Uncapped with Jack Altman
· Garry Tan
[00:00:00] S3: I think it’s dead. The thing is, if you run a SaaS company, you don’t have to be dead. All you have to do is embrace Claude Code. You have to embrace, top to bottom, an agentic view of how everything is going to work. Put it this way. The same week that I personally realized everything was different, I funded a team from Meta Superintelligence who had left. They were pointing out that Meta has 20,000 people working on Reality Labs and Alexa has 20,000 people. I thought about my experience. I didn’t even have 20,000 people. I had five people. Why did it take two years? Because I knew what the architecture was, I knew what I wanted to build, and then I had to farm this out. I had to have meetings, come up with a doc, and then other people have other opinions. We’d have five meetings about the architecture and argue about it. Two weeks later, maybe something happens. If you’re in a big company, it’s three months later and maybe something happens. But we don’t have to do that now. We could just try both. Go into plan mode and then just do it. Literally an hour later, we will have something done that would have taken two weeks, or two months, or sometimes two years. If you’re not a tech company and you’re an incumbent, it’s two years or never that you would even make that decision. The speed of making that decision, how decisive you can be… Going back to the transformation thing, that’s actually what I feel like I learned at YC. It’s not that I didn’t know how to do it. I was employee number 10 at Palantir. We were moving fast. I was sleeping at the office. The big difference was realizing that instead of getting 20 basis points of Palantir… That’s now an astronomical amount of money, by the way. We didn’t know it at the time.
· Jack Altman, Harj Taggar, Garry Tan
[00:00:00] S0: We’ll run some math. We’ll put the math up on screen. That’s a thumbnail. That’s perfect. [00:00:00] S2: Right above Garry. [00:00:00] S3: Basically I wanted 97% of the company that I started. I went from a place that was already fast. Then being the founder and CEO, YC sped me up even more. Because you’re in office hours with people and it’s like, “Oh man, this person actually grew 10% this week, 20% this week. How did they do it? I need to do it.” I think all of this is an accelerant. Claude Code and Codex, being able to make two years of product progress in about two weeks, how could that not make YC more insane? The amount of things you could try and do. You could do two years of work, realize that nobody wants it or there’s too much competition, throw it out, and do it again. That’s not a throwaway. You learn something, you also got better at using these tools, and you get another shot. What’s funny about seed is you could think of people raising two or three million dollars. That used to be a Series A by the way, which is hilarious. [00:00:00] S0: Now it’s barely a seed. [00:00:00] S3: Yeah, it’s like “oh, that’s a small seed.” Are you serious? [00:00:00] S0: It’s a pre-seed. [00:00:00] S3: That’s so much money. This is outrageous. You don’t even need this money. It’s crazy that you can go so much faster now.
· Garry Tan, Jack Altman, Harj Taggar
[00:00:00] S3: That’s so much money. This is outrageous. You don’t even need this money. It’s crazy that you can go so much faster now. [00:00:00] S0: I actually want to come back to fundraising advice because I feel like it’s gotten into an interesting place. But the opposite question of the “is SaaS dead” question is, what do you feel is not AI but safe from AI? Are there areas that you think are insulated from this mega-trend, that you’re happy to back without fear of that? [00:00:00] S2: I think the obvious one first is just marketplaces that are aggregating people. I think Airbnb is very safe. I think there’s a bunch of marketplaces. DoorDash is totally safe. That’s a very clear one. [00:00:00] S3: One thing we’ve been talking about is what the agents want. What the coding agents tell you to do turns out to be itself a really big moat. GEO, making your API docs actually written to prompt inject Claude Code to force it to use you… I’m just joking, it’s not clear that you can do that. But if you could you would because it’s that powerful. People will just say “I need X” and then “what’s the best thing on the internet to do that?” I just think that’s really powerful.
· Garry Tan, Harj Taggar, Jack Altman
[00:00:00] S3: One thing we’ve been talking about is what the agents want. What the coding agents tell you to do turns out to be itself a really big moat. GEO, making your API docs actually written to prompt inject Claude Code to force it to use you… I’m just joking, it’s not clear that you can do that. But if you could you would because it’s that powerful. People will just say “I need X” and then “what’s the best thing on the internet to do that?” I just think that’s really powerful. [00:00:00] S2: The other thing you might say, going back to this SaaS thing, is even within SaaS you might say things that feel like they are essentially databases or systems of record — things like Rippling — feel like they’re going to be in a good spot. Then things where the moat was around the number of integrations built, or data connectors, or that kind of stuff — which you can now just code up in 10 or 30 minutes — that’s brittle. [00:00:00] S0: Do you think it’s system-of-record that makes payroll sticky, or do you think it’s touching money? [00:00:00] S3: Touching money. Regulatory. You have systems that are working, you don’t want to touch that unless you have a really good reason. [00:00:00] S0: Because one of the ones that comes up a lot right now is, are CRMs safe? I don’t have a particularly strong opinion, but it seems like an interesting question. It doesn’t exactly touch money, doesn’t exactly touch regulatory . It includes a lot of important information, but that information can now live in email or somewhere else.
· Jack Altman, Harj Taggar, Garry Tan
[00:00:00] S0: Because one of the ones that comes up a lot right now is, are CRMs safe? I don’t have a particularly strong opinion, but it seems like an interesting question. It doesn’t exactly touch money, doesn’t exactly touch regulatory . It includes a lot of important information, but that information can now live in email or somewhere else. [00:00:00] S2: I think Salesforce is probably screwed. There have been so many attempts to do the Stripe strategy. Get startups in a YC batch using your CRM, because everyone hates Salesforce. But no one could ever really grow into a big company because at some point your head of sales says, “No, I need these reports,” [00:00:00] S0: And to your point, you need all those integrations, badly. [00:00:00] S2: Exactly. Now that’s all just going away. I suspect the next Salesforce is going to come out of a YC batch that sells to all the other YC startups. Investors will say, “Oh, it’s not going to grow because no one can compete with Salesforce.” [00:00:00] S0: Also, “all its customers are YC companies. It’ll never work.” What about hardware? Is hard tech safe? [00:00:00] S3: Hard tech is just hard. The moat comes from it being hard to source, hard to make it work. It’s just another muscle. [00:00:00] S0: It’s atoms. AI hasn’t exactly come for atoms yet. [00:00:00] S3: Robotics is still a little ways off, but we need ASI. I think we have AGI now, and ASI is coming. Superintelligence is clearly just around the corner.
· Garry Tan, Jack Altman
[00:00:00] S3: Robotics is still a little ways off, but we need ASI. I think we have AGI now, and ASI is coming. Superintelligence is clearly just around the corner. [00:00:00] S0: Do you think when we know we’ll know? Or do you think there’ll be a thing to you that would say that a new moment is here? [00:00:00] S3: I think we have limited versions of ASI right now. I saw a demo— [00:00:00] S0: All the bots talking together? That was crazy. [00:00:00] S3: Yeah, that was crazy. That’s the first real example of swarm intelligence. In AI research, swarm intelligence is like this huge field. This was a massive validation of that field. It’s kind of an interesting question. All the main AI labs talk are like, “We’re just going to build the God model, just make it mega-big.” Think about Dr. Manhattan or something. Biological systems and human society are not modeled that way. We have lots of people with lots of diverse hardware, and different opinions about all kinds of things. You come together and see what sticks. That’s what research is, for instance. Swarm intelligence versus god-level intelligence is a very interesting thing and that’s just the beginning of that. That just happened literally last week.