⚡️No, Don't Do Palantir for AI - Brendan Falk, Hercules (AUDIO FIXED)

Watch the full show on YouTube: https://www.youtube.com/watch?v=IYrO9h4KYZc

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:03Speaker 00

  Hey everyone, welcome back to another latest Space Lightning Pod. This is Alessio, founder of Kernel Labs, and I'm joined by swyx, founder of Smiley AI. Hello, hello. Today we have a fun Lightning Pod with Brendan Falk. Brendan, I feel like you and I go back a little bit, but now you're, I guess, introducing yourself as the founder of Zeus.

• 0:21Speaker 01

  Correct, yes. We've gone back, probably like four or five years at this point. But yeah, Brendan, founder and CEO of a company called Zeus, and excited to talk about the pivot that we did about a month ago and everything that we learned.

• 0:32Speaker 00

  Okay, actually you set it up really, really nicely. So I will just queue up the tweet and hopefully figure out the screen share because I just changed browsers again. I'm now back to Arc, the old faithful browser here. Okay, so.

• 0:48Speaker 00

  People can learn about your Fig journey. I was an early Fig user and I now use Warp. Long story short, now you're now you're working on AI stuff. You tweeted about two months ago that you were. We were in conversations for doing a pod on AI needed Palantir, and then you pivoted away. And so we just wanted to cover this because obviously Palantir is a super hot topic. Everyone wants to deploy for deployed engineers these days. Everyone wants AI needed transformations. What happened?

• 1:20Speaker 01

  A lot, and you know, I can really dive into it. I'll rapid fire through my background as to like how we landed on this. And then I can talk through, you know, what the business was, what we learned, why we decided to pivot away. You know, my background, yes, I was founder and CEO of a company called Fig. We got acquired by AWS in 2023, became part of Amazon Q Developer, and I led product at Amazon Q Developer for about a year.

• 1:43Speaker 00

  Which, which by the way, I've heard very good things from other mostly Amazon people, but like, yeah, actually a very good CLI.

• 1:50Speaker 01

  It's very similarish to Claude Code, but you know, some differences and it seems to be going really well. I've obviously left Amazon now, but yeah, it's it's going really, really well. We, I wanted to start another company though, and so what I did was I moved to this crazy team called the Amazon. Private equity business development team. And yeah, that's exactly the face swigs that pretty much everyone gives me when I say that. And it was very deliberate. My role was the global AI lead. So I was working with all of these private equity backed customers of AWS on AI transformations effectively. And what's interesting is Amazon has about, AWS has about 20,000 customers that are backed by PE firms. And why I deliberately picked this role is I wanted to get as close to as many different types of customers as possible

• 2:34Speaker 01

  to sort of think through what I wanted to start for my next company. Private equity doesn't discriminate. They've got every industry, every geography, every company size. And they also, compared to most large enterprises, move fairly quickly. Because at the end of the day, these private equity firms just want to maximize return on their funds, which is usually maximize the enterprise value of sale for these companies. What I ended up doing was working with tens, probably hundreds of executives and like tens of companies on these major sort of AI transformation initiatives. And I saw the full spectrum of companies going from, hey, what is AI? I've heard a lot about it, but I've got no idea how it works. All the way through to like software companies that were just absolutely ripping,

• 3:16Speaker 01

  that are already integrated AI into their products, were going really well. And so I spent a lot more time on these like really large companies. And what's so fascinating is we're in Silicon Valley and we see all of these software companies just iterating so quickly on AI. And we think that that's what the rest of the world is like. But I would regularly be on calls with companies where they'd have a billion dollars of revenue. They'd be in the Midwest somewhere, no one's ever heard of them. And they'd be asking questions like, yeah, so what is Anthropic? You know, like, that's the, the sort of type of question I'm getting. And so, you know, that's one end of the spectrum. And this was obviously like about a year ago. And so things have definitely changed. But we are, my role was basically to work with them on understanding their business, understanding where AI could actually have an impact,

• 4:00Speaker 01

  and then helping them get those into production and actually capture the value of it. And so I was doing this for so many companies and across lots of different industries, lots of geographies. And I decided to go and, and start a business doing this full time for, you know, myself and going after much, much larger companies, going after the global 2000 companies. And I could keep going here because they're like, this is sort of the story. But I, I mean, feel free to jump in if you have questions. If not, I'll, I'll keep going.

• 4:24Speaker 00

  I think you're on a roll, but Lassio, go ahead.

• 4:27Alessio Fanelli?

  Yeah, no, and I think like, what, you know, if I'm in your shoes, I'm like, man, these people don't even know what Anthropic is. Why would I go be the palantir for?

• 4:35Speaker 01

  Enterprises, you know. I honestly got off the call sometimes with these customers and just went, man, you could give me two engineers in like a week or two and we could probably save you a million dollars. But it was just the, and it wasn't this company's fault is they're just not technical companies. You know, I bucketed these companies into tech companies and non-tech companies. And within non-tech, it's are they technical or non-technical, you know, do they have good in-house engineering resources? So take like Goldman Sachs, they're actually quite technical. They've got tens of thousands of engineers, whereas a lot of other non-tech companies are really not technical and they're outsourcing to Accenture. And so one of the key things that I saw was these companies knew that AI was adding value. They saw on the news, hey, Klarna has saved tens of millions of dollars by, you know, customer service automations.

• 5:18Speaker 01

  And Amazon saves this amount and this company saves hundreds of millions of dollars here. So they, they saw that AI was creating real value. They were playing around with themselves, saying it was creating value. And they just, they had no idea how to actually get it done. And so what I ended up seeing is they were all going to Accenture. And what was happening is the board was yelling at the CEO, saying, where's our AI strategy? The CEO was yelling at the SVPs and other C-suite, saying, what's our AI strategy? The SVPs were yelling at the VP, saying, what's our AI strategy? And once you get down to that layer, like the job of these people is not, if you're the head of customer service, you're the head of insurance claims processing, you're the head of something else. Your job is not to be the AI wizard. Your job is to do, you know, your function and your business really, really well.

• 5:59Speaker 01

  And so they were all basically going to Accenture or another system integrator, saying, hey, please help us. And the system integrators were just sort of bleeding these companies dry. They were basically saying, hey, we'll come in, let's do 20 POCs. And none of these POCs got into production. There were all these fun sort of science experiments and they, they sounded good, but it's pretty easy to get to an 80% good solution with AI. AI does most of the heavy lifting, but nothing was getting into production. And so I just saw that, hey, these large, non-sexy, non-tech companies have tons of money. They want AI too. They're actually willing to spend the money on, on AI and to do these transformations. They're probably the companies that could be transformed the most and capture the most value from AI. But they have no idea how to do it in-house themselves. They're paying to outsource it.

• 6:44Speaker 01

  These companies. These are not incentivized to do a good job. The traditional system integrator model is sell more headcount for longer. It's not create a really high quality solution and get it into production. And then on the flip side, the time it's taking for software to build high quality software, to build AI agents is going down to zero. And so I just saw that this is amazing moment in time, strong demand, amazing new supply side. We could come in and build a sort of new business here that helps these large enterprises transform their core operations with AI. And so ended up leaving Amazon, starting this company Zeus. And you know, the idea, pretty simple. Number one, we would only go after the global two thousand. That was very, very deliberate. Two is we are, we were building very custom software.

• 7:28Speaker 01

  So the idea was we wanted to get to a five or $10 million contract with every customer that we worked with. You can really only do that if you go after the global two thousand.

• 7:38Speaker 01

  And then three is we would do the end to end solution for these customers. We would do the strategic advisory, hey, what is AI? How can it impact my business? What should I be prioritizing? Two is the custom software development. How do I actually build this? How do I make sure it passes the evals? How do I even build evals? How do I get it to production? And then three is, once it's in production, how do we do the change management? How do we actually capture the value from all of these AI changes? So we would advisory, the build, the change management. And the idea was if you're getting $10 million, if AI can automate so much of this process, you know, hey, it doesn't matter how custom it is. Doesn't matter what the use case is. Let's just come in and go after and solve these big problems.

• 8:19Speaker 01

  And the, the interesting approach that we took, and this is where the sort of Palantir style comes in is, well, obviously we're very, very hands on. We would go, you know, in, you know, in some cases like to the customer onsite. But two, we'd also have a software platform that we would build our products on. And so the idea was, we're not just going to be a services business. How do we get some sort of high gross margin recurring revenue out of this? Well, we are going to build our, you know, foundry, similar to Palantir, build all of our solutions on top of that. So our, you know, foundry, the Zeus platform was effectively an AI agent builder. You could build, test, observe and monitor and deploy AI agents. And there was also a document intelligence platform. So we built this out and we worked, you know, I can talk through all the customers, but that was the key pitch.

• 9:02Speaker 01

  Surprise AI transformations only going after global 2000, aiming for a five to $10 million contract, any use case that we would go after. And then, you know, maybe you're thinking, oh, man, big scope. And yeah, obviously it was a big scope, but you know, I can talk through what we did. And that's, that was the key pitch that we started with. And obviously we pivoted away from it about six months later.

• 9:20Alessio Fanelli?

  And what was the start not working through the PE firms? Like, if I kind of look at the background, it's like, hey, I work with a PE firm. They want to save money, so I should just work with them to get in all the portfolio companies to do this.

• 9:32Speaker 01

  So this is the, one of the reasons we also pivoted away and same reason with PE firms is, you know, the model really only works if you end up with a really large contract. Let's play out this scenario. Let's say the contract was a 500K contract. All right. You got to, and you'll, as you hear me talk through why we pivoted, all of this stuff takes a lot of time, takes a lot of time to actually understand what product needs to be built and do the statement of work. Number two, it takes time to do the implementation and to do the build out. Three, you have to maintain the products after they're built. Four, things go wrong, you got to factor that in. And then five, there's a lot of, there's still headcount involved here. It's not just like AI replaces everything. And so the, uh, it's the, and I said this in my tweet, it's the amount of work for like

• 10:15Speaker 01

  a 500K contract versus a million dollar contract versus a $5 million contract. It's not proportional. It is like, you know, a little bit more work to get a much larger contract. And so it's just far more optimal to go after these much larger customers. If you are doing this model where there's not really much repeatability, like if we were doing exclusively insurance companies, for instance, then it just makes sense to get as many insurance companies as possible. And if you can get in a small company, go for it, a large company, go for it. But we're deliberately saying, no, no, we're going to build very, very custom software for you, custom AI agents specifically. So it just, it doesn't really make sense unless we can scale up one day to $10 million contracts. And maybe another anecdote is like, if a company is a $100 million revenue company,

• 10:58Speaker 01

  there's just no way they will ever pay us $10 million annually. It's just, you know, that's 10% of their total revenue and you're taking a bet that their revenue doesn't go down. What if they're a billion dollar revenue company? Okay, $10 million, 1%. That's probably one of the larger vendors that you have. What if they're a $5 billion company? That's effectively the global 2000. So we had a pretty hard floor on $5 billion revenue plus. Otherwise, we just could never see a path to getting to, uh, to these, these really large.

• 11:26Speaker 00

  Yeah, and maybe talk about the initial outreach. I'm sure they get a billion emails of we're gonna help you use AI in your company.

• 11:33Speaker 01

  So one of this is the thing. It's if you actually look at the Palantir model, what sort of happens is Alex Karp goes to Davos and he has dinner and lunch with all these CEOs. And they go, wow, we like each other. Let's pass it down the chain. And then, you know, the SVPs of each org have some negotiation about what's actually going to be built here. They do some pilot in a really short period of time. They show unbelievable progress and then they go out and sign a sort of annual contract and show even more progress. And so I'm obviously not going to Davos, maybe not yet at least. But I was one of the theories here was, one, I have some connections through Amazon, which was really good, where we know what customers. Two is I have connections from my last company and through FIG, I have college connections. But also, we picked our investor very deliberately to go after an investor that had these really

• 12:18Speaker 01

  large global 2000 enterprise connections. And not just like, you know, Salesforce and these large software companies is we're going after the really unsexy companies that you've probably never heard of, those types of companies. And so, you know, we really picked our investor to make sure that we could get those introductions. That's a short term thing. In the long term, like, obviously, we have to build our own go to market motion, but we deliberately worked with an investor that had that. So the first like 10, 20 introductions literally directly came from our investors, our angel investors, et cetera, and then some through my personal network as well. Fortunately, solved that problem, but yes, that was a problem that the idea was if we had some really good case studies with these really large customers, be able to get the next five, the next five. And the other thing is, we don't need to get to thousands of customers here.

• 13:01Speaker 01

  You know, if we're going for $10 million contracts, to get to 100 million of revenue, you need 10 customers. And obviously, some of them will fail and some of them will, you know, take longer. But we really needed a small number of high quality intros. And so the investor path was the main way to get there initially.

• 13:15Speaker 00

  There were a few other points that you brought up in your thread. I'm curious if you want to double click on any other one of them. So we talked a little bit about the deployment and lack of use case consistency.

• 13:30Speaker 00

  Any other stories or warnings to other founders that you might want to pass on?

• 13:36Speaker 01

  I can go through, I'll go through the process that we talk to the customers and then for each of them, I'll say like where it went wrong and plenty of stories. And you know, the person is like, you should not do this. It's like, I'll just give the anecdotes and then hopefully founders listening can take it with a grain of salt and decide if they want to do it or not. But you know, so what we did is we would go to a customer, these customers would know that AI was a big thing, but they really wouldn't know how to prioritize. So first bit was the strategic advisory and we do this pretty simple thing, which works really well. And I did this at Amazon was we would basically put together a list of use cases and then we'd plot them on this graph. On the Y axis was the business value that would be created if we actually solve this with AI. And then on the X axis was the technical feasibility. So business value, technical feasibility.

• 14:19Speaker 01

  And what you end up with is this graph in that top right quadrant, you can see like, oh my God, these are slam dunk use cases. You'd often see things like customer service would always, almost always be at that top right of just, look, this is very automatable. The business value is pretty clear. This is usually a use case to go after. And so we would do that for the company and it's pretty quick to do that. That was like pretty easy. The second thing is we would then pick a use case to go after. And so this is something that I actually got wrong and this is if I were doing it again, I do something differently is our thesis was, all right, when you, when you to this industry, how the hell is someone going to trust us with a $10 million contract? Maybe we'll go after the quick win use cases. So like high technical feasibility, but lower business value.

• 15:01Speaker 01

  So it's more like the bottom right quadrant. Let's get our foot in the door. Let's get maybe a million dollar contract out of that or a 500K or something like that. And then we'll quickly expand across the other use cases. And you know what? Because we've done all this system integration for this use case, it's going to make it easier and faster to do these other use cases as well. And what ended up happening is they were okay with that and we'd go after that, but it still ended up taking as much time. And so this sort of goes into the build out where firstly, we'd have to understand the use case, just writing up a statement of work, going through the sort of sales process, working with the customers on what the product needs to be built. It's really hard to do that for one thing, but when you're doing like 10 different use cases.

• 15:39Speaker 01

  Honestly, it's an absolute nightmare. And so this is another thing that we not, I wouldn't say got wrong, but you really have to go after large contracts if you're doing very, very custom use cases. If you're doing small contracts, you just want repeatability. So to give examples of of types of use cases we're working on, these are all global 2000 companies. We had an insurance carrier, we were working on RFPs. This is like a Fortune 500 company, RFPs, so like a community college wants to get a proposal for insurance for all of their professors and they fire it off to the insurance carrier. The insurance carrier quickly comes up with a quote and responds back. That's one. We're doing first notice of loss for another insurance carrier.

• 16:18Speaker 01

  We were doing like a legal use case for a large conglomerate that does a lot of M and A and they wanted to do contract review, like vendor contract review. We worked with that customer. We ended up doing a customer service use case for them as well. We're working with like a large government on like a unemployment related sort of workflows that are needed to help people get jobs and mash them to jobs. We're working with an energy company. We're working with a call center use case for a like a large sort of B two B and a consumer SaaS company that had a large call center for like a core operation for their business. So like Switch, you're smiling. Yeah, we basically ended up building Vapi. We built a Decagon. We built like all the big names is we were building so much stuff just to solve each of these issues.

• 17:02Speaker 01

  And the problem is, and so this gets to the build component is, you know, I'm not an expert on first notice of loss. I'm not an expert on, you know, RFPs, et cetera. So we're trying to get up to speed on these things as quickly as humanly possible. And there are just so many edge cases. This RFP example is a good example. You know, if I asked ChatGPT, what's the most messed up thing that could possibly happen with RFPs? It would give me a bunch of things. I'd go, there's just no way some of these things would happen. And one thing that happened was an email would come through for the RFP and it would say, here's the questions. It's in a word document and you'd click on the word document and there was an Excel spreadsheet embedded in the word document. Like this was a technical thing that I didn't even know existed. Like it was just so.

• 17:43Speaker 00

  And did that?

• 17:45Speaker 01

  Embedded, I'm like, I would have looked, if ChatGPT told me that was a thing, I would have just laughed and just been like, that's crazy. Sure enough, if that happens 1% of the time, but you're a Fortune 500 insurance company and you can process tens of thousands of these a year. That's like a lot of times. And so there are thousands of little paper cuts and edge cases here. And so when it, you know, my thinking coming in is, oh, AI is just going to automate everything. And, you know, software is going to be easy and commoditized. But in reality, you're working through all of these really crazy edge cases that you couldn't possibly imagine. That are not just industry specific, they're company specific a lot of the times as well. So a lot of the time was spent on the messy data problem is we'd look at databases.

• 18:24Speaker 01

  The databases would be missing information, they'd be conflicting information, there would be uh uh wrong information. There'd be edge cases that we couldn't account for. And so this comes back to sort of my thesis around, we could automate a lot of the process. So much of our time is just actually human to human interaction of saying, hey, what the hell is this? How do I fix this? How do I make sense of this? And then, once we finally got the evals working, easy, the AI is really, really easy. But getting the evals, like writing those evals, getting that right, just took so much time. And then this comes back to, well, it takes so much time for a 500K use case, same amount of time for a million dollar use case, basically. It's just, there's so much work that goes into just getting that right.

• 19:04Speaker 01

  The understanding what needs to be done, understanding the product, building out the sort of the, the solution and, and, and looking at the messy data and cleaning that up. Another interesting anecdote here was on integrations. Everyone on Twitter has been saying for last year, MC, not last year, six months, MCP is going to solve integrations. And in reality, like, I do think AI is going to really make integrations a lot, lot easier. But in reality, if your docs for the, you know, the internal system incorrect, the AI has to like wrestle with it. In the same way a human wrestles with, with bad docs to work out what the hell is going on. If there are rate limits that aren't published, you need to work out how to fix that. If there are error messages that are cryptic, you need to work out how to fix that. And so a lot of our time was actually just spent on doing these integrations.

• 19:47Speaker 01

  And it makes sense why companies like Five Tran and Airbyte have hundreds of engineers just dedicated to maintaining these integrations. It's a lot of work and it's not as easy as just like, oh, I'll just ask AI to write the node integration script or something like that. So integrate, I could see a path to AI solving the integrations problem. I really couldn't see a path to AI solving this sort of messy data problem because it's more of a people process problem than it is like an AI software problem. One other anecdote is, okay, we build it, we deploy it, it's in production. The one thing I really underestimated was the ongoing maintenance cost. Same thing, there were these edge cases and even if you think you've sold all of them, they're always edge cases. And when you're a traditional software company and you sell your product to 10,000 people, you know, you have a customer support team

• 20:32Speaker 01

  that serves 10,000 companies and you have, you know, every time you solve a bug for one customer that helps all of the other customers as well. In this case, the ongoing maintenance cost was way more than I expected. You know, you can't really amortize that across everyone else. And so it, yeah, you would just sort of have to like constantly have someone supporting the use case. So Alessio, when you said like, oh, could you just do this for private equity firms and smaller contracts? One of the big things that really happened here is the economics just didn't really work out as well as I'd actually planned out. And I didn't see a path to them getting a lot better. I mean, you could obviously do it, but it only would really work if you either focus on one use case and then go and sell that use case out to 1,000 people, which is traditional, you know, vertical SaaS or just traditional software. Or if you go really, really big,

• 21:14Speaker 01

  you go after those, like exclusively go after one project that can scale up to five or $10 million as opposed to one company that can scale up. I keep saying last thing, but this is also one other last thing is, it sort of felt like there was this existential threat to everything we built for customers. Meaning, you know, I just said, oh, we built Vapi and we built Decagon and we built all these things in our internal software. You know, Decagon and Sierra have hundreds of millions of dollars and they have a ton of expertise in this one specific use case. And so at times I felt, you know, we only did one customer service use case, but at times I felt. We are working so hard to clean up the data and to actually build out this customer service use case, but I can just guarantee that Decagon is always going to be better.

• 21:57Speaker 01

  Sierra is always going to be better because they spend all day, every day thinking about it. And so I worried that we'd be the guys that cleaned up all the data and then at the end, Sierra and Decagon would come in and eat our lunch. And we saw that we did some expert network calls and we saw that even in customer service, within a year, companies are changing their providers. And so this sort of Palantir model really only works if you do a lot of like, you have very low or even negative gross margins in the first year, but then once it's deployed into production, you take the headcount off, the margins go up and you have this nice high gross margin recurring revenue stream ongoing. But if the customer is going to churn, suddenly like all of that just falls apart. And so you either need to have really heavy ongoing maintenance for that specific customer,

• 22:40Speaker 01

  which also just lowers your gross margins, or all of these things, which it takes time to implement way longer than I expected. It takes time to understand what product to build. There's a lot of ongoing maintenance. There's this risk of churn, et cetera. That's why, like, those are sort of the main reasons why we're just like, you know what? It actually makes sense to just focus on one specific use case and go after that rather than trying to do 50 use cases for 50 large, large enterprise companies.

• 23:04Speaker 00

  That's a really good overview.

• 23:06Alessio Fanelli?

  So what's the, what's the learning, I guess? What are you doing now to take all of this, you know, into experience and then doing something new?

• 23:14Speaker 01

  So we decided to pivot and then the three options we basically had were, one is you could keep doing the model, but only go after projects that go up to $10 million. I think that's really the only way this actually works, or $5 million, not even at the company level. It has to be the one project that can scale up to be a large company. Two is you pick a specific use case and you just absolutely go after it for a one company or for one industry usually. That's basically the sort of the two models here. And I think that the middle ground doesn't really work as well. We decided to go back to honestly where we have found a market fit. Uh, you know, my co-founder and I were at FIG and we built a developer tools company from zero to hundreds of thousands of active users. You know, this was all before AI, obviously. And so now we're back in the AI game. A lot of the tooling that we built out for our customers was pretty similar to these

• 23:58Speaker 01

  AI like app generation platforms. I really hate the term vibe coding, but that's what we built out. It was just a sort of, let's get back into the game. Let's go out and, and we, we call this new brand Hercules. So you can go to hercules.app and it's so similar to what we built at FIG and so similar to where our expertise are. And so we're, we're about four and a half weeks in now and it's really, really freaking good. The AI does a lot of the heavy lifting, but just with our background, everything that we've done before, you know, you can actually just build fully production apps. We've done all the backend. We've done all the frontend. We do all the database.

• 24:29Speaker 01

  We don't do any of these sort of integrations with, you have to sign up to Superbase, et cetera. We just white label all of that stuff. And so we're putting, pushing pretty aggressively on that, but I can see a world where we focus in even more. We just wanted to get back to building and get away from enterprise for a little bit. So, uh, yeah, Hercules is our priority right now.

• 24:47Alessio Fanelli?

  I'm curious if your learnings from FIG and this, so FIG for people that, that I know, you basically had kind of like a command line marketplace, you know, it's maybe like an easy way to grasp it. It's like you had these different skills. You could like do things more repeatedly. How do you see the rise of Claude Code and some of these kind of like CLI based tools? And a lot of them are based on, you know, they have this kind of concept of hooks. They have like MCPs for integration. Instead you're going the web way. Is it just because you're targeting a customer that is like less technical? Like you feel like the CLI base is like too.

• 25:20Speaker 01

  Really not satisfying for you and it's not satisfying for me. When we got acquired by AWS, we actually had a non-compete for our product for two years. And so I've got 12 days until that non-compete expires. So I can honestly imagine we will get back into that pretty aggressively.

• 25:36Alessio Fanelli?

  By the time this comes out, I think the 12 days might, might overlap. So, yes, please go on Twitter and see what Brendan cooked.

• 25:45Speaker 00

  I mean, okay, let's, let's go into that, right? Like there's so many app builders. There's so many CLIs. You're, it sounds like you're going to go back in the CLI game. Um, how do you just, how do you differentiate? Uh, what do you think about the cursor problem or the Claude code problem as, as has been going around recently?

• 26:00Speaker 01

  Meaning the gross margin problem?

• 26:01Speaker 00

  The gross margin problem.

• 26:02Speaker 01

  Yeah. So, you know, honestly, I'm not saying I'm, we're going to do some AI app like cursor, et cetera, but it is a world that we know very, very well. And obviously we use these tools every day. So I'll start with the like AI app builder, more like the lovable's, the bolts, the replits. I think. I like to think about which massive company are these newer companies coming after. And so one world you have Figma. And so a lot of these tools are being used for prototyping and design. And candidly, I think that ChatGPT is going to come in and replace a lot of that. Because that's what they did in their GPT five demo. They were really showing that off. These things are not designed for production. It's just another reason to come to ChatGPT. So I don't think it's going to maybe win the market. I don't quite know, but I think ChatGPT is going to pretty aggressively get into that.

• 26:44Speaker 01

  The next sort of type of company you have is the Wix, the Squarespace, the Weebly. And this is more what sort of Base44 is going after. These are big businesses. Wix has a billion of revenue. Squarespace has a billion of revenue. They valued it like five to 10 billion dollars each. And the way I think about it is I think that market is going to expand. You know, Wix is very clunky. It's drag and drop and it was built in the early 2000s. My brother probably who works in finance is probably not going to use Wix. But he would use some sort of AI native Wix where you could chat and build things just because the barrier to entry has gone down massively. And so now he's willing to pay $20 a month where he probably wasn't willing to pay for it before. And the good thing about those apps is they go to production. You know, you're not just building a website for fun. You are an SMB building a website.

• 27:28Speaker 01

  You are a personal, you know, you have a personal website for your portfolio. There's a lot of e-commerce stuff on there as well. The third world is Shopify. That's a $60 billion business. You know, e-commerce isn't going to disappear. So I think Shopify is another area where these back builders can go into. Another is just SaaS. You know, this is a new category that hasn't existed is you can build Wix is like static websites and sort of basic, pretty simple stuff. Now you can build full SaaS. I think that market will be pretty big. I mean, the overall global SaaS market is in the hundreds of billions. My hunch is it will continue to grow as opposed to just shrink massively. Just because now you can build so many of these micro SaaS apps. Honestly, we're still working out exactly where we want to go. What we've built out over the last month is all of the infrastructure that will allow us to enter each of these things.

• 28:12Speaker 01

  And our approach is very much let's make sure that we white label all of these things. So even instead of going to Stripe and getting an API key, that should all just be within us. If you say, I want payments, should just magically give you payments. If you say, I want to send you an email, you don't have to. Go to SendGrid, it is all just sort of in our application. And yeah, it's it's really, really powerful. You know, you can go to it right now and say, build a Twitter clone and it just one shots it. All of that just like completely works. It's really, really cool. We're using convex.dev under the hood and I'll give them a big shout out. Really amazing, real time. They allow us to white label it. Uh, it's really good for writing code. Founder's Australian, so that always helps as well. So they they've, uh, we've worked closely with them and they've really helped us out, make a really great app.

• 28:56Speaker 00

  Yeah, we've had your CTO on as well. Yeah, Convex is getting some some traction. Actually, it was, uh, I just, I just fired up a little app. While, uh, this was going on and it, um, you know, one shot it the game for me, personalized game. And he created that. I didn't do any backend functionality because I didn't know, I didn't know that was supposed to be one of the things, but it looks good. You know, it's got the standard.

• 29:16Speaker 01

  That's got the terrible pink and purple, uh, linear gradient. That's bad. We'll fix that.

• 29:22Speaker 00

  You can have any app you want, as long as it's purple, right?

• 29:25Speaker 01

  We actually have a thing in our system prompt, which is do not do linear gradients unless someone explicitly asks for it. So we try and avoid the. Yeah, we are, we have some, by the time this podcast was released, we literally did a lot of this integration over the weekend. So. Hopefully by the time this is released, Hercules.app. It's really insanely, yeah, it's like very powerful. Lots of stuff.

• 29:46Speaker 00

  Amazing. Yeah, I think like, you know, uh, there is a race going on for, uh, what is the ultimate vibe code backend? I think, uh, Karpathy talked about it when he was talking about his menu gen app for people who don't, uh, know, they can look that up. Uh, vibe coding menu gen is is the name of the blog post. Uh, um, I think, I think there, you know, there's going to be interest in that, uh, you know, you're going to build it, both, Levelball are going to build it. Um, you know, Superbase should build it. Um, and we'll see, we'll see who wins. I guess, I don't know.

• 30:15Speaker 01

  I think this market is going to be enormous and I think we're just getting started. These models are going to get so much better. This comes to your question though, around gross margin, Swix.

• 30:22Speaker 00

  Yes.

• 30:23Speaker 01

  I'm definitely, there's been so much back and forth. Again, take everything I say with the greatest salt. My hunch is a lot of these apps, uh, everyone's saying, oh, they're going to zero, but you know, I still pay for plenty of stuff, which I could build myself. There's probably some. Free open source clone that I could just download and host myself. And it sort of reminds me of the classic Dropbox hack and use comment, which is just like, oh, why don't you just use R sync and just. It's like people are fundamentally sort of lazy and they're happy to pay five dollars just to sort of take something off their plate. And so when it comes to like cursor and and those apps is, yeah, I mean, they're they're obviously at negative gross margins or whatever they're at now, but. I think they have the brand and I think that actually matters a lot. And as crazy as this sounds, I think brand is going to matter a lot in the future for a lot of these apps.

• 31:05Speaker 01

  You know, if the difference between a thousand dollars a month and a thousand and fifty dollars a month is not that much for most people who are spending a thousand dollars. And so. If that's what enables them to get higher gross margins, I I it's the the brand is the thing I think that will enable them to get higher gross margins. I don't know if they'll get to eighty percent, though. I think that'll pro more likely be sort of infrastructure to be ten percent or fifteen percent plus the cost of the API. And then maybe the large language model providers, they are able to undercut it. But I do just think they have the brand and they have the interface. Where I think it's different with these app builders is you're paying twenty dollars for a website. You know, we do have the toggle, which allows you to select which model you want to use, but we'll probably get rid of that. And so at some point, you know, you just want a website that you want to get deployed.

• 31:49Speaker 01

  That's it. You know, we're not going after the engineering market here. We're going after the the like everyone else in the world who potentially needs a website market. They don't care about Sonnet. They don't even know it exists a lot of the time. They just like AI. And so I think that is what enables us to ultimately see a path to actually pretty high gross margins. Whereas the the CLI based and the sort of engineering focused ones, I think you're in a tough position because everyone always wants the best model.

• 32:14Speaker 00

  I think I think you will be competed with from, you know, the other side by whoever has the best models too, like, you know, like. People notice when you dumb down models, like even during the day when people accuse Ntopic of like serving quantized versions of Sonnet and like they're like, they can feel it. I don't know if they can.

• 32:31Speaker 01

  They can feel it, but yes, you're talking about engineers. You know, I assume if we look at your list of followers, I'm sure it's very engineering heavy. Yeah, there is eight billion people in the world. I think if, you know, my brother was like, this is a quantized model, he's gonna be like, I don't think that's it either works or it doesn't. And yeah, anyway, we'll see. I think the I think the space is going to be massive. I think the gross margins for the like app builder type, the the Wix Squarespace world, the SaaS world will work itself out. I think on the the like on hands on keyboard coders, it's going to be an interesting spot. I think they'll probably be fine, but you know, who knows? We'll see.

• 33:04Speaker 00

  We'll see. Well, thanks for joining us. Thanks for sharing your journey. I always think it's rare to get a very honest review of like a pivot and like a, you know, a thesis that like everyone's kind of interested in, but no one actually spent the time to explore. So thank you for being so transparent.

• 33:20Speaker 01

  I'm happy to. I wish all of Twitter was transparent as well because I feel like a lot of Twitter's hype and videos and ARR numbers and whatever. And it just under the hood, you know, I've done startups for a while now. Everyone knows you're sort of faking it until you're making it. And so hopefully this story helps more people come out and be more transparent about this stuff. And I learned a lot. And as you'll see at the bottom of the tweet, it's like learning, failing fast. It gives me an information advantage. And so like a lot of good things happened out of this tweet for us as well. I, yeah, actually think it's the optimal strategy long-term, just to be honest and upfront. Share all these learnings.

• 33:55Speaker 00

  Yeah. Yeah. I'm curious, you know, as a parting thought, like anyone you want to thank that replied to your tweet and like, you know, got you thinking or helped you out a lot?

• 34:04Speaker 01

  There are a lot of founders who replied and said, you know, hey, like they thanked me and basically said, oh, we've been struggling or we've been thinking about this. They were very, very thankful. And then on the flip side, people like, you know, Bob McGrew, who is the chief research officer at Palantir, OpenAI, and he was formerly at Palantir, we had some back and forth on Twitter. Yeah, like a lot of really interesting people reply. It got a ton of traction. I was like not expecting it anywhere near this much. So it's, I think there wasn't anything crazy that I'd learned from the tweet. It was more me sharing the learnings and people being just very happy that I had like shared these. Hopefully if someone wants to still do the business, I still think there's a great business to be built here. And if someone doesn't want to do it because of me, that's also great. It's anecdotes and people can just choose to, to, to use it however they like.

• 34:50Speaker 00

  Awesome. All right. We'll leave it there. Thank you so much, Brandon.

• 34:52Speaker 01

  Cool. Thanks.

• 34:53Speaker 00

  Bye, Brandon.