The Pomp Podcast - #1256 Amjad Masad | The Company Arming Coders With Superpowers

Episode Date: October 9, 2023

Amjad Masad is the co-founder & CEO of Replit. The conversation was recorded at the BUILD Summit in New York. In this conversation, we discuss the customer obsession that Replit has, importance of being technical, current state of artificial intelligence, culture of Replit, e/acc movement, and why we must accelerate the adoption, creation, & scaling of technology. ======================= Auradine, a leader in web infrastructure solutions including blockchain, AI, and privacy, has unveiled the world's first 4nm Bitcoin mining systems, featuring breakthrough EnergyTune™ technology, setting new standards in performance and energy efficiency. The Teraflux™ product line from Auradine offers best-in-class performance, efficiency, and total cost of ownership (TCO), positioning it as the optimal choice for Bitcoin mining needs. With EnergyTune™, a patent-pending technology, Auradine's Teraflux™ systems enable rapid demand response and optimal energy usage, fostering a symbiotic relationship with electrical grids, and contributing to sustainable energy practices. Designed and manufactured in the US, Auradine's Teraflux™ product line not only ensures cutting-edge technology but also mitigates supply chain risks and provides increased supply chain resiliency. Visit ⁠⁠⁠⁠⁠⁠⁠⁠⁠www.auradine.com⁠⁠⁠⁠⁠⁠⁠⁠⁠ for more information the Teraflux bitcoin mining systems. ======================= Get Better Crypto Data: Do you want faster, easier crypto data? Sign up for Velo Data, a new product that we have been working on to solve this problem: ⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠velowaitlist.com⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠⁠ ======================= Pomp writes a daily letter to over 250,000+ investors about business, technology, and finance. He breaks down complex topics into easy-to-understand language while sharing opinions on various aspects of each industry. You can subscribe at https://pomp.substack.com/

What's up, everyone? This is Anthony Pompliano. Many of you know me as Pomp. You're listening to the Pomp Podcast, which is my effort to find the most interesting people in the world and sit with them for hours while I ask questions in an effort to learn. So it would mean the world to me if you would subscribe to the show on your favorite audio platform, watch episodes on YouTube, and tell your friends and family about the podcast. My goal is to help millions learn from the world's most interesting people. So let's get into today's episode. The following episode is with Amjad Mashad. He's the co-founder and CEO of Replit. This conversation was recorded during the Build Summit, where 700 founders came together in New York City. In the conversation, we discussed the

customer obsession that Replit has, the importance of being technical as a founder and a builder, the current state of artificial intelligence, how Replit works, how they built such a culture of shipping and shipping very quickly at Replit, and exactly what goes on with the EAC movement and why we must accelerate the adoption, creation, and scaling of technology. I really enjoyed this conversation, and I hope you do as well. Here's the episode with Amshad Mashad. Anthony Pompliano runs Pomp Investments. All views of him and the guests on his podcast are solely their opinions and do not reflect

the opinions of Pomp Investments. You should not treat any opinion expressed by Pomp or his guests as a specific inducement to make a particular investment or follow a particular strategy, but only as an expression of his personal opinion. This podcast is for informational purposes only. This episode is brought to you by Auradon. They're a brand new startup led by a number of Silicon Valley legends who just raised $81 million to build the future of internet infrastructure. You're probably wondering what that means. Let me explain. There are numerous new disruptive technologies that are being adopted simultaneously, from blockchain to artificial intelligence to zero-knowledge technologies.

In order to ensure that these technologies thrive in this new world, we need new infrastructure, and that is where Ardine comes in. They just launched their first product line called Teraflux, which is a Bitcoin miner powered by the world's first 4-nanometer silicon chip technology. These air-cooled, single-phase and dual-phase emergent cooling miners have unrivaled speed and efficiency. They have superior uptime, and they leverage a brand new innovation called EnergyTune that allows miners to dynamically adjust the energy consumption and Bitcoin hash rate

based on demand response needs of the electrical grids. Auradine is an ambitious company working on hard problems. I'm really impressed with them. And if you want to check out more, you can go to Auradine.com. That's A-U-R-A-D-I-N-E.com. Go check them out at Auradine.com today. Before we get into this episode,

I also want to tell you about a brand new product called Velo. Velo is faster, easier crypto data. Everyone in the industry is always looking for it. What's the price? What's going on on the exchanges? Where are assets flowing or not flowing? How is things like open interest and derivatives actually playing out in the market?

Well, that's where Velo comes in. it's faster and easier crypto data you can go to veloweightless.com today myself and a couple of friends we invested in the business we're advising the founder and we think it's pretty cool this one is something that keeps me informed on a daily basis so you should check them out at veloweightless.com that's veloweightless.com all right our next speaker is amjad massad from replit uh come on up here we're gonna start off with uh what was paul graham's quote about replit which one he's had a few you know the one yeah he was like uh replit is the next microsoft so paul graham who started white combinator knows you guys pretty well

thinks that you guys are the next Microsoft. Let's talk about why. And not so much from his perspective, but you've been working on this for a while. You have not focused on making money. You have not focused on all the bullshit. You basically have just been building a product that really solves a problem for customers. Where does that kind of discipline and focus come from? by the way just before we get started i was so engrossed by the uh talk with howard he i don't know if people realize but he dropped some really serious uh alpha uh like a lot of what he talked about i i sort of had to learn the hard way you know for example the one-on-ones thing you know i um i used to like get get so drained uh just doing these therapy sessions with everyone and it's like

like is this why you started a company to be a therapist i was like no i'm just gonna pay for coaches for all of you and you can just go talk to coaches but a lot of his uh a lot of what he said is like really deep wisdom that it took me a long time to i was basically like vigorously nodding my head over there uh which is really awesome uh so i hope i can sort of match that but um yeah so i think you know started uh working on replet um or the idea that became replet like back in college uh so i was doing computer science uh at uh at university back in amman jordan where i'm from that was 2008-9 um and you know i i thought that um i could open a browser window and start coding in the same way that i could open a browser window go to my email go to google docs

turns out nobody has built that i was like what what this is crazy uh it's sort of like instead of like finding a hundred dollars on the on the floor of grand central station it's like it's impossible like why is this idea nobody has worked on pretty quickly i found that it's pretty fucking hard that's why nobody's done that uh so that started kind of my obsession with this idea of like um why hasn't this been done before and also what does it mean for the world of actually software creation becomes as accessible as creating a document as posting a picture as all of this stuff and kind of the more i thought about it uh the more the implications felt that it was going to be really big like uh if we like on the order of the of the creation

the printing press like the printing press uh made it such that um uh sort of reading and writing became such a important things thing civilizationally um and you know all the democratic scientific all the revolutions we saw since since the gutenberg uh press in the you know 1500s or something like that were the result of that right like we wouldn't be where we are as a civilization if we didn't have this concept of you know anyone can write and and you can broadcast anyone can read it obviously the internet brought it to the next step well i would take this step even further right shopify has this um mantra of like arm the rebels and they're talking about arming e-commerce uh kind of smaller businesses that want to go sell things on the

internet compared to the goliath of unsaid amazon and you although almost feel like you are arming the rebels of the next generation that basically you're saying look learn to code and then we will give you this sandbox to go build anything that you possibly can and like go take over the world right what are you seeing in terms of that first customers that come in because i think part of what i'm so impressed by your business is the long-term thinking it takes to not just go and say hey i want you know the best coders in the world to be able to go onto this web page and be able to code you literally start with like well if we need enough users we should teach them how to code and then you're so far upstream that you can capture a lot of the value a la microsoft

which actually is going to be bigger than microsoft uh if we if we did it right right because microsoft uh ultimately relied on sort of people already knowing how to build things and how code and what howard said about kind of not letting your vcs dictate your growth rate that's something that really benefited us a lot because um you know the conventional wisdom is that you built a good product okay now you go sell to companies uh you figure out what the companies want and you build for that but actually when you go to enterprises and you ask them what they want they want the past they don't want the future so you're sort of it's a premature optimization that locks you into the ways of the past

and doesn't really let you innovate. Whereas what we wanted to do is actually make the act of software creation radically different, radically more accessible, more fun, more interesting, more exciting, faster. And to do that, we had to go arm the rebels, right? We had to go after the entrepreneurs, the hobbyists, the misfits, the weirdos,

the students, the kids, everyone that was really underserved by the enterprise of operational market. it um and um and so uh you know the uh um learning is just a means to an end uh learning you know it could be fun to just learn something for fun but people learn a code in order to create something and people want to create all sorts of things with code you know from art to science but also companies and we've been so focused on this idea of enabling entrepreneurs um and initially it started on a small scale like you know someone who built like a uh like an app or something like that

and that would go viral maybe they make a few dollars off off of it or game but uh just uh yesterday someone um uh sent me a tweet um built this company called spellbook.ai they just raised 11 million dollars and he's like i couldn't have done it without replet it allowed us to move on a week-by-week basis create new mvps and new prototypes and get it in front of customers and he's like i wouldn't be here without without that the speed that this this gave me so let's talk about artificial intelligence um it is this i'll call it a new technology even though it's not really new um it seems to have caught the world uh by surprise in some degree but also people very quickly are wrapping their heads around, this is going to touch every industry.

You all are heavily involved, both, I think, in terms of many of your users are using this new technology. You all are doing everything from hackathons to incorporating technology into your product. And then you personally are very interested and involved in terms of how this is going to get regulated and some of the pros and cons of current approaches. Talk just about when you see something that is so potentially fundamentally shifting, how do you even wrap your head around what is the technology and like what the implications are and how you can adapt your business to it yeah i think the most important thing is to really understand what the what the technology does um from as close to an axiomatic understanding of the technology as

possible right so for us uh we were focused on the act of software creation uh so what are the things that actually make like so in the ideal scenario you have an idea and you turn that idea it into a product or a piece of software instantaneously right that's you want a zero friction uh you know zero milliseconds between idea to software right so okay what are the steps that actually make make it so that it's not zero milliseconds like it's actually an hour to week month whatever it is um and there's all these complexities that has to do with software with coding things like that and so okay let's let's take the active actually coding the mvp or the prototype a lot of that is just boilerplate and it's just um like keystrokes to just uh help

the computer understand what's in your mind and a lot of that could be automated and we did a lot of classical automation of these things like it's uh you know just running scripts and um and kind of pre-setting up things and configuring a lot of what makes replets special but i always thought that the next level is going to be about generating code and we looked at it when i started the company around 2016-17 and there were some interesting experiments there actually using very primitive language models but uh it wasn't just there look at it again in 2018-19 it was kind of gpt one two it's like oh that's an interesting thing but like hey that generates like a lot of crappy code and syntax doesn't compile or whatever and the gpd3 happened i was

like okay it's now here and went kind of all in and in 2020 21 released our first drive of ai product uh and 22 released ghostwriter our flagship kind of code generation product describe a little bit about what ghostwriter does uh ghostwriter has a lot of different modalities and like the most simple way is like as you're coding it actually predicting what you're going to code next and just doing it for you so i'll present a suggestion similar to you know copilot or gmail kind of autocomplete kind of background suggestion if you like it just hit tab if you don't like it just keep coding that's like a domain modality people use it with the other thing is if you actually want to stop and construct a prompt uh to generate like a function or a class or an entire

file you can do right click generate code and actually give it a natural language prompt and this goes to the largest model that we have or on offering so that we can generate an entire file so this is this has to be accurate the error rate needs to be really low here if you want to talk about any of these things if you want to talk to the ai about what it just generated if you want to ask the questions about any code if you want to plan a larger refactor or change we have a chat modality as well where you can like go and have a have a multi-turn conversation with the ai So as you're using these technologies, what's interesting to me is the people who use the technology is very different. If somebody goes in and uses ChatGPT and it's, you know, your mother, your friend who maybe isn't as technical, doesn't have kind of the hacker mentality, they may ask it somewhat primitive questions.

When you put these tools in the hands of people who have that hacker mentality and have the skill set to leverage them, they can move incredibly quickly and they can really stretch the limitations of what these technologies do. Can you tell us one or two stories of things you're seeing people do that even you who are in this every single day and have a good sense of the technology, you're like, I cannot believe that people are building this stuff? So I kind of knew that this made coding easier. What I was surprised by is like one day, this guy built 250K ARR, single guy on Replit business, right? In a few months. and i started talking to him just trying to understand how he's using the product and he asked me a question uh i like to provide some support our customers as well so he asked me a question it's like how do i do how do i lock a file make sure i'm not editing it while i'm

changing another file he was describing git he was describing version control so here's a guy who built a quarter million dollars probably going to one million rr without knowing git you know and the conventional wisdom is that oh you need to learn all this you go to college and you go through you know you go through the motions you learn getting all that and you start a project get in it whatever all this bullshit doesn't matter what matter is actually being able to create a product and i asked him like how are you able to do it and he's like yeah just the ai and you know i'll go back and forth with ai um you know i'll produce an error uh when the error happens i'll either use ghostwriter or paste it in chat gpt if i wanted like a longer conversation

and he uses all the ai tools he's barred and chat gpt and replica ghostwriter and everything um and it's just this iterative thing between that person the environment and the ai that actually produces the code at the end and this is a non-coder that builds a startup with code. And I didn't think that was possible. Is that really that different? Like one of the jokes, I think if you go hang out with developers is like, it's all just Googling, right? You're laughing because that is the joke, right? And so like, to some degree, are we just now outsourcing the Googling to the AI and it's coming back with the code? Or is there something different that's happening? Yeah, I think that's, I think that's part of it. I think what's different

is that Google doesn't talk back to you. The AI actually can ask you clarifying questions, can provide feedback. It's a lot more personalized. So it's like Google taking to its ultimate conclusion. The other thing that I found interesting in how a lot of these entrepreneurs are building businesses

is thinking about generative AI as a suite of tools that allow you to build fundamentally new experiences. We have another startup founder, Priya, who built this very interesting app that basically takes a document and creates a multimedia presentation from that document. And it is a pipeline of a lot of different LLMs and a lot of different generative tools such as audio, video. And it's kind of obvious, like you want to put these things together, but the extent to which people are pipelining these tools to generate an end creative product. is really surprising and makes me really excited because we're getting to a point where the way to think about these AIs is more like employees, is more like extensions of yourself.

It's more like, you know, team members that have different specialties that you can kind of bring in depending on the task you want to do. And that's really how the people who embrace these tools the most use them as they kind of view them as extension of their teams themselves you all i think more so than any company i've seen uh seem to hire your users or hire your customers and this seems like a great hack to really get high quality people who believe in the mission talk a little bit as to like what that process looks like and do you think other companies should be doing this more often the process is that um it's just so natural it's just so organic Like we sometimes like some users force themselves into the company where you sort of like they apply and they send an email to our jobs.

They apply in whatever, you know, portal we use for jobs. They send emails to individual employees. They go to contribute to our to our open source repos. They'll do everything to be useful. And suddenly we just see them as part of our team. i'm like okay i guess we're gonna let this 15 year old kid like you know join our team and you know a lot of these kids stay for a long time i mean um we've had we've had people join when they're 18 uh and um and actually recently someone left they're like 25 i was like oh i kind of want to

see other things in the world and and so it's like okay you know you've uh you've you've earned your keep in a way and they're rich as well right i think they made a lot of money uh you know they got stock options at replit like really really early on um and uh it's awesome like seeing all these hackers and these people that wouldn't otherwise be getting jobs like get jobs at replit and a lot of them are younger right that you guys are specifically seeing because of kind of the hacker culture and the product itself and you're teaching people to code but can it be replicated in other businesses um i think so so we've had it we've had instances where where people look actually i would say that everyone who almost everyone who joins replit

has used the product and has loved the product and loved the mission and the vision um and so there's like some selection effect that you can apply onto a lot of other companies So I think the way traditionally recruiting works is you hire a recruiting team or recruiting agency, and they go out into the world and they cold email a bunch of people. And you kind of need to do that, at least early on. But, you know, you have a lot of low-context people coming in, and you have to sell them on the vision. They have to actually believe it. And, you know, some seasoned executives will get themselves to believe anything, right? to you know they just want to be team players but it's not as strong as when you actually

get people who believe and i think you could do that by actually just interacting with your customers all the time uh with your partners uh and you're going to be building these relationships again that will make it very organic to just hire from so as you guys are building there's a philosophy along with the actual product itself. There's been a rise of EAC. And it's this idea of just like everything needs to accelerate rather than have people tell us to slow down or to over-regulate or to stop using energy. And we can go down the line of all the stupid arguments that are being made in the world. How important is it for the philosophy to be aligned with the business? And how do you take that and incorporate it into the company culture?

because i think that really if i talk to anyone at replit like there's a it's like of course you work at replit and that seems like a really important part of the success of the business yeah so um yeah i think there's like a radical streak at replit and what we're doing which is how do you take uh this sort of arcane art of software creation that's been sort of created a lot of wealth typically for a small population of people typically in silicon valley and actually make that so that anyone in the world can put spit in it and my story where i came from and being able to come here and become a founder uh and build this company that has a big impact i think that story could be done a hundred a thousand a million times and a lot of the reason it hasn't

been done is because of artificial constraints, whether it's constraints on the tools and how accessible they are. And a lot of times it's regulations, it's certain ways of doing things that perhaps not so explicitly, but sometimes implicitly is about trying to centralize power. AI is where we see that happening right in front of our eyes in a very accelerated manner, um where there's like a clear power grab by incumbents you know going to dc telling them um hey i'm building something very dangerous you should create a mode around it so that i'm the only one building it why would they say that i mean it's kind of let's just get it let's just get into it it's kind of comical like if if i was if i was like um if i'm like making a sci-fi about

like an evil executive that wants to uh right to do records to capture it yeah i wouldn't go as absurd as what we're seeing now with these ceos going in front of congress and literally saying saying that so before we get so naked yeah before we get to the critiques of what they're doing let's maybe put ourselves and for those that don't know maybe some context is helpful so uh ai is huge uh there's a number of companies probably three four or five of them um who have been building various components or technologies around ai and rather than take the normal stance that most technology companies have done for ever and then like hey government you stay over there we'll stay over here and like you know leave us alone we'll leave you alone these people are

basically going and knocking on the door in washington and they're saying regulate me bro regular me daddy yeah which like if you kind of think about it like what why would you do that and so like what is their thought process or what is the advantage what's the incentive for them to go do this i it's pretty obvious like i i kind of like um look there are true believers that believe that ai is actually going to kill everyone uh it's like a fringe theory started in like berkeley they wrote this blog post a blog um site called less wrong you can go read it it's pretty insane um and they misnamed it it should have just been wrong yeah just wrong just keep it simple um but and i think those those guys are true believers i actually have more respect for them

than uh you know they you know a lot of the a lot of the people um that are actually asking for regulation are using the same arguments um and mark andreessen um actually wrote this blog post about ai recently and um he kind of drew parallels to the period of um uh what's the period where they banned alcohol uh prohibition yeah the prohibition and he he said that during the prohibition there was actually like true believers and there were actually profiteers And so what we're seeing a lot of the executives that are asking for regulation are profiteering from the natural fear that people have from AI and the very potent message that some of those true believers have actually put together over many, many, many years. And the immediate need to create a mode is because open source is actually catching up at a rate that never happened before. You know, typically, sort of, typically there's a lead time between software and open source competitors.

uh in ai we've actually that that lead time has been shrinking well and thanks to people like zuck who's emerged as a hero uh in the situation with llama and llama code and llama 2 um you know the lead time is only a year or so or a year and a half like you know i think gpt4 level uh open source ai is probably llama 3 which i know they're training right now as we speak so uh there's a there's a impeding threat that would eat into the bottom line of a lot of the the ai ugly ugly poly right now so um and this is compounded by i saw a report recently that chat gbt traffic is down for three months in a row do you feel like um there's kind of these short-term pressures of obviously there's more of these products on the market some of them are open source some of them

are just closed source competitors and so people kind of got through the threshold hey we're the market leader at the moment and they want to close the door behind them and that's really what they're doing yeah i mean i mean it's pretty obvious like do you have another explanation for it like i can't i can't think of any like even if i want to be extremely charitable like do you actually believe that those ceos are actually like uh think they can kill everyone in the world like existential risk they actually like if they thought that why are you working on it you know um and and so i just can't i like when i really try hard to kind of think for an explanation that is actually um as tradable as possible like the best i can come up with

if you're actually worried about um say um like effective ai on fraud effective ai on critical infrastructure, you would regulate the applications of these, right? So, for example, the US government can say we can't apply AI to utility companies managing electricity or water, whatever. That would be a reasonable thing to say, or you have to kind of meet a certain threshold for applications in X, Y, and Z. But what they're asking, the policy proposal that OpenAI, Anthropic, and others have put forward is that AI that has used X amount of flops of compute should be banned or should be banned from the open. And only a small group of people should be able to train that.

And again, that's like a pretty clear, they're not even trying sort of like pull the ladder type of approach. And look, I mean, a lot of these people I respect and are my friends. We've worked with them for a long time. but my uh my ultimate goal and the reason i'm working at replit and i've done a lot of open source i was part of the react.js core team react native and you know most of my career has been done in open source and i think open source is one of the best tools for freedom and software self-determination and opportunities for people to build businesses all over the world

it just decentralizes the opportunity in software in the same way that that replica as well does and an ai is the future of software it's literally just software right and if if the future of software is bottlenecked by these regulations then they get that's going to be really bad for the world we're basically cutting off a potential much better future you know the other day i was I was looking, there was this research that come out after like multiple years of study showing that metformin during, when you have symptoms of COVID, if you take metformin, it reduces your long COVID incidence by 40%. The thing that generated the hypothesis for this is an AI. So in April, 2020, a group of scientists at some university used computer simulations

and actually natural language processing to understand how COVID kind of works and what potential interventions. And they hypothesized at the time that metformin could be actually a good way to, I think, based on protein synthesis, whatever, a good way to actually intervene. And it turns out it's right. AI is already generating scientific hypothesis. It's generating scientific research. We had the alpha fold with protein folding. It's amazing for the future of healthcare, for the future of education. You and I are investors in synthesis. They're working on an AI tutor. AI tutors solve the two-sigma problem. Bloom's two-sigma problem is that what he found is there's no education intervention in the world that could be as good as one-on-one

tutoring. And so the whole idea behind the startup is to provide one-on-one tutoring for everyone in the world. And that really raises the global IQ. And so AI is the future of software. It's the future of humanity, in my opinion. Now, back to the EAC question, EAC takes it a little further, which is the idea that we need to accelerate all sorts of human progress. And it's not just AI, it's also energy, what's happening with the de-growthers in Europe and the anti-nuclear people uh and and the idea behind iac is how do you build a coalition of technologists and pro-humans pro-humanity ultimately uh folks that actually are countering the the sort of degrowth uh death cult that has plagued the west uh and the idea behind it is the kind of

pure teal formulation is that some people kind of divide the world into developed versus developing and the implicit assumption behind developed is we already were developed we're done it's time to cover the resources now uh whereas the e-ac idea is like we're not developed at all we're still primitive as to where we could go as a civilization which camp are you in uh take a guess let's talk about training data real quick um i recently saw that a bunch of uh very well-known best-selling authors got together and they filed a class action lawsuit i think against OpenAI and ChatGPT. And they basically were like, hey, you took our work and used it as training data, among other things, and you should have paid us for it. We've seen Elon Musk really

kind of lock down Twitter. It seems like Twitter was a huge source of training data for these models. Where do you think it all shakes out? Will anything on the internet be fair game for the training models? Will people have to, if they want to train a model, go buy a bunch of data? how does this work out um so here's where i'll show i'm not like a total like anarcho-capitalist free radical um i i think that uh there's probably uh um some uh there's there's probably something a government here could could uh could could actually provide clarity on uh i think um at least uh the the the problem as it happened with crypto is when there's so much regulatory uncertainty the players in the market become kind of deranged and so no one

knows what to do there's a bunch of lawsuits and maybe eventually we'll arrive at something but having some clarity around what does it mean and I think that there's like fair arguments on both sides those those people affected by the training where the value of the labor is cheapened because the The output of the labor went into the input of training of this model. It's actually like a very, like there's a moral argument there that you've taken my labor, you obviously mixed it with your own labor, but now you've reduced the need for my labor by this new model that you created. What we've done at Replit is we trained a large-language model for code. it's the smallest most economical fastest model for code it's a three billion per hour model we

train it in april what we did is we trained on open source data and uh we released the model as open source because we thought that it's the ethical thing to do if you train on public data at minimum what you should do is have your model be publicly accessible publicly download downloadable and then we had we wanted to keep some advantage for replit and so the data that replit has and collects as part of our service, we trained a model that was 50% better. And that we kept for our users or customers, right? I think that's a fair trade. If you're opening AI and you're training all the data of the internet, you should at least contribute something to the internet as part of the open source. And maybe there's some advantage you keep for your company and this is

the kind of value you keep. The last thing I want to talk about is AI agents. I think when people first kind of realized we're going to be able to do this there was an it felt like every single day multiple things were going viral on twitter where it was just like holy shit look what this can do what do you think is actually like commercial uh available maybe five years from now versus hackers on the internet you know make it order them food or kind of do cool things but there might not actually be like business opportunity there i actually think agents are going to be huge um but um like a there's like a lot of infrastructure to build we actually three or four months ago took a pause on a lot of the ai features we're building because we just realized

that the amount of hacks that you have to do to get anything working was just like growing exponentially and the ai was getting worse over time and so we took a step back and what we do at Repl.it is build abstractions. So we build an abstraction, we call it a chat service, not a very inspiring name, but basically it figures out, it's like a multi-component system

where there's like an intent detection layer without getting too much into the details. But I think a lot of people are just thinking about one model and maybe you recursively kind of prompt this model and this is an agent. But what I'm saying, agents are going to be a society of models.

There's going to be different models at every part of the chain doing different things. And some of these models are going to be hyper-specialized and trained to be more agentic. And maybe at the core, there's the mega models, the GPT-4, the Gemini's, GPT-5, whatever it's going to be. But it's a system. And I think a lot of researchers,

the way they think about AI is they think about just the model and the end-to-end learning of the model. As engineers, we need to think about systems. And to think about systems means that you would need different tools for different part of components of this of the system so i think agents are coming um i would probably i probably think by next year we're going to have pretty compelling uh use cases both in the consumer and enterprise side and on the software creation side um but i would probably give it another uh year or two probably 25 26 a lot of our lives are going to

be touched by ai agents uh such that you know everything will you whatever you're interacting with robots today it's kind of sucks like when you get when you're on the phone and kind of press one for customer service whatever this this sucks it'll get a lot better you'll be able to just talk to it um a lot of the back office stuff is going to get automated um and uh and so i think I think everything in the economy will just like get behind the scenes better, but also a lot of consumer interactions will be a lot more delightful. It is funny that you call a phone tree and it, you know, asks you to talk now,

or you just press zero a bunch of times and you feel like you hacked the matrix a little bit. Right. And then they just like loop you back and you're like, fuck. All right. They're, they're smarter than me. Um, how can, uh, how can businesses leverage Replit or like how could some of the founders who here who um either have businesses or think about certain businesses leverage the product so i don't think we're ready to run entire businesses on a replica yet but what i will say if you're if you start from scratch i would just do it a replica like if you're starting a product from scratch you would move 10x faster by doing a replica uh if you're a bigger company

you want to start to adopt it think about a microservice or a back office task or internal tool or a slack bot or whatever that could be an isolated thing that people can spin up really quickly that uh uh that way you can sort of start adopting it it can also be helpful for internal prototyping and sharing between between people inside the company um but yeah definitely if you're starting uh if you're a founder starting from scratch like i used to like for the first time i'm like comfortable saying it would be a bad decision not to not to do it on a replica so uh because recently we just um release our deployments feature and so you can actually deploy and scale up to infinity if you'd like uh any kind of application so i think it's

uh it's really wise to start there thank you so much for doing this Thank you.