0:00All right. All right. Let's do it. Super excited about this. We obviously are a YC company. It kind of feels like our DNA. Let's start off with YC. Maybe we can talk a bit about the current batches. Um, can you give a rough idea like of the categories, the breakdown?
0:17What sort of companies are you seeing come into the batch?
0:19Yeah, definitely. um you know it's about 75% software 25% hard tech which is very noteworthy for YC because uh I think even going back 5 years maybe that number was only 3 to 5% hard tech like and so that's a really big shift um you know if I had to guess it's really this anticipation for uh you know more or less AGI loops in things like Navier Stokes like literally pushing forward
0:47the ability to do brand new R&D that's never really been done before. It's the kind of the kind of research that at one point was done in universities. Um the people working on hard techch by and large have PhDs. about um a third of the founders working on hard tech have full PhDs you know not even ABD like they you did their dissertations you know and so it's really it's a different moment right now which is actually very exciting right like software is great
1:16but hard tech is like the thing that creates abundance and actually like does it is it because your confidence in software is lower or your confidence in hardware is higher so what I mean my confidence in YC in uh in software has never been higher higher. I mean, this is the golden age of being able to build. Like, I got in trouble with you guys for uh saying like one person can do the work of 400 people. I retract the 400, but like I think you can do like eight times, 10
1:45times, maybe 50 times. I mean, I might go as far as like there are probably people in this room who can do the work of a 100 people, right? Um you know who you are. So, it's everyone here. So then what do the models bring that gives you so much more confidence in hard tech that you're willing to increase the number of companies in the batch? I guess the wild thing to us is uh I mean the most surprising trend that we're seeing is that not only are the hard tech companies applying and uh the
2:15quality of the people working on them are like so you know this PhD level you know all of the right you know it it's not you know necessarily like the the software companies are all like a wide range like 18 year olds all the way up to 50 year old I mean 60s I mean just like you know classically much younger on the software side but you know, on the heart side, like people coming from national laboratories, people coming from, uh, you know, the Bell Labses of
2:44the world. Um, and what they're realizing is like, oh well, I can't get the compute. I can't get the support. I can't use AI the way I want to use it in like literally the universities and the R&D centers. And I think that's probably the biggest shift is just like you actually need compute in that agentic loop like uh you know GP6 Astro with Navier Stokes like that kind of thing is going to be happening all around in society and I'm really excited about that. Uh you know it costs tens of
3:11millions of dollars um to achieve that result. Uh but the really cool thing about the scaling law is that you know that same tens of millions of dollars today this year might only be a million dollars next year. It might only be $100,000 the year after that. and you know sometime in the next 24 36 months like does that mean that almost every serious problem facing humans uh could
3:39be addressed and you know some someone in this room might pair with like a PhD researcher and like go and make that happen like I'm really excited about that. Are there some categories of like hard tech that you think are like nearer term some that are longer term like let's say the Navia Stokes example because math you can wrap some formal methods around it so suddenly it's very good but when you get some unconstrained maybe physics or something like that they might be a bit harder like how are
4:08you trying to judge the companies that oh these guys can achieve it within this time frame rather than in the next 10 years. I mean the reality is with a startup like these are not research organizations necessarily the you know we're trying to be pragmatic so you know the best version of these things um have some sort of real pragmatic uh use case in like today right I was meeting with a you know I did office hours just now with a photonix company and they told me
4:36this crazy story about how data centers today run on fiber optics but the very largest data centers with you know hundreds you hundred 000 GPUs like the Colossuses of the world. Sometimes they only run 30 minutes because too many lasers burn out during the operation that uh you know the actual number of GPUs turned on and that can communicate to the main network is like you know 20 or 30% of them are like totally offline
5:05at any given time and it's literally like because there are tens of thousands of lasers and so you know how cool is that that you could sit down in front of a computer come up with a new type of laser technology and then that's very pragmatic. like if you could go into data centers and solve that problem for uh for that set of customers like that's insane, right? And like these are things that teams of two or three people uh with frontier models uh maybe developing
5:34your own like child models around like your own version of alpha fold for photonix or whatever it is. Like I think that there's just going to be like this AGI loop around um you know these frontier models that have literally you know super intelligence available on tap like just you have you can have just a team of intelligence working on well here's the next step like you're you're supplying it what you need or what you want it's like no no I actually need you
6:01know 500 m lasers I need you know one watt lasers you know it's not physically possible today but if you wanted it how would you get it and then you would have you know you would have incilico simulation and then you have uh experimentation um and you know I mean periodic labs for instance a fantastic company doing that in material science like that's you know the low they were working on a what post training 01 I think and then they said you know what like this is the lowest hanging fruit we're going to go do that
6:30it's like I think that model is going to happen across all of science and engineering and that's really exciting do you think you have the most interesting job in the world where you just can sit down with people and basically see what the future is going to look like.
6:43I mean you hope that's what the future is. I mean we give them money and we try to give them a network and uh you know often they'll do it which is awesome like what that can happen at all is like the remarkable part.
6:54Your office hours now I imagine are so much more interesting. You're like we were talking to Michael backstage and he was saying that you know in the past a lot of the same problems and now he's learning a lot from office hours. I can imagine it's the same with you. There's a lot of new stuff coming in that's just super fas. I feel like I've always learned a lot every single time is like, you know, uh how the the funniest thing about society is uh how many conversations you have. Like I think you can tell how much fun, you know, I I get to have by like
7:23how many dumb questions I get to ask of people who are, you know, way more knowledgeable about something than me.
7:29And so in in in which case I have the best job in the entire world because I I can ask infinitely dumb questions and learn everything from people who really really know that stuff. It's like um what's the word? There's like a a word for like not uh being sad that you can't live like a thousand lifetimes or something like that. I forget. It's like saod or it's like one of those type of Yeah. It's probably a German word. Um, and it's like this is the perfect job to
7:57like quell that feeling because it's like I, you know, I can only live one lifetime, but when you meet like the best technologists in the world who are obsessed with a thing that other people need, like I mean that's just the beginning. Like seeing them giving them money and a network and trying to help them like that's very rewarding. It's super fun. So now like in these office hours what advice are you giving to the founders and especially contrasted to
8:25two years ago? So watch what is different from two years ago in terms of the advice that you're giving to the founders.
8:31Remarkably it's like all the same. I mean uh you know what's funny is Paul Graham comes and speaks to every batch and then the the number one question everyone in the batch always asks is like what advice did you write about that like you know changed because AI is here and he's always like nothing. Like it's all the same. There's nothing changed. Like you still have to understand your user. You have to actually talk to them. You have to create, you know, an initial version of it. You have to put it in front of them.
8:59And it's like, hey, did it work? Did it not work? Like it's the fundamentals are all the same. Um like my favorite part increasingly, uh I think you know you guys do this really really well. It's like the meme marketing part, right?
9:11like being able to uh have brain you know get brainworms into people's heads so that like when they have that problem they remember oh like I need to use superbase right like that's actually really important like that's one of the things that uh brought everyone here it's like well there are lots of ways to store data but I just know you know this is the shelling point like everyone on the internet knows like this is the way to do it actually and then what about during the batch I guess mostly on the software side Um,
9:41are you finding people a big part of our batch is like finding product market fit or every YC batch? Are people able to find product market fit faster now because of the models?
9:52Yeah, that's happening way faster. Um, I mean that's the thing. You used to have to sit there and code the thing whereas now you can open I use Cappy.ai now for all of my agent coding is like you know you can open 10 windows and just like any speculative thing. It's like I want to do this like let me do another project over here. You I mean you can basically have in parallel like 10 branches of like all what are you building?
10:19Oh uh I mean let's see. So ear in January I built uh garyslist.org which is um me attempting to recreate Posterous uh but like as sort of a news site and like a community site. Um, and then I spent a lot of time like learning about retrieval and like I it ingested like my entire X feed and all I I was kind of focused on politics and so you know all of the uh white papers, all the articles ever written about like you
10:47know pretty gnarly things like the uh you know the d drug crisis uh down in Tenderloin like you know who are the players, what's the non you know which nonprofits did what, what are the policies like what are the um what's the research behind behind it. And so it was sort of this uh agentic rag project. Um and then I built that and then what's funny is uh then I discovered OpenClaw.
11:11And so I had written uh what I call um a Foxcon factory for my agents, right? So I wrote, you know, half a million lines of Rails code using Opus 4.5 and 4.6.
11:23And um I treated LLM calls as the very I mean I don't know it was very expensive like at the time because I was like oh like I have to spend this much like you know you're you I tried to like chop up the use of intelligence and like control it in a really intense way kind of in the same way that a Foxcon factory worker probably wakes up does calisthenics like their lives are like um planned out to like the second and then uh I met Peter
11:51Steyberger uh he came by the YC office. He's friends with our partner, our newest general partner, um, Rafael Shad.
12:00Um, and I tried OpenClaw for the first time. I'm like, "Oh, I was making Foxcon factories where like deterministic Rails code uh, tries to control what the agent does." And like here's the input, here's the output, here's like a bunch of evals and like everything's all like locked into straight jackets. It's like I I have total control over what the LLM does. And then when I'm when I saw OpenClaw, I'm like, "Oh, this is like do less."
12:30Yeah. No, it's like an essin.
12:32It was like a bright open field. It was like free range intelligence is like the int, you know, you tell it what it want what you want and like it's able to code what it needs. And then I heard about Karpathy's LLM wiki and how he uh, you know, has a memory system. And so then I created uh I had created GStack as an attempt to understand like how to get the most out of um uh you know cloud code and agent coding. And it just was like it actually started as uh two
13:01markdown files that I would just find myself pasting into cloud code all the time. And it was like CEO review and uh review and it was like make sure that their tests were running and that it was like properly tested. And then Gbrain was like taking my Gary's list uh agentic uh retrieval stuff and then making it work inside OpenClaw. And then suddenly like that took a life of its own because what I realized was um you
13:29know we already have a form of super intelligence in the working memory like our whole society like if you think about every um standup or every status meeting or every exec meeting that you ever have it's built around like squishy humans who can only keep uh seven plus or minus two things in your head. That's why phone numbers are like the number of digits they are, right? So, you know, like we're kind of limited as human beings. Like I accept that now, but uh
13:58you know, at the same time, an LLM can have a million tokens, which is like I like to say uh three Harry Potter books that it's like totally usable like you know, you can generally like with 95% accuracy pull out like any part of like three books of Harry Potter. Um but it's not a library. It's not like if you could imagine a room uh of this size as a database and it's full full of books, right? So what's the bridge between what
14:27LLM can do today and uh the amount of data that is actually something that represents your life actually represents your company? Uh and then to me like I actually think there's something very elegant and interesting about uh markdown files and the LLM wiki because like you don't you know you don't the the agents just like speak it inherently right um and by default it uses GP but it turns out if you add you know graphs if you add um additional I mean
14:57basically a a full rag uh you know graph rag sort of framework on top of it and you have uh dream cycles and all of these other like syntactic sugar things that make it better. Um, that's what I want Gbrain to be. It's like, and so I use an open claw with Gbrain. It has half a million markdown files. That is literally like my entire life. And uh, it makes my life really simple. It's like I, you know, I don't I generally don't even have to open Slack and like click through anymore. Like the second
15:26someone like if there's a crisis at one of my companies, it's like, "Oh, hey, most important thing you need to do right now, it just like, you know, tell it telegrams me right away and I can respond to it within like 30 seconds."
15:37Like I, you know, I'm going to walk off stage here and if anything happened, like I'll be able to respond right away instead of like, "Oh, next time I look at Slack, like, you know, I'll eventually get to that DM, right?"
15:48And it's open source.
15:50Sorry. And it's and it's full open source and it uses superbase.
15:54Actually, I I presented that in the keynote. I don't know if you remember, you messaged me and you're like, "Oh, edge functions cut out after 2 seconds and I can't do uh chunking and embedding." So, we just upgraded so you can do running because of that message.
16:10I'm going to get some uh some agents on that. [laughter] Yeah, perfect. Um, so yeah, that's that's the you know, I've been doing a lot of open source and it's like uh I didn't GStack has 130,000 stars now and then brain has 30,000. It's more popular than Superbase, I think.
16:28Okay. Actually, it's a company now as well, right?
16:31Oh, I made a hosted version of Gbrain called Gbrain.io. Just for, you know, I just had too many friends of mine who were like, I can't figure out how to get this going. and was like, "All right, let's just do a super simple hosted version for you."
16:46Okay, going back to one of your spicy comments that you said um on hiring people and that you wouldn't need to.
16:55Actually, I'm curious then how did how should a founder they're getting started today, maybe two founders, co-founders as is classic YC model. When would you think about hiring now compared to in the past? What type of profile is the best first hire? If AI is going to solve a lot of these things that generally a couple of hands would do, how should people think about hiring just in general?
17:24I mean, I think I I I love hiring people who have actually uh like you've you've actually known. I mean, that's definitely the best first 10 hires even.
17:35I mean the classic view would be uh try to take as little like team risk as possible in some sense you know like the first 10 people determine everything about your company. Um and so you know what what I wish I did with for my startup for instance was like um I had just worked at Palunteer. I started my YC startup in 2008 and I was too afraid to go and uh get my old team to come work with me at my new startup. Um, and
18:05that was a mistake. Like I, you know, my startup ended up being fine, but it would have been 10 times more successful if I was very aggressive about going to the people who I knew were awesome. And it's like, look, you should probably come work with me cuz we're going to go change the world. And like I would have been able to have a 10 times more large impact if I had done that. So that's probably like the number one thing. It's like uh there you know if you're starting a company ideally you're one of
18:33the smartest people in your space and then game recognized game like you probably know these are the five people who is like if I could snap my fingers I need them like go get them actually that's probably the most important um and then aside from that it's like I think uh we're seeing this other thing that's interesting like the first wave of a startup I mean I think you can see this with um some of the data startups uh that sell to the AI labs for instance. So Meror for instance famously
19:01is um selling a lot of a lot of data to labs and they're a little bit like a labor uh labor marketplace where on it's almost like Airbnb for smart people to train the models. Um and then if you really look at what how they do it, they actually have like an army of people like I think they have thousands of employees also. Uh, and that's sort of like usually the first wave in like a lot of these businesses. And then, uh, at YC, we're seeing this wave of data
19:29companies that don't do it that way. Um, like they might have like I mean, they're able to do like similar amounts of revenue in increasingly, but uh, without having a thousand employees like they are actually automating their own tasks. They're actually writing software. And then not only that like because data itself um can be operated on using AI like they're building their own AI and agentic systems that uh
19:58provide better data. And so you have the same company selling to the same customers but instead of thousands of employees and having to raise uh billions of dollars you have companies that are 10 or 100 times more efficient that only have 20 or 50 employees and a thousand agents. And so you you know in sort of real terms like that I think is one of the evolutions that's happening period. It's like you so often you do need to hire but the ideal thing is like
20:28you're hiring someone who understands agentic tools and loops and graphs and you know and and you know even if they haven't done it like they're willing to learn like hiring that person is worth hundreds of people um in any other context very much like systems thinkers things like that people who are willing to adopt the tools for thinking.
20:48Yeah. We also talked backstage um because you were talking about these companies getting to these revenue numbers. We were talking about the VC stage like how there's a bit of a pullback in the VC market but it's actually a good thing for startups.
21:03Yeah, I think it's great. I mean uh you know at demo day routinely the hard tech companies are raising 5 to10 million which is like a little bit unpre you know that the median amount that someone raises at YC is usually someplace between uh 2 million to $3 million. Uh and so the hard techch ones are very interesting and that not only like I mean this was not true 3 years ago. 3 years ago like the hard tech things were oddities. It was very very hard to get people to fund them. I think you know
21:32we're just in a different moment like SpaceX is showing that actually you know I mean Starlink is an incredible product and like you know the ability you like it is possible for very very smart people to have uh to to build these really dominant um hardware companies like Andreil is another fantastic example um and so yeah that I mean that I think we're at an interesting moment for all this stuff but You're
22:01also saying that they're pulling back from some of the software maybe the software businesses but they're it's actually great for them because I mean I think that that's why what's happening like um I think my uh my old VC firm Initialize was one of the first like you one of the first things that Brett did at Initialize was you know I they didn't like pull back software entirely. They still do a lot of great software, but you know, when I left it, it was like a like 95% software. And then he like he turned a switch and he's
22:30like, actually, it should probably be 50%. Because um Agentic, you know, agents are coming and uh software that is like just in time is coming. And you know, I I see it like that's to me like uh that's even open claw for me. It's like almost any piece of software that I need, I can get it just in time. like any analysis that I need like it will just write the minimal amount of TypeScript and write some markdown files and then I'll be able to just use that
23:00over and over again. Like we had a hackathon for GStack Gbrain this weekend. Um, I was able to like kind of feed in uh the last hackathon I did, the last ratings and rankings that I did by hand and then OpenClaw is able to take that and then create um both Typescript and Markdown that replicates what I did phys, you know, sort of manually, right?
23:23And I still want to try all the products and talk to the people who made those things at the end. But now I can do it by and now when I do that the five or 10 people I end up talking to are all really awesome. They're all prevetted by this markdown file and you know I should probably open I don't know I'm it's it's a debate whether to open source it because then maybe that ends up being a thing that people start like you know it's a what is it God Godwin's launch something it's like the second you
23:52release a measure then it makes the measure totally useless right so anyway that that's an interesting moment right like uh software is going to be on tap and it already is like and you don't need open claw necessarily for it you can do like all my friends do all the same things that I do in OpenClaw now with codeex now with clawed code like it sort of doesn't matter what harness you use anymore and um you what does that mean like it it means that people are
24:21worried about SAS they're worried that SAS is going to go away personally I think that's totally ridiculous and I think that the VC world is um it's like in a bit of a mania right now like you know you can talk to some VCs and they like behind behind closed doors will straight up just say like we just want to fund Neo uh Neolabs, right? It's like I just want to fund things at like a billion to5 billion valuation and they have to have worked at OpenAI or anthropic.
24:51Yeah. And um I mean and I get it like if you guys um go on Twitter there are like tweets out there about how uh if you look at the amount of value created by open anthropic in their coming IPOs um it is equivalent to like all of the returns from venture capital in the history of venture capital right um but to me I'm like that's great like that's abundance right there like you know yes do Neols but also I think um you know
25:21the white pill for SAS if you're working on SAS in this audience is that like we have YC companies that are going from 0 to 20, 0 to 50, 0 to 100 million in revenue, but they're 50 people, 30 people, 15 people. And so, which is like very unprecedented to us. Like you used to have to have $100 million to get to like half a billion dollars. And you know, you had to raise hundreds of millions of dollars to hire the best
25:49executives and to build out the team to just be in the running to be one of these mega SAS companies. And so, you know, I I I'm sure people remember like the the one person trillion, you know, billion dollar company meme. Like, I don't think it's going to be one person, but I think it'll be like 20 people. I think it'll be 50 people. like you know what uh Instagram had when it got sold, whatsApp had when it's when it sold like I think that that will become
26:18WhatsApp was 55 people I think it was when it sold for a billion and I think that's going to be like increasingly the norm and that's if you're a founder and you're working in SAS like this is amazing like the YC companies are out there. They're getting this revenue and they're continuing to grow. They're just you know basically writing more agent software. They're writing more loops. They're getting better at their markdown files in Typescript. Um, and then, uh, they're financing themselves off of revenue.
26:47Like, they're they're profitable, right?
26:50You know, I I I just had an office hours just now with a company that got to $15 million in revenue. It's two two 19 year olds.
26:58Yeah, it's amazing. Gary, I could honestly ask you questions for another hour, but we've got to wrap up. So, thank you very much, Gary, for coming to Select.
