Live @ YC Demo Day | Garry Tan, Delian Asparouhov, Matteo Franceschetti & Many More

23:11All of those things are manual. So what we're doing is we're bringing AI agents to all of your developer tools outside of the code editor. Okay. Walk me through how that actually works because in cursor I'm in the IDE. Are you puppeteering the AWS dev interface or are you operating at a lower level? What's the actual interface? Are you rune-pilled? Is this like a text is the universal interface play? Yeah. So basically you land on a developer homepage and you see all of your tasks from across those tools. So like exceptions from Sentry. And then you see an autofix button for each of those tasks.

23:48So it's like a priority inbox that prioritizes everything across those developer tools and then gives you one-click actions. Got it. How much of what you're doing requires just building on top of an API for something like PagerDuty or actually doing a deal with them to integrate at a deeper level or just puppeteering and computer use? They're never none the wiser. So we're actually doing both, right? So some of the dev tools haven't built out a lot of agentic features. So we're building on top of their API. And then other partners, like one of our first integrations is with Sentry. And we integrate with Sentry AI.

24:25Yeah, okay. I remember Sentry. Yeah, Co-Found is actually here. Oh, cool. Did you guys come in with this idea? Did you have to iterate to it? What did that look like? Yeah. Yeah. We did. We came in with it. So before this, Daniel here, at Netflix, I was on the team that built the internal developer portal, which was the most used engineering tool at Netflix. That helped accelerate the company. You probably heard they shipped ads and live streaming and all these things, and it really helped the organization move faster. Everyone thought that they were going to have to partner with Microsoft, and I think they wound up doing it internally because they were so fast, right?

24:58Yeah, that's a great narrative about Netflix. Exactly, right? So it really helped accelerate the team, and that had no AI. No AI in it, right? And you can squeeze all that productivity. So then Jonas and I were chatting, and we're like, wait, what if we did this developer first? We did it with AI. how productive could you make engineers? You just go from Cursor to Starsling, and you're just talking to agents. You're slangin'. Slangin' bits. Slangin' the stars. What is the go-to-market motion? Cursor obviously just goes bottoms up directly to the dev. Is this something that a DevOps engineer can bring in, or do you need to go through a CTO, get approval before you do an enterprise deal?

25:34Any developer can actually just sign up, connect your accounts, and start using it. The beautiful part is this isn't for DevOps engineers. This is actually for all the engineers that have to touch those tools. And so it's like pretty much every engineer at the Modern Software Engineering org can just sign up on their own without talking to anyone else and still using the product. How's the progress? Are you live? Are you growing? What metrics did you share with the Demo Day crew today? Yes. So less than a month ago, we launched. And we have over 400 companies that have signed up for the private beta, including DoorDash, Snowflake, Golden Sacks.

26:09Golden Sacks? Yeah. Let's give it up for big clients. Everyone wants agents. That's crazy. Where are you guys going from here? Did you close the round already? Yes, we have closed the round. There we go. There we go. Yes, yes, yes. We're happy to give you these ramp-ups. Awesome. Thank you, guys. So we put the money in ramp. Yes, exactly. Awesome. That's incredible. What's next for the company? You guys hiring, scaling? We're hiring very slowly. How big is the company right now? It's just the two of us. Just the two of you. The classic, the old school YC. This is the way it was. Now there's folks coming through with 25 employees.

26:48Yeah, no, no. We incorporated right after getting into YC. Really? No way. Awesome. We're like, all right, the two of us. That's great. Build the MVP. What were you doing before, by the way? So I started a company called Stackshare. It was a developer community. We scaled it to over a million developers. By the end of the journey, it was used by over 40 million developers. I think we did like an interview or something, right? It was an interview of the Soylent Tech Stack. Yeah, that's right. I was like, how the hell are you shipping all these calories? And so we talked about all the tools. Yeah.

27:16So yeah, it was a big developer community and then sold the company last year. Awesome. Congrats. That's amazing. You guys are in an incredible position. Yeah. I'm feeling really good about this. Yeah. You should be confident. It's going to be hard, but I think the confidence should be high. It's a good market. I mean, it's amazing how much you can get done now just with like two people. Yeah. Right? Yeah. It's never been done before, so we're excited. It also feels like an interesting market in that we've seen it's just less monopolistic. It's not like trying to break through a social network where it's like you're either a trillion-dollar company or zero.

27:48I feel like in DevOps, in enterprise, you can understand the roadmap ahead. Chop wood, create a great product, and carve out a fantastic business. That's why we see so many companies going public every year, so many Decacorns in this category. So congratulations. Thank you. Who else should we keep an eye out for in this batch? Yeah, we're huge TV fans. Come on the next time you raise money. We should be watching if you're a founder. There we go. Have a great rest of your day. Good luck. Thank you, guys. We'll see you guys soon. All right. And we will bring in the next crew. Stoked for you guys.

28:22Oh, yeah. Bring the phone out. We're going to run out of those. How many more do we have? We have a variety of goodies. Oh, I love the sweatshirts. You're owning a color. there we go yellow pink you can have those if you want wheel free limited edition to this demo day you can only get them this year the colors are working together perfectly roughly the same saturation roughly the same saturation on the orange background yeah this is jazz berry it feels kind of vintage it feels like I've known it What are you guys building? Yeah, break it down. Oh, we're building an AI agent for bug finding. Okay.

29:04So right now we have a PR bot. Yeah. So you make a pull request. We'll take your code, clone it into a sandbox, and then we let an agent just go ham at it. Okay. And then we tell you how we break it. Okay. There we go. How much of this is just about speeding up the pace of development versus like is there a pen testing angle here or is that just a completely separate cybersecurity play? No, I think we do some pen testing. Okay, yeah. So we want to basically find any kind of bug. So a lot of people take a limited approach at bug finding. So they ever do coverage testing or they try and find integration bugs.

29:40We really want to basically build an agent that can do any of that or kind of what's best for your tool. So you like when people are vibe coding because they're just creating bugs all the time? Keep vibe coding. We love vibe coders. We're here to help you make better code. No problems with it whatsoever as long as you buy our software. Yeah, yeah. You VibeCode, we'll test it. Then you take our output, you put it back into cursor. Just feed it back in. Talk to me about the prompts that you're using to actually have the agent go and hammer it. I imagine that that's not just try and find a bug. You've probably gotten very good at designing flows.

30:19It's VibeCode on VibeCode. There's probably a lot of work that goes into that. What goes into actually getting an agent to effectively hunt for bugs? Yeah, so like the most important part is just to have like a sandbox where it can like it can run code Can compile your code it can it can run like unit tests Yep, and so then we just get the agent to yeah go ham and each time that it like runs a small experiment on your code It learns a little bit more okay, and it's able to do run a better test the next time Yeah, it's able to search a repository able to run commands to you know See if you you know changes you made actually were propagated through all the files So we found the bug, which was that someone updated a path, but didn't update it everywhere.

30:56And so that was those sorts of things. So the agent's able to run these like any command that a person would. Yeah. Are you doing stuff like trying to stuff multiple variables in a single function, like that type of stuff where like there isn't as much fault tolerance built into the code? Maybe they need like a, you know, if else try accept clause in there or something like that. Is that the type of like bugs that you're trying to find? Or is it more about like scalability of code? Like, okay, you're making a database call right here. It looks fine now, but if we scale this up and there's a lot of demand, you're going to get cooked.

31:27I think it really depends. So we use the pull request as kind of the initial seed. Okay. So what change you make there kind of determines the path that we use for testing. Sure. So if you're trying to scale, then yeah, we'll kind of test as if you're trying to scale. Sure. But if you're making path changes, we'll test as if you're making path changes. One of the things we find is like vibe coding often, there's a different flavor of bugs that are happening because of vibe coding. Okay, interesting. So LLMs don't make the same kind of errors that people do because people, with people, code grows organically.

31:59LLMs is like one-shotting things. So it'll often like forget to add functions. So it's not as easy as pointing to a line and going like this variable is wrong. It's like, no, no, you like fundamentally missed like this whole section of things you were supposed to implement. Got it. So, yeah, I think it's... Okay. Talk to me about the go-to-market motion. Is this just a landing page you're driving traffic to it? People sign up by themselves? Are you doing founder-led sales, all of the above? No, no. We're landing page. Like, this is our go-to-market right now. Go to Casper. Go sign up. Yeah, you can just go.

32:33You install the bot. We have a seven-day free trial. Okay. Consumption versus seat-based pricing. What are you thinking? Yeah, it's seat-based pricing. Okay. So for every developer, it's$20 a month. Just simple kind of flat rate. Are you running into cost problems? Because we've seen this like, you know, the latest and greatest LLM comes out. It's really expensive. GPT-03 just dropped by 80%. So anyone who is having a problem with their cost is probably fine right now. But how are you thinking about that side? Yeah, we found that like actually for just like, you know, running lots of experiments on your code to find bugs, that it's actually better just to have a really fast and small model.

33:12And so we've actually, yeah, we haven't had these sorts of problems yet. So what does that mean? Like LLAMA, fine-tuned? We've talked to LLM training companies that have trained even smaller models just for JSON to formulation or just translation models or just profanity finding. Are you thinking about going so small you could run it on a gaming GPU or are we still talking about the big boys? Yeah, so right now, we've actually gotten a lot of mileage out of Gemini Flash. We started by, we fine-tuned with RL a model that was specifically good at using tools. And so, yeah, we're getting ready to do that.

33:50Once we find all our pain points exactly in our current architecture, we can train the exact right thing. And the smaller, faster models that are targeted for the specific use case are better. What were you guys doing before YC? We were both researchers. So I was doing research in software testing using large language models. Cool. And Mateo was doing his PhD in reinforcement learning and formal methods. Very nice. Yeah. So our kind of research has come together to make this happen. Trouble. Trouble. How are the metrics? How's the raise coming together? How's the pitch for Demo Day? What are the goals?

34:20Excited about Demo Day. Goals are raise a lot of money here. Let's go. Let's go. We're going big. Bring some big napkins around. Yeah, yeah, yeah. Just do it here. Perfect. Um, and then, yeah, it's, we're slowly kind of growing. Well, I wouldn't say slowly. We've doubled our growth kind of every, every week for the past kind of couple of weeks. Nice. So that's a better way to frame it. There we go. But yeah, we've gotten, I think we're up to like 18, um, different kinds of companies using our tool. That's great. So yeah, it's been awesome. Cool. Well, good luck to you. 1800 soon. Yeah. After this, when all of you go subscribe, yeah, then we'll be at 1800.

34:56Fantastic. Well, thanks boys. Have a great rest of your life. Yeah. It's been a pleasure. Yeah. Cheers. We'll talk to you soon. Show the Crocs off, too. He's got the whole fashion. They've done everything. Let's bring in the next team. How are you guys doing? Welcome to the stream. They've got the shirts tucked. Oh, you have the rare hair on? Did you miss out? Do you need one? I need one. Thank you. Take it. Welcome. Eddie. Nice to meet you guys. Nice to meet you. Nice to meet you. Nice to meet you. Nice to meet you. Looking sharp. Thank you for tucking your shirts in. We're keeping it respectful out here.

35:25Introduce the company. What are we building? So we're CodeTool. We are building AI agents for security teams. Cool. Sorry. We're building AI agents for security teams. Okay. Is specifically cybersecurity teams, are we talking DDoS? Are we talking somebody goes in and tries to steal secrets, steal data? What are you talking about? So, like, you know, you can think of cybersecurity as split into, like, AppSec and OpSec, where AppSec's, like, you know, defending, you know, your deployed software out into the world. And then OpSec is, like, operational security, right? Where it's, like, you know, protecting your employees from phishing.

35:52We're definitely more on the automation side for the OpSec side of house. So think of, like, basically, like, allowing security teams to, like, triage their tickets a lot faster for like impossible travel problems or for like, you know. Oh, impossible travel problems. That's like a buzzword for what happens. Oh, yeah, it's like, you know, Eddie signed in from Singapore. Is that legit or not? Yeah, it's like, well, he was in the office earlier today. He couldn't have possibly gotten there because hypersonic travel doesn't exist yet. Exactly, exactly. Not quite yet. Not quite yet. That's going to make it really complicated.

36:19No, no, it's going to get off. When you get from LA to Tokyo in two hours, everybody's like, well, I guess he did log in to Tokyo. He was going for sushi. It's funny because I think a lot of people have this hacker aesthetic in their mind when they think of cybersecurity, which is a lot of in the movies, fucking around on a terminal or something like that. And it's just like, when in reality, most of the time, it's people triaging tickets day in and day out and that kind of thing. Our goal is basically to automate a lot of the BS that these teams have to go through and a lot of the annoying stuff and then allow them to get back to work.

36:49So what's the go-to-market motion? Are you doing enterprise deals, founder-led sales, selling to other YC companies? What's the scale of a company that needs to use your product? customer is ramp. Oh, no way. Wow. That's hilarious. I mean, where do you have it? That's tough. You're kind of starting. That's good. We got to keep in the family. Time is money. Say both. Come on. There you go. That's nice. I mean, Eric is just an incredible CEO. Yeah, yeah, yeah. He's YC alum. Yeah. But the motion is we really want to bed with this, like, the most cutting edge, like the best teams, like ramp, that can actually, like, inform us as much as we're helping them on how the best teams operate.

37:25Wow. And so we're looking for tech forward. Enterprise would be awesome. But anyone that's pushing the boundaries of what's possible automation. And what does the integration point look like? Is it a single security person at a company can get set up? Or is it something that needs to be deployed throughout the enterprise and has a much more staged rollout? Yeah. I think basically where it is, ideally what we would have is we'd probably first start working with your detection and response team. and get that team plugged in. Basically, we're heavily leveraging AI tool calling. So the idea is we plug into all the different points in your Slack, think like Okta, think like Panther, think like all of these different products kind of thing.

38:08And basically then what we're able to do is deploy out, like allow these teams to write their own agents. So they're in there customizing their own system prompts and all this kind of stuff to go out and tackle their tasks day to day. So I think the initial go-to-market motion is work with these high-tech teams who are really used to automating already. And then basically kind of build up a nice collection of agents, kind of start getting a good network effect, Ikea effect, where people are building their own agents, sharing them out into the world. And then from there, we can kind of move out to wider and wider and less technical teams, where it's basically like we can kind of start to plug into people's stacks and wholesale automate out of the gate and solve a lot of these problems.

38:40How did you two meet? So we grew up together in Santa Barbara, California. We have lived together since we went to school. But yeah, our third co-founder is just outside. He's probably watching this right now. That's great. But yeah, we were really stoked when you guys watched our launch video. Oh, yeah. We did on the show. We were the mole. We were the mole. Wait, you guys are that company. Wait, wait. They did the Rippling deal. That was the single best ad. That was incredible. That was the single best ad I've seen this year. 100%. No, we were pumped on that. Because there were so many ways to do that and have it be just bad.

39:13Oh, yeah. It could have been bad a million ways. But it was tasteful. It was funny. It was funny. It was perfectly timed. Yeah, it just made it funny. Where did that come from? Well, it's funny. We were scared it was going to flop, to be honest. Totally. We didn't know how it would land, and we're happy with how it did. But no, we came with the idea. We worked with one of my friends growing up does comedy in L.A., and I live in L.A. But she knew a director who was amazing, and his friends who were also actors and stuff got in and acted for us. And so it all came together with friends of friends.

39:45It looked incredibly polished. Can you give us an order of magnitude on the budget for that thing? It was, let's see, five digits. Five digits, but not six.

39:58It looked like a hundred thousand dollar project. But it was actually pretty funny. We showed the original idea came from Gary, actually. We showed him the demo on the first week of, and he's just like, we started showing him how you can query Slack and that kind of stuff. He's like, oh, you guys got to do the rippling thing. And we're like, oh my God, we got to do the rippling thing. And then we showed him the final cut the night before we went with it. He's like, you guys should tweak these couple of things. And we're like, fuck, okay. and stuff like that. It was funny. It's funny because it's like YC on YC on YC violence.

40:29But it's really just like, look, they're going to sort their thing out. You guys are just having fun. Did that ad catalyze around for you guys? It did. In some ways, yeah. I think a ton of investor inbound off that. Yeah, exactly. No, no. It really grounds it because not everyone works in cybersecurity every day. It can be a little abstract. There we go. There we go, boys. Let's go. What do you do with that? Are you pushing it up or something? Yeah, we're figuring it out. We got a lot of these. Congratulations. I'm super excited for you guys. I think we got to give them a different hat for this.

41:01I mean, it's the best. Oh, man. Yeah, we can give you two of these. Let's go. Let's go. It's a special edition. Congratulations. Thank you so much. Really great chatting with you guys. Yeah, great chatting with you. Super excited for you. We'll follow up. Yeah, we'll talk to you soon. Cheers. That was fantastic. I'm so glad we got to talk to those guys. That was such a good video. Out of the year. Out of the year. Out of the year right now. And right at the perfect time, because we were just hitting like peak vibe reel, right? Yeah, and if they had dropped it even a month later, it would have been gone.

41:28Welcome to the stream. I'm John. Nice to meet you. Nice to meet you. I'm John. How are you doing? Are you going to PMF or die season so soon? Everyone died. Everyone died. Did you ever hear the story about PMF or die? They died. They died. It happens sometimes. Never lock yourself in a room for the 90 days. Instead, move to sunny San Francisco. do YC. That's our recommendation. Basically the same. From here on out. Anyway, please introduce yourselves and the company. Yeah, I'm Tom. I'm Eric. How are you guys? You guys look, you're not related at all, right? We're not related. You could go by brothers.

42:05People ask us, people ask us. Yeah, yeah, people ask us. What does the company do? We're doing a Gentic people search. So we have a database, people database of around 200 million people. Interesting. We use a Gentic search, a search over that for companies and businesses to do like sales recruiting uh gtm uh so so are we talking about specifically like i am a recruiter and i need a salesperson and i'm going to go to you to try and hire them no no so basically like you can put in a a criteria like essentially what we do is hold hold the microphone up here here push the microphone a little bit yeah i mean essentially what we do is we like like like deploy an llm okay assign every profile to that llm and given given your criteria, that could be like one paragraph long.

42:47We would just ask the LLM if this profile meets that criteria. And then we built a load distributor to run 10 ,000 LLM calls in parallel. Wow. To do that at scale, we can search over like 5 million profiles in like 30 minutes. Okay. So walk me through one of the key examples. I'm sure this is live, you have customers. Yeah. What's an example of someone using this? So for example, we had a query come in yesterday, actually. It was just like every founder that was a CTO of their startup was acquired by Databricks or Snowflake in the last three years. And they still work there. And there's only probably 25 people in the world that fits that profile.

43:25And we found all 25 of them. Because we can use and have an LLM going. Because we can throw as much compute at the problems we want. Interesting. And then LLMs get better the more compute you throw at us. It's interesting to be able to find information that is historically effectively impossible to find without doing all the work that you guys did ahead of time. Makes a ton of sense. So what buyer or what buyer archetype within an enterprise or a business is most excited to buy your product? Yeah. I mean, like mega recruiters by far. Recruiting firms? Yeah. No, just like people that, you know.

44:02Or like a big recruiter at Meta, for example. Yeah, I mean, we work with Mercore right now, for example. Oh, Mercore. Okay, cool. So, you know, just like people hunting for talent, but it's like a generalized type skill, right? So, for example, you know, like in AI labs training a new voice model and they need like people that speak Cantonese. Sure. It's like, okay, I find me every Cantonese speaker in the U.S. that has like a podcast presence. Yeah. And they're like, okay, great. These guys can come train our voice models, right? Like that's kind of bad. Or they might even need someone who speaks Cantonese and also is an expert in biology.

44:28Yeah. So they can talk about the biology terms. And that's something that, how are you going to search for that on LinkedIn? It's so funny because I run these type of queries in my own head where I'm like, we need a videographer who's in L.A. that has experience in film. But, you know, he's also part of Teapot. That's basically what we need. It has a sense of humor. How would you even know that? Well, if you look through their posts, I'm sure you can figure that out. It's basically like as close as you can. If you just give like a criteria to a human recruiter. Totally. So you can imagine like 5 ,000 human recruiters like manually looking over profiles and then, you know.

45:02Okay. All right. Business model. Most recruiting firms, they charge on a per fee basis,$30 ,000 to place an engineer somewhere. Are you doing a seat-based pricing, consumption-based pricing? This sounds expensive if you're talking about running 20 million LLM queries at the same time. It's actually not as expensive as people think because open source models have gotten so good. I'm guessing any deep research query costs like$10 maybe. but then we recharge for conversion with our B2B customers so that's probably pretty expensive and also we have a platform that's just available to everybody so they can pay us a subscription fee search as much as they want and then do email enrichment phone number I feel like we could use this we might be potential customers we need a guest that doesn't hate tech that can talk about this how did you two meet what were you doing before YC Yeah, so we met in elementary school, actually.

46:01So we were originally from Canada. Let's go. We went in elementary school, and we became close friends. Brain drain. Yeah, it's been great here. But we became close friends when Eric, he uninstalled Windows on my computer. Nice. And I was really pissed at him for a day, and then we fixed it. So then we became great friends after that. He uninstalled Windows? That's our great. You just couldn't load anything? Yeah, I had homework to do that. No, I had homework. It was during school. and then ultimate prank, uninstall Windows on your best friend's computer. Boys being boys. And then we spent one semester at college each.

46:33So I was at Penn, he was at UC San Diego. And then come January, we were both like, why are we even there? Let's drop out. So you guys 18, 19? We're both 18 right now. Wow, here we go. Let's go. What's the youngest team that you've met here besides yourself? 17. 17. They're a high school senior. Okay. I think they got here. I'm doing some robotics research. It is getting younger. Yeah, yeah. Yeah, we have a bottle of wine here. We're going to give it to you. We're going to give it to the youngest. Hopefully, you're not going back through YC in three years. Hopefully, you're at the NASDAQ or something.

47:08Talk to me about traction, metrics, anything that you're sharing here at Demo Day, anything to get the venture capitalists excited. 270 paying customers. Wow. So, I don't know if you guys heard about Linked. Linked? No. Yeah, yeah. All right. So, a very early version of this product relaunch, which is like ranked Stanford. like Stanford rank so we basically built this we basically scraped the entire alumni database of Stanford I'm sure they love that

47:37we put it online and then we had this we had this app where people can like see two random alumni next to each other and then you vote for who's more cracked who's more cracked and that version of the app picked up like 80 ,000 That's amazing. The users was how we got into YC and stuff. Very cool, very cool. Yeah, and then we came down here, 207 paying customers, about$16 ,000, a monthly recurring revenue since three weeks ago. Yeah, wow. Is the team just you two right now? So there's three of us. We're two co-convoters, our founding engineers from high school. He just graduated. The same high school.

48:19Yeah, we all went to the same high school. That's amazing. I'm super excited for you guys. Congratulations on YC. We're going to get on the product. I actually have a bunch of people to send this to. Yeah, this is fantastic. You should give it a try. Fantastic. Thank you so much. We'll talk to you soon. Awesome, guys. See you. See you guys. Fantastic.

48:39Let's bring in the next team. If we have one, come on down. How are you doing? Aaron, what's up? Aaron, so this is a YC alumni. He just raised a Series A today. Congratulations. Yeah, we can tap it in. Series A. Here, sit here. Yeah, so we were trying to make this happen for the last week. Amazing. But he just announced his Series A today,$17 million, right? $17 million. Congratulations. Thank you. Thank you. There we go. I mean, we have stuff to celebrate. I already got the hat. Yeah, I got the hat. Hey. Congratulations. Seems like you've done that a couple times. Yeah, yeah. This is the biggest round of the day.

49:13This is the biggest round. Who did the round? How far along are you? How big is the company? Give me some stats. Yeah, yeah. Okay, so 8VC did the round. By the way, guys, this is awesome. I did the interview with VentureBeat, and I'm way more starstruck by this. So I'm not just saying that, but I'm like, all right. Thank you. Yeah, so 8VC led the round. Cool. Let's go. Joe Lawson, absolute dog. What partner are you working with? Jack Moskovitz. Nice. He's great. He's fantastic. Yeah, so how far along? So we're... And were you in the last batch? No, we were a couple batches ago. We were summer 23.

49:45Cool. Great vintage. Was that still remote? Excellent vintage. or was that back? No, we were the first back. Back, there you go. We had an advantage. Very nice. But yeah, so Outset does AI moderated research. Okay. So what does that mean? It's like when big companies have to do a ton of research, like Nestle has to go figure out whether they should launch this weird version of DiGiorno pizza or something. They go do research. So either they're running massive surveys and they're getting very low fidelity data. They don't really know much. It's just a bunch of numbers. Or they're running interviews and that's where it's a one-on-one interview with everybody, right?

50:15Which is not actually cost effective. Yeah, totally. So now you can run AI-led interviews. That means AI is leading the conversation, digging in, following up, probing deeper, and synthesizing all that for you. Got it. How big is the legacy market? So the research market is like$140 billion. It's actually a really big one. Yeah. Just like, yeah, there's like huge companies that you probably may have never heard, like Ipsos and Contar. Yeah. There's like big guys that are just kind of sitting there. So there's a lot of opportunity. And yeah, we work with like Microsoft, Nestle, Weight Watchers.

50:47So we've been super enterprise focused. And yeah, we're a couple of years in. What are the interaction patterns that you like or you think are kind of unlocked by AI? I imagine that voice phone calls is top of mind versus like forums, which was always like available and already computerized essentially. It's actually, I think the cooler thing is what you can get from participants and synthesize. So it's like, so now we do video interviews and then do audio and they can even share their screen. Yeah. Right. So all of that is being ingested. and then like people ask like to say i have an avatar when it's interviewing and we actually did research and people don't want that they don't want the avatar no they don't want the avatar yeah that was something like weird uncanny valley stuff where i'm like and you're just like people are used to phone calls yeah people used to phone calls and they're fine sharing their video but like if you suddenly see an ai you're like i'm trying to make you know 15 bucks doing some research totally like you suddenly see that you're gonna be all weirded out so it's actually much better to have ai come through with voice and text yep talk to me about recruiting you said 15 bucks that's enough to get someone to jump on, but how do they even find out that 15 bucks is on the table?

51:43Yeah, so we partner with a number of different companies that do nothing but panels. And recruiting. Nothing but recruiting. So you have user interviews and prolific. And they probably get some demographics and some basics. They get all that. And they have millions of people that are ready to... It's like gig work, right? And so they connect. Okay, and you know, 100 ,000 people that drink energy drinks to give me feedback. Exactly, exactly. And you could do all that through our platform. Is there a certain unlock when an AI doesn't need to, like, you know, if a human is scheduling research calls, you know, you can imagine they do like four 30 minute blocks and then they take a little break and they do some more or something like that.

52:18I don't know how it works, but an AI could hypothetically talk for hours and hours and hours and hours, like kind of a hundred percent. So there's like, you know, I think a lot of AI stuff is like, Oh, where can it replace the human thing? And now it's cheaper, but actually like this, this is, uh, taking on the stuff that like humans can't physically do. We only have 24 hours in a day and so it happens that people use it to like interview 500 people right in a day or two and that's like it's literally not possible it's like just a lot of physics don't allow it yeah and so they're able to do that they can do multilingual they kind of do it all at once so you can wind up like unlocking actual like stuff you've never heard can i curse on this oh we usually don't but all right stuff that they like would not have uncovered anywhere else yeah um and they actually are able to get that like in this way yeah what's your what's your stack under the hood what models are you using getting the most value out of?

53:07Yeah, we're using a combination. So it's like we're hitting multiple models constantly, but we're using a lot of Azure, actually, so OpenAI models under the hood through Azure. Is their model router helpful? I saw they announced that at Build. I haven't talked to anyone that's actually used it. I don't know yet, actually. So I don't think we've played with that yet. So basically, what we need is just incredible amount of reliability because what we'll have is three customers are all running 500 simultaneous interviews. So we have to actually scale very, very quickly. And also, like a lot of our customers, like the Nestle of the world, they care a lot about super safe, super reliable.

53:42So we kind of wind up needing the kind of Azure. Yeah, and then you're also probably going through peak LLM usage hours, too, because you're not doing it in the middle of the night. No, no. We can't just process it on our own time. It has to happen in that moment. But then we also fall back to OpenAI. We use Gemini for stuff. So we're hitting a lot. Yeah, talk about the, is the voice modality beyond the uncanny valley at this point? Is that why voice is valuable? Are you using voice a lot? Because you can imagine that this might have been possible via text interactions. You're texting your responses back and forth.

54:14Probably get more out of a voice interaction. Are we just, if we talk to you again in like two years, do you think it'll be like, okay, yeah, now I'm a believer in the avatar thing because it is photoreal? Okay, so two things. So one, right now, what matters most is obviously the modality of the participant. Like if you think about like all we care about is the most deep, like, like thoughtful, in-depth data that we can like pull from you. That responds. So that matters way more. And that's going to be way better than text. And that's way better than text. Our first product that we like. You got to get them talking.

54:43Yeah. Our first like pilot with Weight Watchers back in 2023 was like all text. And it was actually pretty good. But like it's nothing compared to what people actually say. Sure, sure, sure. And so that's what matters most is like video screen share voice from the participants. from AI, yeah, like, so we use voice a lot. So there's a voice-to-voice mode where there's, like, no buttons. It's just conversation. And, like, people like it, but, like, not as much. And I think there's still just, if it's not a person, then it's still got the minor imperfections, the kind of slight bits of latency. And so I do think in two years, we'll talk more about that.

55:14But ultimately, like, people know it's a computer. Like, they know it's AI. And, like, they're kind of cool about it. Like, we've done a bunch of this research. We're like, do you care that it's AI? They're like, no, like, I get it. Yeah. Right? In fact, they have even expressed that they'll share more because there isn't a person. It's an idea of social desirability bias. You're like, I don't want to come off as a person that is X, Y, and Z to another human. I will just tell you, look, I want this to be faster. Or I want this to be better. Or the real reason I'm not using your product is this.

55:44I'm not going to say that to a PM or a researcher who's at the company. But instead, you'll tell the real truth. And that was like our initial pilot with Weight Watchers was all about weight loss. And like that was like a thing people don't want to share a lot about. But with AI, they like shared everything about their life. Yeah. I mean, it sounds like you've done like you've worked with really big companies and that's where the money is. But when you went through YC or if you're here, you know, there's a big theme of talk to your customers. Yeah. Has there been any pull at the lower end of the market?

56:11Yeah. It's funny. Sometimes you get a founder reaching out like, I really want to use this because I don't want to talk to my customers. Yeah. I might be like, talk to your customers. We don't want you as a customer. We're not super intelligent. Yeah, exactly. Yeah, there's basically, I think a good reason not to use an AI moderator is like, if you could sell to that person, probably don't outsource that. Maybe to build the relationship yourself. So the truth is, especially all the B2B stuff, there's not as much. You usually turn around and say, come back when you're like a couple hundred people and you're kind of scaling that out.

56:41But yeah, with some consumer stuff where you really have millions of people you're trying to learn from, it can make a ton of sense. Talk to me about the data processing and the intelligibility that goes into once you have all that data. It's very nice to be like, hey, we have like 5 ,000 hours of video. Yeah, we prepared a 200-page report for you. Yeah, but nobody reads that. So how are you thinking about actually compressing it down? Because that seems uniquely suitable for AI. But there's still a lot of art in terms of like a lot of people say, oh, yeah, I did a deep research report. And then the next query was give me 10 bullet points about that because I didn't actually want the full report.

57:16Yeah, yeah. So, all right. So our vision here is like, we should be building, we are building the like deep research, but for primary research. So you think about deep research and like, all right, all desk research is now commoditized. So what if you just ask a question, you just want to like learn from real people that's much more up to date that whatever, like your own proprietary data, it should be the same kind of idea, right? End to end agents are doing that. But today, like the premise of AI kind of scaling your qualitative research is that you also have to help them do something with all that conversational data, right?

57:45Because otherwise you're like left with just like hours and hours of transcripts. I'm like, I'm not going to review that. The application of this is I know this is not what you're focusing on. But if I was a VC that wanted to write a multi hundred million dollar growth check, being able to like get live interviews with like 500 customers. I mean, people are going to do this with GLG. Yeah, that's right. No, this is a huge opportunity, actually. So, yeah, I have a whole side thought. Like, I think one of the challenges there is with the expert networks is all about finding the right people. But if you're in a situation where you can find the right people and you can find 100 of the right people.

58:19Well, we just had Play-Doh AI on. You should talk to them. They're in this batch. They do, like, LLM-based people search. So you can find, like, every person at Databricks that was a former founder. No, there's an opportunity to connect all of that through that. But what I was going to say with the synthesis is what we do is basically take in all the video stuff. And then we, like, process it. And we give you, basically, reports. But it's, like, not just a, like, here's what it's telling you, like, deep research style. But what you could do is start slicing and dicing it. We quantify it. We give you breakdowns.

58:48We build highlight reels for you. And then people can cut it and segment it. And so it's like a whole analytics suite on top of qualitative data, which is not a thing that has ever been done. What were you doing before this? So before this, I was VP of product at Triplebyte, if you know. Oh, yeah, yeah. Harsh Tagar. Yeah, Harsh hired me. Yeah, exactly, exactly. Yeah. Anyway, so I worked at Triplebyte. at Jumpstart, another company. So I was leading product and design teams. But earlier in my career, I was a consultant where I was doing this work nonstop. Are you a nominative determinism guy? Yeah, he's looking for the capital cannon to fire a billion dollars into research.

59:30I have not capitalized on that. I was looking at that too. We're big into the names or the truth. I like that. I have a friend who always calls me the loose cannon. You can never know what to expect. That's a little bit too much. Dial it in the capital cannon. You can trust it with your capital. Just give it to him and fire it in the capital cannon. And boom. I'm going to pivot off of loose cannon onto capital cannon. Aaron, capital cannon. There you go. Anyway, this has been fantastic. Thanks so much for having me. Great to meet you guys. Congratulations. We'll have you back on sometime. $17 million Series A.

1:00:06Let's go. Fantastic. Biggest round we've heard yet. Let's bring in the next guest. Operative. There we go. The shades are going on. The pit vipers are going on. The pit vipers are going on. Welcome to the stream. How are you doing? Hey, how's it going? Good to meet you. I'm John. I'm Eric. Eric, nice to meet you. Nice to meet you guys. Good to meet you. You're going to want to keep those mics close to you because it's noisy here at YC Demo Day 2025. Good to meet you. Can you introduce yourself and the company that you're building? Yeah, we're operative. I'm Chris. I'm Chris Settles. This is my co-founder, Eric Quintanilla.

1:00:41Very nice. You can introduce yourself as well. But we're friends from high school. We met at West Aurora High School in Aurora, Illinois in 2014. We were doing our first Java programming class together. Wow. And 10 years later, here we are. Deep, deep. What did you guys do between then and now? Yeah. Do you want to learn other programming languages? Do you want to tell more of the story, Eric? I mean, we went to college. Maybe you still write Java. I don't know. Yeah. We went to college, worked at a couple of big tech companies, We're a couple of startups. Yeah, yeah. Let's hear it. Big tech. Amazing.

1:01:15RIP. Chris was at Uber. You can talk about that. There you go. Nice. Did you wear the pit vipers during your main pitch? Sorry? Oh, you're asking where the sun is? Did you wear them while you were pitching the full? Oh, no. I just got these from Marco. I don't know. We saw Marco in here wearing these. We're like, Marco, you look so cool. Yeah, yeah, yeah. So break down the company. Yeah, what are you guys doing? All right. So we're working on web app code generation for internal APIs. Awesome. So you can imagine having like a lovable plus a retool for your company where you can on demand just create any kind of application that you're thinking about, whether it's like a new dashboard or it's some kind of like app to manage your airflow.

1:01:55And you want to have a nice UI with it and you want to connect it to all these different pipelines. I mean, that's just one use case you can think about, but you can build any kind of front end application that connects to existing APIs that you have in your company. Yeah, I remember when you set up a Django website, you kind of get the Django admin out of the box, and you can just investigate all the different classes that you've defined in the database. What are some examples you could give us of these internal APIs that typically float around in companies? Or you can either give a precise example or just a general example.

1:02:29Yeah, do you want to give one? Yeah, I mean, a lot of the use cases are centered around customer service. So people want to interact with a database. People who are doing support want to change payments and stuff like that, data visualization, stuff like that. Okay. So, yeah, I mean, is it unique about bringing different services together? Because a lot of smaller companies would say, you want to look at the payments admin? Head over to your Stripe dashboard because Stripe's already built that. Where are companies currently falling short? Is it that they're not bringing the data together across services?

1:03:02or is it that they're developing their own databases and their own tables that just don't have an API out of the box or don't have a web app in front of their API out of the box? Yeah, so definitely startups are super easy to just say, okay, yeah, check out the Stripe dashboard and manage that. And then that's what Stripe does. That's a typical customer service flow. It's like you go to Shopify for this, and then you go to the CMS for this, and then you go to Stripe if there's a problem over there, and then you go to payroll. It's all SaaS products. Yeah, but one of the things we like we noticed, especially at our jobs in big tech, Yeah.

1:03:36is that eventually your product gets really complicated and you actually don't, there's not a SaaS product that those services offer. Sure. And so you end up needing to build something on top of it in order to manage it. And then you end up needing to build all these different internal tools to like have all that organized in a nice way where you can connect it all together because the traditional SaaS doesn't support the features that you're looking for. Got it. And so then as a result, you hire a team to build out those kinds of applications and then spend a lot of money. And that's what we're trying to bridge that.

1:04:07What was the aha moment for you guys? Were you seeing the kind of explosion of tools like Lovable and things like that? And then you saw, wait, why aren't people doing this internally? Or what was the kind of moment that you guys decided to focus on this? I mean, I think for us, we saw like we launched this like Lovable type product and we saw a lot of people from companies using it. As a consumer product actually at first. Yeah, to build internal tools for their company. So we actually, we were like, oh, we should just bring this directly B2B. Cool. Yeah. In some ways it makes sense being able to quickly generate something, test it.

1:04:38It's very different than these ephemeral products that exist and ultimately need to be re-repelling. Sometimes completely rebuilt. Talk about the go-to-market motion. It sounds like this is not a company where you need to sell to every other startup in YC. Who are you selling to? How are you actually convincing them to take the leap and go with you? Yeah. So we're starting from the initial sort of traction we saw with the consumer product. We decided we want to work on this enterprise direction and or larger organization direction because of these use cases we saw with people signing up for all of their work emails and building apps.

1:05:14So we're starting some design partnerships with large organizations. I can't say, I won't say the exact name. And we're planning to kind of just leverage our warm intro network to be able to talk to some more organizations. And I think there's some, there's actually a large percentage of companies that even will try to use like Retool to build internal apps. And so we think we can work with some of the other YC companies that are doing things like that and using that. but maybe have... Do you guys work with Retool's demo day? Are you guys working with Retool? Or are you competing? Kind of competing.

1:05:53Good luck with what you're talking about. He's an absolute dog. He's a monster. But good luck. It's a big market. Yeah, we have a pretty cool direction that's pretty unique at least at the moment. So we're excited to go after. Talk to me about the actual instantiation of the web app. I'm sure you're using AI and code generation. What's working? how much are you leveraging O3 or Claude or Open Source or Gemini? What are you looking for? Where are the models falling short? Where are you trying to stay ahead of the puck because everything's developing so quickly? You want to take it? Yeah, so I think on the platform I think the foundation models provide a nice foundation but we try to leverage good tools, so like viewing a file tree we had a browser agent on top to test and validate the app.

1:06:41The model's far short in, I think they're good at going zero to one, but I think if you go zero to one and then there's a bug, going back and finding what file it is, what's implicated in it, I think it's very hard for the models and it's better to just start over. Okay. Yeah. Have you benchmarked against how long it takes someone to build a web app for a, on top of a private API, using cursor and just saying, I'm going to vibe code this myself versus your product? because it feels like it's getting faster and faster. And there's a lot of, I mean, there's like, there's whole open source projects for like, build out the API docs around an API just from, you know, this was programmatic.

1:07:19This wasn't even AI. You know, you just have like, oh, you want to put this, you want to instantiate this as a blog. Okay, here you go. What are you benchmarking against? Actually, maybe a cool story, like on the backstory of how we came across this is that we actually started with people wanting to develop apps, inside Cursor. And so one of the kind of unique insights we had is that today, developing any kind of front-end requires that you ask Cursor for a prompt and then you open up the web page and go click on it and just verify that it looks how you want it to. And then you can go and then tell Cursor, okay, I want to move the blue button inside the black box.

1:07:57And then you have to just keep prompting it until it does that. So actually, the first thing Eric and I worked on was actually an MCP tool that opens up a browser agent that can go and view changes that a coding agent makes. And it will just test to see visually if the app actually works. Actually, there's one of our users over there. He's just giving us feedback on it. And we grew that to around 1 ,000 GitHub stars. And so people got really excited about that, but we wanted to bring those same tools into a product like Lovable or like a code gen and be able to have all of this code generation with testing built in.

1:08:38So that way we can allow users to like just go from prompt to a working webpage without any interaction. Very cool. Are you guys still raising? Raising right now? We're still raising. We have, we've been back by Weekend Fund and also a few other angel investors. And we are, we're still raising and open to chatting with other investors. Oh yeah. We have an angel investor and Retool investing in us. Play the hits. Play the hits. Play the hits. Find something you love and just keep. Do you have any metrics that you shared here at Demo Day? Yeah. As I mentioned, we had the GitHub repo. We scaled to 1 ,000 stars.

1:09:16That's pretty good, yeah. We just launched the consumer product. It got around 2 ,500 different users using Operative to build apps. Yeah, the usage was crazy, yeah. Yeah, that's great. And right now there's about a 4 % conversion from free to pro plan. We have a pretty generous free tier, so you can go and use the app however much you want. But for the power users, there's around 4 % of them converting. So we have about 1.5K monthly revenue at the moment. Amazing. And that's with the consumer product, but we're planning to scale more revenue. Yeah, I think the enterprise, you could probably get that on your first contract.

1:09:47Exactly. Thank you for stopping by. Please enjoy these hats. Thank you so much. Have fun out there. It's great having you on. Awesome. We'll talk to you soon. All right. Let's bring in the next crew. We are at YC Demo Day. Hey, how you doing, Matayo? What's up? What's up? It's time to talk sleep scores. Let's talk sleep scores. How did you sleep last night? It was brutal. I was at the Rosewood. The Rosewood doesn't have eight sleep. It's a big problem. It's a big problem. Not yet. Yeah, yeah. 90 % of the guests are good. Good to see you. What's happening? Great job, guys. I want to talk F1. I want to talk F1.

1:10:21You just did an interview with Charles Leclerc. Yeah. Break it down. What did you learn from him? His obsession for every single detail in his preparation. He goes for two weeks at the mountain in January to prepare for the season. Is that just for elevation or? Elevation training? Elevation. And then he does every sort of skiing for like eight hours a day. And then he goes to the gym and then he brings the pod. He sleeps on the pod while he's there. That is when he really prepares the season. Yeah. and then his obsession for every single detail when he travels, all the gadgets that he brings with himself from devices for recovery, ice bath, obviously the pod.

1:11:04Yeah. Is he doing the leg compression thing? That's a big one right now. And I don't know if you saw it. So Charles was talking to us about the fact that when he's on the grid, it's really hot. Yeah, yeah. And a lot of athletes in Formula 1 now, they use this vest. the cooling vests. Oh, interesting. But they don't really work. Okay. And so the way you want to really cool your body, you need to cool the palm of your hands. Oh, interesting. And so I was with my co-founder, Max, and we said, what if we build cooling gloves for you? He said, oh, that would be interesting. So we built the gloves in three days.

1:11:40No way. No way. That's awesome. And then one of us, one of our people, they flew to meet him in Barca. In Barca, they were like 35 degrees Celsius. It was super hot. And so Charles started using these gloves, cooling gloves that we built for him. And then as he goes in the box, Ferrari takes a picture of him, just, you know, for Scuderia Ferrari, blah, blah. And so he was on the cover of the Instagram account. With the A-Sleep. And so he was pre-branding for us with the device we built for him. And so now you see that with the tires, they have the tire warmers, then you have the hand coolers. Every different part of the F1 machine needs to be temperature controlled.

1:12:19But then other athletes saw him using these. And so, paddle players, tennis players, they all say, oh, can I use it? And say, but man, we didn't. It's a one-off. It's a one-off. And so, now we have a team that is building this cooling globe for a few athletes. That's incredible. Great partnership. What has YC Demo Day been like for you? Are you here just trying to sell eight sleeps? Or are you also doing some investing? What are you getting out of YC Demo Day today? Yeah, I do a bunch of investing. I invested like probably 20 companies in this batch. Already? Wow. Amazing. You got to get it. I'm sure that you got to get them in here.

1:12:55You need to start early or then you're out. Oh, yeah. You got to get it early. You got to get it early. And so in every batch, I try to invest in around 20 % of the batch. Oh, cool. More or less. That's great. And I have been doing this for two years. Have you seen anyone doing cool hardware, cool consumer devices? I know that there's a ton of AI. There's some hard tech. Defense tech's becoming cool. But I always like the aid sleep because it's a gadget. You can give it to someone for Christmas, and they experience it every day in a very different way than just nap on their phone. Have you seen anything?

1:13:28I'll answer your question, but it's cool because I walk around with my hat, and people stop me, and they say, oh, I sleep on your product. I just met two founders, and one had a 75 score, the other one a 48 score. Oh, no, that's weird. 48 the night. We're never there. We're never there. That doesn't work. I've been smoking Jordy this week. I beat him twice in a row. I'm up in the 90s. I'm doing great. I've had a rough one. But there is not a lot of consumer hardware. It's hard. It's so hard. Yeah, it's really hard. I was just talking to a few founders. Even when I look back, other companies starting with us, they were struggling.

1:14:02So we have been so lucky to be here today. But you see a lot of really cool stuff now in robotics. Yeah, you have Matic robots. We've had those. It's been really cool. Yeah, we have them in the office and at home. And so usually there is an hardware section, and that is the first section where I go. So hard tech or hardware, that is my passion. Yeah, yeah, yeah. And then AI is everywhere. Yeah, yeah. I mean, you see a lot at CES, you'll see the AI-connected oven and the AI-connected toaster. And YC has kind of stayed away from that because it feels like it's hard to build an independent business around.

1:14:32But I am optimistic that once the software side of AI is so commoditized and so just ubiquitous, we'll see another revolution in hardware, another turn. And I think with the robots, things will get easier because robots will explode and a lot of consumers will buy different shapes of robots. Yeah, this is what Nat Frieden was talking about. He wants the robot that picks up the leaves one by one instead of leaflets. And it's like a funny idea, but it feels like, yeah, it's only a couple of years away and it'll just be a company. Even for security, I have this idea about a robot that goes around your house for security.

1:15:06Yeah, so just putting cameras everywhere. With cameras and lights. I would immediately buy that if it could go around my house and make sure that... Especially if you could put a gun on it. I mean, Amazon did launch a drone that will... Or maybe they just launched a video about this. There's a drone that takes off from a base station where it charges, and it can fly around your house and kind of patrol inside your house. But yeah, I don't know. When the fires were happening, I figured out that a lot of fires in LA just started from a single ember landing in a backyard or on a house or get stuck.

1:15:36and you just realize that there's probably an opportunity for drone-based fire fighting. Yeah, there's actually a couple companies that did water turrets that you mount to your roof, and then if they see fire, they can just spray right there. It's literally just a hose on an articulated arm. It's not that complicated, but you could sell a lot of those. What about at 8? Any other hardware coming down the line that you can talk, you can kind of allude to yet? So actually, we have an office in SF. So I just landed, I came here, and I'll be here for a couple of hours. Then I go to the office to see the new products.

1:16:07Very cool. I'm already sleeping on the next generation. There we go. Sleeping on the future. That's such an edge. That's such an edge. You're on pod five. I'm already on pod six. All my friends, they always now make jokes because I always sleep on the next generation and there are new devices. So they buy the latest, but I'm already one generation ahead. That's your edge. This is how you win. We started working with some new sensors that are incredible in terms of computer vision. So we really want to double down on body scanning and scan your body while you're asleep to save your life. And that to me is one of the most exciting things because if we can convert your bed in a health platform and save your life, that's pretty sick.

1:16:49Yeah, that's amazing. What are the companies that you're most excited about in the batch without picking too many favorites out of the 20? Anything that you're most excited about? At the end of the day, I just get excited with founders. I mean, I just met this guy, and he's a high school dropout. He's 18. And he's like, what? Yeah, because I was at school. I decided to drop out. I thought there was this opportunity for AI to help teachers. And so I dropped out. I built it, and now I'm selling it to my teachers. I said, what? That's amazing. So when you see these people, I don't care what happens with your company, but I admire you so much.

1:17:28Of course, because I want to back up the founder. Exactly. And so that is what excites me, to stay young and see these generations shaping the future. That's fantastic. Amazing. Well, thank you so much for stopping by. We will see you soon. Great job, guys. Thank you. Yeah, yeah. I love you, dude. It's been great. Let's bring in the next team. Welcome to the YC Demo Day 2025 stream. You good? I've heard so much of the confetti in the selfies. Yeah, yeah. It's okay. I'm working through it. How are you guys doing? Oh, you guys already got the hats on. We'll see if there's other surprises in store.

1:17:59We might need to blow the whistle. Welcome to the stream. How are you doing? I like how you guys just said made in SF, not made with love. Made with...

1:18:10Making shots. We have a made with love in San Francisco as well. Oh, you do? Okay. Well, what is Throxy? Break it down for us. Explain the business. We're building AI agents that help people selling into traditional industries, such as manufacturing, distribution. There are sectors in which selling is very hard, especially from a B2B perspective. So we're helping those companies prospect into legacy industries. And we manage the entire outreach to those. Whenever someone's interested in speaking with my clients, I connect them with their buyers to help them grow. So that's like selling super technical products.

1:18:49What does that look like? Yeah, it's selling super technical products, but also professional services such as consulting. The important thing is we help people selling into traditional spaces like manufacturing. Okay. So there's a manufacturing company out there. They're making microphones, for example. I am a company that's going to sell better software design software that runs on this or automates the facility. You're going to help me find clients to sell to. Is this more about the prospecting and developing a lead list, doing the actual outreach? Is this an AI business development representative SDR play?

1:19:30Are we going to see a billboard on the 101 for you pretty soon? What's going on? So it's a bit of like everything you've mentioned. So like we don't like to migrate ourselves as AI SDRs because they have like a bad reputation. They have a terrible reputation. And we think like focus is a multiplier on like the work we do. And that's why we're like serving these traditional industries, which are like traditionally like underserved. Yep. everyone's ignoring them, but they're like huge market opportunity and like really high average contract values. So we can like have people serving these industries, both with AI, but with a bit of human in the loop right now.

1:20:05And now we're automating those humans in the loop as models get better, as agents improve and like rolling them out. So talk about the co-pilot era. Where is it important for in the sales process to keep the human in the loop right now? what is the most automatable part of that process so the most automatable part is qualifying the whole companies so like actually like we just like pull a list of companies from like our crawlers and then we look for like specific stuff with AI agents and we just qualify them could they even be a buyer do they have budget are they big enough are they declining in sales are they about to get rolled up into some private equity thing the thing is like because we're focusing only in manufacturing companies we can check for very specific stuff for like does this machine any specific specifications?

1:20:52Do they have this grind type? Do they do CNC milling? That kind of stuff, which is really important where horizontal AISDRs fail and they fail to serve them because they're too generic. Because if I'm selling CNC software and some manufacturer doesn't use CNCs, why should I even talk to them? So you're saving me time that way. How did you guys get into this? So I think it's, I was in sales. I was doing a lot of manual work myself. I've been an SDR three companies completely different value propositions. But I think one of the big issues was actually finding the company I can sell to. I was doing all of that very manual research.

1:21:27Do they have bank power? Do they have this specific certification? When I was selling into hospitals, I had to check what their team staff looked like to see where they fit my software. And I hated my life. I was like... But you were still performing, right? You were putting up big numbers? Yeah, I actually was.

1:21:49But yeah, I generally felt like my job was going to be replaced soon. So I was like, let me replace myself. Talk to me about what it takes to actually qualify a lead. What are the data sources? I've heard LinkedIn is extremely rich, but it's very locked down. And they don't let you play with the API anymore. Obviously, you can crawl around on Google search. And some companies have a lot of information on their websites. Some of these manufacturers barely even have websites at all. How are you getting data? That's our insight. Our insight is LinkedIn doesn't serve these manufacturing people because the owner of a manufacturing company is not on LinkedIn.

1:22:25Their prospects are not there. So we've built our own crawlers, which are crawling the whole Google knowledge graph from scratch and then sending agents to qualify these people which have access to specific tools with directories of other manufacturing companies or API access to machine specification. access to vision so we can like check out the content of the website like take a screenshot oh interesting take a screenshot hey that's a cnc exactly this is a cnc company talk about traction when did you guys come into yc with this idea and you've just been working on it or did you iterate towards it yeah we came in with this idea uh in six months we've gone to 1.5 million in I knew.

1:23:05Let's go. That might be the biggest one we've heard yet. That might. Let's go. 1.5 million, baby. Let's go. Congratulations. That's sick. That's amazing. So the round's already closed, I imagine. Absolute dogs. Absolute dogs. That's great. That's fantastic. What's your guys' backstory? How'd you get to YC? We met in high school, and we've always been, like, building stuff since we were kids. Awesome. Awesome. Love it. Pablo was like, I hate sales, but I love money. I was like, let's go build a company together. I was working in AI at JP Morgan. It was like the most boring thing to do ever. Working in the back.

1:23:46Even for the money. So I was like, let's go do it. For the money. That's amazing. That's amazing. How big is the team? I mean, it sounds like you've already scaled this business a little bit. Yes. So we're a team of four right now. Okay. Focus is very important. so when you're asking where does humans perform better than AI that's what we're testing whenever AI does better we're automated but if not we have us actually doing that work we want to be very thick into the workflow understand the problem and at the end of the day it's my reputation down the line it's not the AISDR of Throxy it's me so we're showing face in front of our clients and ensuring that this works yeah that makes a ton of sense well good luck to you amazing guys Thanks for popping on.

1:24:31Cheers. We will talk to you soon. And we will continue our coverage of YC Demo Day 2025. We will bring in the next guest. I see some people out there. Come on down to the Palace of Party Rounds. Tell us about your company. How are you doing? Morphix in the studio. I see one hat. I think I've got to give out a second hat. How are you doing? Welcome to the stream. Good catch. Come on down. Good catch. Who are you? What do you do? We are Morphic. I'm Avi. And we build open source multimodal search for AI agents and applications. Okay. Open source multimodal search. What are, give me some examples of the multimodality.

1:25:11What are we searching? Because some of these data sets that are imagining if you're trying to search over YouTube videos, YouTube is going to shut you down if I'm trying to search across that. No, that's a great question. So for multimodality, it includes not only like just plain textual documents, but documents that might have pictures, photos, images embedded inside of them. What a lot of other people approach it as is trying to do OCR parsing on top, but that doesn't work because documents are scanned. Yeah, PDFs are a nightmare. The PDF spec is a disaster. We know this. It's a disaster. We've known this for decades.

1:25:41It's not getting any better. Adobe, clean it up. What we do is we embed pages directly as images and retrieve over those. That is much higher accuracy, much, much better performance. Yep, very cool. Are you guys going to be able to fix the photos app? Searching photos, very hard. We left with Apple who gives us a chance. Yeah, yeah, yeah. Give them a shot. Give them a shot, Tim. It does seem like they're trying to do that type of multimodal search across of it, but they're just not there yet in terms of like, I'll be searching for a dog jumping on a couch. I know that I have it in my camera roll, but it's a video.

1:26:13And that scene happened later, and they haven't indexed it properly. So lots of opportunities. Talk about where people are implementing this. Is it enterprise private data sets? Is it public scrape the web? I want to search everything. Narrow it down for me. It's a little bit of both. The most adoption is in the legal tech space. Oh, interesting. Also in the health tech space. They have like lots of documents with like tables, charts. Patent drafting, for example, has like a lot of technical diagrams in there. Yeah, of course. And it works really well for those people as well. Okay. Talk to me about the other major players in the space.

1:26:44We saw Glean yesterday raised, what,$150 million. It's$7.2 billion. This feels somewhat adjacent. and you're kind of eating off their plate a little bit. Is that a direct competitor or is there something different where you can play nicely with them? Yeah, so I mean, the kind of benefit for us is that we provide APIs and we're much more of a developer tool. Sure, yeah. And the way we see this going is actually not as a competitor to Glean, but like more of like a provider to Glean. Okay. Right, and not just Glean, but to people that are building like Cursor for X and you need to deal with multimodal documents.

1:27:15Sure. That's exactly what we can help you with. So we're starting with search, but that's not it, right? One of the things that Cursor, the reason why tools like Cursor and coding agents have become really good is because code is low entropy, which means you can predict well into the future what code is going to look like. One thing that we can do is help you get that low entropy with multimodal information. And so what video editing looks like, kind of like three or four steps in the future, if I cut a clip and cut another clip, I kind of know that I'm going to delete the thing in the middle.

1:27:45right and so uh if you can use that and uh essentially provide that to models as like code uh that ends up like performing a lot better and uh you can start building cursor for video editing and you can start building how did you guys meet what were you doing before yc we're both brothers uh oh no way yeah before yc i was a software engineer at mongodb oh cool and this guy so the open source thing tracks dropped out of where right exactly cornell nice very cool Yeah, talk to us about the open source strategy. How closely are you mimicking MongoDB? We heard earlier on the show that MongoDB didn't have a managed service, a SaaS product, until they almost went IPO.

1:28:25Are you planning to monetize the open source program earlier? What is the play between am I using the open source version or am I paying you look like? Yeah, good question. The way we see it is for people who, if it benefits a single community member, then we want to open source it. If it benefits a team, then we want to leave it closed source. Interesting. So things like SSO, things like connectors, like Google Drive, et cetera, or maybe like speeding it up for making queries much faster. This is all going to fall apart when people start building one person, billion dollar companies because you're going to be like, it all has to be open source because you only have one person.

1:29:02That's true. That would be a good problem. That's a good problem to have. But digging in more into the open source, what's the traction been like? Do you have a GitHub project that has a lot of stars? What are you tracking in terms of rollout? We're at 2 ,600 stars today. Congratulations. That's fantastic. 4 ,000 monthly downloads. Amazing. Thank you, Jordan. And 200 active deployments in production. There we go. There we go. The audience is probably mad at me in the chat. It's amazing. Yeah, it's fantastic. You guys came into YC with this specific idea, or did you iterate to it throughout? We came in with a much broader idea.

1:29:42We weren't too focused or indexed on multimodal before. We were like, hey, we want to be the data layer. What sort of helped us was trying to narrow down on the multimodal aspect, seeing how people are building more and just building on from there. Did you always want to start a company together? Yes. That's been a dream since we were kids. Incredible. Incredible. That's great. How's fundraising going? Fundraising is going well. We're kind of close, but yeah, looking to do it faster. Well, congratulations. It's been great chatting with you. Good luck on the next stage of your journey. Very exciting.

1:30:17We'll be following along. We're excited to use cursor for video. Whenever we ever build that, send them our way. Yeah, they'll probably build it on top of your company. So thank you so much for stopping by. Awesome, guys. Thank you so much for coming on. Bye. See you. Do we have anyone else? There's a little bit of lunch break going on, giving you some inside baseball here at YC Demo Day 2025. But we have one more team. They don't need to take lunch breaks. They're owning purple. They're working too hard. They're verticalizing purple. And the Pit Vipers have made a return. I think it's all one pair of Pit Vipers and keeps recycling.

1:30:52We know the Pit Viper founder. There's not enough color going on here. Let's put on some yellow hats as well. Let's just really get crazy with the fever dream that's going on with the orange, the purple, the yellow. We love this. Cliety, Cliety, Cliety. All right, what are you guys up to? What are you guys building? We're building automatic technical documentation for Codebase. Okay. So saving 30 % of employee time. Okay. Yeah. Explain the first customer. Are you going after larger enterprises? Are you selling to other YC companies? Yeah, ideally to enterprises. Okay. But we're starting off with Series A, B, onwards companies.

1:31:23Okay. Majorly because that's when you start onboarding people. You need internal documentation to actually give technical specification for your product, so on. So, yeah, initial focus is series A. Specifically for companies that are delivering APIs that they sell access to? Not just APIs, but any sort of, like, right now it's a web application, but any sort of application in theory. So, like, API is part of it. Yeah, there's a number of open source projects that kind of allow you to stand up a boilerplate for open source documentation or API documentation. How is your product different? What are you hydrating?

1:31:58Because at a certain point, if there's kind of internal sacred knowledge around how an endpoint works, you have to get that from the person that designed it? Or can you instantiate everything from the code? Yeah, that's the plan. Like the idea is for us to not depend on one person. The idea is like, we actually go across the organization, all the code bases, and understand those like technical specifications like in the background. So there's no dependency. How are you dealing with security? I imagine that if you're going across all the code bases, all of a sudden there's like secrets that could leak.

1:32:27There's internal tooling that maybe they don't want to have out there in their documentation. How do you think about that? So we make it completely self-hostable. So especially for an enterprise, where they have to be compliant, regulations and so on. So we completely make it self-hostable. People can actually just use our product, like it's packaged. So you can spin it up, put the repositories in, and nothing leaves their network. How did you guys get to YC? Like in terms of the place? Yeah, what's the backstory? That's right. You wanna go? Car, plane, what was it? Yeah, I mean, through plane, that was the last time.

1:32:58But I've known this guy, we've been through so many places. I've known him for about 10 years. We met in uni, hackathons, tons of projects. Very cool. And to get to YC, just a couple failed applications, and here we are now. There you go. There you go. Always a showdown. So many, everyone has one at least. How's the traction been? What are you sharing today? Yeah, so we launched about two and a half weeks ago. We already are around 3 ,000 in revenue. Congratulations. Let's go. There we go. We have about six, seven customers now. Oh, amazing. So we are, yeah, it's going well. We're increasing over 30 % week on week, actually.

1:33:32There we go. Week on week, week on week. Just do that for the next, like, 1 ,000 weeks. We get to go, hopefully, yeah. Then we need to raise again. Nice. How's fundraising going? You guys in the midst of it? Yeah. We're about 40 % there, 40, 50 % there. So we're getting close, hopefully. Yeah. Is it just you on the team right now? Do you have anyone else? Keep it that way for a while, or do you think you'll start really scaling up? We are thinking of getting people on board, but we want to keep it lean. We're trying to automate documentation. You've got to live it. That's great. Well, thank you so much for coming on the stream.

1:34:12Congratulations, guys. Thank you so much. Have an awesome day. We'll talk to you soon. Bye. And we are ready for our next guests coming into the Palace of Party Rounds. Welcome to the stream. The hum of YCDemoDay has died down as people move across the street to lunch. Good to meet you. How are you doing? What's happening? Nice to meet you. I'm John. Nice to meet you. Pleasure. Nice to meet you. How are you doing? What's happening? What's happening? Big day. Keeping up. Yeah, it's a big day for us. Yeah. How did the pitch go? We're going to be in the afternoon. Okay. Okay. How are the nerves? Honestly, we are used.

1:34:50Yeah, alumni, they do a good job of kind of like getting you so many reps that it just feels like anything else. Yeah, honestly. They work you up to it very easily. Introduce the company. What are you guys building? So I'm Francesco, the founder of Kua. Alessandro is the CEO. So we are building computer use AI agents. Okay. Meaning AI agents that can solve any problems like a human would do in terms of clicking, typing, scrolling. Okay. What layer of the AI stack are you working at? Are you sitting on top of just like a Chromium instance? Or are you actually sitting on top of something like a browser base?

1:35:25Or do you can keep people with them? We wrap an entire operating system on an isolated environment. Kind of like a Docker for user agents. And that means that we can use system level events for injecting these commands like click type. And really any Bash or PowerShell commands. Yeah, and then what's working in computer reuse right now? There was a theory that you just read the HTML at some point. Now there's more multimodal image generation, like actually take a screenshot, process that, understand where to click. What's working? Yeah, so screenshot is working way better than accessibility tree for operating system.

1:36:03In general, you don't have any HTML DOM to parse really. You don't parse any of that. You just throw it all out. We just use screenshot and pixel based models for that. And it's also been proven by research that's working way better than interacting with DOM. What categories of agents are you guys seeing the most, having the most excitement around traction? I think anytime you're building agentic infrastructure, you've got to have companies that are building great products on top of you. That can be a challenge, but I'm sure there's a lot of other companies in the batch. Yeah, so we haven't chased any verticals, meaning that...

1:36:37You haven't chased any verticals yet. Any verticals. Okay. Meaning that we really like the infrastructure. Just build a tool, figure out what people want. Yeah, exactly. Because, like, actually, we, first month in WAC, we were getting the most esoteric ask from the users in terms of, hey, I have this bunch of contractors. They're simply trimming videos on video editing software on macOS. Can I use Kua for that? Sure. And, like, really, we couldn't converge to, like, very common workflows, and that's why companies that are our customers are chasing those verticals for us. Okay. Okay. Do you see the market fragmenting?

1:37:08Do you think you will find a vertical and niche down, or do you want it all? Honestly, we want it all. Okay. And we are here to... Then what are the key deliverables that you have to optimize against? Is it speed, reliability, price, some sort of combination? How do you think about that? But probably for computer user agents, we are still like six months away from the Chagibity moment. Maybe for browser user agent, that's still the moment is today. So once we get that level of intelligence for models, we need to come prepared with a very good infrastructure to scale this isolated environment. Because our leap of faith is that five years from now, most of the AI agent, a multi-agent system, we rely on API, maybe 80 % of the scenarios, and the other 20 % are going to be based on browser and computer tools.

1:38:01What were you guys doing before this? I was working at Microsoft for over five years. Oh, cool. Awesome. Myself, I was in Notion, and I built also a few startups in the past. Nice. Are you guys Italian? We are Italian. Nice. When did you come to the US? Three months ago now. Three months ago. For YC. Yeah, for a study. Crazy. How's it been? Living up to expectations? Yeah, definitely. Definitely. People ask me... Ferrari or Lamborghini? Yeah. People ask me, how is San Francisco now? Is it better? And I don't have any comparison. How old were you when you knew you wanted to do YC? Was it...? I think it was pretty recently, like three years ago now.

1:38:42Okay. Yeah, for me, maybe since I was 16 and I'm 26 now. So it's been like a long dream for me. It's remarkable. I don't know, I don't know if it's reality or not. Yeah, living the dream. What's the go-to-market been like? So the go-to-market right now, we've been focusing mostly on startup and scale-up because we wanted to prove that we were on the right path and eventually fail also faster. But we have an open-source framework over 8K stars. 8K stars? Oh, you hit that. You buried the lead on us. Buried the lead. Please go and start out here. Yeah, yeah, yeah. Head over to GitHub right now. Give it another star.

1:39:23Yeah, try Kua. That's amazing. Slash Kua. Yeah. So yeah, I mean, what's the monetization strategy around that open source project? Yeah, so we, like first month in YC, we were all these customer inbound that were basically asking us, how do I even productionize on a computer user agent today? So we are providing a pathway for the user from the open source to bring in the same workflow that are working locally for them and scale them on cloud. So we're really charging only based on compute today. You simply have to input an API key on our platform, but what's going to happen also for us, we're going to become an LMA inference provider for these computer user models.

1:40:01Because maybe the public sense is that there are only two computer user models, maybe the one from OpenAI and Anthropic, and that's only because they have a better PR office than other models. but honestly like even on Hacking Face you'll find model from ByteDance UI Darts 1.5 that's also out beating OpenAI and Anthropic. On computer use benchmarks? On computer benchmarks like OS Word. Sure. And they are so hard to set up and also hard to discover and also we're gonna be the go-to catalog and platform where. Yeah how good are the evals right now are the benchmarks for computer use it feels more abstract than just you know do some math problems.

1:40:40Yeah So I was actually working with the Windows team when I was at Microsoft doing evals for computer user agents on a benchmark called Windows Agent Arena, which is derived from OS Word. Really like this benchmark, the tasks are not very meaningful. Like for instance, you will find tasks like, hey, can you go and open VLC and add subtitles? Who's using VLC anymore? So that's the question. Even like using LibreOffice instead of like Office or even Google Docs. Yeah. So what's going to happen, my leap of faith here is that the next generational benchmark will measure like real-world task. Yeah, do you think there'll be a sort of like LM Arena style benchmark where a human is watching two computer use models use computers and there's kind of like a vibe check almost?

1:41:27Yeah, or even like a wiki-race where you have like a - A wiki-race. Yeah, yeah, yeah. Even like a wiki-race? No. Okay, so WikiRace is like, you know, you start with Y Combinator and you have to end at Christopher Columbus. And how many clicks do you have to click to get from one to the other? So you might say YC has San Francisco, San Francisco is America, America is Columbus. And so you try and race through. And yeah, it's an intelligence test, but it's also, yeah, great computer use test. That's hilarious. Yeah, but also I imagine that there could be like a big model smell, like a vibe check on the computer use.

1:42:01because if it looks very jerky and it looks confused, that's something that might not even come across in a quantitative benchmark, but a qualitative human might evaluate it differently. Interesting. Yeah, and also there is this whole problem. Okay, I have this workflow that now is working maybe 80 % of the time. I want to make sure that I'm able to reproduce the same kind of workflow all over again. So we're also working on episodic memory that will let you basically use RPA 1.0, the old-fashioned RPA, new automation for workflows that are very deterministic. And then say that you have a deviation from one trajectory due to noise or changing on a web page.

1:42:40Then that's when you use like full computer use. But it really makes sense. That makes a ton of sense. Any other Italian YC founders? We haven't met many. Are you guys hometown heroes back in the Italian tech scene? You got to go do the local press in Italy. Yeah, do the press. Now you're in YC. Delian does this in Bulgaria. In Bulgaria, he's like a celebrity over there. Yeah, yeah, yeah. And over here, he's from the DC with a couple hundred K. You know, there is a WiseKey WhatsApp group now. Okay. They're aging like these week on the mountains. Oh, yeah, yeah. WiseKey, they call it. WiseKey. WiseKey.

1:43:15That's reason enough alone to apply to my company. Yeah, yeah. Mostly in group chat. Yeah, yeah. Anyway, thank you so much for having us. Congratulations, guys. Good to be here. Thanks for coming on, guys. Thanks so much. Let's bring on the next. Do we have a, who do we have next? Oh, Delian. We were just talking about you. What's up? How are you doing? Oh, you got the hat on already. I got the hat. Oh, shit. Oh, my gosh. Let's go, baby. Shot's fired. Let's go. Shot's fired. I can't believe you. What the, we've been doing that when we're here. It's like big numbers, big ARs. We didn't bring the gong.

1:43:47Should I, so I was a Y Combinator, summer 14 company. Yeah, that's right. Do you want me to go through my summer 14? Please, give us the pitch. So hi everyone, my name is Delian Asparohav. I'm the CEO of Nightingale. We build software that helps autism therapists that work with young children to basically help them capture data on those children's behaviors, pull that into reports that you then send off to your insurance companies to get reimbursed. Today all this is done by paper and pencil, super manual, takes these therapists like 30 % of their workday just doing a bunch of bureaucratic actions.

1:44:19We cut that down, make it super fast, they get reimbursed more quickly, and they get to spend more time with the kiddos rather than on a bunch of paperwork. It's a cursor for autism. I was about to say cursor for autism. It's a cursor for autism. So if you ever wonder where all the autism came from, it's because my first company, literally all I did was spend time with autistic kids. Yeah, that's fantastic. And so if I had some in me already, it got amplified. How did your demo day go? Did you raise? Was it hard to raise back then? It feels like everyone here can put together a million-dollar seed round pretty easily.

1:44:44What was the comp that you used? A lot of YC companies obviously like to give some type of like, you know, and a roll for dog walking. Yeah, and like summer 14 like you have to remember like at the time It's not like there was a ton of like healthcare sass things that it worked It's like you had like epic is like an EMR They're obviously done Epic the story fully. Yeah, and like we weren't really any MR. So honestly, I think like you know I remember like Sam at the time because he was like the head of YC back then, you know Two weeks before demo day basically told me like your pitch is fucking trash And then I went and you know, we're gonna say this on live stream, but um Yeah, I took some psychedelics on a weekend and really thought about my pitch for that weekend.

1:45:21Made a lot of improvements. And then Sam afterwards told me, he was like, relative to where your company's at and performance, phenomenal presentation. But phenomenal presentation in 2014 turned into like 600K seed round. That was a fucking grind. It was like still two months after demo day, like doing individual calls. And it's just like, when you look at the amount of capital that's like, whatever, the next door building. There was like a two, somebody said there was like a two hour wait this morning. If you tried to show up like right at the start, There was like the line was like two blocks down summer 14 like we like filled up like a small little like area Like cocktail area and like the computer history museum down there So it's like I just if you just given 10 grand to like every batch member with you, you would have a billion dollars Yeah, I think summer 14 did have some hits and I have to go back and remember Summer 2012 my batch has a ton coinbase Instacart that was a good bear.

1:46:11Yeah, it was a good batch A bunch of good ones. But yeah, it's interesting to study just like, I feel you can use Demo Day. This is my first time coming to Demo Day in like I think seven years or something like that. Partially like in 2018, I went, spent a bunch of time on it. And they're like, honestly, it just like didn't lead to any investments. And so I was like, at some point, what am I doing here? Then obviously COVID really killed it off. And so I'm excited to be back, but it's also an interesting just like marker of like the industry's sort of maturity of just like the amount of companies, what the companies are working on.

1:46:38Now also the amount of comps that like you can like look at. Like in 2014, if you had said like my comp is like, I would like to be a hundred billion dollar publicly traded company. It's like, okay, there's like basically two of those in the entire history of technology. Now you have to come because people are doing Varda for X. Last fact there was an Indian Varda. So, you know, I got to track those guys down and figure out how to like acquire them. So we have like a Desi Varda. That's great. Have you seen any of the hard tech companies yet? Any of the pitches? Gary said like 11 % of the batch is hard tech.

1:47:07Yeah, yeah. It's interesting to see like, I mean, I remember in like 2018, 19, when I was doing like hard tech, industrials, defense, aerospace, it was just like so uninteresting to so many people. Like they would all like literally like, you know, we talked about this like ever at my former colleague, he's just like, oh, like if anything is like, you know, negative gross margins and like highly capex intensive, send a deli. And he like said that as a joke. And then everybody's realizing like actually capex intensivity is like the best mode ever. Cause like if you just build software, AI slop can, you know, replicate it basically overnight.

1:47:34So yeah, I definitely looked at it, you know, shameful of them. I think it's cool to see that like YC is leaning into this. Cause it's not something that they've sort of. They had a couple hard runs with like Pebble was a really big like hard go because like they got so Sherlocked by Apple with the Apple Watch. Totally. Yeah, what are you going to do? But at the same time, like there's been now like Astronus and Oklo and like a few really even harder tech companies that have made it through and like scaled. Astronus I do think is like the best. Like Oklo obviously like, you know, sort of trading, but like hard to figure out.

1:48:04Yeah, hard to figure out, but still. But Astronus feels like it's like, it's got some, you know, sort of real meat. Yeah, BioWorks is another one. Yeah, it's like 300 feet down. I admit though, like, look, I think like, I mean, if you look at, you know, the like YC deep tech, you know, sort of portfolio and like hit rate and outcomes relative to like the FF or Dellian seed deep tech industrial portfolio. There's definitely one that I would want to buy in the basket of and one that I wouldn't. And so, you know, I mean, maybe I shouldn't be speaking. Maybe Gary's going to kick me out. I think like the bull case is that like, like it is great to have exactly what YC is, which is an incubator, an accelerator, like a pool for that type of talent to come out of.

1:48:41And if you wind up picking over at it at some point, like that's fine. And that actually part of the ecosystem. And yes, like you're going to get earlier stuff, but it's nice that there's at least an ecosystem. Some of the people that go through YC with hard tech, they might become employees at these companies. They might become acquisition or acquirters. It opens the Overton window for like the average state of grads and like not just work on AI. How many multi-stage VCs have pulled back from doing demo day investing at all and just saying, you know, we're not going to try to go? I mean, I feel like in like, you know, sort of 2019, it was like the consensus thing.

1:49:12Like at some point, like everybody was like, why are we doing this? It's just like way too many companies. They're like, you know, sort of seed rounds, you know, feel like, and again, you know, maybe Gary's going to kick me out for this. But it's like the seed rounds felt like they were really overpriced. And especially as a multi-stage firm, it's like, well, you can just wait for the Series A and like on a risk and reward basis. it's not like the companies are going to die if oh yeah there's so many there's plenty of players specific funds that would come in and just write 100k check into tons of companies so like everyone was getting their rounds done but yeah i mean yeah it just speaks to like how they're playing also there's this weird dynamic where yc used to say like i don't know if you got this advice but it was like don't pitch any investors until yc demo day and then everyone was like oh well like the game theory is like if i'm the only one pitching before demo day i should do it all the interest and so and so like there's still like the i'm sure like you will you will take a pitch with a yc demo day founder a couple weeks earlier because they're inventive and creative and they got to you before totally you just might not find them through this i do think there's some amount of like a lot of the like you know companies have already closed their rounds by like the time this day happens which was very much so not the case in 2014 2014 was like oh maybe one company the entire batch yeah and by the way for what it's worth like if you study those like early batches there was basically like an extreme negative correlation between like the ones that like you know sort of raised early and it actually ended up being like the like you know sort of batch returning you know sort of outcomes i remember there was one company in yc summer 12 that had a uh super viral video because they were in the viral video making business and so they got a ton of attention and they'd also like kind of ramped revenue by like saying like yeah well we'll produce a video for a hundred thousand dollars like that's not really like like durable scalable revenue before clearly the company that made the viral 60 million dollar round at before demo day and everyone was like what's going on did you see that did you see that ad it was company called co-tool they did like a skit based on the deal rippling thing no way you gotta see this video it's hilarious but they're in this batch yeah yeah they they have a they have a service that will monitor slacker chats for honeypots and create honeypots for you and do all this different cyber security stuff it's great i mean i think back and like i just had this memory from 2017 when I was at Coastal Ventures where like, you know, to basically have like the junior team go through and like crawl through the entire batch of like everybody, like the week beforehand, basically.

1:51:24On the Sunday night before like the like, you know, she had a Tuesday demo day. We would invite like 15 companies to present at KV on Sunday from like noon until like literally midnight. We would be there. Like I remember like letting founders in at 1130 at night to like come pitch the firm. And it actually felt like we had this like really deep R. It was alpha. Nobody else was doing it, et cetera. And then literally it's like like you said there now there's like these infinite you know sort of YC funds Where like I actually just feel like the alpha there on like doing the pre-work, etc Getting into it.

1:51:50It's not to say there might not be some good outcomes But like many many more people run that strategy And then even the fact that like something like this exists like the idea in 2014 that there'd be like a live show that Anybody would go fly about it actually want to tune into to talk to YC Like we could barely get anybody like you know Pay attention to any of us let alone the idea there's like whatever gonna be like 15 ,000 people on Twitter So yeah, I mean this entire industry just gotten like, you know, I mean like this is we are like the new Wall Street It's like kids grow up, you know, in like 2008 being like I want to be like an investment banker Yeah, the kid right here just said I knew I wanted to join YC when I was 16 Yeah, 26 now.

1:52:21So it's like decades ago I mean we talked about this in relation to like Teal Fellowship recently where it's like in 2012 They're like off-track thing to do was like drop out build a company join an accelerator work in technology Now this is the track. Yeah, yeah, we like I need to practically There's a high school dropout in this batch. Yeah Yeah, yeah. I mean, at this point, like, I mean, there's an article in Business Insider today by Julia Hornstein, who wrote the article about, the first article about us, talking about how, like, just not even going to college at all is the new dropping out.

1:52:51It's, like, just the default. It's just the default. Versus, like, in my year at MIT, I remember there was, like, three of us that dropped out. No, I think it was risky. He has applied to college, get accepted, and then don't grow. Be like, MIT acceptee, but that's all he needs. I had a whole riff on this. It's like, yeah, Delian, you dropped out of MIT. I knew that I would drop out if I went there, so I didn't even apply. Exactly. I did go to a different college and graduate. But it's just like, yeah, yeah, you're a sucker for even having gone for a day. I didn't even go to that. Exactly. You paid the application.

1:53:18What a sucker. Paying a full year of tuition? God damn. Boomer. Want those$45 ,000 back? I could have put that in fucking NVIDIA and I'd be a billionaire. Boomer mode. Anyway, thank you so much for stopping by. Good to see you, boys. Thanks for almost laughing in the face. We got some big news, a new investment from you coming on. Oh, yeah. Looking forward to that. See you tomorrow. We'll see you back-to-back days. Welcome to the demo day stream 2025, good to meet you. How are you doing? I'm John. Nice to meet you, Dave. Dave Munichello. Did you photograph us out there or was that somebody else?

1:53:49No, I didn't photograph you. Someone else at GV, Han. Okay. Han's taken a photo picture. Oh yeah, we love the paparazzi. It was good. It was good. Can you introduce yourself for the stream? Yeah, so Dave Munichello, Google Ventures, my managing partner there. Fantastic. I've been there about 13 years, investing in AI and enterprise software from the early days. Lots of good stuff to do here. I was just standing in a room downstairs hanging with a bunch of folks, and Gary came in and said, hey, you got to go do this. Amazing. Thanks for coming on. I'm here. Excited to hang. Yeah. What trends are you following?

1:54:25What are you seeing that you like? What's interesting? Well, first, break YC into chapters for us from your point of view. I'm sure you've invested in companies at this point, maybe not on right around demo day, but in every batch. I was in Cambridge in the early days in grad school when like Paul Graham and those folks were in Cambridge doing their thing. So at that point in time, YC was like the cool place to be connected to. And I think, you know, a lot of us, I met a partner of mine that now runs our life sciences team and I co-lead our digital team here at GV. we met in Cambridge and we used to like get really excited about any Cambridge YC events.

1:55:01That's amazing. So the place has totally evolved. Oh, totally. It's a machine. Multiple chapters. Totally. Was fortunate enough to do GitLab in the early days. So like new Dalton from day one. And like Sid has always shared his enthusiasm. Patrick Collison is a portfolio founder, comes in and talks to the batch. And then we hear the cool stuff from him. What was the perception around GitLab at the time? I feel like I remember it and it was kind of like discounted because it was open source. GitHub was already such a success. It was like, how are they going to do anything when there's already this winner?

1:55:34I literally just stepped out before this, stepped out to do the earnings call with the CEO and the CFO to like do the investor callback because we're public market shareholders of the company. Oh, wow. So we invested just after the Series B. We actually passed on the Series B. And we had a bunch of concerns. And if I look at those concerns, they were - Hold the mic a little bit closer or just angle it. They were one of the first companies to be remote only. Oh, yeah, that was the thing. It was like WordPress and GitLab. Yeah. We're like, this remote only thing seems strange. How the hell are they going to run a company like this?

1:56:05Yeah, yeah, yeah. Sid is extremely technical. Sure. And I think remote only works really well for him because of the way that he shows up as a human. But he writes everything down. And so that works for a subset of people in society. So he wrote down GV, passed, but then he's still in. I mean, we talked constantly. Like passing out a company is just like, it's just like this stage isn't right for us, but we want to continue to talk to you over time, right? What's your approach to demo day now? I'm assuming you don't have a lot of FOMO. You guys write bigger checks. We've met something like 30 companies so far.

1:56:35You are writing checks even at this early stage. We write checks into YC companies. We love YC companies. Yeah. So even though we're multi-stage, like$100 million into a public company, which we're just talking about, we also do seed and everything in between. A is our sweet spot, A's and B's. Great. We do an event a few weeks before demo day every year with all the YC founders. And then we know all the group partners, right? So they constantly ping us with ideas. That's great. That's great. What is the integration with Google like now? We're entirely separate. Entirely separate. Yeah. So we raise capital from Google.

1:57:08They're our sole LP. It's an LP relationship. Yeah, because it's such an interesting dynamic. Because on the one hand, that could be incredible value at it. It's like, hey, we'll introduce you. You can sell into Google. The incredible thing is we have a stable LP that's there forever. Permanent capital, essentially. With plenty of capital and an appetite for risk. So we have a great LP that we have a half an hour conversation with every couple years and plenty of capital. We don't spend any time pontificating about where we think the market is. It's why I had to call our comms people before I came on here.

1:57:36It's like, we don't usually do marketing or comms events or podcasts or stuff like this. Excited to talk to you guys. Of course. But yeah, I think we're unique in that we spend all of our time with founders. Yeah. So I shared the like, I was in Cambridge. I mean, it's such a hack. I mean, I think you're the envy of every, no matter how successful our GP friends and the GPs on the show, having a dynamic where 100 people could call you on a Saturday at 5 p.m. and you kind of owe them your time to some degree. And that's like that LP, you know, dynamic. I mean, I talked to obviously tons of friends at the GP level across different firms.

1:58:12It does keep them sharp. Like they get a lot of pressure from LPs, constant questions, deal flow, connectivity. I think that's really positive. but we don't have to deal with a lot of that stuff and so we spend time connecting with founders and other GPs. What do you think the nature of those questions usually are? Is it just like trying to understand markets and trends and how the fund is positioned? Yeah, I haven't really sat on that side. I've been to like one or two LP days for big funds. So what's this Transformers thing? I didn't want to say it but I think that's kind of the vibe. I've seen the movie but give me a look.

1:58:44I mentioned this idea of like we invested in you know, get live in the early days. And then now they're a public company. Hedge funds invest in the stock. Sure. Call us to ask us for our perspective. And they're thinking about like what happens in two weeks. And I imagine LPs are like slightly farther out. It's like 90 days. You know, what's what's happening today? What's hot today? Are you in the right deals? That sort of thing. Our LP doesn't care about any of that stuff. We don't have that conversation about like what what we're in, what we're not in. It's like do you have a generational founders?

1:59:14Do you have a particular understanding of how the startup landscape is interfacing with the Mag7, the big tech companies right now? We're in an interesting time. The big tech companies, they all kind of went from$100 billion in market cap to$1 trillion very easily. And it was kind of like the easiest 10x of their entire career in kind of some sort of unexpected way. You would think that would be the hardest one, but a lot of them just did it. At the same time, it feels like there's more opportunity for startups than ever. but the big companies have more resources than ever. What are you talking to founders about?

1:59:47Specifically around AI, we're wondering constantly, is AI going to make incumbents stronger? Yes, it seems like a sustaining innovation. Totally. And yet, I'm talking to companies that are doing millions of dollars in ARR over two days. That's right. And I think the question for us is, does each incumbent, each big SaaS company, do they bolt on AI? Sure. Do they acquire AI? Yeah, we're seeing this with the windscreen. Your thing does every big tech company need an investment. It's a new hot structure. Yeah, it's the new M &A. I'll take 49 and all your best people. I could do this valuation or double it, and you get half.

2:00:24It's like the same number. But I think having loads of capital right now in a time when the seats are shifting in tech is quite interesting. The MAG-7 might be the MAG-70 in the next 10 years, right? You might have a difference. Add a zero to it. I would like more big tech companies. We're on the side of big tech, and so we want there to be as much big tech as possible. I mean, this is actually little tech. The 70s would be little tech growing into the 80s. No, but we're fans of all of it. All of it. Just tech. High tech. We're the fan of tech. Yeah, yeah. Totally. Yeah, I think this idea that people want to say AI is good for big tech or it's an extending innovation, but it can be good for both, right?

2:01:02It can be so transformative. The internet was good for some big companies that adapted well to it. It's always good for a bunch of completely novel ideas. And so we don't have to pick one and pick a side. It can be an extending innovation, and it can also enable all of this. I think we're in this exciting place where the most high-agency humans in our lives that we're all connected to, they're empowered not just to have the underlying cloud be present for them, but intelligence be present for them. And so now they're building. I met a company out there that was one person, no engineers, already have tons of revenue.

2:01:36I think he told me it's going to be the first billion dollar company. Oh, he's going for it. He's going for it. I always get hung up on that because if you attach yourself - It does feel like a vanity metric. It's a total, I mean, it'll be amazing when it happens, but a total vanity metric that can guide you towards bad decision making. It feels like it might happen accidentally. Yeah, it happens accidentally. It shouldn't be something that's maybe delivered. Or the way to do it is just get to a billion and then lay off every other person but yourself. Yeah, the private equity guys are really going to be the ones that do it.

2:02:07They're going to be like, yeah. That's not how we go with companies. We bought a 10 ,000-person company, and we fired everyone except for one person. There are a few VC companies right now looking at, VC firms that are looking at the PE roll-up. Sure, yeah, yeah. Buy something that has massive distribution. Have you looked at any of stuff like that? We have looked at them. We have passed on a lot of them. Passed on a lot of them? I think it feels like a PE play where you're trying to get the multiple to change as a result of the industry that you're going out there. It's all a question of how big the impact is, right?

2:02:30A lot of the founders that are running that strategy, they just should change the structure and just say, like, we're going to run a PE playbook here and we're going to be two in 20. We're going to be two in 20. Yeah. Because if you can raise 20 million on 100 for the strategy and keep like a lot of the economics, sure, it's great for the team. But it should probably be like the basically the comp incentive structure should be look more like private equity. Totally fair. Totally fair. Yeah. You guys are friends with a buddy of mine, Sean McGuire. Oh, yeah. He's been on here a couple of times. Of course.

2:02:59Sean introduced me to a founder on Friday night at like nine o 'clock over text. And I responded back. And the guy's like, do you want to meet tonight? Nice. Do you want to meet tomorrow morning? Like, I'll come to you. I live like an hour away. I'll drive to you. I'm like, that kind of high agency humans can be empowered by, you know, an entire set of tools, a whole software suite. You know, AI will be the thing that powers all of these incredible humans that we're sort of meeting today. For sure. Yeah. The thing I hope we get some more of these companies on. We've had a lot of developer tool teams on.

2:03:32I think this idea of business automation is being heavily explored around agents. But we were talking with Dylan Patel last week, and he says the next category besides consumer tech with LLMs, like ChatGPT, and you have CodeGen as big revenue categories, and this third category of just actual business automation. So how do you just make the machine work? and everybody's thinking about it in the concept of like agents, which is like kind of, I think, like almost a simplistic way to view these things and like what is the next iteration of that where it's like actually like an autonomous system that doesn't have to interact like a regular employee.

2:04:11It's sort of, so we'll see. Yeah, I mean, we're seeing it across an organization like a Sierra or a Decagon. You know, Harvey is a legal AI company that we're invested in. There's some medical AI company. So like every vertical has AI, but it looks a lot like a human. It's like replacement of a human. It's replacement of a human, but how do you replace 20 humans at once? Cross-functional? If you do you worry, that some of the hotter companies at Demo Day don't end up always? I mean, we're investing in super hot companies at the A and the B, and the most sought-after companies get marked up incredibly.

2:04:58So it's a sign of the heat is a sign of enthusiasm from great people. Talent, opportunity. Totally. And so if you're choosing to get exposed to that asset class, you have to be willing to pay the prices to enter that asset class and play in that asset class. hear the challenges that you have like three months of data. Yeah. And so you're talking to a company. Yeah, it's like the company that grows revenue the fastest in two weeks, three months, whenever they actually launch, is not necessarily the company that's going to grow revenue the fastest over two years, three years. We've seen that a bunch of times.

2:05:29We reinvest in a YC company, and then after demo day, it kind of chills out a little bit, and then it re-accelerates. Yeah, that's good. There's a founder in this batch. Yeah, totally. Founder in this batch that raised, I think, like almost$8 million of uncapped convertible notes. And so it's just like humans. Dog, an absolute dog. He's an incredible founder, incredible team. And I understand why people are sort of leaning in to work with him. But that makes it pretty difficult. Yeah, yeah. You're thinking about that next call in a year or two with Google trying to explain entering a party round on an uncapped note.

2:06:06It's easy to justify if it works out. I mean, RLP literally has no visibility. Yeah, sure. That's good. That's good. So maybe it's easier for you. But for us, the thing that we care about is we want to be your lifelong partner. Like the thing that we think about for founders is how can we be there for them over every round and into the future. What's your messaging around signal risk? Because I think there's this like if you're truly a multistage fund, you could do an early round. Yes. And you could say, you know, we're good with our ownership right now. And like we're going to help you raise this round.

2:06:35but we actually, I think this concept of signaling risk is totally possible for you to want to invest in a company today and three years from now. Just because you don't do the in-between rounds or specific rounds. This is the exact conversation I was having when Gary walked up and said, hey, should we talk to these guys? So I was talking to Neeraj from General Catalyst. Oh, yeah. And we were chatting about, do you play when the price gets to an uncomfortable place? What signal does that send? You're essentially sending a signal to the market that your firm is excited about this company. They have a separate seed program.

2:07:05so they can kind of categorize this as like this is a seed bet we're gonna put a million dollars in this company We'll see how it goes. We don't have a separate seed program when we do an early stage bet We we use similar criteria to what we use in the a obviously They don't yet have a metrics, but the question is like if they had a metrics would we lead the a in this company? So there is a big signal when we do it So it makes it harder for us to do like 30 or 40 companies. Yeah, we talked to 30 or 40. We might do a handful Yeah, it makes sense. Very cool. Well, thank you so much Thank you guys.

2:07:33You're welcome. Fantastic. Great to have you on. Thank you. Take care. Good to meet. And we are ready for our next team or individual or investor or yapper. Who will it be? Who will it be? We will be surprised because we are at YC Demo Day 2025. Welcome to the stream. How are you doing? Good to see you. You already have the hat. Hi. Hi. How are you doing? You're looking around for the camera. You're on live. You're on a live stream. You'll need to introduce yourself. You can wear this hat. Okay. You can wear your hat. You don't need to wear that. Whatever. Yeah. Don't wear that. I'm John. John, what's up?

2:08:04Andrew, Jordy, how are you guys? Introduce yourself. Hi, my name's Andrew Lee. I'm a partner at Andreessen Horowitz. Fantastic. Both working in the Games Fund and also at A16Z Speedrun. Very cool. Are there any games companies here today? There are no games companies here. Which is somewhat sad. You suck them all into Speedrun. Is that your guys' fault? It's your fault. It's your fault. There's about four or five of them. You took the game. I would say for YC, there's about four or five consumers and companies, which is pretty good. So you can look at those. For sure. For sure. Four or five in the whole batch.

2:08:33There's like five or six-ish. Stealth. But here's the good thing. I mean, there were a couple companies who previously in the past were like B2B companies. They're like, you know what? This is really boring. Yeah. I don't want to do something fun. And a number of them ended up in gaming, and they ended up in entertainment. So that was exciting. And then also, I just generally think that if you're going to build something that your mom or your 17-year-old degen friend is going to go ahead and play, then yeah, it makes sense for them to go ahead and build B2B stuff as well. Yeah. I want to talk about the Games Fund.

2:09:03I've been super interested in how hard it is to get into the hot games because they kind of blow up out of nowhere. Yeah, you're like getting on a plane to Sweden. It's more like investing in movies or something. By the time you get there, they've doubled ARR. Yeah, there was this Bellatro game that was a massive thing. There was Among Us. Was that the one? Yeah, Among Us. Yeah, I mean, that was probably one of the top ones ever played during COVID. During COVID, these kind of like flash-in-the-pan games that become very viral. but I don't understand if they're like necessarily good businesses.

2:09:33So like, how are you thinking about finding great games or games related companies or even just consumer companies? Like, what are you looking for? Because the signal seems so much more noisy than in, if there's noise around a, like an enterprise dev tool company, like they're probably selling that and it's probably going to be pretty sticky. Whereas like Bellatro, I don't know if it seems like local funk, I think his name is fantastic developer, but who knows if that's like a, you're going to turn into like Activision. I think fundamentally for us, the way that we tend to think about it is there's a couple different things to care about.

2:10:05One is that if there's a large audience that has a bunch of folks that are going to play, if you do one innovation in that audience, that makes sense. For example, big fans of League of Legends, if you're all out there. If you're trying to create a MOBA, that sort of makes sense. But I think that for us, honestly, the big wave we see is in the world of AI. Whether it's changes in 3D animation. We fundamentally think that the future of entertainment is actually going to be, there's going to be an AI Pixar. Mark my words, someone's going to make an AI Pixar. It may not actually even be Pixar, right?

2:10:31Because usually what happens is we tend to think that, like, an incumbent, for example, like, you know, Disney or something like that. Actual Pixar. Exactly. It's going to make, like, the next thing or potentially take the next wave. But I don't know if you guys see any of these VO videos. Oh, incredible. I mean, the Bible Bros, whatever, like Stormtroopers vlogging all over the place. I'm like, it's astounding. And I don't think that's something that could ever occur. There's, like, a sort of innovator's dilemma that's occurring with the existing folks. even just from like a PR pressure like I don't know if you saw that show the studio, there's this whole sequence where they're very worried about the casting for this new Kool-Aid movie and then it is revealed that no one cares about the casting all they care is that they used AI a little bit and it's this really hot button issue there's massive PR backlash and so that's this counter positioning where if you come to the market and you say like yeah we're an AI Pixar we're just making AI stuff you don't have the expectation, people expect Pixar not to use that.

2:11:23That's right. How are you thinking about investing at the game layer versus kind of the infrastructure layer? It feels like there's a bunch of new tooling around just creating these generative worlds that can turn into games. And is the Roblox of AI Roblox, or is it something new? Yeah, it's interesting because I think the thing is what we've seen is, well, we go where the founders go. And it seems the number one area the founders are going is into more tooling and primarily using AI. I mean, there was a bunch of stuff. There's still a ton of, I think, experiments happening in the Web3 layer, potentially in the content layer.

2:11:53But the main problem is, honestly, is that distribution is tough. I don't know if you guys have invested in any consumer companies lately. It's hard. It's hard. It's hard because there's a reason why every DDC company has basically had all their margin eaten by all the big tech companies. And fundamentally, I think that's the same problem. And Steam, which is usually the platform for folks who want to build games, is not nearly as liquid as it possibly could be. So as a result, I think the thing is, the one thing that I think we've seen that's a very positive out there. Liquid as in there's not enough demand coming from Steam itself.

2:12:23Or really that like your ability to get more bang for buck, right? Like that it's fast enough that you don't have to pay a huge amount of marketing spend. It's gotten a lot tougher and you have to be very good at it. So there's great teams that are able to do that. There's a lot of gaming going on on there too with like wishlisting, driving early sales. And so you're paying people to wishlist your product. And it's like there's kind of schemes on top of schemes. Right, right. And the thing is like, I think we could do it. I mean, like there's definitely people who do that but it just seems that if you create a 10x product that is able to grab some great cut consumers out there that seems to be something that has natural growth i mean i'm sure we're all familiar with mid-journey um that was built it's built on discord yeah yeah right still and built discord it's not as if it was something that naturally would have grown so the thing is i think the the and obviously i'm talking about video because that's like my latest thing that i'm just doom scrolling on every single night but um when you can basically grow from zero to all and 120 ,000 followers with four videos over the course of like two days.

2:13:21That's amazing. That's an astounding thing, right? And if someone is able to do that, that shows me that there's like a natural consumer poll. And that's the thing that's pretty interesting. So mark my words, I still think that there'll be an AI Pixar. I also think that fundamentally, the world of AI, tech, and entertainment are gonna converge. And we're gonna basically see somebody create that 10X product that then hopefully just basically skips all the distribution problems. Sure. But I don't know who am I to know. Are you seeing enough? Are you seeing enough weird stuff? There's this concept of trying to leverage AI in this existing paradigm, but when you think about the intersection of tech and AI and entertainment, I can just imagine entirely new ways of playing games.

2:14:06You already have seen this on some of these people basically creating games within games. but yeah, are people being, are people being like, is the average pitch, you know, really trying to rethink things from, from the ground up or is it, Hey, we have a, we think we can build the next candy crush. Yeah. I think a lot of, well, for us as a fund, we'd want to obviously have people who are going to be willing to build the next unicorn company, but that doesn't mean that I think great developers out there won't create amazing experiences. I definitely agree with you that you have to basically aim for the weird, like you have to do things that are basically in the frontier of what's possible because otherwise then and what you're doing is you're most likely just basically doing a me too, which is then hard because in a distribution channel, we have to basically pay for your customer acquisition that's super hard.

2:14:49I think something that's also pretty interesting that I was telling some folks about was that the one thing that's interesting about both consumer experiences and games is that if there's a limitation, take for example on, let's say that you can only network 20 people on a server together, right? Well, the way you do that then is you just make it so your game has 20 people playing all at the same time, and there's something called Fortnite, and then 20 people have to kill each other, and then it eventually ends up with only 20 that you'll be able to go network with each other. So the good thing is you can just basically change.

2:15:16It's not like a bug. It's actually a feature, and that's kind of interesting. But I think also generally what we're seeing in AI, though, is when it's, you guys are probably familiar with this, which is the sort of cycle between B2B to the app side, which is basically you have B2B infrastructure that allows you, for example, to create really powerful video models. That then allows you to have apps that use those video models and then create content that create new network models, for example. And then that leads into other things happening with the infralayer. We're seeing a lot of that happening.

2:15:44Unreal Engine was a B2B play. Exactly. Birthed Fortnite. And then there's another company in the Andreessen portfolio that does kind of like world-scale sharding so that you can basically build a big MMO that you can walk through. I forget what this company is, but they were doing some scientific computing and some economic analysis. It was years ago. But there's obviously a lot of work done on the infrastructure layer. and are they going to be the one to build the next great consumer game? Maybe, maybe not. Maybe it's built on top of their platform. What are you seeing in VR? Oh, yeah. In VR, let's see here.

2:16:17So there is one great company that came through before from Speedrun in the past, and around the first one was a company called Trask Games. Okay. So it's one of those things where basically I think just like people grew up in the mobile generation, you need to have people who grew up in the VR generation. Sure. So if you talk to anybody who's under the age of like 16, most of them will be like, I play a lot of VR. Really? And it's crazy. Like they all go home. They just play VR a lot. Yeah. And they have to obviously have the device installed. Yeah, of course. But for them, they spend a lot of time there.

2:16:45So we backed this team that was like a bunch of teenagers who like, they won a bunch of Apple awards. And then they're just like, look, we just spent all our time in VR. So they built a game that basically is taking one of the bigger ones, attacking one of the bigger ones, which is Gorilla Tag. They made a game that is astounding. Their most recent game, actually, it's called Yeeps. Yeeps actually has, you know, I can't describe it because I'm too old. But basically, it's like, you know, we're all in a world together, and we get to go and do stuff together. But then we all collect Yeeps. We yeep at each other.

2:17:19We're yeeping. Yeah, we're yeeping. It's all the Yeeps. All the shout-outs, all the Yeeps out there. We're yeeping maxing. But they're doing great because the thing is, I think, ultimately, it's not just a hangout spot for a lot of these kids. But it is ultimately taking advantage of, you know, what VR and the general install base. I still think that VR still needs a shot to the arm in terms of, we all know this, that it probably needs more installs. It probably needs a fine way there. It needs lower churn relative to the device sales. That's right. Or the device sales have to be way less expensive.

2:17:50Or find a way to just get as good of sales as the meta Ray-Ban glasses potentially. We need VR games that are truly addicting, not just novel. Yeah. I was really hoping for the first... You probably shouldn't. I mean, I remember, I remember. I like EVEs though. And the kids love it too. With the original like PlayStation, like Metal Gear Solid, that game was like a hundred hours or like Final Fantasy VII. That was a hundred hour experience. GTA IV. Yeah. That was like a hundred hour experience. And I've played a lot of VR games, have yet to find one that it's like, okay, the progression in this game is so addicting that I need to keep putting it on to play.

2:18:26It's like, it's like more like, okay, it's a cool demo. I set a high score. Okay, I'm done. I'm not like I need to finish this. I need to know where the story goes. And it's hard because it's a very expensive investment. I mean, I think that's why it's easier, honestly, if you're gonna innovate in the world of entertainment, to innovate on that B2B layer, right? Which is where we're seeing it. We're seeing a bunch of AI sort of video creation tools. We had one company, Hydra, who came through, who was honestly just astounding. When the Studio Ghibli sort of content started exploding, people were like, well, how can I animate the Studio Ghibli content?

2:18:53And then everyone was using Hydra as a result. Oh yeah, I saw that. I saw your partner do that. Can we have a can we have a studio so the magic of studio Ghibli was that it could one shot these beautiful outputs? And and are you has there been are you anticipating that kind of moment for sort of ephemeral gaming where I could like take a picture of the three of us And say like make a boxing game where we can like fight and you get swords and then it's like it creates that and it's like fun and viral Can that happen in the near like are you expecting that at all? Well, never say never. I think that ultimately we have to get past this sort of distribution problem.

2:19:31But my hope is that it gets a mid-journey moment like we've had in the past. And the good thing is, I mean, I've talked to a lot of investors about this, that I think a lot of folks are in that cycle right between B2B back to the sort of like app layer is a bunch of the investors are pretty interested in obviously the B2B AI side. And I think that will then drive a lot of innovation, which then gets you the 10x here. and hopefully if someone builds a good network, then you have sort of like unwarranted or really sort of differentiated customer acquisition. That makes a ton of sense. Well, thank you so much for joining us.

2:20:01This was fantastic. Come back on to you guys. Come on when there's big game news. Of course, of course. Games correspondent. I'm going to take this hat. Please do it. Enjoy it. All right, see you guys. Let's bring in the next person. This guy is absolutely eating data. We are at YZDN Day 2025. Eat data, make chunks. Make chunks. Welcome to the stream. I'm John. Nice to meet you. Nice to meet you, John. Can you introduce yourself? My name is Shreyush. I am the co-founder of Chunky. Chunky. Chunky. That's the end of the company, yes. What do you do? We take really complex documents. We split them up into meaningful pieces, such that one piece is one idea.

2:20:32And then we send your LLM only the data it needs to answer questions. Give me an example of a really complicated document. Financial reports. You've got graphs. You've got actual text data paragraphs. You've got tables. And if you're asking about... Tech games are so annoying because there's so much boilerplate, you need to just skip to the right thing. Exactly. Yes, exactly. And most of the time when you're asking questions to an LLM, you really only need one table. Or maybe you need a summary. That's it. Why don't I just throw all of that in a big context window, Gemini, one million tokens or something, and then just ask it?

2:21:00It'll work with maybe one PDF, but you have a whole database of PDFs. You've got 100-page PDFs, thousands of those, 10 ,000s of those. You've got schematics, which are really complex. Models get confused. We actually ran this eval yesterday after the price drop on 03, which is we took relatively simple documents. We took classic literature, you know, David Copperfield, Lewis Rob, everything like that. And we gave that to O3. We asked very pointed questions. O3 got a retrieval accuracy of 75%. Okay. Great. We chunked the data through Chunky. Then we asked O3 the same thing. Always chunk. 100%. Always chunk.

2:21:31100%. Always chunk. This is my favorite name since last YC batch, which was a company called Pig. I'm sending a trend here. You just like large animals. I mean, it's just so, it's just, it's going to stick. We're going to be talking about this next demo day. Talk about the pipeline. I have a bunch of huge PDFs on S3 or something. I feed it into your system. Am I getting a Postgres table? Am I getting a MongoDB, like unstructured JSON? Am I getting embeddings waiting so it's like a vector database? Yeah, so you get embeddings out. You can put it on your own vector database, or we can also wrap around your vector database.

2:22:07That's totally up to you. It's really developer-friendly. The idea is to just make a dev tool that people just enjoy using, and they can have it be two lines of code, five lines of code, whatever it is. So what's actually happening with, it's not open source, right? It is open source. It is open source. Yes, we have an open core strategy. Okay, interesting. And so we are open source first. We started as a side project on the open source and we love the open source. So is this something that I should be running in like an ingest process as I'm generating new large documents, I'm chonking them and then loading them into my vector database, which I may be also hosting on AWS?

2:22:38Yeah, you can async chonk, but if you're building a code gen tool, then you want a live chonk. Yeah. And so if you're doing like co-gen on the fly or if you're like, you know, things, if you're working with a corpus that's changing all the time, then you do want to, you want to do it live. Did you come in? You came into YC with this idea. You're already had it as an open source project. Yes, yes, yes. We had an open source project all set up in like February and we came into YC with this idea. How many people on the scene? It's just me and my friend from seventh grade. It's just the boys group chat made it out of the group chat and then we're now at YC.

2:23:11Chonky made it out of the GC. How many chunks have you chunked? How many stars do you have on GitHub? How much revenue are you making? What do you got for us on the quantitative metrics side? So the metrics I really like are we've got over 180 ,000 downloads. Very nice. And we've got over 200 projects using us. Wow. We're a core dependency on projects like Llama Index. Oh, cool. And we've got like 10 to 12 batch companies using us. Cool. Good inbound coming in from there on. Fantastic. Round's already done. What was that? Is the round already done? Round is almost done. Almost done. We're trying to wrap it up this week.

2:23:42Thank you. Preliminary. Yes. Just wing our options and just like making sure by Friday it'll be done. That's fantastic. Amazing. Well, good luck out there. Well, thank you so much. Thank you for having me. Never stop chonking. Never stop chonking. We'll be following you. What's the domain? C-H-O-N-K-I-E dot A-I. And if you want this merch, it's shop.chonky.ai. Oh, he's already selling merch. He's selling merch. He's selling merch. Let's bring in the next participant of the Demo Day stream. If we have one. We got Godzilla. to demo day 2025 we are live from y combinator in san francisco great to meet you nice to meet you i'm cinnamon hi i'm john nice nice to meet you can you introduce yourselves yeah i'm cinnamon i'm a mechanical engineer by background uh built hardware for apple google slack uh wow yeah hardware hardware hardware for slack well yeah yeah what do they have rfx oh sorry the stanford linear accelerator center.

2:24:38Oh, that. Not SLAC. Not SLAC. Okay, that makes more sense. That was like, what hardware device did I miss? Good, clarified. Yeah. Cool. Awesome. So I'm Abhijit. I worked as a researcher at Stanford, a researcher at Harvard. I was like an intern at Intel. Cool. Built a lot of stuff, I guess. Can you pull up on the mic a little bit more? Yeah, absolutely. Yeah. And then tell us what your company does. Yeah. So we're Godella. We're building a frontier physics model for mechanical engineers. Okay. So currently, AI models can't handle physics accurately. because a lot of them are language-based. Ours are different.

2:25:10Ours are built to handle physics accurately, which means it can be used as a faster, cheaper replacement to simulations and physical prototypes. This is something that I think a lot of labs like to promise, right? They like to bring up this idea of solving these problems in physics. So walk me through. It sounds like you're actually training a model. Is it all reinforcement learning with verifiable rewards, or are you generating a whole bunch of training data? Is there a human data labeling component? Like, what is the pipeline to create what you're creating? So at a core, we extract embedded physics from data, and that makes it generalizable.

2:25:45Okay. Abed, did you want to share more about that? Yeah, so we don't use reinforcement learning or anything. It's basically like this sort of, like, you know, encoding framework where we actually, like, take the mesh itself. Like, we work with meshes, right, because we're, like, in simulation and stuff. So we take the fluid mesh, and then we, like, encode it into, like, this lower dimensional space, and that allows you, like, learn the actual physics of the system. Interesting. And this allows, And when you do like symbolic regression on like the latent space, you get like a lot of, you know, like you get a lot more generalizability than you would get with like regular models.

2:26:12Okay. Yeah. When did you guys start working on this? Did you bring it into YC or did you pivot to this? We brought it into YC. So we were TAing a computational mechanics class together at Stanford, which was teaching undergrads in mechanical engineering how to use the traditional simulation software. Yes. Kind of bred a hatred for those softwares. Meanwhile, Abhijit's doing insane research at Stanford to use ML to model the physical world in these crazy accurate ways. He's like building ML models for Intel that are replacing months of trial and error in their plasma edge process. And I realized like, hey, there's this huge opportunity to, out with these old simulators, let's bring physics-informed ML to the broader group of engineers who could really benefit from these faster, cheaper answers.

2:26:52Okay. Try and make it a little bit more concrete for me. I'm familiar with like CFD, computational fluid dynamics. I have an engine, and I'm trying to simulate how the air will flow over the jet that I'm building. Is that an example that we could use to kind of build off of? Is it just faster inference than calculating everything deterministically? Is that the goal? Yeah, absolutely. Like, all of simulation, every time you start with a simulation tool, the engineer is starting with a question, right? You've got a question in your 3D model. You want to know, what does drag on that? Yeah, exactly.

2:27:22How is that going to change as I change thickness or angle of attack with my airfoil? Now imagine instead of needing to learn a simulation software, you can ask with natural language, drop in your CAD, and get simulation quality results instantly. When we say instantly, it's 4 ,500 times faster than a benchmarked GPU accelerated software that we tested against. Wow. Okay, very cool. What's the go-to-market like? I mean, imagine you're selling to very large aerospace defense companies, or who else is building stuff? I mean, Apple, Google, these companies could buy this? Yeah, exactly. I think the huge benefit of physics-informed ML is we can tackle problems that traditional simulators cannot tackle, multi-physics, multi-scale, make it extremely feasible to tackle those problems.

2:28:03And we can also fill in the gaps where your idealized equations don't suffice to capture the complexity of the problem. So these problems that Apple, Google, maybe aerospace are throwing millions at in terms of R &D and building and testing, we can give you accurate physics models that can replace your need to physically build and test a product. And they can probably reality check your faster results with the traditional system that they have in place whenever they need to. Yeah, and build that confidence in the product. Or run that overnight instead of while you're live designing. Talk about traction, selling it so far.

2:28:34It's been great. So we launched two weeks ago. We entered a 25K a year contract with an engineering firm, Gerbilis Ansys, which is$30 billion incumbent. However, I think our stronger pull right now is we have some exciting opportunities with enterprise customers to, again, tackle those highest value problems where there is no solution for. For modeling something like drop simulation, there is no great simulator that gives you that fast, accurate answers, even though it's governed by... Simulation software gets 70 % accuracy or something. It's also very slow, though. It's like two weeks to compute a drop simulation on a 14-inch MacBook Pro today.

2:29:08So like these really high value problems for enterprise customers, we have the opportunity to go apply our software and give them more accurate answers. Very cool. How big is the team? Where are you going next? How's the fundraise going? Yeah, it's three core, one advisor. The fundraise is going well. It's great. Well, it's very exciting. We're just amazing. Whoever's bidding, I'm sure they're going to watch this. Well, congratulations. It's a fantastic demo. We've been giving out hats to folks who come on the stream. Thank you so much. Nice to meet you guys. Great to meet you guys. Congratulations.

2:29:39We will talk to you soon. Yeah. Let's bring in the next team or whoever it is. We got Dan. Don't forget to go to Vanta.com, Adio.com, NumeralHQ.com, AdQuick.com, Aidsleep.com, Wander, Go to every website right now. Look at the bottom bar. What's going on, guys? Hey, how you doing? Dan. Dan, what's happening? Good to have you on the stream. Nice to meet you. Welcome to the show. Thank you. Can you introduce yourself for the stream? I'm Linus. Linus, pleasure. It is a pleasure. Do you want me to go into any more detail? Sure. Yeah, a little bit. Let's get Justin's name. My name's Justin. And what are you guys building?

2:30:15We're building DEN. Okay. What is DEN? DEN is Cursor for Knowledge Workers. Okay. Break it down. We're an AI native Slack replacement. If you load up to the app, you'll see a bunch of AI agents. Slack replacement. Slack replacement. Okay. So I'm not in Slack at all. I'm just in DEN when I'm doing knowledge work. Okay. What is the typical knowledge worker experience with Slack? because I feel like there's a lot of just like managerial overhead status updates that are going on. This sounds like something a little bit more mature than that. What am I doing in Den? In Slack, there's a lot of kind of lost threads, a lot of kind of lost information, and just a lot of communication ultimately, but no actions.

2:30:52With Den, we're all about actions, and we work backwards from what you need to get the task done. Okay, give me an example. Yeah, so I guess in Slack, you would be asking your coworker for or how many users signed up last month? That's gonna be like an async process. That literally just happened in the Slack. In one of my Slacks I'm in, it was like, are we reviewing last week's numbers, or the week before, or the week before, was a developer on a team talking to the CEO? Yeah. For sure. In Slack, that task might just get lost in the background. You don't know what the status is in DEN, an AI agent's just gonna pick that up.

2:31:26So we're all about that multiplayer aspect. Your inputs are building the agents, and configuring your tools, and configuring the tasks. And then we kind of provide like the multiplayer environment where you might be orchestrating like thousands of AI agents. Yeah, talk about the path of AI agents, long-running agents. We have this idea of like 10-minute AGI, 20-minute AGI. O3 Pro seems to work for 13 minutes every time you kick something off. Are you putting these things on cron jobs? Is there some sort of like long-running process that can run through my den installation and say, Hey, like every single hour, I want you to check on things that aren't getting done.

2:32:03Pretty much. Yeah, we break it down into three things. Like you've got your ad hoc tasks. You've got your, yeah, cron job tasks, your scheduled tasks. And you also have things that kind of respond to triggers. It's like, oh, hey, run this task whenever I receive an email. Got it. Okay. There was this idea, I think it was in AI 2027 around the idea that like an AI would just spin up a Slack instance and use that to communicate. AIs would use Slack to coordinate with each other. And I can see a world where that makes sense. What was the catalyst for you guys to realize that you needed to kind of rethink the communication stack from the ground up to serve agents over, you know, first and foremost over humans?

2:32:41I guess what we realized was that there's tools like Zapier, tools like Relevance AI, but they're external to your communication source. Knowledge workers spend more than 80 % of their time in Slack and Notion. And so we wanted to bring the agents to where people actually did the work. and the most important thing is now the agents can escalate tasks through like the CEO, the head of customer success, whereas they couldn't before because they were siloed. So we had to build it from the ground up for agents in that way. Do you think we'll still call ourselves knowledge workers in 10 years if knowledge is instantly accessible by all machines?

2:33:10Taste workers? There will be taste workers and agency workers because knowledge is commoditized now and intelligence is too cheap to meter. Like taste curators. Yeah, exactly. Taste makers. That's the only job that will remain in the future maybe who knows how um you guys are creating a platform that other agents can work on top of how do you rate how do you rank the quality of different agents across categories obviously there's like coding agents we're friends with the cognition team um and and and the factory ai team and things like that so coding agents are great uh i i haven't heard a lot of people saying like i love my ai bdr yet right maybe there's some use cases maybe they don't want to talk about it but what are the categories that you guys are most excited about?

2:33:54Deep research is obviously another category. Agented workflow. I think it's going to slowly move down the stack. I spoke with Sholto Douglas, who is a researcher at Antomar. We had him on the show recently. We have great coding agents and math agents because researchers like math and coding. Interesting. It's not just a verifiable reward thing? It's that as well. It's that as well. Yeah. I think we're kind of building in the infrastructure. that's gonna allow those iterations to happen where you do actually get a really good BDR agent. We're already building customer success agents that haven't existed before because we're providing the primitives and the building blocks.

2:34:32And I think the verifiability will come. It's just, yeah. Interesting. How's the traction been? What's the rollout? What's the go-to-market? Go-to-market is handing out 500 business cards to everyone at Demo Day. So you want every company here, all the small startups, the early stage companies, that feels easier than ripping out it's a massive Slack installation, right? We take a lot of inspiration. Start with Den, stay with Den forever. Got it. Okay. Seat-based pricing? Yes. Interesting. Agent-based pricing? Yeah, do the agents have to pay? What about a thousand agents? Okay. So the one-person company, you're cooked.

2:35:05Exactly. Good luck, though. You'll figure it out. I'm sure it'll be fine. Nice. Have you shared any numbers? What were you guys doing before this? I previously started a company when I was 19. I would go to like zero to five million ARR, 50 people. That's good. Congratulations. Thank you. You know, I started this three months ago now, so it's been awesome. Awesome, very cool. Three months right before YC. Exactly. Justin, what were you doing? Yeah, I mean, Linus is a beast. He hired me actually at his previous sub, so I was able to kind of go from like that 500 ,000 ARR to 5 million. And we just loved working together.

2:35:37That's awesome. This is what we wanted to spend the next 20 years on. Yeah. Amazing. Fantastic. Thanks so much for hopping on the stream. This was fantastic. Awesome. Good luck, guys. See you on the next one. And we are ready for our next guest coming on in to the Palace of Party Rounds to YC Demo Day 2025. Welcome to the stream. Eloquent. How are you doing? Nice to meet you. Very nice to meet you. It's Tuche. Tuche. Hey, Jordi. Nice to meet you. Hi, I'm John. Pleasure. What's happening? I'm kicking this over. Would you mind starting with an introduction on yourselves and the company you're building today?

2:36:07Absolutely. This is Tuche. I'm the CEO of Eloquent AI. And I'm Aldo. I'm the Chief AI Officer of Eloquent AI. Fantastic. And what are you building? We are basically building an AI platform for financial services to automate complex regulated operations. Okay. What's an example of that? For example, in a bank, if you want to unfreeze an account or handle a reggae dispute, run KYC, KYB, this usually goes currently to support teams. Sure. And you have a big queue, right? Yep. It might take up to two weeks to open a new business account. Happens all the time. So we solve that problem. Rather than that going to support team, it comes to our dedicated AI operator.

2:36:50Sure. Our AI operator takes on the job. And here's the real magic. It basically navigates your existing core banking portal just like your support team does. We don't need any APIs or any engineering. Exactly right. So this is coming from Aldo's research for more than five years. they developed this technology that allows us to do computer use and browser use in a reliable way for very specific tasks. How much of what you're doing is purely enabled by the advances that the foundation model labs versus fine-tuning or any sort of scaffolding that you're doing on top of the state-of-the-art models?

2:37:31That's a fantastic question. Do you want to tell a little bit? Of course it's a little bit of both. The foundation model, of course, gives us a lot of synthetic data on which we can train on. But we leverage multi-agentic architecture to actually perform the actions reliably. What's the response been from big regulated financial institutions that are not traditionally the earliest adopters of new technologies, but everyone's talking about AI every day, so I'm sure they're excited to at least talk to you. What's the response been like? You know, things have changed. We are finding it actually we reached half a million ARR in four weeks.

2:38:06Congratulations! I love that. Congratulations. And the reason is because I think all these banks now have the mandate. Wait, did you say four and a half? Four and a half? It was four weeks. Half a million ARR in four weeks. There we go. All right, just do that every four weeks for the rest of the year. Well, at this stage, we actually have a waiting list. We have more customers than we can onboard. Because as I was saying, things changed in the financial industry. They want to bring this cutting edge AI in-house as quickly as possible. That's amazing. Very cool. How's YC been? Honestly, in any metric, it massively exceeded my expectations.

2:38:47That's amazing. I'm a second time founder. So I thought, okay, do we really need to come to YC? We already have the investor networks. But what was really beyond my expectations is the customer network, especially in financial institutions, is incredible. A lot of YC alumni are now running big fintech companies, banks, and they have been very welcoming. We can definitely feel the love. That's amazing. That's great. How big is the team? Where are you going next? I'm sure you're raising money. What's happening down the road? What's in the next 12 months? Absolutely. So we raised the big seed round.

2:39:20We closed last week. Congratulations. Do I get another?

2:39:28We closed the seed round, everyone. I love the spirit here. It's amazing.

2:39:37That still surprises me every time. It does still surprise me. I love it. We closed the seed round. At the moment, it's basically heads down. We are building. We are six people. Aldo is leading our technical team. I do sales. and we are hiring 10 more people. 10 more people. If anyone is watching who is looking for a new role, they are very welcome to contact us. Reach out. That's amazing. Well, thank you so much for coming. Thank you so much. Very nice to meet you. Take care. Let's bring in the next guests. We are live from YC Demo Day 2025. Book of Wander. Find your happy place. Find your happy place.

2:40:16Bring them on in. Go to wander.com. Hey, guys. Welcome to the stream. How are you doing? I'm good. Good to meet you. How are you doing? Good to meet you. Why don't you kick us off with an introduction on yourselves and the company you're building? Sure. I'm Somi and I'm Achuk and we're building Ulearn. It's an AI tutor for students. Very cool. What's the go-to-market? I tried to build an edtech company back in 2012. That was actually the first company I applied to YC for. It was a disaster. It was extremely hard. Never really made a dime off of it. I think I got like 500 installs on the iOS app I built.

2:40:48Let's give it up for 500. Yeah, yeah, yeah. Not exactly one of these. How are you solving that? It's notoriously hard to sell into education. It is, but I think we're seeing a change, especially with AI. Yeah. We have over 200 ,000 students active on our website. There we go. A thousand more than what I got during my YC test. Thank you, thank you. You guys can keep talking. Okay.

2:41:14200 ,000. 200 ,000. We love it. Absolutely wild. So what is the channel? Let me guess. You got a bunch of TikToks. Yep. Instagram, TikTok, YouTube shirts. We have, I think, over 500 ,000 followers on Instagram. That's amazing. And how long do you take to build that up? We do a side project for a while, but I think eight months is like the past few eight months. Awesome. And how do you, I'm picturing kind of like an LLM style chat interface. Is that the wrong? Yeah. So essentially, students upload their learning material, like textbooks, and we give them concise notes. You can have a conversation with an AI tutor as well.

2:41:53You can have quizzes. You can create exams. All the study tools. You create podcasts based on... Yeah, so right now it's a conversational experience. We want to make it more proactive. So you can listen. And are you getting to the point where teachers and schools are kind of asking their students to get on the platform and use it? Yeah, so we're talking to a bunch of school districts and stuff. But I think our main focus for now is students. We just want to get the product done for them and then move up. And what level are we talking? Middle school? Where's the option been strongest? High school?

2:42:23Yeah, so undergrads are number one right now, and then it's higher than that. And now we're seeing a lot of high schoolers as well. Are you monetizing yet? Yeah, so we have a freemium subscription. It's free for a limited amount of time and then$20 a month. I have an idea. Surge-based pricing, exam season. Right before exam season. We give a discount on exam season, and it worked really well. Even post-exam, there's always a summer sale. Yeah, yeah, yeah. I think it just works out really well. That's amazing. Wait, so how long have you actually been building this? I missed that. So we started eight months ago.

2:42:57Eight months ago, okay, and then YC. Okay, great. So have you been focusing on specific growth metrics during YC to kick off demo day? Yes. Just growing everything. Yeah, so MRR is like our key metric. Fantastic. And then retention as well. We want to make sure like a key part in learning is like retaining. Fantastic. Did you share an MRR number today? Yes. Can you share anything with us? Yes, we can. Fantastic. Let's hear it. We are at$75 ,000. Let's go. Congratulations. Wow. I mean, honestly, I got to say sorry. Yeah, I know. 83. 83. What if there's one million? That's amazing. That's amazing.

2:43:32You're close. Yeah, you're close. Next week. Honestly. Yeah, yeah. Put another sale on. Go to every person in the back. Get a subscriber right now. let's get into one mill we're not going to let you leave until you get to a million of our hours you've got to stay in this room last question obviously tons of developments from all the foundation model labs what's working, what's most exciting what are you taking advantage of are you focused on cost optimization looking for open source, do you want to use the best are you a beneficiary of the O3 pricing drop, are you a beneficiary of O3 Pro 100 % Every model improvement benefits us directly.

2:44:11Yeah, we want to focus on accuracy the most right now. We want to make sure everything that a student gets is as accurate as possible. Are students cooked by AI? Are you uncooking them? I think they're leveraging it more just to study smarter, and that's what our vision is. Also, an interesting thing is they really like our voice mode, so it's not just text-based. They can have a natural conversation. Yeah, so is it like you leave it open on your desk when you're studying? Yeah, you can do that. Yeah, and it can create mind maps for you. It can create flow charts. You can create whatever you want.

2:44:44It's very cool. How are you seeing the competition? Did you guys drop out? In the process. In the process. Legally, not yet. School for you, but not for me. How are you seeing the obvious competition between just using ChatGPT, 20 bucks a month, that's right at the pro tier or like the plus tier. There's a lot of, Tyler Cowen's been writing about, I used to feed in a snippet from a book and ask Chattop about it. Now I just say, hey, summarize this book and it already knows. Or you can just go out and find it. That feels like the logical competitor. But how are you differentiating in terms of actual UI design?

2:45:24Because it seems like as we move to the application layer narrative, there is a world where aggregating demand around AI tools, around a specific niche works, but there's a lot of secret sauce that goes into the UI. What are you doing to stay ahead? So yes, definitely a UI is a big component, but the main thing is we just ask our users why do they use us or ChatGVT. Sure, yeah, what's the answer? And they say, like, we have, like, a feature called the quizzes and the flashcards, and we create that, like, very accurately for them. Interesting. And what they can do is they can upload, like, a thousand-page textbook, even 2 ,000 pages.

2:45:55Someone uploaded, like, the entire... Imagine ChatGVT hallucinates, So you just fail your exam. I remember in college, I had a textbook and I wanted a digital version. I took it to a scanning and they scanned it all. And they were like, do you have the right to do this? This sounds illegal. I would try and download illegal PDFs. It was very sketchy. How are college students uploading 2 ,000 page documents? Yeah, so if it's available digitally, they can upload it. Okay, like ePub or anything mobile. Yeah, so I mean, they can upload anything they want, but it's on them. Okay, okay. They have to figure out that part of it.

2:46:28Okay, got it. but eventually maybe you could do partnerships with publishers. We want to partner with e-book companies. That'd be very, very cool. Yeah, yeah. Amazing. Well, congratulations. Massive news. Yeah, thank you so much. Great to meet you. We will see you later. And let's bring in the next team. Let's do a lightning round. Let's speed these up, right? Speed these up. There's demand. Is there demand out there? Okay. Hey, we only have two seats. You guys can squeeze in. Oh my God, there's six. We are going to light. Okay, come in. Come in. Come in. Okay. Realistically, only one of you or two of you can talk because there's only so many mics.

2:47:01Keep going. Can you guys move the camera to show everyone for a second? We got a small army here. Just tilt that camera. Look at this army. Yeah, yeah, yeah. Tilt over a little bit. Go to the wide. I want to show you this whole team. Are you guys, is this the founding team? Yeah, it's all the founding team. Six founders. That's the whole record. Who's the founding engineer? All of us. There's one guy who's not cutting. Let's go. What are you building? Break it down for us. So we are building AWS for AI agents. Okay. Imagine, we are 20 years ago, it was mostly infrastructure for software as a service.

2:47:35Yep. Now AI agents are becoming like the de facto new software. So we are building this infrastructure for the agents. What is important about infrastructure that needs to be different? Why would I not just deploy this on AWS? Yeah. Or Microsoft. I mean, you saw Satya Nadella at Build. He was saying like, we have model router. We have everything from DeepSeat to Llama. They vend every model. How are you going to stand out? Yeah, we stand out because today, first, people who are building agents are not cloud architects, cloud infrastructure experts. So a new generation of software developers, software engineers.

2:48:11The second thing is agents are going to build themselves their infrastructure. So they need new interfaces to interact with the infrastructure. And the type of workload that they're using, like the technology they're using, the lifetime of the computing platform, is not the same. So serverless is one thing. Traditional virtual machine is one of the same. Even going back to the YC story, Heroku was that story of like, yes, you could always spin up EC2, but Heroku made it easier. Exactly. And right now we are doing the same for agents. So when your agent is generating code, it needs to have access to some resources to run this code.

2:48:47So we provide these kind of resources called sandboxes to help them to run this code for 15 minutes, an hour, or maybe months or years. When did you start the company? Did you start it three months ago when YC started, or you guys have been at it for a little bit? Yeah, so actually we both all worked in my previous company. We sold to a cloud provider, so we know how to do it. The traders hate over here. It's like Intel again. What happened? They couldn't contain any of you. You couldn't be bought. I see what happened. You couldn't be bought. Now it's the six of us. That's amazing. Congratulations.

2:49:22You guys stick together. I love it. How's adoption been? Who are you selling to? Who loves this product? So we launched seven weeks ago. And like most of cogeneration companies from this batch, even in the next batch, are starting using us. Really? Really? They're paying customers in production right now. The economic model, I assume it's consumption based on top of the underlying cloud platforms that you're building on top of, is that correct? Yeah, exactly. We started actually to do something with a unique subscription, with monthly subscription, just to be sure that people were serious about using us, being sure that we get the right traction.

2:50:00They are using actually in production our software. Now we want to expand, so we are basically offering more just a usage-based model for the next batch. live in the office? So we leave, we sleep when we can sleep, yeah, but we leave in the office. It's a lot of mouths to feed. Have you raised money? Yeah. Rounds closed? Yeah, rounds closed. Congratulations. Thank you. Thank you. Congratulations. I think Jordan's got something for you. I don't know why. Spray them all with it. Spray them all with it. Fantastic. Well, congratulations. Thank you so much. Thank you so much for coming by. Thank you.

2:50:38Thank you guys. Fantastic team. Look forward to following your journey. We got to. Have a great time. We'll talk to you guys. We got a good name coming in next. We got Waffle coming in. Let's bring in Waffle. Waffle. Come on in to the Palace Party Rounds. Oh, yeah. They're a customer. That's great. Waxo customer. Love it. Grab some seats. Put the microphones as close to your face as you can because we are live from YCMD 2025. You guys are building agents for agents. What are you building? We're building an AI operating system for small to medium businesses. Okay. So what that means is we help you build your website, set up your business email address, your phone number, your bookings, your payments, all that kind of stuff.

2:51:16Are you guys replacing Google Workspace or are you building on top of it? We're starting off replacing Wix. So we've just launched our AI website builder three weeks ago. We've had 700 projects built on there. Okay. And yet people are building websites. Okay. So you build the websites and then you also are able to instantiate all the downstream stuff. But I imagine you're not rebuilding everything. So who are you plugging into? Like you're not rebuilding Stripe, right? No, we're not rebuilding Stripe. It's like wrapping on top of these developer tools. Because there's so many of them now. There's so many and they're so cheap.

2:51:45Yes. You know, like if let's take just like outbound email marketing. Yeah. Someone could use something like MailChimp. MailChimp. Yeah. And that's much more expensive than how much me and Diogo pay when we set up a project with Resend. Yep. Because it's a developer tool. It's super cheap. Oh, interesting. And then we can make money on the spread. Okay. Interesting. How's adoption been? Adoption has been really interesting these first few weeks because we've done no marketing. We're just like full-time coding, working on the product. So we've had people come in from the YC launch, from the Twitter.

2:52:14And now in the coming months, what we're doing is like going in and doing more go-to-market stuff. So one of the things we're doing is taking people's existing Wix or Squarespace website and just cloning it. And then just sending it to them. I've seen a lot of Wix and Squarespace. You see the ads, the Super Bowl ads, and it's always a small business, a restaurant or a pub. And do you have a couple of customer avatars that you really like or want to use to fuel your go to market? Yeah. Right now, I think what's really interesting is like these small family run businesses, like the small family run travel agents or law firms.

2:52:50I think long term e-commerce is really where it's at. They care a lot more about their website. Yeah. Yeah. How important is design in that? Are agents good at design yet? It doesn't seem like it's something that would come native yet. A lot of the art that's being produced is beautiful. Yeah. Yeah, I mean, yeah, it's interesting. The default Claude, like if you just ask Claude to generate some UI, it's actually quite good, but it's very unopinionated. Unopinionated. It's generic. Yeah, it's just like some generic type website. Sure. I think if you start giving examples and just really working the system from, you can get it to a point which is pretty good.

2:53:28What we've been doing is using 21st Dev. It's a tool, they have a lot of UI, and they're building out their magic chat, which allows you to kind of take a UI component and have remixes of it. So they work on that specific technology and we're working with them. So yeah, it's actually pretty good at UI, but it's making stuff which is still quite generic and we're trying to push the boundary. What were you guys doing before this? Yeah, so we've never worked full-time jobs. Yes, neither. Let's give it up for unemployment. I was kidding. You guys are founders? Yeah, yeah. So we both did computer science.

2:54:05I was at Oxford. Diogo was at Ecole Polytechnique, Lausanne. Oh, very cool. Met up in London. Yeah. I moved in with him. And then we've just been working on startup. And we got you out of there. We got you to America. Yeah. Great. Great. They said it was impossible to build a company in Europe. And you're a testament to the fact that you got to come to America. Congratulations. Yeah. It's been insane. I was listening to you guys two days ago. Oh, really? When the Cluley founder was on. Oh, yeah. And now we're here. It's important to get attention, but there is a limit. Where's the line for you?

2:54:36Clearly has their line around what they're willing to do for attention. That's pretty crazy. You know, YC goes the other end of the spectrum where they're like, just don't focus on your pitch. Let the numbers speak for you. Yes, yes, yes. And they're like, just put your graph up and that's what matters. I think it's good. Did you have a number that you shared today? Yeah, yeah. So we got 700 projects built in the last three weeks and then that's grown 14 % over the last week. Congratulations. Well good luck with the rest of Demaday. We have one last hat here. There might be more out there. You guys can share that.

2:55:02You can swap. There's more hats. Let's give them two hats. We're running out of hats. Are we out of hats? Yeah, yeah. We don't want to run too much. Anyway, congratulations. Thanks, guys. Great to come out. Congratulations. Let's bring in the next team. And let's tell you about Linear. Linear app. I'm guessing at least half the batches. I think so. Welcome to the stream. How you doing? We're good. Sorry, guys, just my hand in water. I'm going to not shake your hand. Good to meet you. Let's start with an introduction. Who are you guys? What are you guys building? And please, microphone as close to this as possible.

2:55:34Yeah, I'm Daniel. Cool. And this is Musa. Cool. And we're the co-founders of VibeGrade. What do you guys do? So we help teachers save time grading papers directly in their existing learning management system. What's the response been like? Do the teachers love it? Yeah. Actually, all of our customers are paying out of pocket. Really? Wow. Crazy. Okay, interesting. You said in their LMS? Yes. So who are you plugging into? Who's the most dominant platform providers right now? So the main one, the best integration is on Google Classroom. Oh, interesting. So a lot of K-12. Is Blackboard not a thing anymore?

2:56:04Not that much. We got a request for it, but most teachers are on Google Docs and Canvas. Sure. So like most K-12, which we serve, are on Google Docs and Classroom. We have Tottle as well, but it's mainly those few. Makes sense. But for universities, Canvas is great. Like they can, it'll go and highlight like actual sections. So can you add that on as like one of those, like if I'm in Google docs, I can, I can do an add on. It's a Chrome extension. Yeah. Okay. So we realized that teachers, they don't want to have to learn a new tool and go to a new platform to do that. Okay. So we built an integration that works directly with Google docs and directly with canvas.

2:56:40So it's easy as clicking a button and then we open up a window inside Google docs. Yeah. So walk me through the typical like workflow homework. I assume it's like the student uses ChatGBT to write the questions, the student uses ChatGBT to write the answers, and then the teacher goes back and uses ChatGBT to grade it. You guys grade it. But what's the actual workflow? No human in the loop at all. Are students sending Google Docs links to their teachers when they're done? So Daniel was just in high school, so he really knows how this works. Oh, okay. You dropped out of high school. Get the bottle of wine.

2:57:14Get the bottle of wine. Congratulations. Yeah. High school dropout. So the word is Chad. Yeah. Yeah. You front ran everybody. Everybody that was like coming into demo day. Oh yeah. I dropped out of college. It's like, that's so played out. You had to go farther. Yeah. I mean, like selling, I think we heard about you earlier. You're selling it back to your teachers now, like that you, at school you dropped out of sort of, I mean, they don't really like me that much for some reason. Yeah. Like on my final exam. Well, I mean, I skipped finals to go fly to, we went to this ed tech conference in Orlando.

2:57:47So I guess they don't really like me because of that. But we had to bring it to more teachers. They'll like you when you're the commencement speaker in a couple years. Maybe. Hopefully. So the workflow is where teachers, usually the students submit on Google Classroom or like Canvas. They submit a Google Doc or a PDF and then the teacher opens it up, the doc, and they grade it within by like highlighting certain sections, adding comments, a summary of the feedback. So we do the whole process right in the Google Doc. And then we We have the teacher review everything. They can speak to our system to understand their tone and their style.

2:58:19So then we add those comments. How is the actual testing and grading and homework process changing? Because I imagine that the solution to ChatGBT homework is probably everyone's on a laptop typing the answers in the classroom. And there's a monitor watching that you're not just AI-ing it. And then it can be AI graded. And then that's a win for everyone. Is that roughly what's happening? Yeah. Like the thing that we've seen is like people say all the time, you know, like students are writing with ChatGPT. Teachers are going to grade with AI. Like what's the whole point? But well, the objective of the student is to learn.

2:58:52And if you're just using like ChatGPT and mindlessly submitting something, well, then you're not doing the right thing. There's always going to be a way to get around it. Yeah. If you really don't want to learn. But like for teachers, the main objective is to give students the best feedback and instruction. Yeah. So what we're doing is just enabling them to do that a lot better. because even in school when like our teachers would give us just like generic copy paste feedback yeah so we just want to give students the best feedback possible okay give teachers their time back and they really get the value because they're the ones experiencing the benefits yeah talk to me about the top why would you not at one point just give the tool to students yeah as well yeah 100 that's actually what we're working on next so we just signed three contracts with schools last week.

2:59:36Okay, congratulations. And they want to be able to give this tool to students so that while they're writing, they're going to get real-time feedback on their Google Doc as if the teacher were adding comments in real time. So it's sort of like, you know, you submit a rough draft to your teacher, they have to grade it, and then give it back to you, and then you can do a final draft. Instead of that, why not just have, like, the assistant give you feedback in real time on the Doc. Yeah, that's great. Talk to me about the top of funnel. How are you telling teachers that this works? I imagine at the price point, you can't do hand-to-hand combat sales.

3:00:10How are you getting in front of teachers these days? So we have 100 teachers that are paying completely out of pocket. And all of that has been organic through either the Chrome Web Store. So they find us when they install another extension like Grammarly. And then they tell their friends about it. So word of mouth has been really strong. We got 600 ,000 views on Instagram and Facebook in the past 30 days. There we go. We just started posting like memes and things like that. And teachers really, they share it and then they look at our page. And so they find us there. That's great. So that's the top of funnel up until this point.

3:00:41And we're going to start pumping out more like UGC style content. We're actually going to the biggest EdTech conference in the U.S. Let's go. At the end of this month. The Super Bowl of EdTech. There's going to be 20 ,000 teachers there. And we've got a really big, we've got a 400 square foot booth there. And we're going all out for it, trying to reach as many teachers as possible. Congratulations. How do you guys make your first money on the internet? How did we first make... I used to make websites for random people. I charged this pastor in New Jersey like$3 ,000 to make his church a website.

3:01:13That was back when I was maybe 16. I made a mental health chat app in grade 10. And a couple of people, like one person in Japan found it and they installed it and paid for it. They paid for it. Let's go. That's amazing. Congratulations. Thank you so much. Great to meet you. Thank you. Appreciate it. Really rooting for you guys. High school's cooked. Excited. Call us when you meet the first. Yeah, we're good. We got middle school dropouts next. With the next middle school dropout. That's how I was going with that. Welcome to the stream. We are live from YC Demo Day 2025. Good to meet you. I'm John.

3:01:44Nice to meet you. Welcome to the stream. Nice to meet you. How are you doing? Good to meet you. We're going to have you sit down here. We're going to have you pull this microphone as close as you can. Talk directly into it. Introduce your company. What do you do? We are building AI agents for insurance claims operations. Okay. How's that going? Is business ripping? Yeah, we're at 200K AR now. Let's go. Let's go. Let's go where the money is. That's why people rob banks, because that's where the money is. That's right, John. Go to the insurance industry. It's expensive. Who are you selling to? I assume that the insurance industry is very oligopolistic.

3:02:17There's a few big players, but is that not right? We're selling to third-party administrators, which are these claims outsourcing companies, of which there is a surprising number. It's like 42 ,000 across Europe and America. Did you realize that you guys loved insurance? I'm six years old. I wanted to process claims. No, I'd seen it in my job before. I sort of had some exposure. My roommate is an insurance adjuster. So I was like, look over his shoulder. You're like, I'm going to put you out of jobs. We can hang out and go to more movies on the weekends. I'm like, you're working late. I want to play video games with you.

3:02:49That's the best. That's awesome. So yeah, walk me through what you're actually building, what the product experience is like, how it plugs in. I assume this is some sort of like co-pilot experience right now, or is it fully agentic? It's a fully agentic experience. So the goal is to replace like autonomously parts of the workflow for the adjuster. So the thing we're live with with our customers is like a claim intake agent. So when you crash your car and you call in, you basically ask you a set of structured questions and then create structured data and writes it into the system of record. Got it.

3:03:18Okay. And previously that was handled with someone on the phone talking and typing everything in. My roommate. Your roommate. There you go. Did you get him as a customer yet? No, not yet. He works for a very big carrier, so that's going to be a longer SaaS. Playing the long game with that one. And what were you doing before? I was doing product consulting and then worked as a software engineer at Agentive, which is another YC startup. How's the round coming together? Closed it yesterday. Closed it yesterday. Congratulations. Well, thank you so much for hopping on. Good luck to you. We're moving into lightning rounds.

3:03:53I'm going to give you one of these. Let's bring in the next team. We're moving through these quicker. Thank you so much for hopping on. We are live from YCDemoDay 2025. Welcome to the stream. Come on. Sit down. What do we got? Text AI. Text AI. That's a good domain. Are we supposed to suit up for this? No, no. Of course. We don't have the right gear. Pleasure. Pleasure. Pleasure. Nice to meet you. Nice to meet you. Nice to meet you. How are you doing? Sorry, we don't have an extra chair. I briefed like hot like 30 seconds of your two-minute pitch. Give it to us again. What's the pitch today? Yeah, we're AI agent right in your group chats.

3:04:29Okay, in the group chats. Yes, sir. How are you plugging in? Because I feel like iMessage, they really don't like when other companies plug in. 100%. We're carrier based. As long as you have a number that you can just add to a group chat instantly, it's right then and there. Are you a beneficiary of the RCS update? Yes, we are. Explain how. Yeah, we're working through a couple of providers that give us a number. So working with all the carriers like Verizon, T-Mobile, AT &T. Okay. And then we're applying for the RCS. It's still in the beta stages, even though it's rolled out. But SMS infrastructure is one of the best infrastructure in the world.

3:04:58So that's working. Whoa, hot take. I feel like everyone's annoyed by it. That's why Twilio exists. But, you know, is it good now? It feels like it was terrible for a long time. No, I feel like the big question around adding an AI to your group chat is security. Security. Things that happen. Yeah, single group chats. You want to be very careful about who you are. Yeah, yeah. We've seen this in the news. Don't add journalists to the group chat. How are you handling security? Yeah, yeah. No, great question, guys. We don't sell your data. Okay. You can kick out the AI whenever you don't want it. You bring it in, bring it out.

3:05:26Nothing compared to like Met AI where it's always living in there ambiently. It's completely under your control. So we don't sell any of that data either. What's the most obvious use case? Like the family group chat creating a grocery list or something? Give me some examples of how people are using this. It's actually a lot of friends using this to get a restaurant recommendation, roasting each other. When do we need to actually all meet up? Have reminders in there. That's our first agentic flow. Yeah, because when you're planning trips and trying to go out, there's always that one person that's like, no, I'm too busy or this doesn't work for me.

3:05:59And someone has to put in the screenshot. Someone's always taking the lead. Simplified it. It's like adding agents to group things. It feels like I saw this meme that was like, I joined a Zoom call and it was seven AI agents. Do you see this one? Everyone see this one. Yeah. And it was like the one poor girl. And then like this person's note taker and fireflies and Lou and this one and this one and this one. How are you thinking about building the agents yourselves? It's definitely a new flow. We're just letting people build on top of text AI. I think two things. Number one is the quality control that has to come in along the things, like you said, on the data piece.

3:06:30We're a consumer company. We're like text.ai. People are interacting with us. You want to be very careful of allowing third parties to come in here and see what they do with the data. That's prevalent. And second is we build it ourselves so we can really figure out specific use cases like calendaring, for example. Working with other people's calendars, giving you an exact answer like, hey, between all five of us, when's the best time we can actually grab coffee later this evening? And it can actually find it done and dusted right then and there. How much of the go-to-market is just pure viral growth because someone gets added, you realize that it's text AI, and you add it to the next group chat, and it just kind of goes from there.

3:07:03Can you add a Studio Ghibli machine to the group chat? That is a great question. We have Ghibli images working. I wish I can put it on live stream right now, but we actually have collaborative AI where we can actually start doing images. People can edit it together. I mean, that was the beauty of MidJourney is that they use Discord, right? And then so if Jordy writes a great prompt, I can kind of spin off that. It makes sense that that would happen in iMessage and everywhere else. There's so many fun, and I'm already thinking, like, anytime somebody shares any photo, just immediately make it, like, a Star Trooper version of that.

3:07:29That's, like, half of our group chat. It's, like, somebody should just be sending Ghibli's of all the other guys in the chat. Isn't it exhausting that everyone, like, cuts and pisses? Yeah, yeah, you go back and forth, back and forth, back and forth. Make it so that everyone is accessing it equally. Yeah, yeah, yeah. And the beauty is you can create an image, he can edit it, and someone else can edit it on top of it. Yeah, yeah, so it's much more collaborative. There is no need to really copy-paste images. You guys want to do a fun use case? Yeah. This is the best part of being a consumer, right, is that you get to put this out there, and you get to see what people do it.

3:08:00Hinge dates. Oh. Someone goes in, brings them into a text thread, says, hey, by the way, I have my AI staff here. Do you have any restrictions? Do you have any favorite restaurants? Oh, interesting. Okay, yeah. And so you can imagine, like, this is well above beyond use cases you could have imagined. Yeah, yeah, yeah. It's going to places where people are being really thoughtful about how do you bring it in to make it more easy for us to have a human connection. What were you guys doing before this? So I was at Tesla for four years leading their... Can you explain what Tesla is? No idea. Your guess is best as mine.

3:08:33No, I was leading the digital supercharging team there and the vehicle subscriptions team. That's great. And then Parar. Yeah, I was leading an engineering team at Walmart. And then before that, I used to... Let's give it up for big retail. I wonder what Walmart is. But before that, I was at Eventbrite and OpenTable doing a lot of consumer personalization. Some of the same workflows. So that's why I think one of the main reasons we founded Text.ai was our rich sort of consumer background. Yeah. So, yeah. That's very cool. Yeah, no, I've had a blast of being in media for a long time. Most recently, I was the CEO of a B2B media company down in Los Angeles.

3:09:09We exited it. I sold it over in October. Cool. I've known these guys for eight years. Yeah, yeah, nice. And then Rishi's first job was at a crypto startup my wife was at. No way. She hired him, and then he wouldn't leave our house. He just showed up. And so I've seen him do everything. And then Pahar, of course, these guys went to school together. And so the opportunity after going in and doing a lot of executive jobs at publicly traded companies and whatnot, I think if you want to know what's happening, you've got to get hands on keyboard. There's no talk the executive talk and pretend like you understand it.

3:09:38You've got to get back on. And so for me, this was a great opportunity to go work with these two guys and start from scratch and just get to know what's happening because this is going to be the platform for the next 15 years. That's amazing. Well, congrats. Thank you very much. Quick question. How's the watch game in your guys' batch? I see you guys each the Texas timeout. He has a better gatch. He's leading the way. I might be the oldest founder ever to go through at YC. What are you, like 32? Oh, thank you. Yes, yes, yes. Lucky guess. I'm going to stay with that. I'm going to 50, then I am to 32.

3:10:10Okay, okay. Cool. Pleasure. Pleasure to you guys. We'll see you soon. Bye. Up next, we have another team coming into the studio. Welcome. Good to see you. Great action. Come on. Welcome to YC Demo Day. I see two sweatshirts. Pull the mic up. Hey, hey. What is that? What? Preliminary. Just preliminary. You just assume that they're blowing out their metrics. I just assume that you're crushing it. Pleasure. Good to meet you. What's up? Me, you. Can you kick us off with an introduction on you and the company that you're building today? Hey, guys. We're building ValuMate. We're automating real estate appraisals with AI.

3:10:45When did you realize you wanted to evaluate real estate appraisals? Yeah. So a little bit, you know, family background, like things of the sort. We kind of discovered this pinpoint. We both study AI at CMU. Families had a background in it. And, you know, seen appraisals and things of the sort. Super inefficient process. And we were like, it's one of those industries just perfect. Like just ready, you know, to become a lot more efficient. And that's kind of how we got into this space. What is the structure of the legacy market today? You guys are providing a tool for people that do appraising. Yeah, exactly.

3:11:21So the current competition has literally existed for 40 years. And it's just like appraisal reports is like a snapshot of a property value. The human goes in and looks at things, takes some pictures, and writes down the quality of the trim and floorboards and mold. So we bring this entire thing down in just a scan. Okay. So you scan it. We built a 3D model, 2D floor plan, and Computer Vision takes notes. Sure. And then we pull data from all these various sources. Got it. Right? And then, you know, we use AI to obviously fill out this kind of report. So it's a pretty time-consuming process that we're able to bring down.

3:11:55Is it a partnership with Zillow to do a better Zestimate or truly augmentation co-pilot for the existing true appraisal reports that are used in underwriting? For the true appraisal reports that are used in underwriting. Got it. because the issue with Zillow, and every appraiser will kind of tell you this, this is super inaccurate. Oh, yeah, totally. Like, people are like, ah. Yeah, it's mostly based on, like, recent sales in the area. It's not actually the quality of the building. Exactly, right? So what we're able to do is... Doesn't even take into account remodels. Exactly. Because they don't know.

3:12:25Right, they don't know. They don't have the data. And what we're able to do is have this digital twin of a property. So it's actually, along the way, as we're building this and selling software appraisers and things like that, We also have the most valuable property data set that anybody's going to have to build future models. It's super accurate. What were you guys doing before this? Before this, we were students at Carnegie Mellon University. We dropped out. Nice. There we go. There we go. Yeah. Yeah. I'm sorry. There was a couple guys. Dropped out of high school. It's not really special anymore.

3:12:58But, you know. Not cool. Yeah. You did. The error form here. Talk to us about traction. How are things going? This has been going super well. We started selling 20 days ago, and we're at$124 ,000. Congratulations. You were correct to bust the confetti. Did you guys get the round done already? We are still filling out a round. Late stage talks with some leads. Good luck. But it's looking like we're going to fill it out. Congratulations. Last question. When did you first discover YC? Oh, yeah. That's a good question. First discover YC. So my freshman year roommate from Carnegie Mellon was actually YCF 24.

3:13:38Oh, there you go. Told you about it. In my mind since then, I was like, okay, I have to do this. You got to do it. You got to do it. And then, you know, we did it. Well, welcome. Fantastic. You're in the league now. You're in the league. Playing the main. Welcome to the league. Welcome to Silicon Valley. Yep. And welcome to Demo Day. Thank you for stopping by. We will talk to you guys. Thanks, guys. Good to be here. Let's bring in the next crew. Thank you so much. We got a line. We got a line out the door. We're going to bang through these. These guys got QR codes on their sleeves. Bloom. Yes.

3:14:03Bloom. Are you building Bloom filters? No, no, no. What are you doing? So the easiest way to do it. Bloom and Onions. Nope. No? Okay, break it down. No, so lovable, but for native mobile apps. Okay, cool. Oh, interesting. We'd love to just, like, show you. Please, please, please. Okay, so have you guys tried to build mobile apps before? Three touches. What is that? I've never seen that interaction before. He's got a UI from the future. So have you guys ever tried to build mobile apps? Yes, yes. I wrote Objective-C. It was terrible. I wrote Xcode. It was awful. Perfect. We love Apple, but we're very excited about the Anthropic partnership.

3:14:35Yeah, so typically you'll have to write code. You'll have to build the app in Xcode or something. You'll have to submit it for apps to review. You'll have to get your users to download it in TestLite, enter an invite code. Okay, so with us, I can literally just talk into my phone, build a native app, and then I can send it to you via literally bumping phones. Do you have an iPhone? Yeah, I do. Wow. You'll just load the app right under the home screen. Okay. Malware installed. All personal information extracted. I don't know. I got one. We got half. This is just using AirDrop. Do you need to turn it on again?

3:15:13Oh, no. Live demos. AirDrop. Maybe we need another iPhone. Are you on the same Wi-Fi? You want to try? Yeah, I'll try. Let's see. AirDrop. AirDrop is on. Oh, contacts only. Yeah, that's probably the issue. Okay, I got it. Okay, cool. Let's try it now. There we go. All right. Skill issue. Skill issue. Maybe. Okay. Is it going? I got the wave. You saw the wave, right? Yeah, yeah, yeah. I only have RAM, public, adquick, eight sleep, wander, and bezel installed right now. Those are going to be absolutely happy. Let's give it one more shot. You can take it. Otherwise, we'll have to try it on your phone.

3:15:48Yeah, it's okay. It's not like this is live or anything. Okay, you try. You try. We'll give it a shot. Anyways, keep while we're... Yeah, so... So when someone tries to build an app with Bloom, right? It's doing the way, but it's not doing that? I think it's the internet. The internet just sucks. I think the internet is not great. Let's blame the internet. Anyway, so what you would see is you'd be able to open this app on your phones instantly, right? And it uses app clips under the hood. Oh, yeah. Okay. So that's how you're getting the app on without me actually installing the whole app. Exactly.

3:16:18Oh, so you're really hacking the app. Exactly. That's awesome. Okay. Exactly. And so the other thing is when you prompt to build an app with us, we also automatically deploy a backend for you. Yeah. That real-time syncs data between devices. As I understand app clips, it's like there's one app, like the Uber app exists, and then there's an app clip that ties to Uber. Are you creating custom app clips? It's all around one iOS. Oh, wow. It's one Bloom app clip that then loads apps from anyone. So it's like an app store within an app store. Yeah, kind of. Wait, wait, wait. I don't want to say that.

3:16:50We've talked on our show before about this idea of ephemeral apps. Yeah, yeah, yeah. Memes. Exactly. There's a lot of things that should exist, but only for a day. Yeah, yeah, yeah. We've had so many ideas for funny apps, but we don't want to actually go pay, find someone to do this. There's so many people have an idea for an app, and you're like, well, you realize it'll cost like a million dollars to build that. It's not actually valuable. Exactly. With us, we built an app for Demo Day in five minutes that everyone in the audience could just scan to vote on our valuation in five years, and it had a graph that updated in real time for everyone, because it has a back-end connected to it.

3:17:23And so what we want to enable is a creator economy, but for software. Okay. Under the hood, what are the best co-generation LLMs that you're using? What do you like? What's exciting? And how is that market developing? Yeah. So, I mean, right now we're obviously for the smartest models, we're using Cloud 4 Sonnet. Cloud 4. But we're also experimenting with, you know, a smart mode and a fast mode. Because sometimes people just want to make like a quick edit to their app. And like literally you can just speak into our phones, right? You just go into this edit mode, and then you can just type, oh, add this feature or whatever.

3:17:55Very cool. And so sometimes you just want those changes fast. Okay. And what's cool is if you had this app open, it would also hot reload on your device. Yeah, yeah, yeah. Right? Very, very cool. What use cases are you most excited about? What are you seeing? What categories broadly? Yeah. I mean, so right now we're seeing people that already think of software as a creative outlet use this. So developers, designers, and entrepreneurs. Sure. So especially for designers and non-technical entrepreneurs, it's great because they can just like be their creativity gets unlocked with this um and so we're seeing people build all kinds of things like personal apps but also apps that i wouldn't have imagined before like someone in africa building like a wildlife tracking app for their conservation uh and then people building like funny apps for weddings to find the most delicious animals to go after to conserve the animals for for my own hunting for my next hunting But yeah, what I'm really excited about is like the apps that I can't even imagine, right?

3:18:51It's kind of like when the YouTube founders put out YouTube, I'm sure they weren't imagining vlogging. They didn't imagine MrBeast or MKHD, just like, you know, you put out images in chat GPT. They didn't really imagine Studio Dolph exactly happening. And for us, I mean, it's software, so you can literally do anything. Congratulations. Thank you so much. How's your round going? Oh, round is closed. Oh, there they go. Let's go. All right. All right. Pleasure, guys. See you. Good stuff. Bring in the next team. Welcome to the demo day stream. I'm breathing in the microplastics. Oh, yeah. It's really, this is violating everything I know about you.

3:19:24You're putting your life on the line. How's it going? Hey, what's happening? Hi, nice to meet you. I'm John. Pleasure. Hi. I don't have a long time. Yeah, no worries. No worries. Can you introduce yourselves? What are you building? Sure. So we're Morpho AI. We're building a software tool for engineers that are building new robots and new machines. Oh, interesting. Yeah, so I came from the manufacturing tech world. We both met at Harvard. I was an MBA. He was a postdoc, and you should talk about yourself, too. Yeah, absolutely. And I met her when I was at Harvard. Like she said, we got an introduction from a mutual mentor.

3:19:54I did my PhD at MIT, focused on automating the design of robots. So it seemed like I wanted to bring that into some sort of product, show people in the world how useful it was and how it could change the way people design. And she was super excited about changing all the pain points in manufacturing. So we're doing this. What is exciting in terms of robotics in manufacturing? There's a lot of noise about humanoids, but we've talked to a lot of people who are just saying. Robotics and manufacturing, or is it a tool to accelerate the manufacturing of robots? The second one. Yes, the second one. But the main app, Beachhead Market, is really in industrial robots.

3:20:24Okay. So give us some examples of what those look like. So the crazy stat that we found is if you're actually buying a robot off the shelf, you can't actually just put it in the factory floor. Like 90 plus percent go through a customization process. So that's like six months of lead time, and it takes up a lot of engineering hours. So one of our customers, they were trying to build a full new industrial robot arm. Sure. Six months gone in, you know, arms not lifting. They came to us, and in two days, we basically redid their entire hardware design. Interesting. So quite literally, we're allowing engineers to build new robots overnight is the hope, and in the future, in a matter of minutes.

3:20:57How concentrated is the robotic arm market? Are there just a few companies that you really need to integrate with deeply, or do you need to create something more generalizable out of the gate? So we're starting out with a little bit more of the OEM side of things, but also these integrators that are buying these arms and saying, I now need a new hand. So out of like the what, 50 ,000 robots that could deploy in a given year, most of them have a new hand that's made. And so now you have a generalist engineer trying to make a new hand for four to six weeks in a new design. So really plugging in in that early stage of what do I build before we even get to the programming part?

3:21:29Yeah, the whole idea is you just have to input the test specifications that you need to solve. And then as much as possible, we're automating of the mechanical and somewhat the control design side of things. Is there a lot of data that you need to feed in? Is there a lot of data that you need to pull from the manufacturer before you can? We work with parts that manufacturers have and that they like to work with. And aside from that, it's task specifications. But unlike a lot of the Gen.AI companies out there, we do employ some Gen.AI solutions. But they're all trained from simulation and not from data sets.

3:21:57And that's really important because the data for how do you go from design to how well it's going to work, that doesn't really exist. Nobody takes logs and curates data sets of things they built and how well they worked. Sure, sure, sure. Talk about traction. Sounds like you guys already are in market. Yeah, so we actually just got a grant from the UK government, two and a half million British pounds the last couple of months ago. Non-dilutive? Fully non-dilutive. Congratulations. We love it for that. Jordy. Oh, my God. My ears. It's going to happen. Hey, non-dilutive. That's less than what we didn't hear than out there.

3:22:30Non-dilutive funding from Europe. I mean, Europe just shooting themselves in the foot. Not getting the captain. They're going to be like, why didn't we get equity? At least like a quarter point or something. anyway congratulations thank you none of us are British but they basically we're setting up an office in London we're really excited about working with them and setting up an office of course there's benefits to the research community we're talking about collaborators there as well it's fantastic it'll be an amazing outcome for them last thing bull bear on humanoids what's your take? I'm bear medium bear if you're going to build a humanoid Which humanoid?

3:23:10Like people come in all sorts of sizes and shapes. A construction worker is not the same as like a toddler or something. Ballerina. Okay, not a toddler, but like ballerina. Exactly. Linebacker looks different than a sprinter. So you're going to need custom designs regardless of what you do. But I mean, the other side of things is just like, what is the application right now of the humanoids that isn't solved better by re-engineering the things around the humanoid? Sure, sure, sure. And there's a long tail there. So we hope we can help people to design the humanoids of the future. Just don't think it's here yet.

3:23:37Yeah, yeah, yeah. Yeah, you're not saying bearish in a hundred year timeline. Or even maybe a ten year timeline. I just think it's a little... I completely agree. Very reasonable stance. Awesome. Anyway, congratulations. Thank you so much. Thank you. I'm going to give you a salute so we don't have to deal with the whole... Or we could do a fair fist like it was COVID. Boom. Okay, perfect. Anyway, let's bring in the next team. We're putting the word out. Thank you so much. The table is now covered with confetti. The UK is giving out millions of dollars for free. Yes, head over across the pond. This could be the greatest capital extraction event.

3:24:06The greatest wealth creation event in history. In history. Since the Boston Tea Party. Anyway, good to meet you. What's up? Welcome to the stream. My name is John. How are you doing? Can you introduce yourself and the company? Yeah, absolutely. Yeah, so I'm Rajith. We're Prism. Okay. Prism is an agentic observability company. Okay. So we let developers configure agents to watch their production systems. Okay. So read logs for them, watch videos of their customers using software, and then enable them to take certain actions. So like create issues on Linear, GitHub. Yeah, Linear. Sponsor the stream.

3:24:37Exactly, yeah, I did that on purpose. And I sent reports to Slack. That's great. How has adoption been? How are you selling in? Are you going medium-sized scale-ups, other startups, YC companies, enterprise? What are you thinking? Bottom-up. So this is a tool that every developer needs and every developer can use. So we're starting with the YC community. We're completely self-serve. So there's 23 people on the platform and counting. More people sign up and start using it every time. Were you guys iterating through the batch process? Oh, yeah, yeah, we were. So the first version of the product, so it's like we watch videos of people using software, right?

3:25:09So the first version of the product was these two watching everyone's videos and just giving them issues and like sending them Slack messages and stuff like, guys, this is broken. You need to fix this. Yeah, yeah. So we were basically like a services business, but we didn't tell everyone. We told them it was AI. And it was just these two. So that was the first version of the business. Gotta start somewhere. That's awesome. Yeah, true. How's fundraising going? You guys in the midst of it? It's going great. We're in the midst of it. Cool. You know, still trying to finish up the round. Yeah. Good luck.

3:25:36But yeah, we're excited. Excited. That's fantastic. Awesome. What were you doing before? I was at Greptile, which is a code review company. Oh, yeah. I think you guys have their website. Crazy website. We love their website. That's amazing. It's a great website. And you two? I was a software engineer at Pound 2. Oh, cool. And I worked in product at Johnson & Johnson. Oh, they're cool. Alex, Alex. Big Pharma. Yeah, yeah, yeah. So, Landon and I met in high school and then all three of us were at Georgia Tech together. Oh, cool. But Alex actually had a post on LinkedIn this morning about how he left$850K a year behind at Palantir to come work with us.

3:26:09There we go. All right. I think you'll make it all back. Burn the boats. Just burn the boats. Anyway, good luck with the rest of Demo Day. Appreciate it. Nice to meet you guys. Cheers. We are ready for the next guest on the YC Demo Day stream. We got 10 minutes left. 10 minutes left, and then we got to go? Yeah, we can speed through this as much as we can. We got 10 minutes until we got to go to the airport. Is that right? Or start packing? We've got 10 minutes. The show's going to be over. We've got 10 minutes more. Come on in. Don't even bother introducing yourself. Just tell us what you do.

3:26:39What company are you building? How are you doing? We are Clarm. We are perplexity on internal documents. Okay. So search is being heavily disrupted by AI at the moment. As you know, Google is being disrupted by perplexity. And we're doing that for internal enterprise search. Competition with Glean. They just raised a bunch of money. How are you thinking about it? We're going to replace them. You're going to replace them? They're just getting started. You're replacing them. Lexi's replacing Google. We're replacing Glean. Okay, okay. So how do you do it? What's different? How do you actually integrate?

3:27:09Are you building a bunch of integrations? Are you building on top of an integrator? We have our own integrations. We found that the only way to do this, to actually build AI agentic search, is to build it from the ground up. And that's why all these legacy players have to do that as well. What's more important in integration with Google Docs, for example, or integration with the data lake? That maybe somebody has a snowflake installation. What's more important? Yeah, for you. It's important to have both of them so that you can connect them together. So you integrate both and you connect them together.

3:27:38We have about 45 connections. 45 connections. Okay. How about customers? How many of those you got? We have eight. We only launched two weeks ago. Congratulations. Congratulations. What's the biggest challenge? Is permissioning hard? Maybe at smaller companies it's not as much of an issue, but as you go kind of upmarket? It's dealing with all the different types of data, and this is why B2C companies don't play in enterprise space because when you have to look at Salesforce data alongside messy Excel sheets, then it becomes harder to connect them together. But that's the value in it. Where can companies go to get started?

3:28:12They can go to clam.com. They can try our live demo now. You own the five-letter domain already. Congratulations. You're up to five-letter domains. Congratulations on Demo Day. They can talk to us directly. And fantastic JLC Reverso. Thank you very much. It's a fantastic watch. Love it, guys. Awesome. Congratulations. Thanks so much for hopping on. Let's bring in the next team. We're doing lightning round. lightning round lightning round lightning round at yc demo day 2025 welcome to the street what do you do what are you building ai co-pilot for solopreneurs okay get out of here good job how's it going uh how many customers do you have what what data are you sharing today at who's more cracked he's definitely the most cracked engineer we have ever seen he's the most So we have gone from 0 to 300k ARR.

3:28:55It just lasts 9 weeks. What? So that's what we're doing. Absolute dogs. And we're seeing the future where solopreneurs are going to completely wipe coat and run their entire business on Cactus. So that's what we're building. Great name. What were you guys doing before this? We built a previous YC company. He was a founding engineer. We scaled it up to 2.5 million years. That's it. I'm the exact guy. He's a crack sales guy. I'm the crack engineer and together we are the crack founders. So crack sales guy, how are you actually selling this thing? Is it hand-to-hand combat with these solopreneurs or are you doing like viral marketing?

3:29:29We've seen levels get a lot of attention for solopreneur stuff. How are you actually attracting people? Absolutely. So it's mainly through word of mouth that's been spreading. What we also do is outbound where we call them. So think about a caterer, a private chef. They're busy cooking all day long. Call them. They don't pick up as expected. They're busy. We leave them a message saying, hey, you just missed an opportunity, guys. So yeah, they get back and then they set up cactus, they get incremental 10 to 15 K in a month revenue. And the biggest thing is the headache for them is gone, right? They don't have to answer the phone.

3:30:00Yep. That's the best. That's amazing. 300 K R ARR. How's the fundraise going? It's going great. We just got completely oversubscribed. Well, congratulations. Thank you. We'll bring it in. Thank you. Have a great rest of your demo day. Come on down. We're live from YCDemoDay 2025 and we have our next team in the building. Nice to meet you. I'm John. Nice to meet you, Ashani. Good to meet you guys. Hi. Would you mind introducing your company? What are you building? Yeah, for sure. So we are building Lumari, which is essentially helping go-to-market teams build tools internally instead of having to buy super expensive SaaS.

3:30:38What tools do go-to-market teams need? Like golf clubs? Yeah, yeah, yeah. Golf club buying machine on subscription and steak dinner booking. That would be nice. That would be nice. I wish we did that. But we are helping them from anything from deal scoring, qualifying leads, making contracts, anything down the sales pipeline. Got it. Got it. Okay. What does the future of the stack look like? Are you guys going to basically, are you trying to verticalize effectively and allow people to build custom software at every point? At every point. Yeah. I really think we're going to look back to this era of SaaS of having all these generic tools that you're using and think that was really silly because why wouldn't you have software that's custom built for your company, for your process?

3:31:21And so I think exactly that. We're going to start with replacing some of these really point solution software, but I think in the future every company's going to have their own CRM. What were you guys doing before this? What were you doing before this? I was at Stripe. Sam was at Google. Sort of a non-traditional background. Who's the key person actually using the tools? Is this for a non-technical person? It's a non-technical person within these go-to-market teams, often like a revenue operation, sales operations. That makes a ton of sense. Awesome. How's traction? It's been great. We're at 90K ARR in three weeks.

3:31:55Congratulations. Obviously, we're out of confetti. Yes. Yeah, we are. We're pretty much out. We got a couple left. Okay, okay. I'll let you say it. They're kind of busted. Anyway, thanks so much for coming on this week. Thank you so much. We really appreciate you. Goodbye. Let's bring in the next team. We have five more minutes, right? something like that. Let's go. Five more minutes. Let's go. I'm losing my voice. Welcome. Welcome to the stream. Introduce yourself. Introduce the company. Good to meet you. What are you building? Chairs all over your shirt. I'm Yoav and this is Shurya. We're second time exited YC founders.

3:32:28Addicted to startups. Addicted to startups. We're building third chair. It's agents for in-house legal teams and we're starting with media and entertainment companies. Interesting. Very niche down. Not just like legal AI, but you've actually Do you want to dominate a small market? No, it's a massive market. Oh, it is. But we're starting by dominating media and entertainment. Sure, sure. We're starting by helping media and entertainment companies find IP infringements. Oh, interesting. And we collect evidence around it. And then that's revenue driving immediately, right? Exactly. It's a revenue generating workflow.

3:33:00Is that the business model? You take a cut of whatever you get or is it more seat-based? We have a software model, but we also take a cut of what we get. So it's kind of a mix. Two bites of the apple. I like that. Yeah, yeah. We're exploring a little bit. How's traction? What have you shared? You're finding the IP infringements and then you're actually sending letter, demand letters. Someday we're going to be a vertical AI law firm. And yeah, we're handling the entire workflow right now. And a lot of it is being done through agents. But our customers just think of us as people who get things done.

3:33:29They don't care about how the other, like the black box is working. That makes sense. Yeah. What metrics have you been sharing? How's demo day going? Wait, is this your third demo day then? It is. Second demo day. Second day of the day. Second day of the day. Second day of the day. Were you co-founders together last time? What's that? Were you co-founders together last time? No, we had separate startups. Different companies. Rivals joining forces. Finance companies before this, social media analytics, now we're in league. So YC Alumni Network is strong. Yeah, that's great. We teamed up and yeah, Traction's going great.

3:33:57Cool. We just crossed 100 ,000 ARR. Congratulations. We're working with the biggest media entertainment companies in the world now expanding to brands, doing stuff like marketing compliance. So yeah. That's great. Well, congratulations to all of the progress. Thank you guys so much. Thanks for stopping by the stream. We will talk to you soon. Let's bring in the next team. How are you doing? QFX. I like this hat. That's some capital. Here, TC, introduce yourself. How are you doing? I think Solar Founder. We've got to give him a hat, a TBPN hat. Oh, yeah, definitely. Introduce yourself. That's amazing.

3:34:29Perfect. Let's swap. Yeah, so hi, guys. I'm a 9am CEO of QFX. Cool. We're making a 24-7 stock exchange. Okay. So we're going to let institutions and retail trade traditional assets like US equities and commodities real-time 24-7 without brokers with loads of leverage just like you have crypto. Loads of leverage! Who doesn't love leverage, right? How's the traction been? It feels like it's really hard to get these things off. What's the underlying... Is this a sneaky blockchain company? There's no blockchain. There's no blockchain. It's completely off-chain. It's basically... It's like a crypto exchange, but without the blockchain parts.

3:35:06We've taken those improvements and moved them over to the traditional assets world. Okay. And the traction was good. We've been running it. Well, we've not been running it to the YC batch because we're not licensed yet. Sure. But we've been running some internal experiments. Yeah, some internal experiments, the current YC batch. And due to those experiments, we're launching in a couple of months. Congratulations. With a license. Are you launching offshore? Offshore. Offshore in Bermuda. We'll get you back. Bermuda. Bermuda is a nice place. It's a good bit. Yeah, it's a great place. You're going to be taking a lot of trips there.

3:35:37Yeah, exactly. Free rum if you come visit our office. There you go. Fantastic. There you go. How did you get into this? I was at Quant, yeah, at Tower Research before. Okay. And my co-founder was at Citadel. Well, you're our Quant now. Yeah, exactly. Well, congratulations. We'll talk to you soon. It's a call of progress. Yeah. Great to meet you guys. Cheers. We'll talk to you soon. Thanks. Come on down. Loads of leverage. Loads of leverage. We love to see it. Welcome to the stream. Tell us what you're building. Wait, were you here last? time? Were you not? No, no, no. Sorry, sorry. I thought I recognized another company.

3:36:08Minerva, yeah. It does seem like a similar name. Anyway, introduce yourself. What are you building? Yeah, I'm building an AI accounting for small businesses. Okay, cool. How's it going? Do you have small businesses on the platform already? Yeah, yeah. We have a few small businesses. We're just trying to automate all of their bookkeeping. So, you know, chase people down for receipts, make calls. I feel like there's so much that's already built into the accounting suites. Do you have to sit on top of QuickBooks or do you actually pitch people, hey, let's rip out what your existing accounting solution?

3:36:40No, we're sitting on top of QuickBooks. There's no point replacing this software. That makes 10 sense. Yeah, we're just replacing the service side of things. Oh, yeah, yeah. So consolidating all the scattered data. That makes 10 sense. That, you know, sits in all different kinds of places, your WhatsApp, your email, your Stripe, everything. How were you doing before this? Sorry? What were you doing before YC? I was building a health tech business. We were selling AI patient triage software to clinics. Awesome. Cool. How's traction been? How's the raise going? How's demo day been? Yeah, demo day's been great.

3:37:10We are raising$2 million and roughly like three quarters finished. So just trying to wrap it up. Fantastic. Classic. Two out of 20. That's really funny. Love it. Congratulations. Thank you so much. Have a great rest of your demo day. Last one. Bringing it in. And closing it out strong with SimStudio. Are you simulating things? Yes, we are. I'm simulating. What's going on? Hey, how are you? Nice to meet you. How are you doing? Great. I'm doing great. How are you? I'm great. Break it down for us. What are you building? It's an open source platform to build AI agents. Okay. Yeah. So it's developer focused.

3:37:44It's like a Figma-like canvas to build agents. Interesting. How many GitHub stars you got? We got to ask. 50 to 4 ,000 in the last two months. Wait, wait. You were at 50 and now you have 4 ,000? Yes. Wow. There we go. Congratulations. Thank you. I appreciate that. Thanks. Yeah.

3:38:014 ,000 GitHub stars. You'll love to see it. Yeah. Congratulations. We've got two more. I can't believe Gary let us bring these in. Oh, yeah. He let us. The cleanup after is going to be interesting. Anyway, how's Traction been on the sales side? I imagine you have a product that you actually sell on top of it, not a nonprofit. Yeah. Yeah. So, yeah, we can't disclose revenue numbers, but we have a lot of great customers. So yeah, Department of Defense, Merkur, Epic Global. Oh, wow. Yeah, a bunch of cool names. Huge, huge, huge. So the round's definitely done. Definitely done. Yeah, definitely done.

3:38:32Let's do another one. The round's done? The round's done? Oh! That's a game. Yeah. What were you doing before this? So I was at Berkeley at my co-founder. At Berkeley? Oh, let's go. Let's hear from Berkeley, baby. Right across the bay. Go Bears. Right across the bay. We love it. We love Berkeley. I'm completely covered. It's everywhere. Congratulations on the playoff. I appreciate it. Yeah, thank you. What's next for you guys? Next is just making our customers happy. What do you get after the Department of Defense? Where do you even go from there? Yeah, I mean, just making developers happy, like building the product out more and more.

3:39:12And I think developers love us and big customers love us. So we're just keeping that energy going, keep launching new products, building the team. We have a really great killer engineering team from friends at Berkeley. So, yeah, we're excited to just keep growing and keep building. What's the key value prop for building a platform for AI agents? Is it like interoperability between different models? Is it the ability to scale? Is it just price and cost and speed? Yeah, I think the question I have is like even last YC batch, there was a bunch, there was companies with this pitch. And I think the challenge is everybody wants to build infrastructure and pitch and travels for agents.

3:39:46But then it feels like you can build good software, but it's maybe even harder to get the kind of companies that can build high-quality agents that actually have value. What's the secret to finding even customers that are not just going to sign up but actually get value out of the product? Yeah, that's definitely true. I think a lot of people, especially right now, it's very in vogue to adopt AI. And so there's a lot of top-down push from companies to adopt these AI implementations. But I think the biggest thing for us is that we're focused on not creating easy abstractions to make it easy to use.

3:40:16And I think that's where a lot of people fell short. It's like you're creating these abstractions that make your platform easy to use, but it's actually not powerful enough to really put AI into your production system. And so for us, we're really focused on actually might be a little more like might be a harder to use. But because we remove the abstractions, you can actually power relatively complex applications like real world simulations, deep research, data transformations. Because a lot of companies are sort of going after the sales and marketing use case. and for us we're more focused on the developers and actually production systems.

3:40:48Interesting. How much is real-world simulation scaling right now? We've talked to some of these companies and it seems like it's almost like video generation. It feels very nascent. We really haven't had this breakout Studio Ghibli moment for real-world simulation. How is adoption going there? Yeah, it's actually going quite well. What are the applications? Yeah, so the applications of that are essentially simulating like i i guess like one that would be interesting would be like international affairs so like understanding how like global events are going to play out like we're using agents um yeah so that's like you know a broad uh topic that we're yeah i've heard about that with like economic modeling exactly like housing prices you can actually have agents running and it was always like just like grand theft auto level right but Now it's like, oh, what if each of them has an internal reasoning engine powered by an LLM and has 130 IQ instead of turn left, I'm running A star.

3:41:46It's much better. Yeah, and I think the thing that really our platform unlocked was being able to run thousands of agents in parallel at the same time. And I think that was one of our grounding thesis was that we wanted to create this environment. Like SimStudio comes from building simulations of things. So launching 10 ,000 agents at a time, and perhaps some of them are going to come back and give you an accurate result, or some of them might inform you in an interesting way. And so you take those aggregated results and you go do something with it. Very cool. Yeah. Thank you so much for coming.

3:42:14Yeah, it was a pleasure. Thank you for having me. Yeah, yeah. Great meeting you both. Fantastic. And that is the end of our demo day stream. Thank you to all of our partners. We will be back in Los Angeles tomorrow, live from the TVP and Ultradone. And thank you to Gary and the whole YC team for having us. Yeah, thank you to the light combinator team. It is fantastic. San Francisco's back. Gary Tan's back. He never left. YC never left. The Rosewood is back. It's just a fantastic time in San Francisco. Fantastic time to be here at YC Demo Day. Have a great afternoon. Thank you for watching. We will see you tomorrow.

3:42:45Goodbye.