Inside the $11M YC Startups That Took Over 2025

Frizzle uses AI to grade handwritten and electronic math for teachers. Being able to kind of just grade hundreds of worksheets in under a minute.

Floot is a vibe-coding tool for non-coders to build serious apps with AI.

Sagekit is lovable for automations. We enable you to build full automations just by chatting with AI.

April is a voice AI executive assistant. It helps you manage your email and calendar completely hands-free when you're on the go.

With Cacao, we're helping millions of remote workers, freelancers, creators get paid in USD, convert instantly to reais via Pix, and spend worldwide with a Visa card, all powered by a non-custodial wallet.

Flywheel is Waymo for excavators. We essentially enable remote control on existing excavators for construction contractors. That increases site safety and productivity for them.

Welcome to our latest NEW ECONOMIES episode, where we are talking to the hottest YC founders from the latest Summer 25 batch in an office hours format. We find out who they are, what they're building, and how you can learn from some of their biggest mistakes and early wins.

Why are we doing this episode? Well, we want to help you build great and huge companies. But as every founder knows, building companies is really hard, and finding thatright talent is even harder. That's why we have partnered with Oceans for today's episode.

Oceans connects founders and execs with highly vetted, U.S. caliber talent at 70-80% less hiring costs. I'm talking everything from SEO Specialists, Executive Assistants, Performance Marketers, Financial Controllers, you name it! I wish I had this access to this type of talent when I was building my first company, as it would have saved me a ton of time, money, and of course we had that trial and error always to go with it.

So if you're scaling and looking for talent that can make an immediate impact, head over to oceanstalent.com. The link is also in the description. Now let's find out what these YC founders are building.

Frizzle is the AI grading platform for math teachers. Math teachers spend over 10 hours a week grading homework, and existing tools struggle with math, especially handwritten work. That's where Frizzle comes in. With Frizzle, teachers upload student work and get that same personalized feedback in seconds.

Frizzle provides feedback in red ink on the page for each question, for each student, just like a teacher would. Thousands of worksheets have already been grading using Frizzle, freeing up teachers to do what they do best: helping students learn and grow.

Frizzle is the first grading system that actually works for math. Try it out today at frizzle.com.

Abey, welcome to the show. Great to have you here.

Thanks for having me. I really appreciate it.

So what are you guys building? What is Frizzle all about?

Yeah, so Frizzle uses AI to grade handwritten and electronic math for teachers. And the reason we built this is because teachers told us that they were working 70 hours a week, with 10 to 20 hours of that a week spent just purely on grading.

And it seemed like an easy opportunity to give them that time back, especially now that AI has, you know, developed so, uh, rapidly. And we also talked to these teachers, and they told us that classrooms are going to actually back to pen and paper.

Because, you know, students are using AI to cheat, and if you have to handwrite something, it's much harder for you to copy and paste something in AI. So we also want to support that handwritten piece for teachers as well, because they're going to now have stacks of work to grade themselves.

That's pretty cool. So if I understand the product correctly, essentially the student does the homework, the teacher uploads the student's homework to Frizzle, and your AI tool will automatically grade the student's homework.

Exactly. And it gives back reading feedback on the page, just like how a teacher would mark it up. So not only is it saving teachers' time, it's also helping students get this personalized level of feedback so that they can learn from where exactly they went wrong.

That's really cool. How many hours are you saving for the teachers?

Yeah, I mean, we've heard 10 hours on average, for sure. I mean, so, you know, with smaller, like for younger class teachers, like where they only have one class, it's kind of less time they're saving. But for like high school and middle school teachers, where they have multiple sections and like that's like 150 students, we are saving them like over 10 hours a week.

Yeah, I was going to say, I mean, if the teacher's doing 150 students, they will have homework once or twice a week.

Yeah.

If it only takes 5 to 10 minutes, like per student, that's a lot of hours.

And that's exactly what we've heard. And like being able to kind of just do all of that in under a minute, like hundreds of worksheets in under a minute, has just been like mind-blowing for them.

That's so cool. How did you come up with the idea?

Yeah, so, so me and my co-founder, we always had a passion for like math education,right? So in high school, we started this math nonprofit called the Midwest Math Circle, which is like Chicagoland's leading math circle, still runs to this day.

So that's kind of where our passion for math education came from. And then, like, me and my co-founder, like, we met in high school, then we kind of split ways for college, and then reconvened back in New York City as roommates.

And here, you know, you always hear about the bankers and the consultants working 70 hours a week, but you never hear about the teachers working that much. Because, like, I interviewed a bunch of my close teacher friends, and that's what they told me.

And they said that they actually left the teaching profession very recently to do like recruiting or sales, just because of the burnout they faced. And after interviewing a bunch of them, and they telling us, like, grading was like the biggest time suck for them, that's kind of how we came across this opportunity and wanted to build and help them.

Andright now you're only math-focused, or?

Yeah, so we're only doing mathright now, and that's for two reasons. One is, there's a lot of, like, essay writers out there because, you know, LLMs work really well with text, but they don't work the best with math.

And so it's kind of like a competitive mode for us to really build this out, to be able to grade math, especially handwritten math, accurately. And then the second piece is, you know, AI is kind of the biggest barrier for adoption for AI is just like human trust.

And so we wanted to start with math because math is an objective subject, so you can prove,right? Like, Frizzle graded this accurately versus not. And so we wanted to kind of build that trust with schools and teachers before doing more subjective subjects, like, you know, essay grading and history and such.

Huh. So are you starting with the hardest subject?

Yeah.

And then kind of working away, working your way backwards? Okay. That's kind of crazy, but I kind of like it.

Yeah.

I actually think there's probably an interesting opportunity where maybe you guys have thought about it. The parents of the students probably also want to know how well their kid is performing. So is there a way you can actually share that live, like grading data with the students' parents?

Yeah, that is, that is what we're working towards,right? So like in a classroom model,right, there's like the teacher, the parent, and the student. And like historically, these three parties have been misaligned in terms of like, you know, information, in terms of like alignment, in terms of like curriculum, and working towards a shared goal.

And so, you know, grading was kind of the first step for us,right? Because now, once we grade a class's work, the most important thing that we're getting here is assessment data. And that's the most valuable data you have in a classroom,right?

Because it tells you, you know, how America is doing against like, you know, these other countries, how each state is doing against other states, how each school is doing against other schools, and then student versus student. And so now, using this data, we'll give teacher analytics,right, to help them understand, okay, the class is doing really well on these topics and struggling with these ones, which they can splice by student or by question.

And we'll also give parents like, you know, like a weekly summary report where we'll tell them, okay, your kid learned this, they kind of struggle with this, maybe this is, you should help them with this at home. And so now they can kind of, you know, close the loop at home as well, so that their kids can learn faster outside of the classroom as well.

That's cool. How accurate have you found it so far?

Yeah, so the fact that I mentioned that like we had that math nonprofit from back in the day, that was really helpful for us because we were able to get this proprietary dataset of thousands and thousands of worksheets that we've used to really eval our model, make it more accurate.

And soright now we're at like a 98.9% accuracy, which you can argue is about accurate as like a real teacher as well, because, you know, real teachers do make mistakes. And we do know that trust is the most important thing here.

And so we've kind of designed it in a way as well where it's like, if we're not confident that we're grading something accurately, we'll actually just flag it for the teacher to review, rather than give like, you know, inaccurate feedback.

That's cool. How many schools are signed up now?

So we are live with three schools already, and we have, you know, over 50 in the pipeline in terms of like, you know, going through the sales conversations, the pilots, and such. So we started at the beginning of July as part of the YC summer batch.

And so the thing with that was, you know, schools are closed over the summer, so there really wasn't much math to grade. And so we're really kind of starting to pick up now in terms of, as we're getting into the deeper end of the school year, teachers are now getting burdened by all this math grading, and now they're really seeing the value of Frizzle and how it can help them.

Cool. How do you think about pricing?

Yeah, so, so a lot of public schools, they get budgets on a per-student basis. And then for us also,right, it costs us like a cent to grade a worksheet. So, you know, the more worksheets we're grading, the more it costs us.

And so worksheets are kind of correlated with each student,right? Because each student does like 100 to 200 worksheets a year. And so we grade, we price per student per month, just so that we kind of match their budget constraints and then it matches like our cost constraints as well.

That's great. How many, how many students are on the platform, or how many pieces of homework have you graded now?

Yeah, so we've graded already over 8,000 worksheets so far. And yeah, we really started to, it was kind of like slowly increasing, and then it kind of had that exponential hit. Once that, you know, like third week of August, when, you know, schools started opening, homework started being graded, so.

Awesome. I, I can imagine in some cases the teachers are probably not as well equipped with AI as people like in the tech world,right? How, how do you bring them up to speed with actually how to use the product?

Yeah, I mean, so on the website itself, we put a bunch of demo videos. We also have like a support thing where like if any teacher does reach out to us, we'll just like personally walk them through how to onboard and such, which we can do at this stage at least,right?

Just to make sure they're all fully set up. But we really just designed it to be the simplest product as possible,right? It's just clearly like a drag and drop, and then it's just like an import-export. So, and then to even make it simpler,right, we're building out their integrations with like the LMS, like the learning management system, like Google Classroom, Schoology, and then even with their SIS, like the student information system, like PowerSchool and such.

So then it's like, they don't even have to do the work of like putting in the worksheets. We can just get them in from the LMS, grade it, push it back to the LMS, and then push the grade to the gradebook as well.

So they don't have to do anything at all.

That's wicked. How do you, how, why the name Frizzle?

Yeah, I mean, so, you know, from the Magic School Bus, Ms. Frizzle was the teacher there. So it was kind of like an ode to Ms. Frizzle.

Oh, it's a good name.

Yeah, thank you.

Frizzle.com.

What's been the biggest challenge you have faced that you didn't anticipate facing so early?

Honestly, I think it's just like, like how slow the kind of like K-12 education system moves. There, you know, it has historically this, this kind of experience in that as well. And then I will say it's kind of, we're at the best inflection pointright now where it's like principals and superintendents are like thinking and like looking for AI tools to adopt into their schools just because they know how valuable it can be.

So I think we're really at the cusp of where now this is really going to change where these school administrators are going to actively start pursuing AI tools to onboard. And so that is kind of our bet as well too.

Cool. What's been the, the biggest mistake you've found so far that founders can learn from in kind of, you know, the first few months of building?

I think for us, like we had this hypothesis,right? Like, you know, teachers spend 20 hours a week grading, we can give them the time back. So we thought that we should just go directly to teachers. But I think it's really important for founders to think about like, what are all the other tangential customers that you could have that would also be willing to pay?

So like, you know, as we talked about, parents would be a great opportunity for them to pay even, you know, there's like the private public school distinction. And then there's also just like, you know, like the Mathnasiums, Kumans, like these tutoring centers outside of schools that also have this pain point that I can be willing to pay.

And then even community colleges,right? Like, they can't afford TAs, so their professors have to grade everything manually. So that's like a huge pain point there too. So I think if we'd kind of done that and then ran more experiments in terms of like trying to see if there was more uptake within one classroom segment, then I think that would have been more helpful for us to understand where we could get the fastest growth from the beginning.

I haven't seen anything like this, and if you can just conquer even 10% of the education market, just in the US, that's a pretty big market.

That's the hope. And honestly, we want to go global. We're already working with like a school in Namibia, so they use like a UK-based scheme. So really South Africa, Namibia, Africa, and Asia, really it's all top of mind for us.

Awesome. Can we see where you guys next? Where, where can people find out more about you?

Yeah, so if you want to just try it out for yourself, we do have a free plan. Just go to frizzle.com, f-r-i-z-z-l-e dot com, and try it out for yourself. And then if you have any questions, please feel free to email me at a-b-h-a-y at frizzle.com, and happy to answer any questions as well.

Awesome. Can't wait to see where you guys go next.

Huge, welcome to the show. Great to have you here.

Glad to be here. Thanks for the invite.

So what is Flute all about?

Yeah, so Flute is a vibe-coding tool for entrepreneurs and non-coders to build serious apps with AI. There are quite a lot of these guys, but we took a very different approach where we think like, well, if AI is not going to write most of the code, then all of these existing human-focused developer tools are not going to work the best.

So we created a whole thing from the ground up, and it allows us to do a lot of optimizations to make sure AI can do its best job. And so far, the experience with Flute has been very smooth for our users.

They can get a lot more done than all these other tools.

Why are people choosing Flute versus the likes of Rapid, Lovable, Bolt, and all the others?

Yeah, we thought the issue with all of these current tools, and again, our target users are really people that don't code. So the issue with that really is a lot of times it's either too complicated, so it's very hard for these people to follow and build the things they want, or they focus too much on simplicity, and then you can't really, you like hit a complexity ceiling at some point.

With us, we actually built out this platform that makes it really easy to use, but at the same time, it's really powerful. You can pull in all kinds of integrations and dependencies, and you can build out almost anything you want.

What I found really interesting when I was trying out the product is the, you guys are very generous with the amount of credits you give. If you try the likes of Lovable, the credits get eaten up super quickly.

You and Bolt are currently my two favorites.

Nice, yeah. Yeah, I think we also did a lot of work to make sure we are like credit efficient. So we can still make money off whatever you do, but we don't want that to be like, you know.

I think Rapid is probably the worst where you can spend, easily spend like $100 a day without much time.

Definitely. How many users do you have now?

We have about 20,000 users.

That's amazing. In three months, give or take.

Yeah, yeah. I think, you know, obviously the market demand is very strong, and once you have some product that really works, people like it, they spread the word, and it's really easy to get people interested.

For founders who are building similar products or in this AI SaaS space, how do you get your first hundred users?

Yeah, for us, we did it on Reddit, and we did it on like our competitor's subreddit. It's like kind of, I think there's like two ways to look at it. One is obviously these people are very high intent, and two, we know this space is somewhat crowded.

So the best way to validate that is to see that we can get people away from these existing products. So that's where we kind of start off with our initial set of users. We go into this competitor's subreddit, say like, hey, we do this new thing, you know, flashy screencast, like, hey, do you want to try that out?

We also give you some like initial credits to start, and that got us a lot of users.

I like it. So you piggyback off an existing community of your competitors and acquire users through that.

Yeah.

Smart. Love it. And you didn't get banned from Reddit yet.

Well, some accidents do happen.

I also think that

it depends on the subs. I think some of the subs are, they just don't even care. So I don't even feel bad about doing that. Or some subs are more aggressive, then we just don't do it.

Cool. We'd love to see the product. How does it work?

So this is like, I'm building this little app for like generating a PDF brochure with AI. So there's a lot of things you can do here. Obviously, you can do little, we have this like message thing. So let's just say I want to make this sidebar a bit smallerright now and make these adjustments.

So I can also use this select tool to like select some of these parts and then say, make this sidebar smaller.

So, you know, it would like go through looking at these things and it figures out like, allright, this is how I do it. We also have some other tools like annotate. So this one allows you to express things directly on a canvas.

So let's say I don't like this guy. Let's say, there we go, this.

It's pretty neat UI as well. You know, for three months, I was thinking it would be a little bit more cloggy, but actually it's really nice. It's really easy.

Yeah, yeah. So we put a lot of thought into, because like we really have our target audiences like people that don't write code, and we really try to design it in a way that works for these people. The other thing that really helps is we can build Flute in Flute.

Maybe I should show you that. So this is like Flute in Flute. So it helps a lot when we can like build out this UI and like some of the logics directly within Flute. It saves us a lot of time.

So it allows us to iterate a lot faster.

Yeah, three months old, 20,000 users. Okay, I'm sure that there's like bugs here and there, but here we go. Okay, this is great.

Yeah. So this is the landing page, but I think the more interesting, the more fun one I want to show you is we actually built like, you can see like kind of the whole Flute interface inside Flute. This is a huge project, so it takes a while to load and stuff.

Yeah, so this is like Flute itself within Flute, and a lot of times we can make like adjustments or like redesigns directly here, and it saves us a lot of time.

That's really cool. How do you think about building the product beyond this? So instead of just vibe coding, are you going to go down the design route? I mean, there are so many other ways you could take this,right?

Yeah. So our focus has always been like enabling people to build and deploy apps, regardless of how technical they are. We start with like the non-technical people. That's a great place to start with like the strongest demand. But going forward, I think the focus has always been on like enabling people to build the thing they want to build.

And, you know, going on that route, there's like building blocks we want to build out so you don't have to go out and integrate with, likeright now we already have database and backend, but let's say you want to do like file storage, your student will figure out like S3, cloud inner or whatnot, and like we should just put that as a part of the platform.

Email, obviously, another really clear one. So there's this kind of like building blocks we want to continue to build out. And then we want to kind of really improve the AI experience too. Likeright now, Flute actually doesn't have access to the internet.

So if you ask it to do something that's too new, it might struggle. So we want to add that kind of capability too within the agent to make it able to like retrieve information and just generally improve the performance of the AI.

What's been your biggest mistake so far?

So when we started off, we actually just gave out this for free, as like, you know, free product so we can get some tests. I think we should probably just start charging back then. And we can probably do a thing where maybe we just give you like unlimited credit for like a flat rate, but we should always be charging people.

Because I think one thing that happens when you give people for free is they don't always tell you like theright feedback you want to hear. Because like, let's say the AI is like struggling with something, they would just like, you know, just spend a whole day letting it run, but like they don't really tell you that, hey, this thing doesn't quite work.

Once you start charging people, people care a lot more, and you get like a lot more of these direct feedbacks.

How do you think about pricing? Your pricing is amazing, I think, for all that it is. How did you land on that?

Honestly, we just tried to come up with something easy that we don't have to spend too much time on. We know that under the hood, we have to pay all the providers, so we're never going to be able to give you like a fixed price.

So we charge you based on how much you spent, and we try to have a margin on top. Then based on that, we just come up with this pricing where you have different tiers of subscriptions, because that's like the easiest thing in terms of implementation.

That being said, I think pricing is always going to be something that we would like look into and adjust as we go.

How do you, something I sometimes think about is, should the remaining credits from your account that month be rolled over to the next month?

Yeah, that's something we want to do too. It reduces the friction and encourages people to stay with us as paying users. So it's a good thing to do.

I think it would be. I'm surprised, I'm not sure if any of the other vibe coding tools, but the ones I've used, I don't think they do it yet.

I think it's been becoming a trend, actually. Like I see more and more of these vibe coding tools doing that now.

I think it should be. And also, if these vibe coding, I mean, these vibe coding tools are not always going to be 100% perfect, but if they design something and it's definitely not what you wanted, you should always be able to be refunded those credits,right?

Like I did it on one tool last week, and I spent like 20, 30 credits, and it was just something completely different I didn't want. And that's like 10% of my usage gone.

Yeah, I think that part is a bit trickier. So, you know, there's like not a very clear line of like you being very specific and the AI does a bad job, or you being kind of vague, and then the AI comes up with something that's not quite what you want.

And there's also always the issue of like exactly how much of that is like revertible versus refundable. So our policy has been not to never give refunds unless, you know, like our system went down and then you wasted some bunch of credit doing those kinds of things.

The other way we look at that is if you are hiring a human developer,

they don't really refund you their time if, like I say, you know, you work on some things together. Because when you're iterating on things together, there's always going to be certain things that don't quite work out the way you want, and you give more instruction to like get that to the state you want.

To me, that's just like part of the iteration or development process.

Is this a pricing war that erases the bottom? I mean, it kind of feels everyone's competing in the same kind of wheelhouse. Surely at some point it's going to be who can offer the most amount of value for the cheapest price.

It's interesting, like you would think that that's the case, and I have seen people doing something like that. But ultimately, like even if this thing is really cheap, if you can't deliver, or like if it's like really hard to use, most people still won't use it.

So I think there's like a price pressure where you can't charge too muchright now, but I don't think people are competing on price. At least people, most of the people are not really trying to compete on price because that's, you know, ultimately people want to be able to build a good experience.

And what were you doing before this?

I was an engineer at RETRO.

Cool. And why the name Flute?

Yeah, so we decided to, so we actually have a different name before, and then we wanted to find, the name was like a little bit hard to pronounce, so we came up, we decided to come up with a new name.

Naming has always been pretty hard. We spent quite a lot of time doing that. But ultimately we landed on Flute because it's very short, it's easy to remember, we can find a .com domain for it that we can afford.

So that was the perfect combination.

The .com, I mean, is this like some crazy expensive domain or was it like $20 or something?

Oh, it's not, definitely not $20. I think it's like five figures, but it's affordable.

Wow. Okay. You changed your name. Quite a few founders think that it's too late to change the name. I kind of think it's never too late to change the name.

Yeah, I mean, we are three months old, so it's not a big deal for us.

What's the big vision? How big can you take this?

Well, so if you think about this, like if you look at AWS, it's powering like developers today, and there's like all sorts of things being built on it. Obviously, a huge entity today. But then if you think about non-technical builders, there has to be way more than, way more of these people than developers.

And tomorrow there will be way more of the software getting built by these guys. And we want to become the foundation for those people. So the big vision really is we want to become the biggest hosting and development platform for non-technical builders.

How do you reach millions of builders who still don't know about Flute? Do you have some like crazy fire on that network of facts or something, or?

I think it's always going to be a work in progress. We started off with Reddit, and during my CEO, we did these launches. And after that, now we are doing a lot more stuff on social. So we just really try to figure out more and more channels and grow the user base.

And we continue to build out all of these features. Like we are trying to think about certain features that can allow us to like encourage people to like talk more about us. Yeah, so this is always going to be a work in progress.

I think like once we hit a certain number, there's going to be new challenges. And, you know, obviously once we do hit a million, the next thing is like, allright, how do you hit like 10 million users? So there's always going to be something new.

How long is it going to take you? 10 million users?

If we do get to 10 million users,

we can easily hit a billion revenue at that point. Or like, I guess 10 million paying users, then that would be like almost a billion revenue. So it does look like it's going to probably take a while.

Go and do it. We're really excited for you guys. Where can people find out more about you?

Yeah, so just go to flute.com, F-L-O-O-T dot com. If you search, or just search for Flute, you should be able to see some of the reviews or links to us. And always feel free to reach out directly in our company page, on LinkedIn, or send us an email via hello@flute.com.

And just go and build. That's the best advice we can give to builders.

Yeah, yeah. And charge it for yourself, whether Flute works for you or not.

Do it. Awesome. You should come wait to see where you guys go next.

Thank you so much.

Tom, welcome to the show. Great to have you here.

Hey, Ali. Great to be here.

So, Sagekit, what are you guys building?

Yes. So, Sagekit is what we would describe as Lovable for automations. So, we enable you to build full automations just by chatting with AI. Another tagline we like to use is, you know, the automation platform that speaks human.

Cool. So, this is also kind of like Zapier,right?

In some ways, yes. In other ways, no.

But the better version of Zapier, what you guys are building, hopefully.

Yeah. Yeah, so I think, you know, how we see our position versus a Zapier is that the Zapier style interfaces are very much drag and drop focused interfaces where you need a user to go in to configure everything.

There's lots of field mappings, there's lots of configurations. And ultimately, they've remained sort of inaccessible to non-technical users and some of those more line of business users that actually have a big need for automation. What we're trying to do is come in, totally reinvent the interface, you know, make it completely chat-based.

We handle all of the automation building on your behalf to sort of make it truly accessible to these non-technical users.

How did you come up with this idea? And what were you doing before?

Yeah. Yeah, so we were building a browser agent, actually. Anders, my co-founder, and I, it's called Magnitude. It had a lot of success with developers. It, you know, has almost 4,000 GitHub stars. And what it did was it allowed you to put natural language.

You know, you could say something like, "Complete the login flow," and it would go in and it would complete it in your browser. But it was a developer framework,right? You sort of had to go in, you had to write code, you had to deploy it.

There was a lot of things you had to do to get it set up and functional. And what was really interesting is we had a lot of people coming into it that were not developers that were like, "Hey, this is awesome.

How can I use this?" And we had to like explain to them, you know, "Oh, you have to set up JavaScript, you have to do this." And we lost, we totally lost them,right? But it really made us realize like this natural language paradigm of just being able to put in, you know, this task and get an automation back is something that would be very much broadly accessible outside of developers, yet it felt like only developers had access to that sort of really easy natural language automation.

What happened to Magnitude? Did you sell it or?

No, Magnitude is still Magnitude. You know, it's got great developer usage. There's actually a few companies in the most recent YC batch who are built on top of Magnitude because it's just such a good browser agent for developers.

And eventually, Magnitude will be put into Sagekit in the future. But for now, Sagekit is an API-based platform.

Smart. You have like 20 apps,right, connectedright now.

Yeah.

For founders who are quite new to this space, who are looking to get similar products kind of connected, how do you go around building those relationships? Is it just API-based? Are you building relationships with all those companies as well?

So, the great thing about a lot of these big companies and apps is they have APIs that are publicly accessible. So, you don't need to have any sort of relationship to build an integration with them. You just need to read their API documentation, sort of pick out the endpoints that you want to connect to, and then build all the schemas and the integrations around it.

That being said, you know, we have had some conversations with companies that are like, "Hey, we'd love to be on your platform. Like we think there's a great synergy here." So, there always is an opportunity to, you know, reach out to people that you really want to promote on your platform and maybe do some co-marketing with them as well.

That's cool. What's the timeframe from landing the API to getting into on the Sagekit platform?

Yeah, we have a very scalable way to build API integrations. So, I'd say that we can build around 5 to 10 API integrations a week going forward. So.

That's pretty good.

The traditionally API integrations have been very difficult to build because of just how tedious they were. There's, you know, 100 endpoints, they all have different schemas, they all have different, you know, ways to call them in different scenarios.

And what's awesome is this is a perfect fit for LLM coding and AI coding to go in, look at these very well-documented APIs. You know, in many cases, they're very well-documented. So, AI is really good at looking at the documentation for these, writing the first 80 to 90% of this integration for you, where you just have to go in and sort of write that last 10 to 20% of the integration and really make it work for your use case.

And there's a lot that goes into the pre-planning step of that, of making sure it's writing it in a way that we want it to work with our agent. But yeah, we're in this golden era where it's very scalable to do things before that were previously very unscalable.

That's awesome. How does the product work? I'd love to take a look.

Cool. Yeah, so this is the Sagekit interface for building an automation. On the left here, you see this chat window. This is where like all of the interaction takes place. On theright here is a visual canvas where it will actually show you what your automation is doing.

Right now, it's just got an empty state that shows you what apps and tools you can use. So, you mentioned before, we have around 20 apps. You know, the Google Suite, Notion, Linear, a lot of these sort of core productivity apps that people want to use.

I think this list will be, as I mentioned, rapidly expanding, but for now, this is what you can use. To show you an example of a workflow, I think I personally have a workflow as a founder. I run every morning where I get all of my unread emails from the past 24 hours from Gmail.

I summarize them with AI, you know, get out some key action items, and then I send a report in Slack. So that like every morning I'm like, "Oh, what is all the stuff I forgot to respond to yesterday?"

Because there's a lot of it usually. And so all you have to do in Sagekit is specify exactly what I just told you,right? So you say, "Get all my unread emails from the last 24 hours from Gmail, summarize them with AI and extract action items, and then send a report to Slack."

This is totally live. You know, this is just you go into the platform, you put in this prompt, and what it's going to do is it's then going to pull in your apps and it needs,right? So I've already authenticated to these.

If you hadn't authenticated, it would be a simple one-click auth. You log in, you authorize, and you're good. It pulls in Gmail, Slack, and AI. And then it will ask you questions too,right? So if there's something you under-specified when you prompted it, it will make sure it has clarification.

So in this case, it says, "What's Slack channel? Should I post it in?" I have a daily email channel I use for this that I wake up to. So you could sort of specify whatever you want in there.

And yeah, so now what it's going to do is it's going to actually pull in all of the information it needs to build the automation on your behalf. So it's not going to require you to go in and sort of tinker with anything yourself.

It's going to go in, it's going to pull all of the schemas for the tools. It's going to pull sort of, it's already done. So it goes in and it builds it. So yeah, you can see on theright, it generated an actual representation of it.

You can see it's getting all the unread emails from Gmail. It's seeing if any were found. And then for each email, it's getting the full content. It's summarizing with AI and extracting action items, and then sending it to Slack.

So it's exactly what I asked. If I wanted to, I could then run this. I could edit in chat if maybe I want to get it from the past 48 hours. Maybe I only want to get star emails.

But like the beauty of it is you can go into chat and sort of edit it however you please.

This is so cool. What's the use case? Are people coming to this for email? Is it for Notion?

Yeah.

Yeah, I guess there's a whole bunch of use cases,right?

Yeah. So obviously there's a whole ocean of use cases,right? There's a million things you could do with it. I think what we're really trying to go hard on at first is really around what we would call prosumer use cases,right?

So there's a lot of solopreneurs, SMBs, startups that traditionally have been locked out of these incumbent automation platforms like a Zapier due to just complexity,right? I think a lot of our early users, like solopreneurs, SMBs, are actually not switching from another automation platform.

They're just switching for manual work,right? Like they might have checked out a Zapier or a similar platform, been overwhelmed by the complexity, didn't have the resources to hire a developer to go do it for them, so they're still doing it manually.

So largely we're seeing a lot of solopreneur SMB usage around things like email management, calendar management, spreadsheet management, and task management,right? Sort of those core things where you're like drowning in them as like an early business and it's the first thing you think to automate is, "Let me help something like fill out my spreadsheets.

Let me help something like go from email to spreadsheet." And it's not like crazy complicated workflows yet, but it's like, "You should just build this workflow here." This would be non-trivial to build in a Zapier style platform,right? Especially some of the logic I put in here, but it's just super easy to build in Sagekit.

That's so cool. And this like took us two minutes. And we didn't do this pre-recorded. This is totally live.

Totally live.

And okay, you have things connected, but two minutes, five minutes it takes.

Yeah. Yeah. Yeah. And it's, you know, you can edit it. And what's cool too here is we actually have a test run. So if you want to simulate your workflow without actually having it do the real sending of the Slack message, you can also simulate it,right?

So if you wanted to build workflows, make sure at workflow first, you can also just use the test mode.

How do you think about pricing?

We want users to build as many automations as they want. We only want to charge you when you're getting value out of it, which is to us when you're running the automations. So there's just a simple per-run automation pricing.

Something we're also doing that I think is different is we're not charging you per action,right? So in this workflow, if it reads 100 emails, that's still just one run when you run it. So each tier just has per-run flat rate pricing.

We don't want you, in the same way that we don't want you thinking about all of the nuances of how to build the automation, we don't want you thinking of all of the nuances of how we're going to bill you and try to nickel and dime it.

So it's just per-run flat rate pricing.

Do you think we're going to go to a situation, especially with these vibe coding tools? Any credits that are kind of left over from the previous month, should these platforms just roll the credits over or are they entitled just to get rid of them?

Yeah, that's an interesting question. It definitely depends on what you originally paid for. If you paid for

a set amount of credits per month, it feels like they should probably roll over. And I think that there's no reason they shouldn't,right? I think if you're as a business, if you are pricing a certain amount of credits, or in our case, runs per month, you feel comfortable with them running all of those runs,right?

So I don't think there's any reason it shouldn't roll over. And I think if anything,right, you want to incentivize people to use your product more.

Right, exactly.

And I know there's margin issues where, you know, certain things eventually you have to put caps on them, but like you want people to continue using your product. And I think you see that with like Anthropic being willing to eat pretty much all of the margin on Claude code by offering it in a $200 a month plan is they want everyone in the world using Claude code,right?

And I think there's something to be said about just wanting everyone to use the product. And like you need to think about a scalable business model,right? So you can't just be, I think Anthropic's the extreme example where they are burning probably more margin than most companies could to incentivize more, you know, usage.

But I think there's a middle ground here where it's a lot of business like ours, like we can pretty healthily, you know, scalably offer, you know, rollovers and those sorts of things. Yeah, I mean, once the workflow runs,right, you can schedule it, you can test it, you can run it.

Eventually, actually, by the time you release this podcast episode, we'll probably have triggers built in where you can automatically trigger these based on an incoming email. So yeah, this is it. I mean, everything happens on the build screen,right?

We've done all of this in live. So people should go and check it out. We'll put in your links at the end.

Sagekit.com, go check it out.

There you go. Cool. Maybe we, yeah, there you go. So how do you, what's been the biggest challenge that you didn't anticipate since you've kind of been building this in the last few months?

Yeah. So what's interesting is how quickly things change in terms of how you need to position yourself. I think that there's even two to three months ago when we were really incepting the idea for like what's the next step of magnitude,right?

What are we going to do with magnitude? And we came onto the idea of Sagekit. At the time, you know, I don't even think Zapier had released their like co-pilot yet, which sort of helps you build things on the Zapier canvas.

But increasingly, there's so much noise being brought into the market with just how many things are being released. And the problem is so many of them are describing themselves in essentially the same exact way. It's like prompt to workflow, natural language automation.

And there's actually huge differences between all of these products in the approach they take, the user experience, how easy it actually is to build something,right? There's actually huge differences in user experience and like the actual product. But there's almost no difference in how they describe themselves,right?

So it becomes extremely difficult to, how do you cut through the noise of all of these platforms that are being released that are claiming the same thing? And I think that's something we didn't really anticipate even two to three months ago, how difficult it would be to really just cut through some of the noise that's out there.

How many customers do you have now?

Yeah, we have a few thousand signups onto the platform already.

That's pretty good.

You know, sort of moving people up tiers now. I think we're making good progress and sort of starting to monetize, you know, have people move up. I think at the end of the day, our north star is really usage,right?

We want really strong retention, you know, as a sort of more of a prosumer product,right? We want people to be increasing their usage over time. So like it is a little bit of sometimes of a long game of we have a fairly generous free tier, but we're seeing very much increasing usage from people.

And do you think, do you see this as a, we kind of described the idea, mentioned the idea, a very solopreneur SMB type of product, or do you see this as a big enterprise player at some point?

I think the way I view it is actually in what we're trying to adhere to is what I would call the Notion playbook. So I think I view it as solopreneurs and SMBs are the ones feeling the pain the mostright now because they don't have access to any other alternative.

They don't really have access to a Zapier and any dinner, et cetera, unless they like hire someone or have staffing for it or hire a developer or contractor. So like they're the ones who feel the pain the most.

So we're appealing to them now, I think, as a matter of that's sort of our wedge,right? That's our way into the market. That's like who's traditionally been locked out of automation. And truthfully, I think that that market is massive.

Like if you think about how many solopreneurs, SMBs, and even startups, you group them in as well, it's an absolutely massive market that I think is underserved. So I think we'll be there for a while. That being said, I think there's a very natural opportunity with a tool like ours to eventually, you know, you have individuals at bigger companies signing up to use it for like personal productivity in a business capacity.

And eventually you have two, three, four of those people from the same organization, well then maybe there's an opportunity for us to reach out to them and think about, you know, how can we better serve you to actually be in the organization.

But I think that's more of like a medium to long-term thing for us. And for now, we're actually focused on providing as much value as we possibly can to these smaller use cases.

How are you getting into these communities? How are you actually reaching customers?

Yeah. Yeah. I think the beauty of

a product-led growth where you have a self-serve model and a free tier is that you get a lot of word-of-mouth growth. If your product is good and it can speak for itself, which is something we've had to work on, I think the first version of our product was not good enough to speak for itself.

But I think we've now just launched a version we're happy enough that we're seeing good word-of-mouth growth. So actually a lot of our early growth has just been word-of-mouth. I think, you know, we had one or two discrete launch events like our YC socials launch and a product hunt launch that got in like big waves of users, but it's sort of just been a nice, nice gradual uptick from there since.

And I think what we're going to do though is we're going to go very hard on organic content. I think with prosumers and what they actually discover most of their products in their social feeds. So I think we want to be in your Instagram feed, we want to be in your TikTok feed even, we want to be in your YouTube Shorts.

So I think we are going to lean into organic content.

Cool. There's probably a massive opportunity to partner up with all these like substackers, newsletter writers as well,right? There's so many who run like solopreneur-focused communities.

Yeah. Yeah. There's a lot of people that are very interested in automation and AI. And yeah, I think another thing is just brand building,right? Like podcasts like this, like your podcast. And then there's been sort of other invites for me to go on, other with automation experts and sort of productivity people.

So yeah, there's a lot of people that want to talk about it, which is awesome, because it's awesome that everyone wants to talk about it. So I think there's a big opportunity there to just, yeah, get a lot of brand awareness to that.

Cool. We also have the biggest community of substackers, of writers. So we can probably put you in front of like lots of people. There's definitely like a couple of you out there.

Cool.

What's the big goal? How big can you guys get this?

Yeah. Yeah. I mean, the big goal is I think that what we're really focused onright now is what we would call workflows,right? And I think it's actually what enough people aren't focused on, because I think a lot of people have jumped to AI agents.

You know, there's this AI agent that's going to do all this stuff for you. But like what people are ready forright now is workflows. They want an AI agent to maybe help them build the workflow and make it a lot easier to build the workflow.

When it runs, they want it to be fairly deterministic,right? They want it to be repeatable. They want to do the same thing every time. And I think the way I would describe it is how many decision points do you have in the workflow where it's more codified versus like an AI maybe making somewhat of an arbitrary decision.

And I thinkright now there's very few of those where it's like one decision point where it's like, okay, maybe I determine if this bug report isurgent or like mediumurgency. And it puts a ticket into one or two paths.

I think where we want to go eventually,right, is the more you can codify into these decision points or AI can actually be making the decision, the more sort of complex processes and sort of workflows you can end up automating.

And I think long-term, we see it as a platform whereright now there's lots of like email, spreadsheet, calendar, you know, these sorts of workflows. But eventually, especially when we put like a magnitude browser agent in it, I think there's a lot of domains it can expand out into.

Like even like, you know, boring old businesses,right? You know, boring, unsexy businesses where they just have tons of process. Andright now there's a lot more human decision-making that needs to go in. Like think about like, you know, an HVAC or plumbing or like service businesses.

They have all this software or, you know, car dealerships, military. I mean, there's just so much software everywhere where there's humans making decisions. And I think as we move more and more to AI being able to make those decisions, I think a platform like Sagekit becomes a platform where it can do some of these more routine roles, some of these more manual roles,right?

And like over time, you end up being able to capture like so many different verticals where a platform like Sagekit is like a very much horizontal platform into doing work in sort of all of these different verticals. So I know that was a little bit abstract, but I think like there's a future where it's doing work in all these different verticals and some of these more sort of really repetitive rote sort of roles.

Awesome. Congrats on the progressive as well. Excited for you guys. I'm sure we're going to absolutely, you know, crush it. Where can people find out more about you? Where can people sign up?

Yeah, sagekit.com. So part of the reason it's sagekit is we have the .com. You can Google it, you can .com, you will find it. So sagekit, it will be there. Sign up, it's free. It's awesome. You should try it.

And I'm tom@sagekit.com. So my email's out there. If you want to email me, you want to have me help you onboard you, I'm here. I'm willing. So yeah, I'm excited for people to use it.

You're going to do it.

And those are the three tips for you to win a hackathon.

Good to see everybody. See you soon.

April, can you summarize the last three emails?

Oh, okay. I might need a break. April, when's my next event?

April, what do we have left for the day?

April, did I send that email to Gordon, the head of sales?

Gia, you have a hackathon coming up in one hour.

Oh, in an hour? Where is it?

The venue is at 660 Market Street.

Oh, okay. Guys, I got to go. I'll see you.

Gustav, your last three emails are Uber for dogs by Luke, Sasha from Clean Trip sending you the details of all your canceled trips, and Ari from Not Hot Dog asking for a YC interview.

Can you archive those three? I don't need them.

Done. You have one more meeting at 6:00 PM.

Can we reschedule that for same time tomorrow?

You have a Coldplay concert at that time tomorrow. Should I still reschedule the meeting?

Yeah, cancel Coldplay. Some of us actually have work to do.

You left that email as a draft, Thomas.

Go ahead and send that draft. CC Iman and the new hire Eric as well. We got to pump those numbers.

Okay, Thomas, I just sent the email.

Thank you, April.

Neha, great to have you here.

My pleasure to be here, Ali. Thanks for having me.

So April, what are you guys building?

So April is a voice AI executive assistant. It helps you manage your email and calendar completely hands-free when you're on the go. So when you're driving, walking your dog, or working out, you can just speak to April and manage your inbox and calendar.

Awesome. So for listeners who may not be familiar with it, essentially imagine it's superhuman but with voice and it's all done for you,right?

Exactly. That's a perfect way to put it.

What's been the best way of acquiring customers, both free and paid? What channels have you kind of started activating?

So we've been doing most of it organically. We've gotten a lot of inbound from Reddit and LinkedIn. We're now trying to work the Twitter game and see if we can get some inbounds from there as well. I think in terms of focusing on the acquisition channels, Reddit does really, really well because when we tell people about the problem and the solution, they immediately connect with it and it feels very authentic.

Everybody's struggling with email overload. We just put across the problem there and then Fit in April and then we get in a lot of sign-ups from there.

How do you also think about product-led growth? So if you think about Superhuman, they have Superhuman like at the bottom of each email,right? Is April kind of tapping into that as well?

Yes. Every calendar invite and every email that you use April to send or modify, we have a SendViaApril signature. And people are very curious, so they end up clicking on it every now and then.

That's cool. How have you adopted a new consumer behavior? As I imagine, for people to send emails by voice, this is still a very new space,right? How do you encourage that kind of whole new consumer behavior? Because I imagine there's quite a bit of work to do that.

It is a little bit of work. So initially, you need people that build trust. So as our users come in, we're not doing a lot of onboarding. It's very self-serve at this point. And we're observing that our users,right, the smart, they use April for summarizing for the first couple of weeks.

And then they start modifying calendar events. And then past this 14-day threshold mark, they're starting to just send emails, play around with mock sent emails, and then they're actually starting to depend on April for sending out real important emails.

I think voice has become such a popular modalityright now that people are intuitively able to figure it out.

It's also like this product Whisper Flow,right? Which I think is for more developers kind of on or maybe just on the go. It feels like we're also entering this world of hands-free tools.

That is correct. Having so much screen time is criminal,right? We've done it because we didn't have another option previously.

Yeah. What were you guys building before this? Or what's your background?

So before this, Akash and I were building a voice agent eval startup, ironically, trying to make everybody's voice agents better. And we're both engineers. I was a lead engineer at Zoho. I helped build out their accounting product for the Middle Eastern markets.

Akash was a software engineer with Apple. He helped kind of block 2.6 billion spam iCloud accounts. He worked on their trust and safety. Yeah. Helpful now when we're like trying to triage emails and stuff.

For sure. What's been the biggest challenge you guys have faced so far? As I imagine, one thing that probably comes up is how do you get integrated into these email providers and how do you avoid them blocking you?

Can they block you? Does that happen?

They do not typically block you because there are a lot of email clients that are out there that just depend on integration with the email providers. The challenge is each of these email providers,right? They have their own security process and they're big techs, so they're really, really, really slow.

And as a startup, waiting for them to kind of approve you for every new email service is kind of frustrating because you're ready with everything and you're just depending on them for permissions. I think that's been one of the challenges so far, being dependent on big tech.

Yeah. So, I mean, can they say no?

I don't think they can say no.

Okay. That's good.

User is authenticating it, so they can't literally say no.

Yeah. That's good.

What's been the biggest mistake you guys have made so far that founders can learn from?

Oh, okay. Yeah. So this is one big mistake that we made. We launched without a free trial. We launched at $30 a month. The biggest traction in run for a YC company is often when you do the YC launch.

So you do the YC, you do the socials around the same time. And we did this launch with no free trial. We kind of knew the product was crazy good, but it wasn't that good that people would just start using it without even having a trial.

I think it's very important for consumer founders, or founders in general, to know that you have to give people a way to touch and feel the product before they actually start paying for it. And then if given a chance, I would go back and give like a one-week free trial when we launched initially.

So let's break that down. As your demo was super good, you just had a free version and people can go straight to paid, or?

I did not have a free version. I had only.

It was just straight to paid?

Yeah.

Wow. Okay. So how many paid users? Did you get any paid users kind of during that period, or?

Yeah. I got some paid users.

Yeah.

But lower than the typical YC launch because of the paid. Once we course-corrected and added like a free trial, that's when the inbound actually started coming in properly.

Interesting. I think it's one of those things where it also takes time to build trust,right, with your customers. And especially like I've tried a few voice AI products, and as soon as there's email, I'm like, I'm not sure if I want to connect to it with a tool,right?

But if you have a bit of credibility, you know, coming on shows like this, hopefully YC and others, you start to hear the name and people start to really love it, but there probably has to be a free trial.

Yeah, I think so. Trust, credibility. I think the consumers care a lot about being able to experience it before they pay for it in general. Even Superhuman, like they have like a one-month free trial, like if someone requires you.

How do you think about pricing? How did you land on that?

We kind of knew the value that we were providing. We were saving so many hours every day. And we wanted to take a very small fraction of how much we were saving time. And to be honest, this isn't the final pricing.

We're still figuring out consumer pricing is a game. Like you just keep experimenting and figure out what lands and resonates with most users. At this point, we feel around $15 is a good starting point. Once we have more integrations in, it's going to be a little more expensive.

Because it is a luxury.

That's good.

Yeah.

That's good. I mean, I feel Superhuman, I think Superhuman is $29, maybe $30 a month. Yeah, I would say it's quite pricey for what it is. And essentially, it's just a bunch of shortcuts and a nice-looking UI tool,right?

It's a beautiful product, but it's quite pricey. But I imagine for you guys, it's voice. They've put a whole bunch of things you can add on on top. Upsells.

$15 is a steal.

I think, yeah. I think $15 is really good to start with.

Right now. How do you think about the enterprise play? If you start getting this into enterprises, are you going to walk into a room and a whole bunch of people that are talking to the email April account? Is that going to happen?

Like, I would find that probably quite frustrating or quite distracting.

So we are kind of looking at this as something that people use when they're not looking at their screen. So when you're driving, when you're doing other things, when you're alone and you have like screen fatigue, that's when you like quickly start like, "Hey, April, can you do this stuff for me?"

And you'd be surprised, like the average person in the Western world is spending 56 minutes a day driving in their car to and from work. That's a big chunk of time that they could use April to get things done.

As opposed to like when you're sitting with other people, then obviously it makes sense to look at the screen. We're also building a screen companion where you can see things. But in a group setting, it's going to be a very different experience.

You don't expect everyone to be talking to their phones in a group setting.

That's pretty cool. If you can help people be more productive while they're commuting, obviously in a safe way, that's really interesting.

What type of users are using this today? Is it builders, founders? Is it people working at big companies? What have you found is that kind of sweet spot?

So we have an interesting cohort. So we have this cohort of investors that are using us heavily because I think investors get a lot of email inbound. And there's the other cohort of sales and marketing folks. Sales, very heavy.

They use us multiple times a day. And then they're like just going through emails, like moving meetings around. So I think it's very heavy for these two categoriesright now. And there's also a third category, which is very interesting.

They're like people between the age of 25 and 35. They're not really managers or anything, but I think they like having a personal assistant or an executive assistant that can do things for them. And these are people from big tech primarily.

This is really interesting. You give me an idea where you have April as almost like the email provider, but then you can also start building your own April agents by domain. One for sales, one for assistants.

That cohort actually made us realize that we will probably have to be doing this going forward. The workforce might be reached vertical. But I would wait to learn more about this cohort of users.

That's cool. As the also just the intelligence of each agent is going to be different for each domain,right? And probably each sector as well.

Like imagine the healthcare system, doctors who are always kind of on call, they're not at their screen very much. Is there maybe something here for like teachers, professionals? There's like a whole universe you can tap into,right?

Yeah. They're very different verticals with very different integrations that are required as well. And the kind of intelligence, like you mentioned, is supposed to be very different. Realtors are another market that really needs this.

Oh,

house agents, property agents.

Yes.

Yeah. That could be cool.

So what happens if somebody uses the product and the voice agent gets completely wrong? Instead of sending it just to one email address, they send it to 200 email addresses.

Well, that is unlikely to happen because before sending any email, April will confirm with you, asking like, "Hey, is this the email that you want to send? Is this the person you want to send it to?" And only when you give the final good to go, April actually executes the action.

Because we're very aware of how sensitive emails are.

Okay, that's cool.

If you say good to go to like the 200 people, then it's going to send it to 200 people.

Yeah. Okay, that's cool. So you have to verify the email before it actually goes out.

Correct. Correct.

Yeah. And that's done by voice, or that's actually like a manual send?

No, that's voice.

Oh, no. That's voice.

April will just confirm like, "Oh, this is what the email is about. This is who I'm sending it to. Are you okay with this?" And then we send it.

This is really exciting. I've tried so many different email tools out there and they're kind of all the same. And we're all super busy. And maybe I think you're probably one of the first kind of voice for email.

I think if you can get that kind of product like growth super early, this is really exciting.

Yeah, I think so too. We're very excited about the space in general.

Yeah. What's the big option? What's the big goal? How big can you take it?

Oh, it can become like everybody's executive assistant, every employee that works, every student, every professor, every knowledge worker. That's like a big multi-billion number of people who are in that category. Everybody gets an EA. Everybody gets to do things by just using voice when they're traveling.

And most people do travel. If they don't, they're at home doing chores. They could use April. I personally, you know, I would love to have an EA who would take care of things for me, but I can't afford one yet.

So for people like me, April is already there.

Cool.

To kind of help triage prioritize and move calendar things around.

Cool. Really excited for you guys. Where can people find out more about you?

In terms of the app, you can download it on the app store. Otherwise, I'm there on Twitter and LinkedIn, talking every day. Yeah, they can find me both places.

Awesome. Nihaal, we can't wait. Congrats on everything. And let's see where April goes next.

Alec, welcome to the podcast. Great to have you here.

Yeah, thanks for having me, Ali.

Your first podcast interview, so this is going to be great fun.

Yes. Yeah, no, I'm excited.

No pressure. So Cacao, what are you guys building?

Yeah, so we are building the Global Dollar app for Brazil. So this is like Wise or Nomad, but we're built natively on stablecoins and we replace the Swift system. So with Cacao, we're helping millions of remote workers, freelancers, creators get paid in USD, convert instantly to Reais via Pix, and spend worldwide with a Visa card, all powered by a non-custodial wallet.

Ah, cool. How did you come up with the idea?

So funny enough, so my co-founder, so I met him a couple of years ago in Costa Rica, and I'm a big surfer. And he was working for a YC company. I was also working for a YC company. And ever since we were bouncing around ideas,right?

And I was back this January. And yeah, we were talking and he actually had hired a bunch of engineers in Brazil. And he was paying them from a US entity and then started seeing the problems firsthand,right? How expensive it was to move money cross-border, how slow the money moved.

And yeah, he kind of pitched me on the idea. And then I had an opportunity to go to Brazil in January. So I went and like that's how we started doing research, kind of understanding the market, the opportunities.

And then by April, Carlos and I, we decided to go all in on this. So then I called up Mike, who I've known for 15 years. This is actually our fourth time working together. He's our CTO. And yeah, we went all in on Cacao.

Some people might say, "Oh, another fintech app or another fintech play." But I imagine the fintech scene in South America is still developing. What's happening there?

Look, I think it's a super interesting market. I actually, so I worked in LATAM and in Brazil specifically in 2016. And how far Brazil has come in terms of open banking and payments is astonishing. So like, yeah, when I worked at Uber, we were trying to figure out how to accept boletos, which are kind of like a version of travelers' checks as a form of payment in the Uber app.

And then essentially with, you know, but in the next five years, Brazil launched Pix. Essentially, Pix is instant settlements, peer-to-peer transactions, but you can use it everywhere. It's like Venmo if it was backed by the government. And with that, like Brazil just has, you know, one of the leading payment systems in the world.

And when I went back this January, I saw that. It was just like, you know, I was actually in shock because it was just, it was a completely different landscape than I knew from 10 years before.

Is also one of those sectors where you have to build trust quite early with your customers. As a new fintech play kind of in this region, how are you developing that trust with customers?

Yeah, you know, it's, what's really interesting about Brazil is, I think just in general, customers are really excited to try new products,right? So Pix just saw this massive adoption and this was like after COVID. But then even like if you look at just, you know, crypto generally,right?

So Brazil's the top five country in the world for crypto adoption. There's over 30 million Brazilians with crypto. I believe like less than five actually own a US equity. So it's kind of crazy how much Brazilians have just like gravitated towards, I think, new financial systems, ecosystems.

Wow. There's also probably this whole adoption curve where, again, the hardest is trying to acquire customers, but then there are a whole bunch of financial products, I guess you could sell down the line. Is that kind of where your thinking of heading?

Yeah, I mean, we're very focused on Brazil. We think there's just a huge opportunity focusing on really connecting Brazilians financially, both to the local economy and also the global economy. And you just see this with like the rise of remote work, the rise of, you know, platforms like Deal, Rippling that are allowing people to be companies to hire all over the world.

So we're very bullish on kind of that trajectory of just globalization. I mean, we want to be a financial layer that essentially connects you to the global economy, but also enables you to seamlessly operate locally in Brazil.

How do you find your first users? Consumer apps are notoriously super hard in the early days. Finding those users is like just the hardest. How have you found them?

Yeah, so I mean, a lot of it came from people we knew,right? People we worked with. And as we started kind of doubling down on that segment, you know, we started getting referrals. And yeah, that's, I mean, the main focus has just been, you know, these Brazilians who are earning from abroad,right?

It's a very large segment.

You guys only started five months ago,right?

Yes, yeah.

How is the, how many users do you have now?

Not something I'm sure I can share on the podcast today, but look, I think, you know, we launched the app in mid-August and yeah, we've been moving hundreds of thousands of volume. And yeah, we're actually seeing just like a lot of interesting use cases.

So in addition to like bringing money back from Brazil, we have users who are actually sending money to the US. So there's kind of a lot of cases that we didn't even anticipate,right? And I think the main thing is like there's a huge opportunity with stablecoins in terms of, you know, being able to just be more connected both to the global economy and your local economy.

For sure. I guess also, well, have you found any interesting network effects? As once you have a customer which is hooked, do they start referring all their friends? Like, what is that kind of just kind of like go-to-market?

Yeah, well, look, if you evaluate kind of some of the biggest neobanks in Latin America,right? They've almost all grown through referrals,right? And I've read statistics that like 50% of customers are coming from referrals. I think for us, it's very early,right?

We launched in August. But as we're kind of developing the product, we're definitely seeing like a network effect here.

That's cool. What's been your biggest challenge to date, would you say?

It's been really just, I think, going deep on this market, understanding kind of the, you know, who we can partner with to make this a reality. We work with a lot of really good partners in the fintech space.

And yeah, so I mean, it's really just like bringing together that with like an awesome team that can ship product. You know, our background, our team's background, it's in, you know, we've built crypto companies, we've built consumer marketplaces, and have scaled to either hundreds of billions of dollars a year in sales for the crypto business that Mike and I worked on.

We scaled that to over $7 billion in volume traded. So yeah, the key thing here is like we want to build something that customers can rely on,right? Like they need to be able to rely on us. They're moving their money,right?

This is like super important. Often it's, you know, it could be their payroll, it could be paying their rent, it could be paying their car bill. So we're just very focused on maintaining like reliability for our customers.

Fintech players also have maybe there's more frustration dealing with the regulators. How have you found that as an early-stage startup? And especially for other fintech founders who might be watching, have you just got to be patient or how do you navigate that?

Yeah, I think, no, it's super important to understand the regulations, to be working with theright partners, to understand, yeah, both like local implications and global implications. For me, this is like something that I worked on at Uber too because I was working on the global payments team.

So yeah, I think it's just, it's important to, you know, make sure that you're always doing things in like a compliance way. And also updating and keeping track of things that change,right? Even just stablecoins in general, there's like already so much change,right?

Just this year, you had the GENIUS Act, like, and then in Brazil, you had a whole set of regulations that kind of changed what the capabilities were and options were for stablecoin companies. So yeah, it's really important to be on top of it.

Where do you see the fintech space going next? It's changing so quickly. There's so much opportunity. What are some of the trends happening, especially in South America?

Yeah, I think, look, it's in the end, fintech, it's like we're building products to help people

financially,right? So that could be moving their money faster, it could be helping them grow their money through investments, it could be helping give them access to new economies, new markets. And like I'm very bullish just on globalization,right? So I think there's going to be a whole suite of products and solutions to just make it seamless to move money, spend money, invest money.

And that's kind of where I see the opportunities.

How do you think also about expanding down the line? Is it planned to conquer Brazil first, then maybe is it Argentina, Mexico, are all the countries kind of the same?

Yeah, I mean, no, I think all the countries are all very different. I think having kind of regional solutions makes a lot of sense, but there also is room for global solutions too. It just kind of depends on who your customers are.

I think there's a ton of space in the market,right? I think, I don't think there's like necessarily one winner,right? There's going to be many players who serve really, they serve a set of customers that have like really unique problems, but that are really big problems.

And so yeah, I don't see, you know, necessarily one app dominating all of global finance. And if you even look in like most countries,right, there's typically hundreds or thousands of banks when you look at any particular country,right? So in our space, I could see similar types of trends happening.

Do you think one of the trends we've seen is very niche players popping up in these kind of like local markets, especially for fintech, healthcare? Is that the way to kind of, you know, grow and build a big business to go hyper-niche versus super broad and global?

Look, I think there's also businesses that are going super broad and global,right? Like look at Wyzeright? But I think these are just different use cases. So it all depends on kind of what your market is, what the opportunity is.

You need to do that research to understand like if there is a big enough market to support, you know, a venture-backed business.

Yeah. We, as we know, building a company is super hard and we will have those city mistakes or just crazy mistakes kind of in the early days. In your first four months, what would you say has been your biggest mistake so far that founders can learn from?

It's, I mean, look, it's been, there've been so many learnings,right? And it's hard to pinpoint one because I think we've made plenty of mistakes. But

yeah, I would say, you know, what I've seen like a trend, especially like in Y Combinator, I think one of the biggest benefits,right, to that whole ecosystem was like talking to other founders, keep like staying plugged in, not being overly isolated.

Because it kind of, it really gives you a lot of drive and motivation when you see like all these people who are like at the top of their game, like working alongside you. It's like, you know, you're all working on different projects, different companies.

But I think being plugged into that network was, it helped us move really fast,right? And it kind of gave, yeah, I think it helps just set the culture for kind of the company and how we operate. So I think, yeah, there's definitely a benefit of staying connected and plugged in.

Because you can, like when you're building a company in the early stages, you can kind of just like go in a hole and isolate yourself. You have to balance it,right? Because you do need those moments to like really move forward, but you also need to stay connected with people as you, you know, are building.

Totally. Get those people who can be your cheerleaders,right, along the side. Super important. Looking down the road, five years, you know, how big is Kakao going to become?

You know, we have some pretty big aspirations,right? So we're adding in a lot of functionality,right, to, you know, to meet our vision,right, to be the dollar app of Brazil. So like we'll have investing, we'll have yields, we'll have like lots of different products that are not in the app today, but very soon we'll be launching.

Awesome. Super excited for you guys. It's super early, but you know, these niche players are starting to pick up in global markets. So can't wait to see where Kakao goes next.

Michigan State Police investigate the death of a road crew worker. It was struck by an excavator and failed on the job. OSHA has now opened an investigation for looking into the death of a 20-year-old man.

Case of a workplace death.

One of the drivers who could not see the person that was in front of the tracks accidentally ran over.

Police say the man was trapped under an excavator that's overturned.

Watch out, there's Mike passing through behind the cabin! Hey, watch out, there's Mike passing through behind the cabin!

Will now be up to My OSHA to determine when work can resume here at this construction site.

Josh, great to have you here.

Good to be here.

So Flywheel, what are you guys building?

Yeah, so Flywheel is a framework for excavators. We essentially enable remote control on existing excavators for construction contractors. That increases site safety and productivity for them. And we do this with retrofits, which works on any existing excavators.

Wow, construction space. So how does the product actually work?

So what we do differently is we have retrofits, which mechanically actuates the joysticks. So we don't have to tap into the system, we don't have to remove anything from the system. And that essentially robotizes the existing excavators, which are millions of them.

And then we connect them through our remote teleoperation stack, and you're able to remotely operate it from literally anywhere in the world.

How do you land customers? Are you working with the building, the building sides, or are you working with the manufacturers of these machines?

We don't have to work with equipment manufacturers because this is OEM agnostic. So we just sell it to the equipment owners, which usually are construction contractors.

How have you found getting customers? How many do you have so far?

We have two customers in Singapore and we have three customers in the US. All the US traction that we got has been in the past two months during our time in YC. And we now have systems deployed in both the places.

In fact, we just open sourced the biggest robotics context data that we collected from these already deployed excavators.

How, where do you see the biggest market? I mean, Singapore, we're both based in Asia,right? There's always construction happening, but the US seems just so much bigger. Is that the opportunity?

Yeah, the way I think of this or robotics in general is robots per capita would be proportional to GDP per capita. So wherever there is a high GDP per capita, you would find it harder to find people who have to do these high-skilled work.

And it could be localized to a city or it could be a country, but any place which essentially checks those boxes is a good market for us.

Where is the next big opportunity in terms of geography? Is it Asia, is it Europe, or just the US?

So for now, we are going to focus on the US and Singapore. But I think it's also like pretty uncertainright now in terms of how the construction demand is differing. You know, like it's been flipped a bit with the data centers because if you have a bit of cheaper land and you want to build more data centers, you're suddenly seeing a lot of construction demand in like places you would not expect to see a lot of public infra works.

And so quite interesting things happening, but like for now, we would be focusing on the US and Singapore.

Construction is a pretty hot business. It takes a long time from a sales point of view. What's your background? What were you doing before this?

Yeah, I have no background in construction. I am a robotics engineer, so I've been building robots since I was 10 years old. But I just started with a lot of humility where I just met a lot of construction founders who had been in the industry from like different aspects of it, just had a lot of conversations with them, understood everything about like the hierarchy, who's liable for what, how does the industry work.

And that's essentially how I learned about the industry. That led to some warm intros as well. Had a lot of like, had to keep a lot of thick skin in those initial customer discovery meetings, you know, because construction's literal sport is to dunk on geeks.

You know, you're trying to sell to people who have been working for many, many years. A few construction contractors might be even, you know, 100 plus years old. And they just have seen a lot of these waves. And so you just have to sit through and actually understand what their problems are.

And if you do that well, you can make a good product. And if you do that, you can sell.

You were building robotics so early. What type of robotics were you building? We want to know it's kind of like robotic teenagers.

Yeah, so I and my co-founder, we have built like different, like all across the spectrum. So like I started building domestic robots, like home robots, which could clean, mop, and also like bring things from one place to the other.

This was back in 2019. I also made robotic dev kits, which were essentially like small washed-down versions of the hardware, which software engineers and robotics software engineers could use to just test out their software, which we were selling to other companies, other robotics companies, and also students and university labs.

I built India's first hot food robotics kiosk, which essentially helped them decouple the kitchen and point of sales and generated like 10 times more revenue for them, which was covered by some national newspapers in India. And so it just worked on a lot of everything from consumer robots where everything needs to be packed in like small space and, you know, has to work reliably in the consumer environment to something which is safety critical, mission critical.

Similarly, for my co-founder, he was one of the early hires in a robotics company called AdVorb, which was doing industrial robots and warehouse robots. And then he spent four years in the self-driving autonomy company Motional, and he single-handedly built out the ML data and eval pipeline over there, which is quite crazy.

And so we had this whole like wide experience, but also like robotics plus autonomy experience to actually do this.

Cool. I imagine for Flywheel, what you guys are building now, actually the biggest selling point is this probably massively reduces accidents,right?

That's true. In construction, safety incidents is a big issue. So safety incidents means whenever an accident leads to an injury or an accident or injury or a death. And roughly about 1,000 workers die every single year in the US.

And with every safety incident, you essentially have to pay a lot of penalties for every safety incident. And your site shutdowns for a few weeks while it's being investigated by OSHA. If it happens a few times in a row, the penalties can go into millions because of repeated fines.

Or you might be blacklisted for five to 10 years from taking new projects. So that is the highest

priority problem for these construction contractors to solve. And essentially, you could solve that with remote teleoperation. And so we are essentially solving a problem which they have today, which is probably one of the most important. And that leads to sort of a path for us to autonomy that we could later on push out.

That's, I didn't realize it's that big in terms of the massive accidents, but also just the severe consequences. These insane fines, you can get blacklisted. So actually, this is a massive selling point. What would you say is your biggest challengeright now?

I think we are in the stage of productizing our retrofit. So essentially, then we could just spend X amount of it because there's a lot of demand for it. Right now, we're just finalizing some of the productizing aspect of it.

And then we just have to scale up manufacturing, which would be its own beast. But yeah, essentially those two things.

That's cool. I think the selling point is so clear. And I haven't come across any companies doing this, I don't think. Super excited for you. I think what you're doing could be massively impactful and go and become the number one player.

Thanks.

Awesome. Great to have you on the podcast. Thank you so much for your time.

Thanks for having me.