Hello everybody and welcome back to another episode of Asians at Work. Today I am with Ben, creator of Polsia. Hi Ben. Hello Jorje, how are you? Doing well. I was really excited when I saw your tweet because it's something that I pronosticated that was going to happen, but a bit later. For those who did not see the tweet, what are you building? So Polsia is an AI that builds and runs companies autonomously. So it's sort of like I built maybe a little early or maybe it's just at the right time. The product that everyone has been talking about, the Sam Altman of the world, the Elon's of the world has been predicting that AI will be able to do everything and will essentially take over most of the jobs. Polsia is an attempt to build an early prototype of that. And so the way Polsia works is like, you know, user signs up, gives it an idea and actually you don't even have to give it an idea. You can click on surprise me and it will research you and find an idea that makes sense for you. And essentially I give each instance of a company, a web server, a database, a GitHub account, an email address, a Stripe account, a meta ads account, etc, etc. So I give it like essentially everything you need for a team to be able to build an online business. And after that, you essentially have an orchestration of agents that are going to like write code, launch marketing campaigns, buy ads, response to emails, do called outreach, do research on competitors in an autonomous way. So like the user can of course guide the AI on like what they want, what they think the direction should be. But autonomously the AI will wake up at periodic, you know, essentially today it's every day, but it could be more often that what's the best next step, right? So if there's a bug in production, maybe it's like, okay, let's, before we do any marketing, let's fix the bug. If there's no bugs and like the onboarding flow and the, you know, the analytics show that the product is working well, then maybe spend more time on call out reach, on trying to do some ads or do stuff like that. So the idea is that like, Polci is able to run any project and essentially enable people that have ideas and I've probably tried to love a ball, the replets of the world and try to tell the I wanted what's different about Polci is that like you don't have to tell you what to do, it will do the work regardless and it will never give up and it will always be enthusiastic about and it will help you with strategy, it will help you with marketing, stuff that like other tools, other no good, you know, have builders don't do right? They just wait for your instructions for you to tell them what you want versus Polci will build what it thinks makes sense if you don't give it instructions. So Polci is the founding team that every entrepreneur would like to have, right? Like I have an idea, I want to start and then usually like there is a lot of tasks for me to materialize that idea and then there is a lot of execution that many times one person cannot carry over so you have a team and in this case the team would be Polci. Exactly. Yeah, I mean, you know, I think that like a lot of like starting a company, it starts with like great instinct about a market, it starts with like understanding, you know, what to build, product sense and taste. But after those initial phase of excitement, there's a ton of grind, right? It's like getting the product right and then once it's right trying to reach out to users or maybe you actually you should just start with a landing page and buy some ads to get some intent or get some emails and talk to customers, right? And it's very, it's in that's one entrepreneurship, it's the most gratifying job, but it's also the hottest because it's really a grind to not give up and to you know, get back up when you get punched, which you get punched almost every day, right? And especially when no one cares because no one cares about anything in these days, right? And so it's really hard to get attention. And so Polci is sort of like a new wave of platforms where the AI is aligned with you as a team and like we'll never give up and we'll be always enthusiastic as long as you give it compute dollars for it to like stay, you know, be able to feed itself and like and continue working. You mentioned that you started like maybe you launched a bit earlier, when did you start the working on this? So and I've been an entrepreneur for a long time. I started, you know, I started my career coding. After that, like I had like more business positions where like I was more like a GM, like managing huge teams of like ops, engineering, product like more of a, in a CEO level. But when AI started, you know, when vibe coding started right pretty much like 2024, I started coding again because I was like, this is crazy. I need to code again. And honestly, since beginning of 2025, I've been, you know, coding from, you know, 16 hours a day, seeing the evolutions of all the different models, we got getting crazy and crazy here, which to me culminated with in December with Opus 4.5, which with the Chrome code integration. Sort of like blue, my, that was me, that was the game over, right? But the long story short is that I spend the whole 2025 building a bunch of different products, but Paul's just specifically culminated in like what I discovered where I was like, as I was building with AI, I was like, okay, actually can do anything. So the most exciting thing to me at this point as an entrepreneur is not to build another SaaS or try to target a specific demographic or problem to solve. It's to build the platform that where I could build a thousand companies. So that it started with this crazy idea. And I was like, you know what, let me start at the end state because we all know the end state is that AI can do everything. So let me build that now and see what breaks, right? So that was the concept. So I started building it in November of last year and pretty much like in a month it was, it was built because you know, once you know how to talk to AI, it can go really, really fast. And so the V1 I launched it mid December and things started accelerating in, you know, in January, I already had very, very heavy users. Initially people like using it, you know, every day talking to the, their, their sort of like AI PM, AI CEO, like daily responding to emails. I had a lot of friends use it, which is the first time in a long time that friends were calling me for bugs. They're like, dude, like there's a bug on like, dude, okay, I'll fix it. And I'm sorry to call you. And I was like, dude, don't be sorry. I'm so happy that like you're calling because it's like, usually your friends are like, yeah, I'll try your product and then we try it. They don't even try it or they try it once and like, okay, dude, I don't care about your fucking product, right? So it's been, it's been a great, great learning. Of course, the first product was very raw and like I've been making it more stable and adding a lot more features and listening to customers. And actually, I made the point right now where like, I'm actually making policy a more and more autonomous itself for two reasons. First, I'm alone. I'm a solo, solo entrepreneur. Even though I've never been a solo partner in my life, I've always had teams. But number one, I think it's a forcing function because since I'm alone, I'm forced to automate every function of the policy of the company, right? Which then teaches me how I can make policy the company builder better for all the users. Because I sort of like eat my own, you know, sort of like learn from like using the product and using how and what I've realized these days is that like, I can actually make policy up right now. It's probably 80% autonomous. And I can probably make it 100% autonomous. Being that like, Porsia will essentially start building itself, responding to user feedback, responding to bug reports. And I'm becoming a bottleneck because actually, you know, right now, whenever, like if you're a customer, like, and you use the product and you tell Porsia, hey, like, build me like, I actually want to add a solo now wallet to like my app, right? So I'm not a crypto guy. So I haven't built that because I'm like, I don't get crypto like that. I mean, I get crypto, you get them saying, I can't, that's not my thing. But there's a lot of users that want that. And so the agent that's talking to you will be like, hey, we don't have this capability right now. But let me put a ticket in a feature request. If there's like enough people asking for the same thing, I have another agent that is looking at that list, the PM and that analyzes what are the features that are most asked and then prioritizes it and then have an engineer build it and then a QA engineer that like runs a unit test integration tests and then says, should we push it to production or should we tell Ben? And so right now I configure it to be very conservative and tell Ben. And so then I look at it and I'm like, okay, you know, is it is it safe to push a production? What's the worst case that could happen? Like, and I analyze, of course, it would my taste like, you know, does that make sense? What is there UI or is it just a backend thing? And the idea is that the loop is faster and faster, right? But at one point, I want to get it to the point where like the whole platform is autonomous because if I'm going to build a platform that lets people create any company, who am I to judge what they want to build? I mean, as long as it's ethical and legal, of course, right? But besides those two boundaries, who am I to judge, right? If they want, I don't know, if they need like an integration to like, I don't know, print on the mantichers. So I'm like, that's not the first thing I would build, but like, if that's what people want to do, well, let's build the feature. And so I love the idea of policy becoming fully autonomous, meaning like it would fix bugs, refactor its own code, build features that users want. So not randomly, just listen to the users that are building real companies, listen to the agents that are building stuff and that, you know, agents when they build stuff for a customer, they also can request features and do bug requests.
right, which means that like even agents are like, hey, when I tell them like, how could you do your job better? So a customer asks you to like, I don't know, build like a specific thing, or do a call-out which campaign. The agent actually initially, early on, they were like, hey, like, actually, I would need like a NMCP, like a Hunter's.io or like a database of professional emails because I'm sending emails, but I'm guessing the email addresses and they're wrong. So give me access to like a professional database that's, and so I built it because an agent told me I need that to do a call-out, which doesn't even the customer asking. And so then you sort of have a system where like agents and humans ask for features and like the most represented features and bugs get fixed right away. And you have a self-healing system and a company like that is like, I mean, as an engineer, it's extremely exciting for me to build a system like that. And the fact that I can do it with one person is insane, right? Yeah, it is. And actually, I think that that's one of the most impressive things that of the world that we are living today. You have been an entrepreneur before. You had successful companies. You had to manage teams of people. So you already did most of the entrepreneur thing before. So you know, not only the outcome that you are looking for, but also like how the path is. However, you said like about two years ago, you were like, okay, this is about going thing. This is crazy. Now I can be a solo founder. What are the things that change? So not only from the engineering point of view that most of the people know like, hey, I asked now to connect to MongoDB instead of Postgres. I have never used MongoDB, but it knows how to do it correctly. And about many of the pitfalls, but also in the business world, what did change? I think the core change has been a couple of things, right? I think the biggest changes has been first of all like sort of like Cloud Code or like sort of like and what anthropic launch with the agent is decay, which now, you know, code access the same thing, which is giving the developers that agent team, which early 2025 was built by the loveable on the curses of the world. And it was all part part of it. And it was like, and you know, initially early 25, I was like, I want to I want that. And it was not available to rebuild myself and like the models at the time were not good enough. I think when anthropic like literally dropped and they're like, you know what, everyone can use it, right? And actually, Cloud Code is like the UI, but also you can if any developer can use the agent is decay. I think that was a huge moment that actually at the time I was talking to a guy on anthropic. And I was like, is it because I was like, this is the future. This is insane. Like you can build so much stuff with this. And I was like, is it working? The adoption is being slow so far, right? So I think that was one of the biggest unlocks, I think the second unlock is just the model of becoming so good, right? I think I think like, I mean, that's an obvious one, but like. And especially like anthropic again, pushing the boundaries of of getting the models to be so good at using tools so good at like investigating so good at finding the resolution. And it's been increased your increase your how good are they are? They are and essentially what what increasingly happens is that you can one shot features one shot bugs one shot everything, right? Which wasn't the case in like six months ago at all, right? You had to like, it was a five shot 10 shot situation, right? Of course, MCPs were a big thing, but essentially MCPs in the sense of like, essentially the models being very good at using tools and just. And you make essentially a portion is a bunch of agents that are that are good at doing a certain thing and then using a bunch of tools, I just give them like a ton of different tools that they can use to do things in the real world, right? I think skills has been a huge one. It was a little underrated for a while, but like, it's absolutely huge in the it's it's especially huge for. From a developer perspective, because now it's sort of like the way I can build an agent in a way that like, you know, if you had to define like a what an agent is, it's essentially it's like a prompt and tools, right? It's like it's like it has a mission and it uses tools to get to that mission. But what's really nice now is that like and it's sort of my routine. It's like, for example, like a builder feature, right? So I can like open code. I'm like, hey, like let's add this this this this improvement to whatever I'm doing. And then when it's done building it, I'm like, hey, now I want you to create a new skill called check that the improvement work in production. So then it knows all the contents of what it changed and how to test the result, right? So then it writes the whole skill to and then that I can just call so simply from the CLI, right? Then I push the production and a day later, I just call that skill and it gives me a report about all the impact of that specific changes. And then and sometimes it's like, actually, I messed up like actually or there's another there's a bug that I didn't find and then fixes it and I go, okay, now they just kill again. So you can then update it sort of like sort of like you can iterate with the agent on like what you wanted to output. And at the end of every loop say just save it to your skill, right? Yeah. Essentially, it's sort of like if you hired an employee and you were like, do this thing and then you see it mess up or not doing it by way, not mess up. Not do exactly what you want, which is that's the uniqueness. An agent is really like an extension of one person because they do it. They are tweaked towards what that person thinks the right sort of like SOP is for whatever action. And the fact that you can like you hire an employee like the first person that first you put like, hey, try try fixing a bug and it's like, hey, oh, actually, you did this mistake of that mistake, oh, you didn't understand that actually this is how the product works. So like that doesn't make sense. They'll learn and they're like, okay, now, you know, save it to your memory. So the next time you fix a bug, you remember that and the agent can do that, right? That was that's a huge thing that like I use extensively and open closed this whole system or so at skills where you know you can have a human say something the human the agent does it and at the end is save a skill so that the next time the human wants it, it's learned right. I think that seems to be powerful. It's literally every two months there's a new thing that like changes the game. I think the last piece is like browser use is like a very, very big one that that that antropic jobs in December, mid December, I mean, it was in bed. I before I was a Chrome extension. It was like, you know, you don't really want to use it as a Chrome extension so much. I mean, at least me. But the fact that you could just talk to cloud code and then it opens Chrome and then I can do anything was like insane because like you could literally like any workflow that before would mean that like you have to set up an MCP and configure it and trust it and this and I was like, you know, open Gmail. Look at my emails. Tell me what's important right now it becomes so much easier right because you can the same thing it mimics how you will tell you and just open my open Gmail and I look at this on my emails. This is the way I work. This is I use a star system for an email. So I would say to all the things that like make things. You said many things that I would like to unpack the first one. It's this idea of one shot versus multiple shots. So before to add a feature you need to, you know, like iterate quite a lot. Now you can just say the future and it just gets built. However, many times what happens is that when I want something like the blessing and course of the agents is that they say what they do. And then most of the time I happen to be wrong because I say like, hey, I want this, you know, like just store this thing and show it to the user later. But maybe it's an incomplete feature or maybe it's something that actually breaks some other feature that I did there. Now that you are one shooting most of the staff, how can you keep the thing coherent? Like do you have any kind of harness or any kind of technique that helps you like well, yeah, we're adding a lot of staff. Does it make sense together? Does it break something? Do we have any invariant that we assume that was always true but with these new feature breaks. And then it actually triggers changes somewhere else. How can you manage this complexity? So so I mean, I can tell you like sort of my recipe of the moment of like the way I work. I think that like so first of all, you know, I've been a product person for a long time. I was an engineer but also a product person. And I think that product changed a lot because now so I've two, I've two different ways to work and depends what I'm doing, right. If I'm building product, so like let's say I'm like, okay, I want to add so I recently added ads. So now in one click, you can click a button that says run ads, you pick your budget and it starts running ads essentially start having launches a meta ads agent that has access to so I brought to and then to a subtitle API and then creates content and specific to the business creates the UGC ads uploads it to like an I made a account set the budget starts it and then every day looks at performance and creates you ads like so that's a pretty complex feature but like I was like, how do I make it so simple for the user? Well, it's deleted run ads, big budget and you don't right and then you use the performance. So for a product like that, it's like a lot of thinking and a rating. So initially I was like, no, maybe users want to click a button to create the a video and maybe this immediate section and then like an ad section. And so this iteration, it's a real process. There's no right answer. There's no wrong answer. It's like literally taste and product. I don't know. But anyway, so essentially I use opus for that because opus is like a very pragmatic, very good at design, very good. It does not the best coder, even though it's like it's really good. It's better coding the code and most people on me included, but it's the there's better coding models that codex, which is like the nerd that like will literally like code like every single test and make sure that everything works perfectly, but takes a little longer and sometimes overbuilt and over engineers. This is great because it's like really. It feels like it's a founder and it cares about pragmatism and just getting it done. And so I would usually use Opus, Claude, for most features, especially when I'm building. So that's like one workflow. Another big workflow is like, this actually like a bug somewhere. So like, you know, the agents report and users, I don't know like when, you know, some reason to get this error. And so, and it touches billing. So I'm like, okay. So they implement a solution. And then I'm again, jamming with Opus. I'm like, okay, so what's the issue here? Okay. So wait, wait, but like sometimes it doesn't have all the contacts. I'm like, okay, but like I've you thought about that. So I jam with Opus. And so we talk and then they implement a solution. I'm like, are you sure? Is it safe? So right now, it's like, there's probably better ways to like make sure you push pressure like unit test, integration test. But like in the meantime, like you just say it's safe. Like what's the worst case that could happen? It would ship, ship this to production. So that's what I would use to do. And it works pretty well because it will understand and like really double check and really have a sense of it. But then what I do is like a copy paste the whole conversation and open codex on like 5.3 extra high, right? And I say, hey, thoughts, what do you think? Like do you think like what do you think? And it will nerd out and like really go through every single detail and say, this is what's good and this is what I would add and this is what it would be missed. Then I take the conclusion, give it to Opus. And I'm like, hey, codex looked at this. What do you think? And then codex and then usually Opus usually says mostly right. But on this and this, that's true. And they pick it spotted real bugs. But the rest is over engineered. So that's interesting. So it's like Opus is like, dude, don't over engineer. It's fine. You don't have like a million users. You have like, because it knows how many users I have. So it's like, it's fine. Like you just don't over don't start like adding all this stuff that that may break something else. So and then go back and forth between the two until it's right, especially when it touches important stuff. And I'm like, and I'm like, I want to push the production fast, but I don't, I don't read the code anymore. So I'm like, I just need to get a fill. Yeah. So I would say those are my two main workflows. Opus for features. And then when it comes to bugs and like sort of like fixes that like touch, I usually do Opus, then codex and then I've them sort of like the bait for a little bit until I feel good. And this is mostly the workflow. Like I talk with a lot of people that are building agents or working with agents. And the people who are not building usually say like, oh no, but you need to review all the code. What if it's wrong? Blah blah blah. I have seen that most of the people who had organizations before they actually have worked at a much higher level. So if you're the CFO of a company, a big company, let's say BP for oil industry, you don't know everything. Like you work at really really high level and you're like, well, what if the data is not right? And you're like, well, we put checks down there to make sure that whatever surface is correct, but I am working at a higher level. And with everyone who is shipping really not only fast, but actually like useful to stop on AI, I have seen this transition. Even the engineers that are becoming like highly, highly productive with it, they are doing this transition really, really quickly off. Yeah, I don't need to see all the output. I need to think in systems and make sure that the system is correct. And if there is a back, so be it. Like the CTO of Spotify before was able to operate saying like, I understand how a Spotify works. But if I tell him like, okay, can you tell me every line of code? He's like, no. And if something breaks, you just go to engineers, it's like, this wasn't look good. Please fix it. And you don't even like the city, all those have to go there to read the code. She just knows the sense and said, like, hey, this is not working as expected. So that's what I have seen the most. And then a small thing that I use personally that the front of mine pointed out and I always shared it with others is this idea of these machines are encoding a lot of opinions. So when you ask them like, hey, what would you do? And they usually come up with really good stuff, but it's incredible how far or how much role playing in proof of results. So when you are doing this switching between colleagues and opus, if you give them a role, like, hey, what could a product person with a lot of experience with zero to one applications do versus what could a staff engineer from a big company do? It really changes and they are able to play both characters. But when you say you, it's really hard to know who is going to come back. So it's just like a small tip that can improve like your your workflow or maybe like have better results on. How to deal with with this stuff. Something that people may be questioning right now is, well, what are the features of the product right now? Because you talk about coding and putting ads and they are like, okay, so just like have an NCP there. One of the best interactions with the product that they had was actually emailing you because interplay back on your behalf. So like, hey, this may be interesting. Ben is busy right now. He's building a multi-billion dollar company as a solo founder. But I will surface this to him. Talk to you later. So what are the features that your product has today? So so what you're what you're talking about is a is a feature I launched on Monday or Tuesday on that yesterday, which is essentially I can talk about the features that Paul, you have. But like what you experience is like a like talking to you know, I, I, I, I, is my co founder and Paul, he has a micro founder and talking to Paul, he, I'm jamming on like what, what's the best strategy to grow to grow faster and to things we should raise more money, right? And I think I already have enough money in the bank because I raised the precede and I'm literally burning nothing because it's just me. And, and I have credits from the big labs that gave me credits to build on their platforms like the GCPs of the world. I don't need money, but like obviously, I think Paul, he is such a powerful platform and I can see the retention and I can see users loving it and I see, I see so many people who are so curious about it and like maybe we should get more, get more compute to go faster, right? Get more intelligence into this product through the self-healing to like, you know, and so, and so what I decided to do is that, but also I need to build, right? I need to, what's most important right now is me making Paul, yeah, autonomous so that it can do, can start building and faster features, faster bugs, but I need to trust it, right? And so I decided to start this two week window where Paul Sia is going to raise money on, on the behalf of the company, right? So it's raising its own round of capital. And to do that, I essentially gave it access to my inbox. So whenever you email
[email protected], it will reply automatically. But what I did is that, you know, Paul Sia is connected to obviously the production database and the production code. And so it has all the context about everything, right? The context about our, you know, the product, the features, the vision, the live data, the retention, like all the pipeline, because all of those documents are in MD files in the code base, right? And so it can essentially answer any question that anyone could have. And for example, an investor was interested, like usually investor would talk to me first, right? But I essentially give them Paul Sia who has all the context and more about what we want to build, the vision, the roadmap, the metrics today, the, you know, everything. And also the live dashboard with that launched, which you can go to, which shows the exact metrics, the metrics right now, how many companies, how many tasks, how many messages, and you can ask it questions also on the chat. You know, all the latest five companies, the latest five tasks was in progress, right? And so you can see the system working live. And so that's what I, that's how, that's why it was, but it's like, even if you're not an investor, it also replies automatically. And it's actually, so I did, that's why that's, that's sort of like what happened when you emailed me. What's really interesting is that I've been willing to automate support, right? So for example, like when customers have an issue, they actually talk to Paul Sia in the app, but when Paul Sia is like, dude, I can build that feature right now or like, or you know, you ask me for free credits, but like I can give you credits right now. Email
[email protected]. And then, but when it lands to post it, superimposed.com, I'm like, right now it's me responding, but I could have another agent respond that has more, more contacts can do more things, can actually give credits, but it's a second layer of sort of like, you know, sort of like an escalation, right? But I haven't done it because I'm like, oh, like I need to configure it, but it's perfect. And what's interesting is that now that I configure it for
[email protected], it order response to people, but I'm like, it's actually a good experience because it responds, it tells, it tells the user, I'm busy, which is always true because I'm always busy making the platform better. And I see the emails. So if it's important, I will apply. I reply to you, you got two emails, one from Paul Sia, one from me saying, yeah, of course, like, let's talk tomorrow, right? So I was like, this is actually kind of nice. And of course, I want to tweak it because, for example, there was this, this higher up at Stripe was like, oh, it's post has amazing. I want to partner together. And I'm like, yeah, I use Stripe. So like 100% I'll do the call, but I wasn't responding. And so there was planning, it was starting to be like, okay, cool. Like, yeah, Ben's available tomorrow. I'd like, and here's his calendar invite. Here's the calendar, right? And it was just making up stuff. And it's like, okay, I need to tweak your prompt a little bit so that she don't make up stuff. But, but it's just really interesting how, and it's sort of like my philosophy here is that like, try the extreme of what you, what you think AI cannot do, try that because maybe it can be.
do it pretty well and then you can decide the limit, right? Or it can be an assist, right? You could do all most of the work and then you finish it. But it's quite surprising and then once you get used to it, you get comfortable with you, like whatever. Like, you know, I'm an AI founder working on an AI first platform. I have an AI responding to my emails. Like if it shocks you, then like maybe you're not the right partner, right? So that's like, that's what you experience with the email. And by the way, like, I told you about the AI engineering teams, engineering team that like fixes bugs and creates new features autonomously. Now you experienced my sort of like support email agent that I can, that I'm probably going to extend to support at post.com, but it will be different sort of like the role, right? And prompt and because it and give more tools. I also, whenever I do marketing, like use some routines, like some skills for for marketing, but like I would want to fully automate it and give it more scope. And I think that like with those three, you pretty much have built full automation because it's like if you have product engineering solved, support solved, and marketing not solved, but like autonomously running. And then you can decide, you can decide in your direction if things should be tweaked. For example, marketing, it's like could be anything. You could do like Google ads or you could like tweet every day or like tweet, find attempts as a day or you could tweet in French or in English. Like that's taste or that's like a perspective, right? But beyond that, you can automate whatever your taste is, is your taste. It's like that's your perspective on the world. But then you could automate that then like, it's sort of like if you had a higher, like a full team that could cost you like hundreds of thousands of years. And suddenly it's a 50 bucks a month. And and actually I have a hot take like I think that like in 2026, companies that are not 80% autonomous will die. Like they will especially new companies companies without revenue that are not 80% autonomous will for sure die. Why? Because if it's a good idea, there will be 10 copycats. And the teams that make their company 80% autonomous will blitz gate at local at no cost, right? So you have like bootstruct people just blitz gating, creating all the features you're listening to users, building with user feedback, accelerating, and using the profits to pull back into more reach and more marketing versus any team that that needs to go raise capital to like start and then as to hire an engineer, then the engineer doesn't work on weekends and also they don't agree with the direction. So there's debates in the company and then all that time is where the team that's 80% autonomous will just build. And that's kind of what I'm trying to do with Pulsia, right? Pulsia is essentially giving anyone this 80% autonomous team. Technically it can be 100% but like people that sign up to Pulsia today at least still like have an idea in mind. So like they guided with the 20% but 80% is what Pulsia does most of the things. And like and over time I will give it more and more capabilities. Not I will like Pulsia will give itself more capabilities who cover more use cases better. It actually learns cross company. So like if if an agent, for example, that's called that reach, it will have a memory file for that specific company. For example, let's say for you, it's like you're like, oh, for your specific podcast, like the call that reach agent, like you never reach out to like very big, you know, very famous people who doesn't have never worked and I don't like to this but like I want you to like to target like more like solo founders that are like, you know, at their early innings because that's what and that work on agents, right? So don't talk to like don't send emails to Sam Altman, right? For example, so we'll learn for your company. But then also as it does call I reach for your company, it will learn that like, Oh, interesting. When I put emerging in the subject line, that got an answer and when I didn't, I didn't get an answer. And so it will save that learning, I don't know musly into a shared memory file that can be read by the next that the same agent the next time it runs for another company. So it will not be like it will not, they will not prepare your data about your company, but it will learn that emojis work well in subject lines. And so over time you have this shared intelligence that makes the product better and better. And that's extremely powerful because that means that like you build your business on Paul, yeah, your teams always remember the best practices, the shared best practices together. So you sort of, you have your product get better and better as there's more companies on the platform. So all those ideas, by the way, they're like, they're, they're, it's like everyone's towing with the stuff, but it's like it's, you know, it's, it's, you start to mimic how teams work, how humans work. This is how it works, right? We already had before, uh, AGI, uh, general intelligence that it's humans or the most generating that we know it's humans. So, um, it was crazy for me to see that many people like, this is teammate that and it's like, oh, agents are going to change everything because, you know, for first time, we have intelligent technology. And you're like, well, when I walk on the street, there is quite a lot of intelligent, biological, uh, technology. And I see them cooperating and say them coordinating. Um, so there is a lot to learn from there. And I think that there is going to be quite a lot of things that we can extract and apply to agents. Something that, uh, from your hot take that, uh, I think that it applies a lot if you are a VC or something like that, uh, or even an entrepreneur, it's this idea of, well, for the last 15 years, we had more or less a formula. Let's say for the last 17 years since YC started like packaging the, the, the staff on how to be successful for companies. And there is a lot of things that it was you were able to use the knowledge as a good entrepreneur and then take advantage versus other teams. So if you are doing a consumer company, uh, or a consumer app, you need like, okay, I need to use ads. Then anyone who was not using ads, uh, was just like, uh, destroyed by you because you're like, well, I have done this so much time that, you know, like that team has a good idea, but unfortunately they don't know how to apply it to the consumer market or the opposite to B2B. And then anyone who had been before doing B2B was able to take advantage of that. Same happened with the specific technology. So cloud, cloud computing, nobody starts and like I need to raise money for a rack of computers. Like if you do that, you, you are dead. So before these things are really applied. So if you don't follow the new technology that it's 10 times better, you are like kind of washed up, but it was per department. So, you know, like most of the time was like, well, marketing, okay, that team doesn't know that, but that the engineering is strong enough to cover up for it. And now with the eye, it's going to be, as you say, like if you don't automate 80% of the stuff and actually it's not only automating it as you could do, it could be better than any entrepreneur by themselves could do because you're going to say, well, I am doing a consumer app is the first time, but marketing it's covered by AI agents and they are doing a really good job compared with the baseline of a group of people that just gathered to do it. So I am really interested to see, I don't know if it's going to be by the end of this year, but I am pretty sure that by the end of next year, so before the end of 2027, we're going to have like $1 billion company by one person that it's not just like a one point trick of, you know, like it was growing really fast, Facebook bought me or something like that, but actually a company with real cash flow. And the moment they showed us, or they show us how they structure the stuff, it's going to be so different to our way of working that we're going to be like mine, well, I want to say, okay, we need to copy this system quite quite a lot. And actually, I think that it may be one of the companies that it's building on your platform. Yeah, I mean, to me, it's like it's crazy because, you know, sometimes I talk to very smart people, but I don't work in tech that maybe work in banking or work on whatever, right? They still, they don't, they don't know because they're not exposed to all this stuff. I think OpenClose was the first mainstream, but I don't even know if it's mainstream actually, because maybe I'm too peeled into the tech, tech Twitter that like I think it's mainstream, but I mean, it's not at all. Maybe Openair will make it mainstream. But essentially, most people, they think AI is chargeybt. Why? Because chargeybt is used by a billion people. And like that's, it's almost like a synonym, right, of AI at this point. But and and chargeybt because they're trying to give intelligence to to everyone, which is a great mission. They and and they give one to give it for free, or like close to free, they sort of have to make sure that like most requests are not too expensive, right? And so the that's why they did the router, which is a very smart thing, which is like, hey, for most questions, like if the model thinks he can answer without looking at the internet, just just respond right away, right? But I essentially got a lot of very smart people tell me like, yeah, I don't know, AI, like, you know, sometimes it makes mistakes, right? So first of all, maybe because they tried it three months ago, whatever, right? But that's in my mind, it's because they're using chargeybt and chargeybt by default. It's always trying to point you to get your requests to be the cheapest possible. And I'm not I'm not I've no insider information. Maybe that's on the case, and maybe that's actually no like whatever, they probably know better than me. But that's sort of like what my hunch is versus when you use cloud code, cloud code by default is open 4.6 thinking ultra thinking with them. And then you usually text a max account of like 20 bucks a month like because it's a professional tool. So then you it's the opposite by default, the product is max intelligence. Right? So you have chargeybt by default is
It's like the cheapest cars because we don't want people to use thinking and pro because it's way too expensive and we want intelligence affordable and dangerous for everyone. And CloudCode, which Huawei has like 10 million users worldwide, I mean, it's blowing up, but like I'm saying, still very small version. So the conclusion of all this is that most people, they don't know, right? They don't know how powerful those systems is. OpenCode is sort of like open their mind because it's like it was such an easy way to interact with CloudCode, right? Just via telegram or whatever, right? And there will be systems like Porsia. Porsia, you just talk to it via email. And it just has all this stuff. It sends you emails every day about, hey, here's what I've done. It's very playful. And there's a lot. But I think that like, I think that like this sure is going to be about democratize, like parks that democratize the power of AI that like most tech people have seen for about six months, right? But most people have no clue. So that's sort of like what's a little scary is that most people have no clue, which means that like there's an inferred vanish to people do. And that's why I'm trying to build a product that is, I'm, you know, right now I'm charging $50 for the subscription. It's like, literally I'm breaking even on that. Because I'm like thinking I want the most people as possible to experience this product. And then I make 20% on every transaction on the platform, right? So if you make money on the platform, I take 20%, which is sounds like an app store fee, whatever it's expensive, like at the same time, I'll make it as cheap as possible for you to get in and give it a shot, right? So I think it's fair. Building a whole business for $50 a month is actually pretty cheap, right? But but essentially like tools I try to democratize this insane intelligence and this insane agentic tools. I think I'm hopeful that like it will even the playing field a little bit and make sure that like creative people get the right tools to build, to impact the world. And it's not just going to be like, you know, a huge hedge fund that like just uses like swarms of crocus and open clothes, whatever. And just like takes over the world. It's really really crazy because even right now what we believe that it's mainstream, it's not. So like in tech, for example, everyone would say, like do you know Figma? Of course, do you know Adelaide? Adelaide? Of course. You go on the street, you ask someone and they say like, I don't know what any of that is. Like they know Apple. But in our mind, more or less, all these is the same level because who doesn't know Figma? Like everyone uses Figma, not really. And then also if you move about countries, like it's even less, that's something that also happened a lot in the Bay Area. Like one year ago, everyone was like, you don't understand how much better son it is than chatgipity 4.5 or something like that or 4.0. Everyone is using it because of that. And you're like, no, nobody's using it. Like most of the people don't even know that there are different models of chatgpt. It's they call it chat and that's AI for them. So now with global, it was the same. Like everyone is like, hey, everyone knows this. You walk into a hospital. And it's not only the nurses and the doctors that they don't know about often, like you ask the people who are in the hospital and they would also say, like, I came here yesterday. So if this is from last week, I should know. I've never heard that name. But when we were in that little level, something that is interesting because it was aligned with the next question was about pricing and cost. So you said that, well, capital is always good to grow if you have product market feed or if you have like a enough pool from the market. But that the cost of running this was not that expensive. What are the cost and where do they come from? Like if someone is working on Asians, like are we talking about hundreds or thousands of dollars per month, versus how are they distributed? It's mostly in coding or it's in customer service. You have like first-selling information of this. Like do you mind sharing it? So you're talking about the company cost or like the Unicognomy of the product, all both. Actually both, because it's really interesting for people who are building today, because they, like, you are really at the age of this. So let's talk about the company level. At the company level, it's really, it gets weird. Meaning, meaning, you know, I've actually raised some capital precede like six months ago. I pay myself a salary, you know, right now I'm in SF, so a few expenses here and there. But like the burn is so low, right? Because I've just one person. And in terms of like AI tools, I use mostly cloud and codex. So I pay 200 bucks a month for codex, max, whatever it is. And 200 bucks a month, I actually have three, maybe they're going to buy me for a please entropic bound by me. I have three entropic max subscriptions, because I use so much of it. And I have the agents running in the background doing tasks that like, you know, midweek, they're like, oh, you finished your weekly quota. And so then I sign out and sign back in. So I have three of those. But that's 600 bucks a month. It's not crazy. So I essentially, let's say three cloud code, three cloud max subscriptions plus one codex. That's pretty much everything I use. That's 800 bucks a month plus my salary plus, you know, I work from home mostly here. I'm like working at like a friend's house, hacker house. So like, I have zero cost, right? And it's like, so that's the cost structure of like, of like how, which is unprecedented that like I can build so much and be able to be actively building the product, actively building the infrastructure. Like right now, it's like, I'm getting so much usage right now because of like the tweet and like the buzz, I mean, the mini buzz around what I do announced me yesterday. The infrastructure is good. The marketing, of course, I have time to do marketing and like, do all this stuff. I have time to do support. And so that's what, so for me, the cost structure is like still, it sounds zero, but it's like extremely low, right? Compared to friends that have companies, where it's like, they have, you know, 10 people, 20 people and it's, you have to pay the salary, you have to pay the office, the offside, the this, right? So, and I think there's, you know, the up and close guy, right? Also, it's just one guy. And I mean, as, I don't even know, I actually was losing money. Same thing, it was like, I mean, whatever it is, right? But same situation, right? Just one guy being able to build a system that like, you know? So that's the cost structure of my company. Now, the way I think about business models in the age of AI is I don't really want to be a token reseller, right? So, if you think about, again, no offense to all these other companies, but like the cursor and the lovable of the world, it, you know, it's like, you sort of subscription for 25 bucks and actually cost you like 12 bucks of credits and then like, you sell, you sell credits and then you have some margin on the credits, but really the entropics of the wall makes most of the money and then like, you know, that sort of, I mean, again, I'm not an insider, maybe they, they're, they're, I'm sure they're worse than this and like, there's different business model, but like, let's assume, right? Well, actually, like, cursor build their own model, that's really smart, right? So like, they now write it to their own model, what they would do all the margins. So that's actually a really smart model. But what I think about it, where I think about it, I'm like, you know what? Like, I don't, I want to make it, I don't really make money, a lot of money, the whole money on like, on the subscription, the access to the platform, which is essentially mostly compute, right? Where I want to make money is like on the economic output. So like, when you like, essentially think about policy, yeah, it's like you hire a team, you hire an agency, a full stack agency, right? And you're like, here's my idea, like, just build it, like just build the website, do marketing, answer to support, like whatever, and like, whatever are the tools that you're gonna come out, like, just use them all and like, just do it, like, make it happen. Like, my idea is amazing, I believe in it, right? I know my, I know my niche, I know my community, I know they need stuff like that. Like, you know, I'm a personal trainer and I know that, my other personal trainer friends, they have this pain point, and like, I know exactly how, I can vocalize the pain point, I can vocalize the product, I just don't know how to build it. So policy as your team, and it's like, you know what, your team is like, it's almost like an agency said, you know what, it's gonna be at cost, right? But like, I'm taking 20% equity, right? Which by the way, some agencies do that, right? Some design agencies, they're like, they do, I mean, it's not like that, but like, it's like, it's like, it's gonna be a hundred games that have 200 games, but like, I take a point, right? I'm equity. So policy is a similar model, where it's like, I'm trying to make the subscription affordable, I cannot lose money on it, right? Because like, I'm not gonna scale a business on losses like that. But for the companies that make money, which they do to my, because it's, policy provides a stripper count. So the money arrives on the stripper count, I take 20% and they keep 80. So like, it's like a free agency, almost free agency. And then it, and so I think that's the right way to look at it because essentially you're aligned with the customer's success. And also you take a cut of the economic output. You take a cut of like the value, because the value you provide with that team is, the value is like building a real business. So I'm taking a percentage of like the final value, right? And I think that's how businesses in the world of AI are kind of converged. For example, if you build like, you know, someone builds like an agency, an AI agency that like offers a service, design services, right? And actually in the back end, it's like AI agents that like, look at your brand and like use like whatever like, nano banana to create a bunch of design assets and send it to you, they're not gonna charge you for the compute, right? They're gonna charge you for like how much would it cost for you to go to a real design agency and maybe charge like 20% of that. I mean, you're saying like charge way less, but like not, they won't charge for tokens. They'll be like, oh, I cut me 20 bucks of tokens, so it's 30 bucks, right? Or 40 bucks. And I think that's the right way to see it. I mean, we're going to be doing this.
was how it evolves, but many times it really packets even one level up. So there's going to be like the commodity thing that only the big laps provide, like nano banana, like I generate images, but people are not going to directly use nano banana. That they're going to use to some service that says like, hey, you don't want to generate images. We're doing advertisement here. Yeah, you need images, but this is not what you are charged. Like you are not telling me generate an image. And this is actually how it works with people and how most of the economy started like leveling up. Like if you see when we move from as we go to the kind of services that we have today, you can see that more and more things are prepackaged. And that's why many people usually fail to predict how things are going to evolve because they say like, well, we're going to run out of jobs or something like that because you know, like they already can package the food. So without people, why do we need people for them? And you're like, don't worry. We can always work at higher levels. And something that I would wonder then is that you mentioned at some point, you needed something like a cloud code. And I know people who were working like they brightly identified this. And then they say like, okay, let's build this tool that you know, like for developers, it takes these tokens and it's tailored to coding. However, they were not successful at marketing it like people see that it's like, well, I don't care much. But then when cloth appears, so someone with a lot of names and like, hey, this product now exists out of the blue people are starting using it. They get mad because they said that was my idea. It's just that I was earlier, something like that. And do you worry that this could happen to your product like big laps? They start saying like, hey, wait a second. So yeah, like this is kind of working, but still most of the people don't know about it. We're going to add it like one of our offerings. Or do you think that there is some competitive advantage, even like specializing on this as a smaller player that then can become really, really big? By the way, this is something that Cursor did. And many people would say, well, big models are going to capture that. Cursor is doing really well. So do you think that the same applies here? I think that like AI is going to take over the whole economy. So like the whole economy. So the question becomes, well, the whole economy be swallowed by the labs, which by the way, it's possible between like the Chinese labs and the US labs and the European labs. It's possible that they want if they're aggressive enough, it all depends on like how aggressive they are. And by the way, if you go back 10 years, you could assume that Google could swallow up also all the economy, right? Because and by the way, this sort of has been doing that little by little. They swallowed search and they swallowed media with YouTube and then they swallowed now, they're swallowing transportation with Waymo. And then this they're doing it. And then, you know, Elon is doing it the same way, like I was swallowing internet communication now. And like he has cars and then he's also doing transportation. And so it's already happening. The question is, there's always has been in those 10 years where like all this massive, you know, brilliant entrepreneurs have like taken taken more and more of the economy, right? Through like their ventures. There's always space for entrepreneurs to take a slice, right? I think my point of view is building like this, this very opinionated ecosystem where like, you know, at the way open, you can ask it anything, Paul Sia is very opinionated. It's like, it's a sort of way to build a business like if you can click on run ads, it does no Google ads right now. It's like meta ads. It may change, but like because people may I it's got features, I may edit, but like it's a simplified world. It's a simplified company building, right? The same way you have square space, which is like a bunch of the templates and and then you have like maybe WordPress, which is like more open ended or, you know, you know, I think you have always have space for different types of people with different types of, you know, Cloud code is a billion dollar AR business and it can technically do what Paul Sia can. But then Paul Sia doesn't talk that could people can use Cloud code, but then charge GPT is like very mainstream, but maybe if they had it like if they added Paul Sia and charge GPT, we became really confusing, right? Because it'd be like, I don't get it. Like I'm here to like ask for advice on my relationship and you're asking me to build a business and ask me to like create a fund of businesses and to, you know, ask you, you know, it's and you send me an email every day on like my business like, you know, so maybe they will and maybe they will stay in their lane, or maybe they will do like, you know, just a friend yesterday would tell me about Mini Max in China, where like they literally rebuilding all the businesses and their goal is like to build every single vertical business so they can train the LLM on every single business business lines. I mean, yeah, I mean, make sense. It's like it's like the empire building like every, you know, it's like the the end set of capitalism, right? Yeah. The good thing is that this like replace always, like it happened before like IBM, for example, own computing. And they would say like, well, of course, they have everything. They have like the best database. They have the best chips buildings. They have the best and then one by one goes goes away. So even if charge GPT tries this or open AI, it will always happen that other people are able to find the cracks on the new new things. And also something that as an entrepreneur myself, sometimes it's hard because in Twitter, they are constantly saying like, hey, you're going to fail. And because, you know, like the AI can do it. But you also need to remember Google has been extremely successful with ads. And then you say like, okay, so they own the market and then you check, okay, so today versus 2017. Sorry, 2007. Do we have more at tech billion dollar companies or less? And you're like, well, actually, we have much more. So how is it possible? I thought that they succeeded like this while of everything. So I think that the same is going to happen with with all these businesses. And so it's funny for me to see in Twitter how people say like, you shouldn't even try AI is going to do that. And then I remember, okay, like if I am one of the at tech companies that it's like valued in intense like Facebook, okay, I will take it. I thought that Google won, but yeah, I will take Facebook. Don't worry. This thing of Twitter, it's actually really aligned with the last couple of questions that I usually have for closing the episode that is saying positive and narrative. Like, what would you like to see more and what to like to see less? Which one do you want to start with? I mean, I would start with the more. Okay. But I mean, it's aligned with post-yes vision, but like I would like to see more entrepreneur entrepreneurship, right? So I think that like the tools like Paul Sia, but also like all the other tools are going to come out this year and all the improvements with cloud code and open cloud all this stuff. It enables people's creativity to come into form in a much more easy and easier way. And again, with the 20 20 80 rule, where like 20% is creativity and taste and 80% is the ground that AI can do. And so I want to see a lot more entrepreneurship. And I think that there's a lot of people in the world that like, you know, have a lot of side hustles and like our freelancers and stuff like that, our influencers. And now they can they can have a lot more businesses and a lot more business lines and try to like put their creativity for their communities into play. And then I really, I really see a world where like we have a company, a company in explosion of like, of ideas of services that are like going to be ultra tailored. And I'm excited for that. Yeah, I think that that's good. Good for the wall too. So in terms of like what I want to see less off, I mean, I do less, less people less on their phones, but I mean, I'm a victim too. So I think if someone figures out a cure to this, that'd be great. Yeah. I am reading the book Deep Work right now. And I am actually in the one of the chapters is quit social media. And I haven't been doing it for quite a while now, like let's say like six weeks, something like that. And your brain was so much better when you don't have a constant input of things that may seem somehow relevant. But then when you look in retrospective, you're just like, okay, so in 2021, I was really connected. What came out of it. And then you know, like it's always like bake stuff. Like no, no, because it's not you're going to miss the stuff. But when you see it after many years, it's like, yeah, it didn't matter. So let's see if we move there. Unfortunately, I think that we are going to a wall where people are more connected than than ever. I hope like maybe with a I would have more free time or something and people that start taking care of their health and especially like mental health with with this. 100%. Perfect. Ben. So thank you very much for coming. I had a lot of fun and looking forward to you know, follow your progress. Sounds good. Thanks, Sorry have a really good one. Bye.
PodcastTranscript AI