the product is breaking, the credit card is maxed out, the API is not working, there's too much happening and I want to hire people and like >> Yasa El Say CEO and founder of Chatbase, he's built it bootstrapped to 10 million AR. >> I was working on Chatbase, the very first version even before Chad GPT was launched. We're not a GPT rapper. We're a harness. >> You're retweeted someone else and you said my moat is that other companies don't have me as a founder. >> I don't know if you remember that, but I loved it. I loved it. Welcome back to the Napkin Math podcast. We've got a very special guest with us today. None other than Yasa Els, the CEO and founder of Chatbase. He's built it bootstrapped to 10 million AR. And yeah, let's have a little round of applause for that. >> Come on. Come on. You beat me. You beat me. >> I I crumbled at six >> before raising. I think I did. >> How long did it take you to Sorry. Sorry. Finish intro. And chatbas is the complete platform deploying to for deploying AI support agent to your business. But obviously Yasa is going to get into that. But Saba you were saying. >> Yeah. So um I mean congratulations 10 million in revenue is huge especially uh bootstrapping. How long did it take to get there? >> We just crossed 3 years like yeah a month ago. >> that's gone by so quickly. When was your first payment? 3 years ago or was it like 2 years ago? >> Like first Stripe payment >> uh you received? Yeah, I think no I we we got our first payment like I think 30 minutes after launch so 3 years ago too. >> That's some good validation. Um something I want to ask is um going back to that time. So I think it was around when GPC4 launched right where this all started to take off >> around early 2023 something like that. >> I'd love to go back the origin story behind chatbase. >> Yeah. So, so we actually like I was working on Chadbase the very first version even before Chad GPT was launched. I was using their uh >> the model back then was called Da Vinci Da Vinci 03. >> Um >> so I've been yeah this is like one of the OG GPT rappers in the in the market. >> How how did you find out about this Dainci Da Vinci API? Yeah, I I was just like randomly working on other projects. Um so I was I was studying computer science and I was doing internships and yeah I just thought like this is not the the path that I wanted to take after doing those uh like internships at big tech because for for a big um like period of my life that was the goal but you know after you you get there you realize actually maybe that's not what I want to do. So that's exactly what happened. And then I just decided, hey, like let's let's try to to build something on my own because I saw I was following people like on Twitter um a lot. People like doing solo projects, people doing bigger projects, people like startups, YC, all of that. And I thought, yeah, like surely I I can at least like do something similar. Um, so yeah, I I always had like some projects I was I was building and then like like GPT3 was a little bit like big of course like nothing compared to what happened with was uh with CHP when when it was launched but there was some like traction. People did talk about it a little bit in the AI space. Um so I thought I had this other project doing like it was called rate my courses. It was like an app for university students to like choose classes to take. >> So I said it's very hard to monetize because my idea was to put ads on it and then like the ads made like $7 every like 3 months. So it was not a good monetization strategy. So I decided I'm going to build a product and sell it to this audience because I had a lot of traffic. So the the product I wanted to build was was uh an essay writer like an AI essay writer um for for students and then I was using the GPT3 API while I was working on that and then like by being in the space like like trying the API like talking to other people like working with the API going into the discord and all of that uh seeing the limitation seeing what's possible I I got the idea for chatbased so I just like stopped everything else I was doing and then >> yeah focused on chatbased >> so so GPT free that's before chat GPT launched correct >> was like okay this is fascinating and you you found out about it because you were just researching this idea of writing essays for you yeah >> did it was it clear that this was going to be a really professional technology. >> Yeah, I think that's that's a very interesting question. I think somehow to me it was um yeah, I don't know. I think I think like to other people it it it didn't look like this is like that special. Uh like of course it was powerful, like it was cool, but like GPT3 was not that strong. like I think maybe 3.5 was the first model that was like actually usable in like actually like um like production use cases but but three was was not that good and it was hard to imagine like how good this models will be because at the end of the day like you can't >> it's hard to like go from like token completion like next token prediction to like AGI you know and like even It's crazy to think that the models we have now are just doing that like next token prediction. Um but yeah, I think it was not very clear that this these models will be this strong, but I thought in like the very first version of Chadbase, I think they were like um close to good enough for that use case. and I talked to a bunch of people. Um but but yeah, I I think like when I talked to them, it wasn't clear to them that this was going to be big. It was like okay, like it's just another idea of a startup. >> So So a common theme for this podcast is to build when you're building AI products is to build products that are going to just improve with each model improvement. did what did you see between when GPT4 launched like how did that kind of change how you view chatbase and the user experience? >> Yeah, like GPT4 was was was was a very big deal. Um and and I think like the idea of building something with the assumption that the models will continue to improve and like building for for that future instead of building for like what we what already exists now. like that idea is is is new like like Sam Alman was not going on podcast saying that in like early 2023. So I don't think yeah people were doing that. I think people were looking at that technology at the models we have now and just seeing like what is possible what's not and then if something is not possible then you you're like not going to force it. But but I I thought the models were improving and yeah like I just built the like the layer on top of it that will like benefit from from that and I think that that worked very well. So when GPT4 was launched a lot of like the problems like the very big problems like you know like hallucinations and prompt injections and all of that it was just impossible to deal with uh with GPT3.5. Um but yeah, four four was such a big step and like the product became usable as the model uh improved and like every model iteration just makes the product a lot more better. >> So there was a bunch of interesting things from this era. So do you remember copy AI and Jasper? like they were >> OG's. >> They were like the OG. Like, oh my god, these guys the these are these guys effectively got crazy viral growth. Went like 0 to 100 million R in a year because they were using the APIs, right, at the GPT3 API. It was mad. >> Yeah, it was it was crazy. And I think Yeah, they they they were very early. I think they like like GPT3 they they before before any any of this before like GPT3.5 I think they were just using the the GPT3 models and I think they got like early access even before it was uh GA and um I think yeah that's like the importance of of timing I think. >> Yeah. So quick history lesson here actually I think this is interesting for everyone. GPT1 was launched in mid 2018, right? GPT2 was early 2019. GPT3 basically took a year and a half to get to. So that was 20. That was basically just before COVID, right? So mid22 and then it was like 2 years to get to GPT 3.5. And I think that's potentially when you entered the room, Yasa, right around that. >> I entered Yeah. like maybe a few months maybe like four or five months before 3.5 >> and then and then not long after GPT 4.5 6 months later is when we get what is now historically known as the chat GPT moment. >> Mhm. >> And then and and also like when I first used Chat GPT I I actually wasn't super impressed. I was like this is really cool but I'm not like wow. >> And then GPT4 came out and I was like oh okay fair enough. Like this is this is this is good. >> Yeah. I just found using chat it was just like the dopamine hit from um the fact they actually could come up some pretty cool stuff and just it was kind of just a format we hadn't quite seen before >> uh back then >> and what's super interesting is like if you look at this timeline content like this idea of writing blog posts and content was unlocked effectively for Jasper copy AI by GPT3 GPT4 uh 3.5 coding was really only unlocked at GPT4 four, >> right? >> Yeah. >> So, my question, Yasa, as someone who is early early early to the technology wave, what is going to be unlocked by GPT5? >> Oh, that's that's a that's a very hard question. Um, >> it is. Yeah. I thought I thought I'd give you an easy one to start. >> We're already on GPT5. Do you mean the GT6? >> We have six. Yeah. So, 5.5. Okay. for six. It's very hard, but I think like coding is still like the biggest biggest thing. And I think we're we're we're getting closer to like like having the AI automate more and more parts of of coding and just like not even just coding, just like the the whole like architecting and engineering and choosing the stack and all of that. Now it's like you still have to guide it now. You you can't just like let it run for like hours or days. Um, so I think of like I I'm pretty sure that's like what the labs are focused on because it's there is such a huge um economic opportunity to like whoever is going to is going to get there first and then they're they're also saying they're interested in uh like scientific discovery. >> After you shipped chatbase originally, how long does it take? You said you got your first payment very short within 30 minutes of launching it publicly. >> Wow. So, so yeah, that's that that was crazy. So, >> was it your mom? >> It was your mom, wasn't it? >> I I thought so. Like I I I I thought someone was pranking me like, "Oh, like this guy is doing a project. Let's let's make him think it's working." >> Um, no. I I launched to like So, so I did like maybe like technically like three launches, but the first launch was was not even a launch. It's just like me like showcasing what I built. The very first version was just uploading a PDF and then chatting with it. But that idea was just so interesting and so new because when you thought of like the the LLMs, you you couldn't just like add more context in it or like that was not something people knew that you can do. It was just like whatever the model is trained on in it like training data and that's it. So the idea that you can give it more data about you, about your company, about a book you like and then like ask it questions that was very new and that's why I think we got like the very first um viral moment with with chatbase because it was the first tool to show people that you can do that. Um, so, so I I did like a small demo of like me uploading a book and then chatting with it. And there was no like there was no product. There was no like payment options. There was no plans. There was just it was just like one page. You didn't even have to sign up. I think uh >> in the first version you just like it's like one page. You upload something and then you chat with it. >> Um, so yeah, that was that was the very first version. that post went like very viral compared to like like I didn't have any any followers. Um, and then I started to lose like money because I put my personal credit card on OpenAI and I I didn't like think that people will actually use this. Um, so yeah, when when I saw that was happening, I had to like actually like >> make a product out of it. So I was yeah I think I I just logged in for three four days to try to integrate Stripe make make some plans and like put some limits on how much people can ask and yeah I think I just merged that and like promoted it and like tested it. It looks good and then I went to the gym and yeah like I think while I was squatting I got the the the very very first notification of of a payment. Um, >> so specifically while you were squatting. Yeah. >> Yeah, I remember the exact scene. Yeah. >> And from that moment onwards, how long did it take you to get to say the first 100k AR? And what what do you have to do to get there? You just locked in from then on then? >> So I Yeah. So exactly 100K I think was like maybe five months. >> Four and a half, five months. So I I know I I got to like 1 million in in exactly four months. 1 million AR. Can you explain why and how you did that? Like what what really what why was this successful? Were you is was it because you were basically doing something very novel with a very powerful piece of technology and you were just early to the market doing something interesting? Was it distribution? Like what like what what did it? So I think it's it's a com it's multiple factors because like the timing of course was was such a big part of it but there was like I would wake up every day and see like 10 other people doing the same thing because there was just no barrier to entry like even if someone like saw my tweet and like I'm posting on Twitter and like tech Twitter so like everyone is a developer that can build something similar. So, like the timing of course was was such a big part of it, but I think like when when I when I looked at like like what happened to to to other people that like had products that they launched at the same time, um I think they just I don't know. I I think the main reason is that like they didn't double down on what they were building. So, a big part of it was like, yeah, the distribution, like getting getting people, getting like doing interesting demos, getting, you know, like doing influencer partnerships, doing some paid ads. It was just like doing a lot of everything all the time. And I had the advantage of like having conviction early on that this this is like an opportunity that I shouldn't waste which I think it's like something hard to to come by like like even even now I think there has to be like a lot of opportunities that like people can do but it's hard to like have conviction in like one idea and say yeah this this thing is going to work if I put in the work and for some reason it was clear to me that this this will work if I put in the work. So I just I just did that. So >> yeah, >> I think um it's quite rare that anyone um has like a viral moment from something that they've created and when it happens you just you just have to double down on it. And um you said you mentioned you were working on lots of different channels and you tried lots of things in that you know like uh ads, partnerships, product demos. What do you think are like the few things that really moved the needle that you tried from that period? I think early on organic demos was was like the the most important thing because like back then everyone was was like it was just as the start of the AI wave and like no one was like now people are a little bit tired of AI but back then everyone was like oh like this is a new cool technology like Chad GPT was was just launched and like people were trying it. Um, so I think like showing people that oh like this is chat GPD which is like this insane technology everyone is talking about but it has like this functionality that is very useful to it. Um, yeah was just very interesting for people to see. And then doing that for my account working with like everyone back then was an AI influencer. So, I just like worked with a bunch of those and you would see you would see like a post like when you write like a good copy and like have a good demo. Um, and you you have like a page on LinkedIn posted. I would see like Stripe go up the same day like like 5,000 MR and it's very clear it's just from that one post that did very well. >> So, are you telling me that influencer marketing is one of the things that made you really successful? I think early on of early on like because there was such a huge like demand from or like interest from just like people on LinkedIn, people on Twitter in that space and like seeing those demos and like just actually like trying to use the tool. It was yeah the peak time for that. >> And and were you how were you and like few questions how like how much how much were these influencers costing? And the second question is would you make the content and give it to them to post or would they make content organically themselves? >> No. No. I I would make all the content, give them all the copy and say, "Hey, this is how it works. Just this video, put this caption and um and yeah, like let me know when you put when you do." Um, yeah. It's like payments. It just varies a lot. Like it it's from like like $300 and to like like three like,000 4,000. It depends on like how big the page was. But but yeah, like some of them were very expensive like like some of them were like $3,000 a post, but when they did post it was like $3,000 in MR. So >> yeah, I I would do that deal. I mean when for us like when Sora came out the which ironically is now gone like some of the big AI creators on Instagram were charging 40,000 uh dollars >> for a post. So like >> actually now now it doesn't work now now like I stopped doing influences because the prices are crazy and also like the the market is like the social media is saturated by like AI content so it's it's not going to do as well as it did before >> when you go from being a student in Canada immigrated from Egypt you've launched something and 4 months later you've gone from being broke to getting 1 million AR. Is it just pure euphoria or was there some kind of other side to this? Like did things break? Was there like uncertainty? Was there uh you know feeling pressure? Like what was that kind of journey like in such a short period? >> Yeah. Um I think like my my my I just didn't register it because of especially early on. Um there was actually no time to like you know like reflect is happening. It was just >> you know like um like customers asking for stuff, customers emailing, I need to respond. The product is breaking the credit card is maxed out. the API and like investors want to call me. It's just there was too much happening and I want to hire people and like >> yeah I don't think I I I intentionally like tried to to reflect on on on this until like I had you know like some some time after >> and even then it just felt also >> I I think even until now it just felt feels like I don't know like a little bit um like unreal. I I don't know. I don't know how to describe it. Um Yeah. >> Was there anything that kind of just broke in this period, this like four to six months period where it's like hair on fire trying to figure it out? >> Oh, every everything. Yeah. Yeah. Exactly. like everything the whole product like the like even even stuff outside of my control because even open AI was like the API would break all the time and >> like I it got to the point that I didn't go out without my laptop because I didn't have like any automated testing or like have any alert system. The only way I would know if the product works is like I need to refresh it and ask a question every 5 minutes. Um, so that yeah, I would go out with friends like when I did go out with my laptop and like that's that was like my my my life for maybe 6 7 months. Um >> and so so when did when did like so what is chatbase today and when did that evolve from how did that evolve from okay first version was this PDF chat tool I mean now you're like you know custom I mean you you tell us what what what are you how did it evolve? Yeah, it was it was very like gradual. Um I actually no like I think in the in in in the early days I needed to make a decision on like which route to take because there was so many options from like that chat with like data start. there was just like so many options and like I saw so many different people go so many routes like like a lot of people that started at the same time just continued to chat with PDF but I don't think that was a good idea because of course that's going to come to the model and um yeah like it's like the rapper was so thin that it just like became part of the what Chad GPT offers so I knew that was going to happen so like I what I did was I just saw like what the market needed And what I saw from talking to customers and from them like telling me like forcing me, hey, like we need this feature and the only way we're going to pay you is if you have like if you can allow me to put this chatbot on my website or allow it to like deploy it on my email so it responds to my email. And just like over time um especially like yeah like after the first year it became clear that the route we're taking is like B2B AI agent like customerf facing agents um and then yeah like it's been that ever since but of course like the definition of an agent and like the definition of uh like what customer support is able able capable of doing uh just changed a lot over like the last two Um, >> yeah. >> So, now like the rebrand that happened is we're not a GPT rapper. We're a we're a harness. >> Yeah. >> Uh, >> yeah. I've seen that. I've seen the word harness is the new uh the new rapper. I like it. >> Yeah. But but harness. Yeah. Harness feels more premium, you know? Feels like you're doing something. >> I liked it when you said like I don't want a sparkling rapper. >> Yeah. I don't want to be a thin rapper. I want to be a fat rapper. >> It is. It is true. I think I think like like all the application layers are they're trying to be like fatter and fatter like kinds of rappers because you don't want to be like uh an update away from um from the models. Um but yeah now now chatbase is uh we're building uh customerf facing agents for businesses. Um so the idea is you we build you an agent for your brand trained on all of your content. So your website, your documentation, and then it connects to your systems. So your CRM, your Stripe, uh your database, all of your APIs and all of that. So it can can take actions and it's also able to troubleshoot issues because we're able to like run it in a sandbox. So like things like you know like like running a CLI tool like for example with like cloud coder and codeex uh to be able to like open files like troubleshoot an issue the same way a human would. Um and then that's that's mainly like the the customer support side. But we're not only doing that we a lot of our customers are using us also for more touch points with their customers. So like sales, onboarding, um like qualification, collecting leads, all of that because the idea is simple. The idea is this is like the interface to to your company that's an agent that's conversational. It knows everything about your business. It's connected to your systems. It can like me like t like ask you a question on Slack when it's stuck and then update its knowledge base. Uh it can like do longunning tasks. Um, so yeah, like the idea is it's um it's an ambassador for for your company to your customers regardless of the channel like whether it's email, chat, chat, uh voice, um all of that. So >> you got >> Yeah, this is >> you got some pretty nice logos on your homepage. So uh Chuck ECheese being uh being one of them. Uh, Opal, is that the camera? Opal camera. >> Opal is the mobile app. It's like the >> Great. I'll take it. MLE, they make very good cookers. Uh, >> F45 training. Are these guys self-s served or are they like on enterprise contracts? >> So, they they they came self-s serve and then we moved them to enterprise contracts. >> Yeah. What part of the business is enterprise versus self-s serve? Like, how are you thinking about that? >> Yeah. So, so of course like early on it was only self-s serve just like PLG. You just sign up, you pay me, you get access to the product. Um, but but I think after a certain point it it makes sense to to go up market and that's what we've been doing like the last like seven eight months. Um, >> is it working? >> It is it is working. I think I think it's working also because we already have a bunch of like like big companies that are not like using the products to it to its like fullest potential. So a big part of it is just like working with them to do that and then like you just end up with an enterprise contract. >> Um >> like that customer success element as a byproduct is also an enterprise discussion, right? >> Yeah. because I didn't know like like when I when I started chatbase I didn't know how the world works you know I assumed like a >> yeah everyone uses like products the same way I use them like because like I I signed up to Stripe I put I I just put like my business information I have my Stripe account I go to Versell I do the same thing it's like the idea of like like long sales cycles and like enterprise contracts and like back and forth with like multiple teams and actually like our tech tech team is like like it's actually like an agency. We don't even like it's not even a part of our business. You have to talk to them. It like all of that was very new to me. But um it just became clear that this is how the world works. And like it's very hard to convince Yeah. like a middle manager at a big company to like try out a bunch of different tools like self-s serve and then decide and like just put put the credit card. Yeah. I think it's they need to like they need a human to like set everything up for them and >> yeah like like go back and forth make sure that the they're using all the all the features and yeah that's that's what I learned and that's what we've been doing uh recently. So but also like I couldn't do that early on because I was I was bootstrapped and that that costs money because salespeople are expensive. >> Yeah. And yeah, um but now now we're operating like we're not bootstrapped anymore. So um we're investing a lot in like the enterprise side. >> And what does it look like? Are you going outbound or are you just kind of picking from the inbound? You've got like like Slack notification saying this person signed up and then like an AE goes and says hey if you need any help let me know. Is that is that kind of how the motion works or is it like more hand raiser right? Is it more like someone then goes to your chat and says, "Hey, can can we talk about you know GDPR whatever?" >> Yeah. So, exactly both like like I think >> we have like different layers. Um like we try to like do the the very warm leads first. So, um so existing companies who are like using us a ton and we think that they can get more value from us and then we can like make them pay more. Uh so those are like I think the the highest priority and then the the second is exactly that like a slack channel that we get like every like high value signup uh that comes to us and then we have like an AI research that company and then like give it a score on like how how how how much of a fit it is and then like we just go through them. Um and then we also like try to get most of the signups to like get on calls with us. um just to like show them that yeah like like any any company that is big enough there is humans here and like we can help you help on board you help set everything up for you. Um and and then a lot of it is like outbound like warm outbound which is uh like people who like interact with our LinkedIn posts for example or people like in our circles or people who are um you know like we met before or like events and stuff like that. Um and then like the the last thing is is just called outbound just like mass outreach. But I think yeah, the warmer the lead is, the I think that's been like working a lot more than just cold. >> Um, Yasa, something else that I like about you, and maybe it's cuz I'm from a creative industries background, is that you're very big on brands. You said before publicly, I read your Twitter, I like your tweets, um, that having a good brand just makes every channel easier. If they already, if you reach out to them and they already recognize you, they're more likely to respond. the conversion rate is higher and so that you have great positioning, copy, design, storytelling. Um, and that not everyone in tech thinks like that, but I think people who do well tend to think like that. But where did your was this just like a something that was sort of natural and obvious to you or was there some kind of outside influence on this? >> I think it was more like like learning given like um the last Yeah. the last few years. I think it was it was obvious. I think also it makes sense that this would be the case. >> Yeah. >> Um but it was also obvious that like the the all the channels just do a lot better if like people already know you and are familiar with you even vaguely. So like paid ads work a lot a lot better if you're just targeting the same people that already see your content organically. M >> uh same thing for for like SEO for example if if they already like have seen you know like your your your uh your brand somewhere and even like you're going to get you know like free back links and yeah like it just makes life much easier in all the different channels and like sales is easier um uh like like uh paid ads is easier SEO is easier even like founder sales is also like now much much easier because people like know me now. So, it's just like there's like an element of of trust and like you don't have that if it's just like a completely random person you're getting on a call with. >> Um, >> yeah. So, so I think that's very important. I think like I think there's a a lot of room for us to grow in that aspect. I think it's yeah a big focus now. >> So, how are you acquiring users today then? So like you said, you started very much with this uh kind of influencer mechanics super selfve. >> Yeah. Self-s serve and then you kind of uh you've been working on brands and like making the product look great, which it does. Like >> how are you acquiring users now? Is it PPC? Is it SEO? Is it AEO? Like what are you like Yeah. What are you really focused on for acquisition? Outbound? Like what is it? >> Yeah. Um so now a big focus for us is still like um organic and word of mouth is is still very big. >> When you say organic what do you mean >> like like social like LinkedIn specifically? >> Okay. LinkedIn organic. And how and how do you how do you make that work? >> I think so we're doing a lot of like EGC like employee generated content. Um so we just like everyone post twice a week every week. about like what they're working on and just like yeah like what they're thinking about and like problems they're facing what's working well what's what's not and I think that's that's been working well because it's the same thing it's also like when when when the AE reaches out to someone but that someone has seen their content before it's just like less less likely for them to to to ignore. Um, so yeah, a big part is is like the organic uh social media, specifically LinkedIn. Uh, a big part is still word of mouth because I think we we like a lot of people know us as like the AI agent builder that like started in 2023 for like uh chat bots. And I think that like narrative is still out there in like some communities. So, um, yeah, we're still growing a lot from that. And then the the third one is SEO and AEO. Um, we were getting a lot of like new signups from big accounts for from SEO and AEO. And we we've been also focus very focused on that for the last 3 months. So, we hire we hired someone who's just focused on SEO and AEO now. And it's um I think it makes a big difference cuz yeah now when I choose any tool I I just ask either Claude or Chad GPD. I want to switch gears slightly Yasa and talk about the product side of building a great harness or wrapper. A lot of people watching this are building one or thinking about building one. And you've mentioned a lot in the past about how you think um harnesses have a bright future and how there's a lot of room for improvement. And um I'd love to get your perspective like what's the difference between someone that makes sort of a mediocre harness versus a really good one that's really effective cuz you said before that sometimes the same model can have vastly different outputs from a great harness versus an average one. >> Yeah. Yeah. Of course. Um, I think like like of course we can talk about the techniques to use and like what works now and what doesn't and yeah like I I'm happy to do that but but I think like that the main lesson is a lot of people like spend a lot of time building you know the harness building like all the tools and context and uh like guard rails and everything around the model to to make it work. and then you wake up one day and you realize like oh like this new like framework or paradigm is like actually the the better way to do things. >> But it's just like one it's it's a lot of effort to like scrap everything that you've built to like start from scratch and also it's like all there's a part of it that's like you don't want to like it's some cost fallacy. You don't want to like uh like waste all of that work and the restart from scratch. So I think like the all the tools that I think are good and like have like what I would call like good harnesses around the AI are the tools that are just like doing whatever like needs to be done now given like everything that we know now and like not sticking to like like whatever they had 3 months ago because they they spent like six months working on and yeah it's just very hard to do that Because every every few months you like there is some new idea or like some new paper or some new like oh now file systems are better now the CLI is better now MCP actually is good MCP is like bad for the context it's just like changes so much >> and like also like the other thing is that you have to like come to your own conclusions you can't just you know like follow whatever is is popular because a lot of it is just like it it is crazy how much content you see from people who don't know anything about anything, but they're just posting for to just like get views. >> It is crazy like Twitter is is full of of that. >> Um, so I think yeah, like one like come to your own conclusions by like like trying trying out the different things and two don't have like too much pride in whatever you have now and like uh at the expense of not like trying whatever is uh is new or like people are talking about now. Yeah. So, I'm hearing that you have to constantly be learning. You need to have conviction. You can't wait around to just see what everyone else is doing. Um, so do you do a lot of user testing and dog fooding to try and help you come to those conclusions? >> Yeah. Um yeah, like we we have like our own like data set internally like we we run like whenever we make a change we run we run like all of those questions on like a bunch of different agents to see how good or bad they are. The problem is like there's just so many variables that change. So like you do something and like it works amazing with like GPT 5.4 before and then you wake up the next morning and now there's GPT 5.5 and like >> now you have to like >> like retest everything, retest the instructions, retest like all the config, retest like how much the model is thinking like I think for example like I think GPT 5.5 sometimes does does a better job when it's not thinking too much when like thinking is medium instead of instead of uh instead of strong. So like stuff like that, you just have to test every single variable that can change and um yeah, like it's it's I think it's just part of of building a good good product is like be willing to do that. >> You said something that you said earlier like caught my attention. You said like, "Hey, when someone signs up, we have a bot that basically does like lead scoring for us." Can you tell me a little bit about how you're like building automations in the company like because I imagine as a company that is like very like AI native right maybe how have you started like built things and like what kind of automations you running how how are you thinking about that is it like are you heavy automation like yeah >> so actually no I I I think when you're too heavy automation you lose like you lose like investment from people like for example we had this bot that um like for every new sign up it like researches them and then it gives us a snippet which I think is good but I think like before before we had that bot like one person from the team would do the research and then they would send a message in the chat and I think like doing that manually is just like produces better results not not not not the text snippet that we're writing but because like oh like it's ingrained in their head that this is like a cool company that we need to follow up on and like if a week later like they didn't respond, it's in their head that we need to like like why what happened with those guys and I think if if like everything is automated like things like that don't happen. So you you like some some things can be automated but shouldn't be. Um but but yeah, I think of course there's like a lot of room for for automation. what we're doing now like for example the we're we're doing a lot of automation for like the go to market stuff so like like all the sales sales motion and like what we're tracking like and like the KPIs and stuff is mostly automated with like an internal tool that that we've built but for example like all the uh signups that come to uh chatbase it like we filter out like the bad emails and then we have an AI I look at like our existing existing customers and like try to see if like if this is like in this industry then like we we know that hey like when we send them a message we want to like say we want to flex that we have this other customer in their same industry and um and other than that like just score them based on like if we think they're a good customer or not if they fit the ICP. Um, but but I think like something that we used to automate that I I we're not automating anymore is just like doing the research about the company and like doing just a text snippet about like what they do and like what the angle should be. Uh, I think yeah, we you need a human to like like have this company in their head and like make sure that they're uh following up with them. Um, but yeah, there's there's a bunch of automation stuff that we're doing for for go to market specifically. I am uh I'm gonna put it out there. I'm bullish on Slack. Like Slack's not going anywhere at all. Salesforce is currently who obviously owns Slack. Salesforce is currently trading at 10 times revenue, 10 times ARR of the whole business. Everyone is so dug into Slack. Like I I couldn't change Slack because I've got so many automations from my HR software, my bots for revenue reporting, lead scoring. You like Slack is Slack's not going anywhere. >> Yeah. Network effect business like that >> network effect and it's like integrations like make it so easy to integrate. I think as long as they uh make it easier to interact with your agents as well, not just employees and stuff, which I think they are. Then I think they're going to be around agent builder. Like how Slack >> I think they do. I think they do now. Do they? >> Yeah. Yeah. Yeah. >> I bet they want you to pay for it though, don't you? >> Uh I suspect so. Um Yasi, you have a very spicy tweet which I I love this. So you're retweeting someone else and you said, "My moat is that other companies don't have me as a founder." >> I don't know if you remember that, but I loved it. I loved it. But it just made me think a bit about um cuz you're a humble guy, right? But you know, everyone's allowed a spicy tweet here and there. But um what how do you think about Moes in like an AI era? Cuz people are saying like, you know, product doesn't matter as much and like now you have to have like a public figure and it's all about brands or about all about taste. Like how do you kind of think about this now in an AI era? Yeah, I think it's it's uh the question that everyone is is trying to answer. So I think it's different from like like I think when you when you're talking about like you know like public SAS companies whether or not they have a mode and like what their mode is. I think maybe that's that's a separate question. What I was talking about is the mode for for startups and like specifically AI startups. I I think the mo I I I honestly like I don't think a lot has changed to be honest even with like all the advancements with AI. Maybe that's the the hot take. I think >> the mo like I think AI just like crazed what everyone can do but but there is still like so much so much you need to do in execution as the founder and as your team and it's like that that didn't change and like there is going to be a big big gap between like a good operator like a good founder and a a good team like a a team that with a good culture that good environment and like they're they're working hard there's is going to be a big gap. Like I think when people say like there's not going to be mode in the product because AI is going to build everything, it's assuming that you just give like the same prompt to two AIs and it's going to produce the exact same thing. That's that's not been the case for a for like ever. And I don't think it's going to be the case soon. I I think there's still going to be like you can get to like 80% of what you want to build, but the 20% is going to be about execution. 20% is so hard, right? Anyone can build a front end and a basic back end for any product now, but it doesn't have users. It lacks execution, right? It lacks understanding the nuance. It lacks the workflows, the integrations, and like you one shot an app, you're like, "Oh god, Vibe coding is so powerful." And then, so you dig in and you're there till midnight and you're like, >> "Yeah, >> oh no." Like it's bad, you know? It's bad. >> Yeah. Yeah. Yeah. So, so I I I I it always happens like I say I'm going to build this in the next hour and like I do like two prompts and I'm already like 80% there and then it's just impossible to make it work. I don't think m have changed a lot like especially for startups uh after AI and and or like after LLMs and before LLMs I think it's just still about execution about you know branding about distribution about like the team like all the all the the product that you've already built the user experience >> um because when people say like now there's no mood because building a product is easy >> I think if if the world worked that way then that means people would be working less now than they used to because like if the if the argument is that AI is doing all the work then yeah like we should be working less but every founder of every company here in Toronto everywhere they're like working like a lot more after LMS than before and that just means >> what they're doing is the mode like all that work is the mode >> I feel like the goal Yeah I totally agree because the goalposts have moved cuz people's expectations change because you can just create so much more and so much faster and >> and on modes so you know I agree with you that like there is still some moat in having a good product still not easy to make a really good secure scalable nuance product with all the edge cases but um what someone can never copy though is your story and I think as a solo founder that's something you do a good job of leaning into like your journey from being a young founder solo founder that bootstrapped this large business you got to use every advantage that you have. And yeah, yeah, that's definitely one of yours, I think. >> Yeah, I I think that's going to continue to be important. Yeah, like the story, the brand, the like how people think of you and like perceive you when they look at, you know, your logo. I think that's very important. >> What do you What do you want people to think of you when you when they see the chartbased logo? like what what do you want them to think? >> Yeah, I I I've been thinking about that. I think there's a few companies that I look up to and I think they're doing a good job. And I think the common thing is um it feels inviting and simple and like I can just start using it. >> Give me some names. Give me some names. What what >> Stripe for example is that maybe like um >> what do you feel what do you feel when you think of stripe? >> So I think um it's not vaporware. It's not like it's not a product that's um just like getting money for no reason. Like it is actually like producing a ton of value to to their customers. Um, a and I think it's it's invite like I think it's easy to to work with and like to set up and to like get get value out of it quickly, but also like it's easy for like someone that's like just starting out, but also it's powerful enough for like huge enterprises to be using it. Um, so I think that's like that's what I want to build because like for me there was like two routes, right? there was like the PLG like you know like sticking to that maybe even like doing like a more of like a lifestyle business and just sticking to that and then the opposite end is you know like extremely like enterprise salesled um just working with like specific type of company but I think what I like about uh like what what I want to build is a product that like is intuitive, easy to get started with and like inviting to use to like smaller companies and powerful enough for bigger companies. I I was going to say I think your your choice of picking Stripe actually says more about what you want to build I think than it says about Stripe. I think what I heard from that is like I want to build like a product that is great for the enterprise that is easy and accessible and that is incredibly robust >> Yeah. Charlie, what do you think of when you when you think of Stripe? >> So I think Stripe are just super reliable. Um they're very user focused. Um it's like infrastructure at this point. It's like an institution now. And um it works with people. They spend a lot of time, they don't need to do this, focusing on like the the little guy, you know, they seem to actually care about somebody who's just starting their business. That's why they start things like um uh they make it really easy to start a deadear company and all of these things that maybe they don't have to do. So I think they really got their ear to the ground. And whenever I've met anyone from Stripe, the first thing to do is sit me down and ask me, "So show me how you're using Stripe. What are your problems?" They all they are customer obsessed. >> They have it baked into them to be really customer obsessed and like um yeah I love that about them. >> I was actually invited by one of the leadership not the leadership someone very actually I think it is like a very anyway a very senior person at Stripe was in London from California and they invited me and like three other people for like lunch at their office in their canteen and it was pretty chill and it was just like how's tribe? How's it going? what you know and it was just chitchat. It was nice. It was really casual and I was like you know that's customer obsession right that is like getting people in the room learning about what they want. >> Uh and also when I think about stripe you know you know when you ask someone like hey think of a color and think of a tool everyone says red hammer you know like when I think of stripe first thing I thought was secure like it's just >> secure just works. >> So I think uh it's an interesting exercise. Yeah, I think yeah, like everyone Yeah. like sees it from a different angle and like but I think what what yeah, I think the point about it being like they they still um focus on their customers regardless of of their size. I think that's that's what I'm saying is like it it is easy friendly to use like like people who are just getting started and at the same time like robust enough and secure enough for for enterprises and >> so something that I'm interested in asking both of you actually is um cuz I remember talking to Sabah just before co they just hit a few million AR and they were talking about maybe raising money and they they you know it was like it's a kind of a one-way door right and so they were talking about the pros and cons and you're in a you know you haven't raised money Yasa but I suspect you could very easily if you wanted to. Um what is your current thought process about that about whether to do it um whether or not and for what reasons? >> Good question Charlie. Good question. >> Yeah, very good question. Yeah. I I mean like a lot of like people who bootstrapped are very religious about bootstrapping and like they're extremely against like it's it's like a more of a moral decision for them. Um like PCs are evil or something. >> I Yeah, it's not the same for me. I think that my my thought process was more um like like I just looked at what my goal is for me, for the team, for the company and I just thought I have a higher likelihood of achieving that goal if I bootstrap. Um because I I think the reason is like when you when you do raise I think the the implic not even like the the agreement not even implicit the agreement is that you're going to like >> work hard on this for the next 10 years to like make it like 100x their investment. Like that's you're hand shaking on that. Um >> technically not true by the way. >> I would just debunk. >> Okay. What is it? I would just debunk that >> like cuz I agreed with you. Oh no no no no. So like just to be really clear um when an investor if an investor invested in you for example this is probably it's the start of being classed as a growth round. So like growth stage investment versus early stage early stage investment is like YC seed round series A right so like up to 10 $15 million. And if you're doing 10 $15 million um checks, then you're likely doing, you know, 80 60 $80 million valuations. Um which which actually means to 10, you know, to 10x uh is is what they're looking for, not 100x, just saying. Um and and the model is based on the fact that like look out of 10 three will go bast will give you your money back. Two will give us a five 6x and one will hopefully should give us a should give us that outsiz return. Now at the growth stage as I said this is probably where you sit now at the growth stage. They're really looking for 3 to 5x uh on a growth round, but they want it to be quite short, right? So like the risk appetite goes down a bit. You're looking for NR, you're looking for um you know enterprise customers. You're looking for quite a robust business. And when I say 3 to 5x, that's like the baseline cuz they if they give you money, 3 to 5x 10 years later is pretty nice for them. But they're also looking for the opportunity for it to be potentially more. >> So yeah. Okay. >> No, no. So I I Yeah. like when I was when I was talking about 100x, it was more like like seed round like like why I didn't raise very early on. Um is because yeah like I just I I just thought that there is a just for my goal of like um uh ju just just like what what I I think success looks like. I I just thought maybe it is easier to achieve that if I bootstrap versus if I if I do raise and then like the definition of success is just just like much much higher. You've mentioned before that you think there's so much more room for improvement in in the harnesses and that you think it could be a um incremental improvement in the models, but there's room for a 10x improvement in the harness and that actually could lead to AGI just through harness improvement as well as some model improvements. Um but this made me think just like what do you think about what how would you kind of define AGI in your own mind and what do you think the path looks like to get there? I I I so I think a big shift for me and like I think for a lot of people is with when when cloud code was launched. So that I think like maybe >> not not launched I think when like it worked well. So I think it was like I don't know like late last year like >> yeah when Opus 4.5 came out as well. >> Yeah. Like I think after like after I think a break, I forget I forget which, but everyone came back and then like the models were so much better and I think the improvement was mainly um the harness the harness just got so much better and like it was able to like giving access to the like giving the CLI access to the model and like having them run in a like a longer loop and like yeah giving them all of these tools uh made them a lot more powerful even if it's the same model and we thought I don't know like in the space we thought like that you know rag is like the now now you did like all this this like cool stuff with rag and then this is it but it was a huge unlock to like do the CLI stuff and um like the file search and all of that and like give access to like all of the markdown files the skills and all all of the things that uh cloud code is doing it was just a huge unlock and I I think It's it's very poss like this this field is extremely early and I think it's very possible that there's going to be like a big step jump like this very soon because >> it also like felt like we we did everything when like >> we figured out like that you know the best way to to like we thought we figured out the best way to do like rag and search and all of that and then yeah like MCPS are like the biggest thing and then all of a sudden like there's a huge step improvement by like yeah like improving doing the harness >> and this field is so early and like there's so many smart people working on it that it it seems to me that >> there there may be a few more uh like big jumps like that soon. >> Um >> and by that you mean how you deal with context management, memory, system prompts, all of that stuff, UX, all of that stuff. >> Yeah. Yeah. So, and there's so much and there's so many smart people like that's the the main thing a lot of people are working on in like different companies and the startups. Yeah. In a lot of places. So, I feel like there's going to be a big jump very soon. Maybe multiple. >> Just on ideas. One of the things that people uh like to hear on this podcast, they like to hear about startup ideas and especially you are Mr. AI harness AI rapper and you've you know had conviction since the early days and you were proven to be correct. Are there any kind of categories of uh rappers or harnesses that you think people maybe should think about building or that are underrepresented so far? >> I think a good idea especially now is is operating like an agency. I think like the first few months you just like >> go to people that you think like those are the customers you want to serve and then like just see like what their problems are try to solve them and then try to solve it for similar like similar uh companies and then >> from there like you'll have like an idea of what the product should look like and I think >> I honestly think like agency business and software business and like consulting it's all converging to me. It seems like like the biggest companies that are killing it in software, a lot of them are operating like with the you know the FDE. It's more like consulting at this point more like consulting with a product um than >> just like mainly the product. And that's what we're doing too. It's like we just like go to our customers and say, "Hey, like this is our product now, but just us your problem and if it doesn't fix it now, we'll we're going to make it fix it." U yeah, >> I think your ability to sort of balance both being fully AI pill with knowing when to have the human collaborative element is fascinating. >> Um and I think that's one of the things that really makes you stand out and I've learned a lot just from this interview. Um I haven't got any more questions and we're already an hour and a half in. Do you want to keep going? >> No, I'm sorry we've we've kept you for too long. Look, let's um let's get you a pay rise number one. Yeah, I'll do I'll send you the picture after >> send I I'll I'll approve it for you. And uh finally, thank you so much for for coming on and and uh telling us how to build incredible companies. >> Yeah. Thank you so much, Yasa. This has been great. Really appreciate it. And uh yeah, hopefully we can get you to London one day. >> Yeah. And I'll see you in California in a month. >> Yes. Yes. >> Yeah. See you soon. Cheers, Yasa. Bye. >> See you. Bye. Bye.