change it on the spot. It's amazing. Michele, thank you so much. This was fantastic. Congrats on everything you guys are building. And this was great. [SPEAKER_00] Thanks for having me. [SPEAKER_00] Thanks for having me. Myself now by, you know, running the entire engineering and product roadmap. Everyone knows how I like to work. Everyone has seen what made us successful. I think at this point it permeates the entire company. I don't even have to remand it anymore.

SPEAKER_01

I recall, for instance, when I changed my role and I tried to address the entire company, telling them what I wanted us to accomplish. I said this explicitly. You know, I talk about not being emotionally attached. I talk about accepting the fact that some projects will fail. Now I don't even have to say it anymore. I think everyone embraces it and I think everyone loves it. It's way more exciting. Every day you know they are going to be shaping something that will really determine our future as a company. It's also hard though, right? Because you have users using your old products. Yeah.

How do you manage sort of the now many millions of users that are using your existing product with sort of the next version of agent? And where do you think about that balance? Likely we have one common denominator with agents, which is the fact that you interact in natural language.

SPEAKER_00

So regardless how much you revolutionize the product surface of how many new features you add, is the transition from one version to another is never, you know, completely, is never something that gets you completely lost in the process.

SPEAKER_01

So the primitives aren't completely changing anymore. Primitives are there. The expectation of you creating software is there. The fact that agent will reply back to you in more or less advanced ways has just kept evolving

SPEAKER_00

gradually over time.

SPEAKER_01

So, you know, we're not going from an interface is based on natural language to something we're just point and click. But absolutely, it requires sometimes to do these transitions. And every single time we go through a big launch, I have to ask myself, shall we keep around both experiences and then allow for like a smooth transition? Or shall we just have a clear cut? It's a muscle that you have to train in a company and you always have to balance between innovation and, you know, smooth transitions. I tend to be more of a fan of innovation as you can imagine. You have kind of an impossible product job, which is that you're building a product for now 50 million users.

SPEAKER_01

500,000 of them are professional users, but your median user is the sort of like the vibe coder type user. You have enterprises that are using your product. At any other company, this kind of range would be product sabotage.

SPEAKER_00

You're trying to build a product for basically everyone. How do you do that? Like, just walk me through your product development process. First of all, you need to grow a very thick skin. Because the amount of feedback you will be getting is honestly overwhelming. In a good way. Like I keep telling everyone at the company, it's great that people care about it.

SPEAKER_01

Even if, of course, we are biased on reading just the negative feedback. Because the positive feedback doesn't last long, you know? You feel supported, you feel loved, and then you go back to the things that are actually broken. But, you know, similar to changing the mindset of the company, being very innovation focused and not getting emotionally attached, I don't think we get offended anymore by feedback because we have seen oftentimes users literally like flipping in a few hours between hide this product, it has become trash because they've done these wrong choices.

SPEAKER_01

And then, you know, maybe we actually got them through the different user journey they had in mind and then suddenly they love it. So it's necessary to be able to digest feedback correctly and react in the right way. It's also a blessing because we are building a product for humans. We're not building a product that tries to replace humans with AI, which I think is a very dystopian view of the future. And if we didn't have people who care so deeply about it, our product would be strictly worse today. So, yes, it's hard, but we are in a place where I never feel lost about what we should be doing next.

SPEAKER_01

If anything, if I could 10x the size of the company overnight and make it run as smooth as it does today, I will absolutely do it. That's as much work as we have in our hands today that we could potentially accomplish. Do you find that your different types of users all love the same things about Replit or is it different things? Different things, especially now that we launched Replit Agent 4 back in March this year. We put a lot of emphasis on the fact that it's a general purpose agent. Yes, building applications, building automations is the reasons why most of the users know and love Replit.

SPEAKER_00

But the truth is you also come to Replit to interact with data sets and extract insights,

SPEAKER_01

to generate slides, to generate animation videos, and this list of additional features and skills will keep growing over time. And so, I think it's a really important thing. It forces us though to really have a as minimalistic as possible approach to how the product should look like. We don't try to tailor it towards a specific type of persona. And one of the principles we follow internally is to, I don't want to hear the word persona during a meeting basically. I want to say we should optimize this for PMs or we should optimize this for designers. Not because I don't allow them as a user base.

SPEAKER_01

I think I actually mentioned two early adopters, you know, among all the other types of employees in a company that really embrace Replit as early as possible. But rather because everyone out there, every single knowledge worker has a need to actually create software. So why would we only make it a bespoke just for PMs or designers, let's say? And it's served us well until now because that's the reason why we keep growing so fast. So in your product development process, are you looking for the things that every knowledge worker needs across every persona versus like, hey, here's the thing specifically that a PM needs or specifically that a designer needs?

SPEAKER_01

Are you trying to look for those foundational primitives? I think many would be curious to understand how you even follow that process.

SPEAKER_00

What does your UXR look like? So our UXR, we go straight to both our power users, to the new adopters, early adopters, especially like seeing how they evolved over on the platform. And we almost learn like open-ended interviews. So we ask them, why do you love Replit? Why have you been building for several months?

SPEAKER_01

Why did you leave Replit? What made you move away somewhere else? Why did you leave a lot of clients? And based on all the signals that we collect, then we simply try to make their life easier on the product. Scrubbing away what they do for a living. We talk about them in first name. I don't say, you know, like, Louis PM. I say, this person told us X, Y, and Z. And their concerns are valid. And we should be fixing our publishing pain. The tone of the agent should be different. This interface is too overwhelming. And we keep making steady progress based on all the feedback that we believe to be useful. And you'd be surprised.

SPEAKER_01

There are always like a universal team across the feedback that we receive. Like a good product is good across a very large user base in our case. A good product is good across a large user base. I love that. I heard you had one sales rep when you shipped Replit Agent? Yeah, for quite a while.

SPEAKER_00

I'm sure you're not. And then we had two and then both of them, you know, had amazing performance during that year. What is the profile of the sales reps that you hired today?

SPEAKER_01

Are they traditional enterprise reps? Are they like product minded folks? The vast majority are not.

SPEAKER_00

And many of them have never done sales in their lives. We try to find people that love the product, can resonate with it, abused it, have fallen in love with it. Perhaps have built something useful in their past job.

SPEAKER_01

Or some of them are reaching out to us because they realize, I want to sell Replit. It's so amazing that I want to be part of the journey. Those folks end up being our top performers. And I think what a buyer wants to hear on the other side, especially today where testing AI products is not that hard. So all of them already got in their hands dirty with Replit. They don't want to hear the basics. They want to hear, why should I buy it? And there's no better person to tell you why than someone who's actually been impacted positively by it. So those are our best sales reps.

SPEAKER_01

And I think we're going to keep using this philosophy because they need to have a good combination of storytelling and being technical to an extent. Like they need to really know the nitty gritty details of why Replit is so powerful. The nice thing about sales is such a meritocratic thing. You can literally measure whether someone's good or not. Like even if you know sort of that that goes against traditional sales knowledge, I'd say, of like hiring your traditional sales folks, the proof is in the pudding.

SPEAKER_00

So everyone's trying to figure out pricing in the time of AI. I talked to a ton of users who are actually thinking of turning off their free trials, for example, because it costs too much compute to serve these users who may not ever convert. You all have a free tier. You have your sort of, you know, I guess somewhat like standard SaaS tiers right now, SaaS pricing tiers. How well is that pricing model working for you? The tiers that we have are, I would say, fairly standard, as you said. You know, we have our core seat, which is running like $20 a month.

SPEAKER_00

And then we have a pro seat that has like a few more features that are very important, like including premium support, better SLAs of the infrastructure. And then from there, you step over to the enterprise plan.

SPEAKER_01

I would say that only describes the part of our pricing philosophy because the truth is agents fundamentally have to be charged by usage. So by something proportional to the amount of compute that they are using or, you know, in the AI jargon, the amount of tokens that they are burning. So we embrace usage-based billing very, very early. And I think by being like trailblazers in this space also when we launched the first version of the product, we were also trailblazers on when we changed the pricing model. And as you can expect, the backlash from the community during those couple of months was, you know, was fairly substantial. But I think over time they realized why.

SPEAKER_01

Now you look around the industry, everyone is doing exactly the same. And we had to do it because by means of how agents work, it is relatively impossible to predict how much, you know, they're going to be literally running for a specific task. So any level of fixed pricing or even purely based on subscription wouldn't scale for a company like ours. And the fact that agents are more and more widespread means that, you know, there is better acceptance of us having to charge in that way. And that said, if I had a magic wand, I would love to figure out outcome-based, you know, pricing for the model, for our product.

SPEAKER_01

The reason why it's going to be practically a research project for us is because the variety of tasks that you can accomplish on Rebbit is such that, how can you even come up with like a rate list of the different outcomes that you can have on Rebbit? So, yes, I know that we want to get there. I do know that it will come with a certain amount of goodwill from the community when we actually accomplish it. But I also know that it's not a short-term goal that we can crack very easily. And I think we need to see some like step function improvement on how the entire AI models are working today in order to make that happen.

SPEAKER_01

Outcome-based pricing has like somewhat similar issues with sort of just like normal tier pricing because you wouldn't be pricing it based off of usage, right? You'd be basing it off of outcome. It's based on outcomes, but at least it doesn't have that perception of non-determinism where, you know, you just write a prompt and you don't know how much you're going to be charged.

SPEAKER_00

Now, we try to put some more guardrails on our product experience today. Like we have three different agent modes, light, economy, and power. And respectively, each one of Dengs come with certain expectations on how autonomously the agent will be running. Like in light mode, we do our best to have a very narrowly scope task.

SPEAKER_01

Even if you write way too much, we carve out the core of your request and we don't allow the agent to roll out for long. If you're in power mode and you write a very long PRD, off to the races, you know, the agent will run literally for several hours and put in front of the entire work being done. Expert users are amazed by that. And a lot of them are using our agent exactly in that way. But if that's one of your first experiences in product, of course, it can be, you know, confusing why you spend so much money and why the agent did maybe more than ever expected. So the alignment between

SPEAKER_01

user expectations and what the agent accomplishes is one of the hardest AI problems that not only us, but the entire field faces today. Yeah. And you guys are thinking not only obviously about monetizing your own product, but you also want to help users monetize their applications that they build on Replit. When did that become a part of the product? How do you guys think about building that infrastructure? We always wanted to do it because the goal of this is to have our users coming to Replit to get work done and then being able to create products that they can monetize, create new sources of income, start a new company.

SPEAKER_01

That is what we are really passionate about. Those are the success stories for us. Now, in order to make payments easier, of course, you know, the first area that we explored was Stripe integration that we work on together in the last few months. We launched that in Q4 last year and it has been growing at an exceptional rate, which makes me both proud, you know, for how easy we made it together in our product, as well as proud of our users that are really crushing it. Those numbers going up really make me understand that the vision we always had for our product is going to be something I'm going to be passionate about, you know, for my entire lifetime.

SPEAKER_01

It's not just building our company. We are actually going to become enablers for like an ecosystem of companies.

SPEAKER_00

You're a big proponent of the one person billion dollar company, the one person unicorn. Yes. I would love to talk to you about this because I've always found this framing kind of confusing because wouldn't you assume for as long as adding a human sort of incrementally improves your company,

SPEAKER_01

competitive forces will always require companies to want to add more humans. It seems like hard to just sort of like stay with the one person billion dollar company if everyone can start a one person billion dollar company, but like a 10 person, 10 billion dollar company. So where, you know, where does this sit with you? I see your point. I don't think there should be like a reward for someone who carries the burden of creating like a massive company just on their shoulders. I think our point is more about you can scale a business really fast on your own orchestrating a lot of agents way better than you

SPEAKER_01

could even imagine six months ago. Because the truth is, even if you are a researcher on the bleeding edge and you really understand exponentials and you're predicting what the future is going to be, mid last year, I don't think many people would have thought a one billion dollar single person company will actually exist. And here we are, like we're starting to see a few entrepreneurs that are on track to make that happen.

SPEAKER_00

On rapid? On rapid, yes. Yeah. So it's going to happen. And of course, you know, there will be maybe hiring more people and maybe it's going to stop at say 250 million ARR instead of one billion. But it's still like very exciting to know that if you have the entire product vision in your mind and you want to be as effective as possible with little resources at the beginning, you can actually do it. Like what matters is that we're removing the friction and the inertia to make that happen. And then, of course, more people will be hired and these businesses will become bigger. But

SPEAKER_00

what I think will generate wealth in the era that we're creating with AI is the fact that way more

SPEAKER_01

people will become entrepreneurs. Way more companies will be born for that reason. What is going to be the scarce good in that new sort of era? If there's an abundance of software and coding and entrepreneurs, where are the bottlenecks going to be? I think it's more and more real good business ideas. Like what should we be working on? It's easy to test them. If anything, it's never been as easy to test, you know, ideas. But if you go and see maybe through the long tail of unsuccessful business ideas, tried on Repplet and other similar coding agents, most of them, you wouldn't be shocked why they don't work out. You know, maybe they

SPEAKER_01

target a very small niche. Maybe they were just focusing on attacking our problem rather than something that the users need. And I mean, it's very similar to what we're facing before launching the first version of our agent. You're left wondering, do people really care about this? Like, should we build it or is it a waste of time? So by far, that it's always going to be the hardest question that you're going to be asking yourself. And the fact that way more ideas can be tested, on one

SPEAKER_00

hand, will even perhaps lower the value of the average good business idea. On the other hand,

SPEAKER_01

if you strike gold, you're going to get there much faster than you ever did in your life. So even more than in the past, I think I will encourage people to not stop at the first failure attempt. It's something very common people say about startups. You know, you shouldn't maybe wear it as a badge of honor, but it's fine to fail as long as you stand up again and try. Now, probably the frequency which you have to do that is going to be even more than before. A lot of folks growing a lot of thick skin. Oh, yeah. Will agents be able to tell us whether a business idea is a bad idea? That's a great question. I think to a first approximation, yes, why not? Especially on the,

SPEAKER_01

if it's a business that has some data aspects where, you know, you can go out and do some level, like a marketing analysis and then you can run user studies on listen labs and you can completely automate that. So I do believe that at least moving the first steps can be completely automated, you know, with agents. Then I love to think that humans and taste there will still play an important role. But one of the moonshot projects that we have a rapid that Amjad and I talk about all the time is, can we go from a prompt to an agent that basically on every single day tries to craft a business idea

SPEAKER_01

at zero to one and then puts it in front of the MVP that is already running, is already looking for

SPEAKER_00

customers, is already like shooting emails and running ads on social media and see, okay, how fast is this idea taking off? Is it worth pursuing or not? So the zero to MVP and the basic steps required to run a business, like including incorporating an LLC with Stripe Atlas, all of that can be easily automated.

SPEAKER_01

From there to understanding where should I bet, where should I be putting managers, I do think that we as humans will still play a key role because after me, a lot of products will still be tailored towards be sold to other humans. And I love to believe that we understand each other much better than machines do. So what was in your, this manifesto that you wrote, that you kind of, where, where, where you sort of wrote out, uh, Replit Agent V0 or V1 and you know, what it looks like now. What's, what does it look like in three, four or five years? We have been creating basically applications since the, the first time we launched our agent. So

SPEAKER_01

website, web apps, automations, always software that often requires a human on the other end to interact with it. And if you think about it, user interfaces existed in the last decades, just because we haven't found a better way to have humans interacting with machines until literally like a few months ago, where agents became powerful enough, where perhaps like even yourself in your daily job today, you probably do quite a lot of interactions that are not point and click based on your keyboard. You're dictating your phone and that prompts end up somewhere in an agent that that runs some job in

SPEAKER_01

background and then gives you back, you know, some results that you care about. I believe that most of the interactions we're going to be having with computers in the near future will be based on that pattern, not anymore, you know, interacting with UI. So what does it mean? In the short term, we're probably going to see even more applications being created. Good for us, you know, we're amazing app builders. So more of them will be born, especially, you know, because Reblet exists. And in parallel, though, I think we are already experiencing an exponential growth of how many agents are being built on the market. I even see today inside Reblet, a lot of the work that we

SPEAKER_01

do is not anymore about, say, creating dashboards or creating workflows. We have agents that scour our entire company and get the job done. And then like there is a human assessing, you know, the results that they obtain. I expect more and more people wanting to build that. So the future that I see for us as a product is not just building applications, but it's allowing everyone to create very powerful and advanced agents with the same use of use that we got everyone used to, you know, with Reblet agent when it comes to creating software. Amazing. I'm excited for that, that future. Likewise. So, uh, Reblet is

SPEAKER_01

known as a very high intensity mission driven culture. How do you hire for intensity? Wow, that's a good one. I, I've been asked by my Italian team several times I would do that. Um, on one end, I think I, I learned to recognize some

SPEAKER_00

patterns from the people that we have internally that are like amazing performers. And then I go and look for, I do some pattern matching basically with the new, with new candidates that I, that I get to meet during the interview process. Definitely something that helps is to find out if they were former

SPEAKER_01

founders. And I would say maybe today in our like engineering team, more than 40% of the members are actually former founders. So like including myself, that always tells you something about that person. You know, they've been crazy enough, at least for a window of time in their career to believe so much in themselves and have a lot of agency and ownership, you know, to try to do zero to one on something. It doesn't really matter if they've been extremely successful or not, but that trait, the level of, you know, uh, intensity is there. And then a lot of other folks that, you know, are

SPEAKER_01

part especially of the team. I, I want to see that they, they work on projects that they care about. And you'd be surprised. Like if during an interview, you ask someone to tell you, just go in depth about a technical project that you work on. What are the hard choices that you made? What kind of confrontation you have to have with your colleagues? You immediately realize if they cared about it or not. And I, I want to see more of that than how good the technically are today. On one end, because interviews are always faulty in terms of the single you collect. Now, not everyone is

SPEAKER_01

amazing at doing interviews. And then they show up and they're absolutely incredible engineers. So, I only partially rely on the, on the technical interviews, but also because I want to see exactly that behavior once they come and wrap it. We give an incredible amount of agency scope to every single IC from day one. So they need to love that level of responsibility, not, not be overwhelmed about it. Going back to that, you know, sort of the, the 2024 launch, I think 2024 revenue at the beginning of the year was what, two and a half million? Yeah. And, and what was that at the end of the year? At the end of the year, I think we were roughly 10, if I recall correctly.

SPEAKER_01

Okay. So we're in the ballpark. Yes. And, uh, if you can share, what are you guys at now? Uh, let's. What's your run rate for 2026? Let's say we're on track for 1 billion at the end of this year. And I'm starting to tell the company

SPEAKER_00

I'd be disappointed if we hit that in December or not earlier. Congrats. Thank you.

SPEAKER_01

Yeah. What are the metrics that you obsess over at this point with this kind of an exponential growth? I care a lot about engagement. Engagement tells me if users are actually funding value in our product or not. Yes, of course, we care about how many subscribers we have. If, you know, if they are the first thing with, uh, they're keep using the product on a monthly basis. I, I care about how many applications they build, how many of them they also publish online. But after me knowing that they find rapid more and more useful and they come back several times a week, that's what tells me that we build something magic for them and something that makes them more productive.

SPEAKER_01

So you're looking at like daily active users, weekly active users? Correct. You know, 2d7, you know, yeah, these kinds of metrics. Are you looking at any sort of surprising engagement metrics or metrics that, you know, other companies might not look like, look at like, you know, what parts of the product they're spending time on or? I think something is very peculiar to Rapid because again, we care about the software creation lifecycle end to end is the fact that we also have a deployment product. So you can take whatever you create on a rapid and then publish it. And it could be within your company. So like an internal tool,

SPEAKER_01

it could be a link that you share only with a few friends. It could be something that you put online for everyone and it goes viral. And the beauty is you don't even realize because everything out of scales magically for the users. Um, that for me is the, one of the most exciting metrics to take a look at on a daily basis because it tells me, did they build something that they care so much about that they're willing to spend more money to actually publish it? Like thumbs up, thumbs down, feedback, everything is valuable, but nothing speaks as loud as money. Like if you're willing to invest

SPEAKER_01

more of your money because you're proud of something that you built, then I know that we actually made something right. And, you know, if I need to dig deeper on what our agent is doing correctly or some debug that we have, I always put more emphasis on the agent traces that brought us to a deployment. rather than those who didn't, because I know that that's what a user consider like end-to-end work. It's like a unit of valuable work for them to be done. Yeah. Amazing. Favorite Replit app? I have to mention things that we build internally because they're absolutely amazing. I hired, um,

SPEAKER_01

a very small team of literally full-time by coders. And I called the lead of the team, my AI chief of staff. And the reason is, it's someone who has to embed himself across every single team of the company, spending like a couple of weeks with them, understanding what they need to build. And they don't have bandwidth because they're already working full steam on their current project,

SPEAKER_00

goes back to his desk and builds with rapid, exactly the internal tool that makes a difference for them. So my support team has like a very standard ticketing system that, you know, we, we always use for many years.

SPEAKER_01

And then on top of that, we built a very futuristic dashboard that tells us pretty much everything about

SPEAKER_00

trends in terms of sentiment or our user base. And what is the, our response rate, if we're doing well on enterprise versus our pro plan versus our core plan. And we have all these dashboards scattered across the company. So, you know, exactly for doing well or not, you know, among all the different teams.

SPEAKER_01

We have another tool for our HR team where we have our entire internet data index, our org chart, our desk positioning, like the moment you onboard, you just go on this tool and you can get everything done in one place. I love to talk about them because oftentimes people think that by coding is just creating like a small set project, you can run a company with these tools, like literally. And then we do it. You know, we, we, we believe so much in it that first we build them and then realize we have to sell this enterprise. You know, it actually works extremely well. Do you buy any external software at this point or is it?

SPEAKER_01

Very little. Like I'm not, I don't want to be part of these, you know, maybe. The sasspocalypse. Yes. Doomsdayers. I think it's kind of like psychosis. And also we have seen the public market readjusting after the drop and I'm not surprised that it happened. Of course, there are very large sass vendors that do have a reason to exist. And I will dare to say most of their value is not in the software that they built. It's in the business processes that they refine over many years. The domain expertise. The enterprise. Domain expertise. The system of records that they own. Like there is a lot of value that is not in

SPEAKER_01

the lines of code. But conversely is a very long tale of like a small sass vendors where by definition they can't really customize what you need to your needs because they need to build a generic tool. And all of them we have really built ourselves or we didn't even buy, you know, as the company grew and we realized that we needed something. I didn't even have to ask most of the times that the specific team that needed something to be built already created that on Replit. Like we never bought a single software to do like order forms. It was built on Replit from day one with a much older version

SPEAKER_01

of the agent. If you see today it's a very professional tool. Like sometimes we wonder maybe we should start to sell it as a side project, you know. Yeah. Your next vertical. Exactly. Who knows. How much of Replit, the user facing product, are you building with Replit? At this point it's quite a lot. Our designers probably spend 80-90% of their time building Replit on Replit. So they do all the different design variations. We have the entire design guidelines of our product, the branding, everything in one place. And they just iterate there. Which is amazing because when I show up at

SPEAKER_01

the other product sync or like a design sync, I see them updating based on the feedback. You give them real time. You know, they write their prompts and we see the interface changing and then we make decisions on the spot. The cycle that it takes to make product improvements probably went down by a order of magnitude easily. So much more satisfying, right? It's just like the software is fun again. Yeah, it's malleable, you know. You can change it on the spot. It's amazing. Michele, thank you so much. This was fantastic. Congrats on everything you guys are building. And this was great.

SPEAKER_00

Thanks for having me.

SPEAKER_00

Thanks for having me.