Jesse Pujji and an interviewer debate whether AI will consolidate around platforms like OpenAI or splinter into specialized startups. They point to Excel spawning Airtable and Smartsheet as a model, argue that vertical-specific problems like legal underwriting will fuel startups, and note Y Combinator's AI investments jumped from nearly zero in 2015 to 90% today.
Chapters
So Benedict Evans is a very influential analyst in tech. He has done this now for— I thought he was still with Andreessen Horowitz, and truthfully, the reason that I thought that is because he's no longer attached to any company. He's an independent analyst who now happens to work for Mosaic Ventures. The fact that he was associated with any company is just kind of a memory of mine. And what's big about him is the level of analysis that he brings into a market. And this year he did a presentation that he called called AI Eats the World. I think it's really interesting. You sent it over to me. I thought we could analyze some of the slides and just talk about what you think they mean and maybe go deeper than he even went on it, 'cause he was just kind of ripping through these slides and ideas.
Fair? Yeah, let's do it.
All right. So Jesse, here's the thing that actually stood out for me. He talked about how startups exist to unbundle use cases, and the classic one is Excel spreadsheets, right? They, they basically take away these big pieces. And then he says, and ChatGPT, and you and I have been talking about this. Will, will that happen, do you think, where startups will unbundle the key features of ChatGPT, or is this an un— a new path where it's going to eat it up?
Yeah, it's a great question. It's a, it's a trillion-dollar question in one direction or the other, right? And I think, right, you know, it's interesting. One funny thing to think about with Excel is Excel existed as this only— it was the only game in town for a long time. And then as SaaS and the internet distribution came about, there were features and functionality that shifted that. So Google Docs being— I mean, I personally haven't been in an Excel spreadsheet in a long time because we use Google Docs for everything, right? And for a lot of things other than hardcore financial modeling. And then let's take it one step further. Then companies like— what's their names? Their names escaping me. But there's a bunch of these spreadsheet companies that popped up. That were making much smarter automated spreadsheets. Right. Why am I forgetting their names?
Oh man, I know the one you're talking about. Not Airtable, but—
Yeah, Airtable. Airtable is an example of it. Smartsheet, Airtable. There you go. Thank you. Like, I think Airtable is the best example of it where it's like, hey, now you have the— it's kind of a spreadsheet, but it's like a— it's taken SaaS or Google Sheets to the next level because there's so many new things you can do. And even Excel kind of, and then there's, you know, of course the CRM in some sense is a version of Excel. Like it's a specialized version of it. So, you know, I like to look at history when I think about some of these questions. So that's like one interesting mental model. Now you look at ChatGPT and you kind of say, what's going to win there? And people, by the way, love to argue this. Like it's, I don't think it's going to be one thing or the other. Like it's going to, it's just a question of how much is ChatGPT going to eat everything? Versus how much are there gonna be solutions used?
So you're thinking it's not eating everything, it's just like, it's gonna eat a lot of things and then also gonna be a lot of other things, a lot of like standalone products that take off a feature of ChatGPT and make it into their own standalone products.
Yeah. I mean, what, what I was trying to do is like, what's the framework? How do you think about that question? 'Cause I think that's like, you know, is it, um, you know, there's a famous example of Jasper AI was one of the first marketing AI companies that apparently went up to close to $100 million ARR and then came right back down to $30 or $40, you know, it's, And it's because a lot of their features and functionality of writing an ad or writing copy, it just, it was so easy in ChatGPT that their tooling didn't necessarily have it. So I think here's my somewhat of a mental model for it. I think, I think you need it to be somewhat vertical specific or pretty vertical specific. I think that's a big part of it.
Mm-hmm.
At the end of the day, OpenAI is still a company like any other company. So if OpenAI wanted to, let's pick a vertical.
I'm going to pick a great one. Note-taking.
That's not a vertical.
That's not a vertical.
That's like a function, right? Like a vertical would be like underwriting insurance. Okay.
Or legal. Legal.
Yeah. So legal has actually— legal right now, there are a bunch of companies out there that are using their models but seem to be pretty resilient. And the reason is, well, legal is very complicated. got its own things. And the point I was gonna make earlier is OpenAI is still just a group of human beings. If they wanna build a legal tool, they have to do the same thing a startup has to do. They have to go find a bunch of people. They have to go get clients. They have to figure out the problem. They have to make up the thing that has the problem. So it's not like, even if their AI is powerful, they still have to do all the same things. And even if you think about how it's probably organized internally at OpenAI, Mm-hmm. there's like, there's the model team who's making the LLM. And then there's some different group of people with a different leader who's probably in charge of like the legal group inside of there. And so how is that any different than a startup being outside of the, of the LLM? Now, if the LLM gets so smart that it can just absorb all those things, that's kind of the risk everyone's worrying about. But I think you're going to have people who build hooks into these firms are going to be experts. So distribution will matter a lot. I think OpenAI might just buy them. I mean, Google and Facebook did this all the time as they were becoming platforms, right?
Yeah.
I mean, Google famously bought YouTube and it was like, oh, well, hey, like these other guys got distribution, they got scale. And then Google said, oh, we'll just buy you. Why would we? You know, they, by the way, they had a competitor called Google Videos and it didn't do as well. Just like OpenAI launched an N8N competitor recently. And they also have a note-taker. They have a note-taker that they listen to.
They do.
So, so it's not a guarantee at all. But I think vertical specific, you know, I think I would pick a problem that was like probably medium-sized, not the biggest. If you go after the biggest problems, they're going to want to go after the biggest problems, obviously. So you have to find something that they're like, yeah, that one's not that interesting for us. So it's like sufficiently, sufficiently expertise, some complexity, and like medium-sized, medium-sized. It could still be a multibillion-dollar opportunity, but just medium-sized relative to like if you're, if you're OpenAI. And, and I think you have a good shot at building something really valuable that will be resilient.
Okay. One of the screenshots that he has, or one of the charts that he brought on is this one, which essentially what it's saying is what Garry Tan told me when we talked. They are huge into AI over at Y Combinator. And you can see that AI back in 2015 was a tiny sliver of the companies they invested in. And now they're basically 100%, let's say 90% of their investments in AI companies. And And a lot of them, Jesse, even if you look at how Y Combinator describes them, they will say they are the lovable of or the cursor of this thing. And so they're banking on that for sure. All right, let's go on to—
Well, I mean, that chart's funny. I'd be curious to look at that chart in like what the other categories were because like, I mean, that chart probably looks the same as what mobile looked like in 2010 to 20— you and I lived through mobile. Everything was mobile. You know, this is mobile. That's mobile. Of course it was mobile. That's the thing where that's the platform. That's the shift that's taking place. So in some ways it's a little like, well, duh, of course everything is gonna be AI cuz you shouldn't be starting a company that's not, that's not AI.
I see what you're saying. Everything is AI. So now let's get beyond it and say, what are you? If you're—
That's right.
It's like saying I'm, I'm a tech company. All right.
Yeah. Then there was a time when people said that I'm an internet company.
Right. And then it became so obvious that you are.
Of course you are.
You're bringing it up. Yeah. Why did you send this one to me? Why'd you text this screenshot? It's, uh, let me just read it out. It's the Accenture, um, new quarterly generative AI contracts in millions. And the chart just keeps going up and up from 2023.
Well, it's almost $2 billion according to this, right?
Mm-hmm.
I mean, I, I think there's gonna be a mass, a massive trillion-dollar-plus mark, you know, services industry around AI implementation. And, and let's think about why for a second, cuz there's a lot of interesting pieces to this. Number one, the nature of building AI tools. Let's just, you know, people call it AI tools, but let's just think of it as automation and software. You know, if you, the average company in marketing and sales, they have ideas or like, oh, I should get this automated. Like it was a very classic thing at Ampush where the team on the other side would go, oh, I need this done. They can't get any dev resources because the product team is using all the dev resources. Yeah. But they'd say, hey, if you could automate this for me, this influencer thing, they'd put it on their chart of impact, confidence, ease. They'd go, you know, it's kind of big, but we're not sure it's going to work. It's going to take us 6 to 9 months and 5 engineers to build it. And we're not sure it's going to work. Like, sorry, we're just not going to make the investment. If you ask any AI-first engineer now, they'll tell you like, I can build you something. It's going to be probably a little worse than it used to be when I, when I would code it up. Maybe 20% worse, but I can get it to you 90% faster and 75% cheaper. And so when you change those variables, what it's going to create is every, every team is going to have a bunch of, we'll call them, we could call it software or tools or automation, but they're going to have a bunch of those different things. And, and, and so you're going to end up having every organization, let's say a marketing team is going to have 30 different things that they couldn't have built 5 years ago. Now there's a— so that's the first thing you have to believe. The second thing is they're, for at least for the next 5 years, they're not going to know how to build it themselves, right? So they're going to need a third party to come in and understand it and assess it and build it. But services businesses are built to get close to a client, understand their workflow, and then build out against it. And then the last thing I would just point out is that this is not a new phenomenon. There's a stat that like for every dollar of salesforce.com or HubSpot revenue, There's $5 of services revenue because people needed to implement those cloud providers. They needed to make sure they worked. And there's big businesses built that do that. So that, I think cloud is gonna look small relative to AI from a revenue for ARR perspective. And I think services will look even bigger as a result of that.
Do you think this goes up and then goes down when something else happens? Or are we looking at a chart that just goes up and becomes the new norm? that companies spend billions of dollars every year on generative AI contracts, on automations?
Yeah, I mean, look, I think you'll never see the down because the up will sort of outdo it. But like, I don't know, you know, what do you think on-prem consulting looks like these days? Like at one point that was a huge industry and now cloud consulting is a much bigger industry and we just never even, I mean, I'm sure in some old document we could find someone showing their on-prem consulting business. going down, right? Or so there are consulting and there are revenues and things as they become less of a thing that nobody— but you never see them from the outside because they're just like declining businesses.
Essentially what he was saying was, oh, it was one of the first ones. He goes, I'm going to find it. I don't care. Oh, here we go. I'm going to share my screen now. And he's essentially saying, are we looking at something that's continuing to grow like we expect, like PCs, like mainframes, like— here, one sec. Here's a chart. Is this going to be like mainframes that grow and then we end up with a new norm, then PCs take over, grow, and we end up with a new norm, web, etc.? And is that what AI is going to be, or are we looking at something that is dramatically bigger? That's the big question, right? And so if it's dramatically bigger, then we can expect that spending for AI software by companies is going to continue and continue to grow. If it's just the next new thing, then it'll grow until whatever comes up after that.
Yeah. What do you think? I think it's definitely bigger than these things, these examples, because it's in some ways I would reimagine that chart. You know, I love my charts. To me, it's more like a pyramid or something. Like, it's— they're building on top of each other and it's actually expanding. So it's like, first I had computing and then the internet sat on top of computing, or, you know, there was PC. Then you couldn't have the internet without the PC. Then you couldn't have mobile without the internet already there. Right. And now— and so you're just getting wider and wider with your capabilities. And it's almost like the world— I mean, imagine ChatGPT. You couldn't have it without the internet because we're all connected to it.
Right.
And of course it could live on your device in theory, but it just, it wouldn't be a great experience. And so I think it, I think it has a chance to be a lot bigger. I think there's a lot of real-world stuff that, that GenAI can touch that nothing could touch up until it. But he also has these great examples that I think are really like helpful. Like he talks about the automated elevators. Did you see that one?
Yeah. He was saying, look at how elevators started to pop up with these people in it. Then it became, and he showed the graph of how many people were working in the space. And so suddenly you had this big, massive number of jobs. And then as automation came in, they were replaced, and now there's basically nobody running an elevator.
Yeah, so I think, I think the AI stuff really hits a lot of different jobs that are really offline jobs or formerly creative jobs, like jobs you couldn't have imagined replaced. Now again, I tend to be the optimist and go like, yeah, but tons more jobs will be created, and the opportunity we can't imagine will be created because of this. And so I think people will be fine. You know, we never heard about any issues with the elevator Like, those people all found jobs. But it's just interesting to imagine how much stuff like that it will end up replacing.
Okay, here's another along the same lines. He was talking about this, where he's saying steam engines gave Britain the equivalent labor of 5 times of its total population. So, basically, the steam engines didn't just take over the jobs of people, but they're basically more people than you could have, and still you end up with more, more jobs coming up.
Well, isn't it the classic thing of the pie growing? I mean, as long as the pie is growing, there's always, there's always more for everyone. Um, there was a really cool thing, by the way, sidebar. Did you see this Ari Emanuel interview? I think Patrick O'Shaughnessy did.
No.
You would love it. I think it'll speak to your heart right now because You know what his, his whole thesis is on AI?
Yes. He says, look, I can't understand AI. I'm not going to get it. I'm not going to outsmart everyone else. I'm going to go the opposite direction. I'm going to go for live. You cannot reproduce live with AI. So that's what I'm going all in on. He's raised a bunch of money to go and buy—
The anti-AI thesis, which is, which is, hey, what if people are only working 3-day weeks and making as much money as they make? They're going to have a bunch of free time on their hands. Let's, let's cater to that.
You know what? And he keeps saying that people already are working 4 days a week. Who are these people working 4 days a week?
I think there's a lot of people who work for— I think remote work plus all this stuff has definitely created, made work easier. But I think his trend, I think I love that trend. I've also had the thought, you know, I'm on the board of a grocery store, like people are going to need to eat. Like, and, you know, they never— no one ever worries about that in the grocery business, interestingly. Right. We never talk about do people need to eat? We talk about the type of trends. But that's like a really cool thing when you think about the resilience of it. We talk a lot about distribution. Is it competing against Costco and, you know, using Instacart? But we never talk about if people are going to need to eat. And that's a good one. And like real estate is an interesting one. Like I'm— I've been spending a lot more time thinking about putting, you know, more money of mine, my assets into real estate because you're just like, well, people are going to need a place to live, right? Especially like multifamily. It just feels like people are going to need a place to live.
I see. Food is the anti-AI. Homes are the anti-AI.
Entertainment.
Uh, I get that. He also actually— Benedict did in this, in this slide presentation. He said, look at how much of commerce has gone to e-commerce. And he just showed a graph that goes up, up, up, up, up. But if you look at it carefully, he says excluding— did you see this? Gas and groceries. They're like the 2 obnoxious big expenses that would ruin the chart if you included them. Um, well, pull—
let's pull that one up. I mean, that's an interesting one because I think the other thing that I think it's just worth noting on those charts is it's still probably not a majority of commerce.
For, for groceries, you're saying, right?
No, no, no. Forget about groceries. Just pull that chart up.
Okay. Here, let's take a look. You're right. You know what? This is the part that I just can't believe. I live in a whole other world.
That's right.
I just cannot imagine. Here, this is it. It's going up every year since the beginning of this is around the year 2000. It goes up over COVID, it comes down, but not nearly even to pre-COVID levels. And then it continues to go up.
But the total number is the most interesting thing, which is it's only 30%.
I know.
That means 70%. I mean, imagine this is a great example of like, what if we just showed you a chart that just showed it going from 100% down to 70%? You'd be like, oh, What's the big deal? But like, you know, massive companies— Shopify, Facebook— I mean, they're all building that, but yet there's still so much more room to run.
Yeah.
Right? I mean, when you look at— anytime I used to look at charts like this, I'd buy more Google stock, I'd buy more Facebook stock, because I was just like, you know, we still got a lot of room. Like, this number could be 40%, and you would still say it's not even the majority. And that 40% for Shopify and Facebook and Google is another like $100 billion. It's probably another $500 billion in market cap, you know, for Facebook and Google and another $100 or $200 billion in Shopify market cap because that's how much bigger these markets can get. And it's just crazy when you think about it.
Okay, how about this one? This is the very next one.
And ChatGPT will catalyze it a lot more, dude, because there's gonna be—
You ask a question and it gives you the answer.
There's just gonna be another catalyst and more likely that there's things we wouldn't have bought offline that we will now buy off— on online because ChatGPT will change the experience.
Look at this. This is robotaxi use in California. You and I were in San Francisco when these robotaxis were being tested in the street, and it felt like such an odd thing. Will they ever come? Will they not? And for a long time it was so frustrating. You couldn't get a ride. They weren't coming. It felt like maybe they were all gonna bomb just like Cruise.
Um, and then did you see that thing that went around recently where it was the thesis for how Google's gonna be a $10 trillion company?
No, what is that?
Oh man, you gotta see it. There's like, they even have like a, uh, a funny image of, you know, did you ever watch the Teenage Mutant Ninja Turtles?
Not really.
Uh, you know, you know, there's like—
I do know them for sure.
There's Splinter and then there's the 4 turtles. So there's this image of them where they're kids and they're little turtles and Master Splinter is very big. And then there's an image of them where they're grown-up turtles and he's kind of the small one and he's He's got his 4, the 4 Teenage Mutant Ninja Turtles. And somebody applied that to Google and basically showed the search business as Master Splinter, and then cloud, Waymo, AI, and I can't remember the other one. But those 4 as basically like they used to be the little turtles, now they're all gonna become the big turtles.
And make search really, basically dwarf search.
They're gonna dwarf search. And if you believe that, then Google is still a 3x from here. And it's diversified, and it's still, it's not like their core business is going down either.
The thing that I took away from this was, it feels like it's kinda here, it's here for special people, it's here on a small basis, and then boom, it explodes. And that's what we're seeing on this chart, that robotaxi trips in California are tiny, they're kinda flat for about a year even once they're up, It doesn't seem like a thing, and then it just takes over so much that it feels ridiculous not to have been in one. And now people are— you could decide not to take one, but if you're taking an Uber, there's a very good chance if you do it on a regular basis, you're gonna end up in one of these Waymos.
I think that's right. And, and yeah, I mean, the, the other crazy chart in this, of just in the spirit of thinking that way, is like the dating chart.
Yes.
And I mean, we should share that. Like, if, if, you know, how many people, how many heterosexual people met their spouses online? In 1995, it's close to zero, obviously. And even 2000 or 2005, it was— it's maybe, you know, 15%. And now it's more than half of people. More than half of people. And anecdotally, you hear that. I feel like an old person. I'm like, yeah, I just met my wife in high school.
Dude, I remember busting Gary Vaynerchuk's chops. Like, he was acting all cool and everything. I go, how did you meet your wife? And it's like online dating. And my like, come on, what's wrong with you message was, you must not be that cool if you had to go online and date like somebody who needed assistance. And then I think about that all the time because everyone is finding their partner that way. It feels ridiculous today to walk up to someone at a bar, which I would still prefer to do if I was single. But of course you're going to go online. Why bother with all those other headaches?
Yeah, but it just shows how fast— like, it kind of reminded me of that chart you just showed, which is, you know, 5 years ago, percentage of people using robotaxis is near zero. I think in 15 years it's going to be more than half of us.
I don't know why I can't find that chart, but yes, I do think that. So that's the big takeaway from this. So what is he really saying here? He's saying that this is really slow, but it could be much bigger than all those other growths. And there's a lot more opportunity than we saw. And one thing that's not in the, uh, in the Google slides that he shared but was in his presentation was he goes, a lot of us are thinking about maybe ChatGPT and AI is going to replace search. But he goes, let me give you an example of something that doesn't replace search. I took a picture of my refrigerator and I said, this is what I have. What should I cook? I would never have put that into Google. It's going to expand that pie even further.
Well, it's kind of like the, you know, you and I got really excited about that Suno app, right?
Oh yeah.
And that's the one that makes music, which, by the way, it's another interesting example of like, will ChatGPT eat that or not? I don't know the answer to that question, but like, I don't think they will because these guys have built a really thoughtful user case study. They're, you know, they've got distribution quickly, and I'm not sure it's that interesting of a market for ChatGPT. So it kind of fits some of the parameters we talked about earlier.
Right. Mm-hmm.
But, but the most interesting thing about Suno as a business is already since we've played with it, I sent it as like a gift gag for a friend. Adrian had someone do it for a client and a client relationship and how their growth assistant is performing. It's really good. It's a good song. So these things just expand. Like we just would never have ever thought to do those things. And then you start to think to do so. Again, this is all pie creation. It goes back to that idea that like, I mean, dude, there's going to be agencies that live on Suno and all they do is make marketing songs. So like every— like I was telling Achint and Adrian, who you both know, like every client, we should send them a song every quarter about how the quarter went and how their growth assistants are performing. And we can also send the growth assistant kudos that way. Hey, here's a song about what you did this last month. Thank you for being a great member. And the person gets to listen to the song. And like, by the way, more than 30% of people— you brought this up— that I've shown songs about— we did it for Larissa, who was our au pair, is now kind of our nanny. We did a song for— she started crying. People start crying. They get very emotional when they hear these songs. And, and it's just a crazy thing that, like—
I'll tell you why. Malcolm Gladwell did this whole great podcast episode about why country music is more emotional than rock and roll. And he goes, rock and roll is so general because it's appealing to a general audience. Country is appealing to a small group of people and they could get specific about the situation that these people are going through. This is like a million zillion times more specific. I specifically had a note in there about how Olivia, when I waved one for her, when our baby got his finger stuck in the door jamb, she rushed to him when I looked away because I couldn't stand to look at it. That's like such a moment in our lives that you can't have some rock and roller or even a country singer do. By the way, I talked to Richard who works over at Zuno, he goes, Andrew, you don't even see the use cases. He goes, first of all, students are using this to memorize things and prepare. I said, oh, I thought I was the only one. He goes, no, they're doing that. He goes, for sure people are sending it to potential clients, walk-on music, and then he goes, there are people who have issues emotionally that they need something to help them get through. They're writing these custom songs that allow them to do that, and these are not use cases. So essentially that's what we're saying. It's going to expand the pie. So if we were to just sum up before we go into anything else here, Actually, if we're to sum up what we took away from this, what did we take away that it's so revolutionary that he said here?
I think the trend is just much bigger and much earlier than anybody actually thinks. And the reason is because we, our brains are like, it's so hyped. It must all be there. And that's not the case at all. The hype precedes the actual substantive growth. That's happened with social. It happened with mobile, happened with the internet. It just always happens. So I think that's a really, really important point. I think the second point is like, you really have to start to think about where to play inside of this map. And like, we offered a framework that I think could be helpful. The third thing is this concept of the pie expanding, because there's things you would do that you wouldn't do when the technology changes and it shifts. And I think all of those are just really, really like, those are, those are my quick, quick hot takes on, on this, uh, on this big AI thing that Ben, Ben did.
Okay, I'm gonna give you 2 of mine. One is, it goes slow and then it pops up. And a lot of these things we feel like we're big on, like Gary Vaynerchuk doing online dating, it feels like, okay, he might be ahead of the pack and maybe this is as big as it's gonna get. No, there's a much bigger pop beyond that. And the same thing for robo-taxis, it just, it goes slow, it seems like it's got whatever adoption it's gonna take, and then it goes pop. And then the other thing is, maybe this is more of the same But different. Maybe there is a much bigger pop here than we, we had seen before. Maybe instead of like one of his graphs that goes like this and it's more— maybe we're now looking at something that potentially can shoot up massively. And if that's what it is, then I want to be on it and I want to— I want that.
All right. Good stuff.
Cool.
Bye.
