Pod Insights Lightcone Podcast 2024’s Biggest Startup Trends About this episode 2024 has been quite a year for AI and startups.
As we head into the holidays and the new year, the Lightcone hosts reflect on this year’s biggest startup trends, moments, and breakthroughs.
Transcript Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Garry Tan? The wildest thing right now is you can start a company that can make tens of millions of dollars literally in 24 months and you can do it for potentially 2 million, 5 million dollars.
0:12 Harj Taggar? A year ago, I remember many of the startups in the batch would get sort of enterprise proof of concepts or pilots in particular, and there was a lot of cynicism around whether any of those pilots would translate into real revenue. Fast forward a year, I think we have all firsthand experience that these pilots have turned into like real revenue.
0:29 Garry Tan? It's still early days, honestly. Like, you know, we sort of breathe a sigh of relief right now in 2024, but it's anyone's game, honestly. Like, these things are moving so quickly.
0:47 Garry Tan? Welcome back to another episode of The Light Cone. I'm Gary, this is Jared, Harge and Diana, and collectively, we funded companies worth hundreds of billions of dollars right at the beginning. So, 2024, what a year. How are you feeling about this, Harge?
1:06 Harj Taggar? Pretty great. I think this is the year that everything broke in favor of startups. What I've been thinking about a lot recently is when ChatGPT launched two years ago now, the immediate consensus view was all of the value would go to OpenAI. And very specifically, do you all remember when they announced the GPT or the ChatGPT store?
1:27 Garry Tan? Yeah.
1:28 Harj Taggar? Like, I remember the consensus was everything that was built on top of ChatGPT was a GPT wrapper and the App Store was just going to be released and crush every single person trying to build an AI application. And OpenAI would be a ginormous company, but there'd be no opportunity for startups. It sounds kind of ridiculous to say that now because...
1:48 Garry Tan? Who even remembers the ChatGPT store?
1:50 Harj Taggar? Exactly. The ChatGPT store itself was a nothing burger, but like, more importantly, what are the big AI applications today? Like, I'd say, outside of ChatGPT itself, the breakout consumer application is Perplexity. The breakout enterprise application is probably Glean, maybe. In legal tech, you have CaseTech, you have Harvey. Prosumer, you have photo room, like there's the point being there are many, many applications that have been built not by OpenAI. It's been a great time to build startups.
2:20 Garry Tan? Yeah, the wildest thing right now is you can start a company that can make tens of millions of dollars, literally in 24 months from zero. And you can do it for potentially, you know, $2 million, $5 million. That's sort of the story of one of these companies, Opus Clip, which never had to raise a real Series A. And that's something that we sort of see across the YC community as well.
2:45 Harj Taggar? Yeah, I think that's a particularly important point that you can do it as a startup without raising tons of capital. Because post the GPT store launch, I then remember Anthropic and Claude emerged. And the consensus view for a while was all of the value is going to go to one of these foundation model companies. And that the only way you can compete in AI is to raise huge amounts of money, either because you've got venture capital, or you're Amazon, or Facebook, or Google with tons of cash already. But that if you weren't one of the big foundation models, there would be no value. And the applications built on top of these things would either be built by the foundation model companies themselves, which is not be that valuable. Again, something that turned out to be completely not true,
3:30 Harj Taggar? right? And in particular, what drove that is open source, like the weird series of events where Meta, like the weights being leaked and like Meta just like rolling with
3:40 Diana Hu? it. Torrent?
3:40 Harj Taggar? Yeah.
3:41 Diana Hu? So I was going to force the hand for Meta to launch Llama, which was funny. And people thought, oh, it was just this cool open source model, but it was 18 months behind OpenAI. And people started doing a lot of derivative work out of it. It's like Vicuña and all these other animals related to Lamas that came out. And it took the, oh, Lamas, one of the companies at YC as well, that enabled people to do local kind of Docker development, like models running on device.
4:15 Diana Hu? It was pretty cool, but people didn't think that they were gonna be able to catch up. And the thing that changed from 2023 to 2024 is that during the summer, it was a turning point. It was the first time that the top foundation model in all the rankings benchmarks was Llama. And that was a shock to the community.
4:41 Garry Tan? Yeah, so it turns out choice matters and choice means that it's not as much about the model. I think the model still matters quite a lot. But once you have choice in model, it means you can't have the sort of idea of monopoly pricing. And you have that model, your competitor also has that model. But all the other things seem to end up mattering a lot more, which is product, your ability to sell, your ability to actually adjust to user feedback, your ability to get to zero churn. All of those suddenly become far more important than capturing a light cone of all future value through the model.
5:19 Harj Taggar? Right. A very specific way I felt this is I remember a year ago, working with startups in the batch that were essentially building model routers, just like an API to call like a specific model. And I remember the, a lot of the motivation for that at the time was reducing cost. It was like, oh, like you don't want to just like burn up all of your chat GPT calls. You want to spread them out across like various different models. And the, the argument against that was just, oh, like the cost of all this stuff is going down to zero anyway. Like there's no value to be had in being like a model router and no one wants to build their applications with a model router. They're all just going to call whatever is the best model. I think fast forward a year, like that's totally not true.
5:59 Harj Taggar? Like from what I can tell, the, the model router was actually a really great entry point into just building sort of a new stack for building LLM powered apps. And most of the, the applications we're seeing, I think they just don't want to be beholden to a specific model. Does that map with what you've seen?
6:16 Diana Hu? Yeah, actually, one of the things we've seen now in the fall batch that just presented at Demo Day, which was one of the trends that shifted from summer to And winter twenty four was precisely what you're saying. Companies started to use multiple models for the applications, like the best one for speed at some point, because sometimes you need to parse a lot of the input very quickly. It's fine if it's a bit more lossy. And then you need the bigger model to handle the more complex task. So a lot of companies in fall twenty four have this actually multiple model architecture to use the best one for the best task, which is similar to the concept of a model router, but it was the idea evolved.
6:58 Diana Hu? Instead of being more of a routing, it was more of an orchestration. I think a concrete example we gave a couple of episodes ago was Canfor. It was a company you work with. They use the fastest model for parsing like PDFs and the more complex ones. They use O one and that's that's how it's done. And other companies that's doing fraud detection, they have this concept of like a junior risk analyst where they just use like a fast and easy G P T four mini, and then they use the bigger one with like O one. Or the other example is, I think Cursor talks about it in their episode with Lex Friedman.
7:39 Diana Hu? They also have this complex multi-architecture with multiple models. And this is why it works well is like they do one very specific for predicting what you're going to type next, but one for understanding the whole code base. So very different tasks. So that's definitely happening now.
7:54 Garry Tan? Yeah, the other thing that popped up for Fall Batch, there's a company I'm working with called Variant. And what they're trying to do is take basically state of the art, open source L L M models that can do code gen and then teach them aesthetics. So starting with icon generation. And so they built this huge sort of post-training workflow that should work on, you know, as the open source models get smarter and better at code gen broadly, they can just, you know, take the next version of that and then take their post-training architecture and data set and then basically teach a given model aesthetics.
8:35 Garry Tan? So what a certain thing is supposed to look like and not in a diffusion sort of way, but actually at the SVG level. And we think SVG will actually translate into all kinds of aesthetics. So it's an interesting approach and like one of the newer ones in that post-training is a whole coherent way to sort of skip the whole idea that all of the value is like accruing into the model, especially because of open source, to your point.
9:03 Harj Taggar? The other thing I I've been having flashbacks to is a year ago, I remember many of the startups in the batch would get sort of enterprise proof of concepts or pilots in particular. And there was a lot of cynicism around whether any of those pilots would translate into real revenue. Lots of parallels to crypto and how anytime there's some new interesting technology, well blockchain more specifically than crypto, but anytime there's a new technology, enterprises always want to run pilots and POCs because it's someone's job to like check off, yeah, we did the like hot new technology thing.
9:34 Garry Tan? The chief innovation officer must have his due.
9:37 Harj Taggar? We've spoken about this one of our one of our episodes, I think. And yeah, and fast forward a year, I think we have all firsthand experience that these pilots have turned into like real revenue. And if anything, the startups in the YC batch now are going to sell into real enterprises faster than they have before and are ramping up revenue and reaching milestones like a million dollars ARR faster than I've certainly ever seen.
10:01 Garry Tan? Yeah, the fall batch just did this actually again, which is actually the first time I think we noticed it was actually the summer batch of this year. And one of the funnier things that we realized was, do you remember when Paul Graham would tell us how fast you needed to grow during the YC batch?
10:16 Harj Taggar? 10% a week.
10:17 Garry Tan? 10% a week. And the wild thing is in aggregate across both, you know, summer and full batches, that's what those batches did.
10:25 Harj Taggar? Wow.
10:25 Garry Tan? So which I don't think ever happened.
10:27 Harj Taggar? Three X over the course of YC.
10:29 Garry Tan? Yeah, three X over the course of YC, which I don't think has ever actually happened.
10:33 Diana Hu? On average.
10:34 Garry Tan? On average.
10:34 Diana Hu? It was only the best companies that did that, which is the top quartile or something, right?
10:38 Garry Tan? So the companies are better.
10:40 Harj Taggar? The general thing that is true is that the time it's taking to reach a hundred million dollars in annual revenue is trending down.
10:47 Garry Tan? Yeah, and not only that, we had dinner with Ben Horowitz recently and remember he was saying when they started Andreessen Horowitz, the common understanding was that in any given year, there'd only be 15 companies that year that would even make it to a hundred million dollars a year revenue. And they said they ran the numbers the last 20 years and every decade, the number of companies that could actually make it to a hundred million dollars went up by 10x. So what was 15 per year, maybe 20 years ago, I mean, we're talking about 1,500 companies a year that have a real shot at actually making that
11:28 Garry Tan? number. And when you combine that with what we're seeing in the summer and fall batches, it's not that surprising.
11:36 Diana Hu? And Jared had a really good argument on our last episode about how vertical AI is gonna enable this to have 1,500 plus companies to bloom.
11:47 Jared Friedman? Yeah, that's why it's growing so fast. It's because the value prop to companies of these products is so incredibly strong that like, they're just flying off the shelves. Because like, companies are smart and they can do an ROI calculation. And when the ROI is fantastic, all these, all these truisms that people believe about enterprise sale cycles and like, how it takes to get big enterprise deals go out the window because companies are smart and they'll make rational decisions. You know, Harj, there's another way that this broke in favor of startups that I was thinking about. It's hard to even remember it now, but a year ago, one of the things that people said a lot was that these LLMs are not reliable enough to deploy in the enterprise. They hallucinate.
12:31 Harj Taggar? Yeah, that was why a lot of people said like, these pilots and POCs like, won't translate into real contracts is because, yeah, it's too risky of a technology for people to actually deploy.
12:41 Jared Friedman? Yeah. And not only is it translating into real revenue, but it's translating into real deployments that are like, being used at large scale, you know, doing thousands.
12:53 Jared Friedman? Of tickets a day and I think it's because we've learnt how to make the agents reliable via the kinds of techniques that Jake talked about when he was here and just the all this infrastructure has grown up around around the models that's enabled people to make them reliable.
13:07 Diana Hu? Yeah, that's actually a big trend. This year is this concept of thinking of AI more as agentic. That is a term that kind of bubbled up a lot this year. It was not in the bubble space of conversation last year. Last year was more about a lot of things that were kind of very chat like. Chat GPT, I mean, that was kind of the riff on it, but now it remixed into a bunch of agents for XYZ.
13:32 Harj Taggar? And we just, I mean, you just put out Gary's put out a great explainer video about computer use from Claude, but just the capability of the models keeps pushing in the direction of just being able to do like complex multi-step things and actually take over your computer and call other applications and and perform complex tasks that just didn't seem possible a year ago.
13:53 Garry Tan? What about regulations? Seems like we sort of dodged a bullet there with 1047 and it looks like some of the Biden EO is not that likely to survive the Trump White House. TBD what that means in the longer term, but certainly one of the things that we were very worried about was that some certain amount of math beyond a certain level would suddenly become illegal or require registration at your local office.
14:21 Harj Taggar? It's certainly been a weird time to be in tech because I've never experienced software and technology intersecting with politics so much. And in particular, I'm not used to genuinely caring about national politics affecting startups in a YC batch or just, you know, companies that are less than a year old. But it did really, it did really for a moment was worrying. It wasn't clear whether startups would actually be able to build innovative AI applications versus suffering from regulatory capture from OpenAI and a few big players.
14:53 Harj Taggar? We're obviously very glad it broke in the favor of startups.
14:56 Garry Tan? Seems like we're still in the early game, right? I mean, it's very easy to see that the platforms themselves really will or could possibly resemble, you know, the Win32 monopoly, right? Windows has access to the APIs. They in fact know all the stats about what's working on their platforms. And guess what? They can build it into their platform. You know, we sort of breathe a sigh of relief right now in 2024, but you know, it's anyone's game, honestly. Like, these things are moving so quickly. I wouldn't totally breathe your last sigh of relief yet, you know? We got to keep working on this.
15:36 Harj Taggar? Okay, so it's clearly been a great year for startups. What else has been happening? Who else has it been a great year for, do we think? There's certainly been some big funding rounds, right? Like, OpenAI unsurprisingly has raised huge amounts of capital.
15:50 Diana Hu? Scale.
15:51 Harj Taggar? Yeah, even within YC, though, we've seen like Scale AI has really broken out this year.
15:56 Garry Tan? Six billion dollars for OpenAI, one billion dollars for Scale, a billion dollars for SSI, the new Ilya Sutskever startup.
16:06 Harj Taggar? Scale, I think, is just worth talking about because it's such a classic startup story. I mean, you were there in the early days, right? You interviewed them for YC.
16:15 Harj Taggar? Tell us what the idea was that they interviewed with and how they ended up landing on what is probably one of the best startup ideas of the last 10 years.
16:23 Jared Friedman? The fun thing about the scale.ai story is that it is the sort of epitome of the like classic YC startup story. There's other kinds of startups that get started, you know, like SSI, for example. That's not a typical YC startup story where like some very well-established people raise a billion dollars with like a PowerPoint pitch. But like scale.ai is like the classic story of how like young programmers can just gradually build a like $10 billion company over time by being like smarter and harder working than anybody else. And so, yeah, when Alex interviewed at YC, he wasn't working on anything related to AI. It was a completely different idea. And the idea for Kind of got pulled out of him by the market.
17:05 Jared Friedman? And it's it's actually still like several pivots because like the the original idea at Y C didn't have anything to do with A I. And then for a long time, he was basically doing data labeling for the self driving car companies.
17:17 Harj Taggar? They applied, as I remember, they applied with like a healthcare related idea.
17:20 Jared Friedman? Yeah, it was a website for booking doctors appointments.
17:22 Harj Taggar? Okay, cool. And then they pivoted during the batch. Do you remember how they came up with the data labeling idea? Because this must have been, was this two thousand sixteen?
17:32 Jared Friedman? Yeah, the way they came up with the data labeling idea was that Alex had worked at Quora and Quora had to do some data labeling for like moderation and stuff. And so at the time, the big data labeling service was Amazon Mechanical Turk. And they were deemed unbeatable because they were like run by Amazon and Amazon could throw infinite money at it. It was always at scale, it was at like, like quite large scale already. But Alex had a unique insight, which is he'd actually used Mechanical Turk at Quora and he knew that it kind of sucked to actually use it. And so he he had this sort of like unique insight on the world. And so he just tried to build a better Mechanical Turk, basically the version he would have wanted when he was at Quora.
18:09 Harj Taggar? And as I remember it, like really they their early traction came almost entirely from one customer, Cruise.
18:16 Jared Friedman? Yes, which needed to do tons of data labeling on all the images that the cars were taking as they were driving around San Francisco. You've got to like draw a circle around the traffic light and thing like that. And the the cool thing about scale is that they've actually caught two waves. So they, you know, accidentally caught the first wave of all these self driving car companies. Because M L took off at that time in computer vision, there was just an unprecedented demand for labeled data for training sets that just hadn't existed before. And so they were able to ride that wave. And then as that wave was like cresting, L L M's got big. And all of these companies needed to do R L H F at very large scale.
18:57 Jared Friedman? And scale was just like perfectly positioned to move into that business as well.
19:02 Harj Taggar? Yeah, I think the scale story is just so interesting because it was pre L L M. It was it was clearly a multi billion dollar business anyway. And L L M, it caught the L L M wave, which is now propelled into probably it's going to be like a hundred billion dollar plus company. And I'm seeing that at the ground level too, where many companies I had that maybe finished the batch even pre the batch didn't have an idea. Pivoted into an A I idea that's taking off. Like, I'm just seeing much more success in founders who waited out and.
19:31 Harj Taggar? Can find an idea that they just couldn't before. I have a company from a year ago. They pivoted, they pivoted the whole batch. They couldn't find a great idea. It actually took them six months after the batch until they realized one of their parents ran a dentist office. So he just decided to go hang out at the office to see like if there was anything he could automate. And they just ended up building an AI back office for dentist offices. And now it's just like the week of a week growth is fantastic. It's, it's doing really, really well. And I'm seeing lots of cases like that spring up.
20:02 Diana Hu? Definitely seeing that as well. I think there's something about the advantage of having all these very hardcore, young technical founders that are willing to kind of just bet the farm and go all in on just a little bit of a glimmer of, oh, this is where the future is going to be. Let me just try it. And then it actually ends up working. Like your story with the dentists, I have a lot of teams that pivoted as well into different spaces where they kind of found that glimmers like, oh, computer use came out. And I have a couple of companies that are working and betting and going in that direction. And it's like, working well. I mean, it's still early. I mean, this is just the fall batch, but that's cool, too.
20:37 Harj Taggar? Okay, so what are some of the, the trends that we've seen? Or some of the, the specific trends and waves that startups have been riding coming out of the batches? Voice AI is something we've talked about. It's clearly maybe the most promising vertical for AI right now in terms of just raw traction.
20:55 Garry Tan? Do you think voice is a winner take all? Or will it be something that has sort of a hundred different verticals that are very tailored to those specific verticals? That's literally one of the questions I get from some of our voice AI startups themselves. They're like, should I be going horizontal? Or should I just continue to grow within my vertical?
21:19 Harj Taggar? It feels to me like voice itself, voices.
21:24 Harj Taggar? And just like AI, it touches everything and there's so many different applications for it that you can, there's probably infinite applications to build where voice is the interesting element of it. I mean, things that just spring off top of your head, like language learning applications. I'm sure there's not going to be just one really cool voice AI powered language learning applications. Probably going to be multiple of them. Remote work, like teleconferencing, probably like a whole other area where there's interesting things to do with voice AI.
21:54 Diana Hu? And even within customer support, we highlighted a number of companies we talked about last time, company PowerHelp, capa.ai.
22:02 Jared Friedman? Yeah, it turns out that customer support is not really one vertical. There's like many different flavors of customer support and there's like very different on the inside once you get into the details.
22:11 Diana Hu? Because I think there's very specific types of workflows you need to do per industry and that's to the point of why vertical AI agents are going to really flourish. I mean, same thing for voice, it's just very different workflows. If you're building the, I don't know, the voice agent to do customer support for our airline, very different than doing it for a bank, very different than doing it for a B2B SaaS company, et cetera.
22:32 Garry Tan? Yeah, I guess that question of is there going to be pure horizontal integration is sort of like saying, will there only be one website?
22:41 Harj Taggar? Yeah, it'd be like saying, like, there's just going to be both. There'll be horizontal infrastructure companies that do really well and vertical applications. Because to say otherwise would be like saying, oh, like Stripe powers payments on the internet and it's also just going to have all the most valuable applications that accept payments on the internet. It's just not how it works. Like, there's enough value at just being the horizontal infrastructure layer. So I'm sure there'll be great voice AI companies that just make it really easy for you to build your own voice AI application while there'll also be hundreds of really valuable vertical apps.
23:11 Diana Hu? What are other trends that we've seen besides voice?
23:15 Harj Taggar? We were talking about robotics earlier. Certainly, we are certainly working with more founders building robots this year than I think any year.
23:24 Jared Friedman? Ever.
23:24 Harj Taggar? Ever. What's driving that?
23:27 Garry Tan? I have an ex-Apple team that's called Weave Robotics, that they're going to try to ship a real robot in 2025. It costs about sixty-five, seventy thousand dollars. But that's actually what it costs to have the actuators and the safety needed to actually have it work in your home. I think it's actually driven by this idea that The LLM itself can be sort of the consciousness of the robot. Like, am I doing this thing that, you know, my owner needs me to do? How do I actually interact with them and the other people in the household? But it's funny, because then the voice-language-action model that might actually do a certain thing,
24:07 Garry Tan? like fold laundry, that's almost tool use inside of the broader LLM consciousness. So I feel like that's one of the things that I'm excited to see. You know, will it really work? And I think we're going to find out this year.
24:23 Diana Hu? I guess the way I think about it, robotics is basically half AI and half hardware. Half of the part of the equation is starting to work.
24:31 Garry Tan? Well, the hardware is still hard. The hardware is still very expensive. Yeah, there's still, there's some evidence that being able to actually do laundry, for instance, like, that might be one of the first things that gets shipped.
24:42 Harj Taggar? I think the dream case for startups is going to be that you can build just the AI or the software piece of it and run it on commodity hardware and do really great things. The opposite case would just be actually if the two things, that if you need to be good at the hardware and the software and they're like coupled together and you need to produce both, then you would expect Tesla to be the obvious winner in the space. And it remains to be seen. I'm pretty optimistic. We have multiple companies, I feel, that are trying to be creative on how to run the models on commodity hardware for specific use cases.
25:17 Jared Friedman? It still feels early. It feels like the robotics hasn't quite hit its, like, chat GPT moment yet.
25:22 Diana Hu? Maybe the moment is self-driving cars have been working in San Francisco. I don't think it's talked enough.
25:29 Jared Friedman? People who don't live in San Francisco or like often don't realize the extent to which these are fully deployed in San Francisco and regular people are writing them every single day.
25:37 Harj Taggar? Yep. I saw Tony from Dordash recently, and he said he exclusively uses Waymo, like, everywhere. I live in Palo Alto and I have no option for it, but I'd love to. It'd be amazing.
25:47 Garry Tan? I mean, the wild thing is there are only a few thousand of these deployed right now in the entire world. And how lucky is it? Yeah. What about big flops for 2024? I seem to remember that we started one of our Light Cone episodes all wearing Apple Vision Pros and Quests. And we have not talked about AR since. Diana, what happened?
26:13 Diana Hu? It hasn't happened. There's this moment for a lot of the hardware that needs to be a lot more lightweight. Like, we need to get to this form factor. But there's actually constraints with physics to fit all that hardware in such a small form factor. And in order to have enough compute and the optics to fit, it's just super challenging. And I think there's still more actual engineering and physics that needs to be discovered. And that's it. I think the algorithms are there, but it's just lots of really hard hardware and optics problems.
26:45 Jared Friedman? It's a tough chicken and egg problem because there's not enough hardware in people's hands for it to be worth it for app developers to build apps. And so there's not enough apps for people to want to buy the hardware.
26:56 Harj Taggar? And I feel like the people who did buy, like, the killer application so far seems to be using it as
27:03 Harj Taggar? a really large monitor.
27:06 Garry Tan? And it does work very well for that.
27:07 Jared Friedman? Yeah, you've actually retained this as a user, Gary, right?
27:11 Garry Tan? Yeah, it's great for watching the movies.
27:14 Diana Hu? Maybe the one device that I think actually been playing and actually feels good is actually the Meta Ray Barn. It doesn't have any of the actual displays, but I really like it for the audio and voice. And one workflow I've been trying out is actually using the Meta Ray Barn and connect it to any of the voice modes for either ChatGPT or Claude and kind of have a conversation with it about a topic.
27:39 Harj Taggar? Oh, I haven't tried that. That's an interesting idea. Yeah, that's a great idea.
27:43 Diana Hu? That is like a fun thing that I've been doing and just chatting with myself. Maybe I look a little bit like a crazy person while you're walking, but it's been fun to kind of learn about different topics.
27:52 Jared Friedman? Should we talk about AI coding?
27:54 Diana Hu? Oh, yeah.
27:55 Jared Friedman? 2024 was the year that AI coding really broke out.
27:57 Harj Taggar? Yep.
27:58 Jared Friedman? I mean, we had the majority of of YC founders now use Cursor or other AI IDEs. They just like exploded over the summer. Devin proved that you could like fully automate like large programming tasks. Yeah, all that was this year. That's pretty wild.
28:13 Harj Taggar? Repli agents continue to improve. Like, I hear more anecdotal stories of people building Repli apps on like their way home from work.
28:23 Jared Friedman? Yeah.
28:23 Harj Taggar? Being really impressed.
28:24 Jared Friedman? Like, Repli took this technology and popularized it among like non-technical people for the first time. That's really crazy.
28:32 Diana Hu? And even more lower technical version is Anthropic's.
28:38 Diana Hu? Artifact where you can actually prototype very simple apps and chat with Claude to build really simple front pages. And then you could prototype stuff as a PM and show it to your engineering team and it's like a full-fledged working version.
28:51 Garry Tan? Yeah, it's wild because it just means that you can, one person can do so much more. And do you think it's going to change the nature of how startups are actually hiring? Are you seeing this yet? Like, some of the founders I've met who recently raised their seed rounds coming out of YC, they're not really approaching it how maybe the classic advice would teach them. In the past, you might say, let me try to find, you know, more, let me try to hire more people. Like, you know, there are certain tasks that normally I have to find, you know, the person who did it at my competitor, who did all of customer success. And I need to find that person who's under the person who runs that function.
29:35 Garry Tan? And I've got to hire that person and promote them. And they're going to come with all this knowledge and people networks. Some people are saying sort of the opposite, which is, I'm going to get my software engineers to write more processes that use LLMs upfront. And, you know, I probably will end up needing to hire that person, but maybe after the series B or C and not right now.
30:00 Diana Hu? Yeah, I think I've seen that as well with companies after the batch where they're looking for engineers that have more upside and they're really fully native with setup with AI coding stack. And part of the, one of the clever interview tricks I've seen is people do pair programming and watch them use the tools. And you can really tell if someone really has tinkered with them. It's actually an engineer cut that is not only good at coding, but also prompting and telling with when the AI output is not correct. I think the part of reading and evaluating the output of all these AI coding agents is actually a lot more critical.
30:42 Jared Friedman? Yeah, there's been an interesting controversy this past year about AI coding agents and programming interviews because AI coding agents basically broke the standard programming interviews that companies have been doing for years. Actually, Harj, I'm curious what you think about this since you ran a programming interview company.
30:58 Harj Taggar? I mean, I guess the, the interesting debate is whether you should.
31:04 Harj Taggar? Penalize or prevent people who are interviewing at your company from using Cursor or one of these tools to ace your programming interview, or whether you should just lean into it and adapt and test to see how productive they are. I generally think the way these things tend to is more in that direction, that I think it will just become, you'll just be measured on your absolute output and the bar will go up. I think like Stripe, for example, were early on this about a decade or so ago, where they recognized that so much of what they needed their programmers to do would like build web application and web software and not do hard CS problems. And so the industry shifted away from the Google style interview
31:48 Harj Taggar? of lots of computer science problems and whiteboarding to just give someone a laptop and make them build like a to do app in like four hours. So I think we'll just see the same thing happen where people will just, the industry will just adjust and you'll just be interviewed using these tools and just be expected to do a lot more in like a two hour interview than you are today. To your point, Gary, around just the startups, like maybe how many people they need to hire or just like how they would scale. It seems too early to see like dramatic effects on that yet. But one thing that I'm interested in is I sort of watched an interview with Jeff Bezos recently, and he said that, well, one, he's back at Amazon working on AI.
32:31 Harj Taggar? And two, that apparently Amazon itself has like a hundred or maybe it was a thousand. It was a surprisingly large number of internal LLM powered applications, presumably to just run Amazon. The last time Amazon took something it ran for internal infrastructure and released it to the world was AWS, which completely changed how startups are built. So I'm curious to see if they have interesting applications to run Amazon internally that they'll just release out and suddenly like there'll be new stacks to just build and scale your companies on. And we'll see the whole, something that we've talked about in recent episodes of the 10 person, the one person unicorn.
33:11 Diana Hu? One of the applications that they talk about is they did this giant migration for an old version of programming languages. Whenever you need to upgrade different versions of database or et cetera, it's like a lot of work and they use LLMs for it. It was like changing hundreds of thousands of lines of code. And it would have taken an engineering project of six months or more. It was done in weeks.
33:32 Jared Friedman? I mean, Amazon is just such a perfect use case for like LLM-powered agents doing back office processes. They must have just like just absolute gold mine of opportunities.
33:40 Diana Hu? And they just launched their big foundation model, actually, that is starting to be top in some of the benchmarks as well. So I think they're trying to be another contender through this race.
33:53 Garry Tan? That's interesting because like from the bottom up, like certainly from some of the people who still work at Amazon, maybe right out of college, many of them do not have access to LLMs or are actually barred from using it from in their day to day. So, you know, maybe that's one of the downsides of organizations when they get big enough.
34:13 Garry Tan? You know, the future is already here, but it is not evenly distributed even within the same organization. But that bodes sort of well for both open source and sort of self-hosting LLMs. Like it's on my to-do list to build my own stack of Apple minis and run Llama on my own little cluster on my desk.
34:33 Harj Taggar? I bought all the hardware to build my own machine, but then we had a baby and it hasn't happened. At some point.
34:41 Garry Tan? I've been pretty excited in that, you know, YC has been operating back in person at San in San Francisco for some time, but we got a real live demo day all the way back. So no more Zoom demo days, no more Zoom alumni demo day. You know, we did alumni demo day right here in this office, right, you know, right downstairs. That was awesome. And then we took over the Masonic Center and 1,200 investors all in one room. It was actually really great for the founders, I thought, because it was about a third as many founders than the summer batch. And it was more than 2X, maybe 3X the number
35:22 Garry Tan? of investors that then who had came to our investor reception party.
35:28 Diana Hu? So it was like a ratio of 10 investors for one company, roughly.
35:32 Garry Tan? So I think all of them had a really good time.
35:35 Jared Friedman? I'd almost forgotten how great the energy of an in-person Demo Day is. Like, it's just not something that you can replicate over Zoom. The YC Demo Day has also always acted as the de facto investor reunion in Silicon Valley because it's the one event that all the investors would reliably show up at. And so they were really excited that we had brought it back because when we weren't doing it, there was no equivalent event.
36:01 Diana Hu? So did the homecoming for Silicon Valley.
36:03 Garry Tan? Yeah, so now it's, you know, four times a year and it's the one time that all the top early-stage investors in the world are going to come back to San Francisco for hopefully that week's festivities culminating in our Demo Day. So it's a real celebration.
36:19 Harj Taggar? It feels like in-person in general is back. That's certainly another theme of 2024. Certainly, the late-stage startups that we've been meeting with and speaking to this year, one of the highest priority items has been figuring out how to get everyone back into person, back into the office. I think the era of it's going to be remote forever is definitely gone.
36:38 Garry Tan? Good riddance.
36:39 Harj Taggar? Yeah, exactly right. And then find like, yeah, in-person is back and then San Francisco is back. Like, a lot of thanks to you, Gary.
36:47 Harj Taggar? The elections recently seem to have gone well. Like, there's a lot of optimism, I feel, around San Francisco.
36:52 Garry Tan? Yeah, we have a new mayor. We're hoping that he does the right things. And, you know, we have a very thin, moderate majority on the Board of Supervisors, but we did get rid of some of the worst people who created a doom loop in San Francisco. So I'm optimistic. You know, we didn't get everything we wanted, but it's tracking in the right direction. And I think as in startups, as in politics, you always, you know, way overestimate what you get, you will get done in one year, but you always way underestimate what's going to happen in 10 years. I think it's going to take 10 years. It's going to take 20 years. But just as startups went from 15 companies a year that could possibly make it to $100
37:35 Garry Tan? million a year to 1,500 in any given year, knock on wood.
37:41 Garry Tan? That, you know, I think San Francisco needs to be the beacon for all the smartest people in the world. And that's actually probably the thing that I'm most hopeful for is that we can actually just keep building. So from all of us to all of you watching, happy holidays and we'll see you in the new year.
Previous Vertical AI Agents Could Be 10X Bigger Than SaaS Next The Lightcone 2025 Forecast Page updated November 5, 2026 · transcript and summary are machine-generated