Apoorv Agrawal on Consumer AI Winner-Take-Most Dynamics

bank accounts to move money, fight fraud, and improve lending now with AI. And without further ado, I believe we have our first guest, a poor vault Romel Timmer in the reream waiting room. Let's bring him in with the TV. How are you doing?

Great guys. How you doing?

Thank you so much for taking the time to come chat with us. Welcome to Finally have you on the show.

Uh obviously I want to jump straight into consumer AI, but maybe you could uh since this is the first time in the show, give us a little bit of intro on uh your role, your background, what you like to focus on.

Awesome. Uh well, I'm a partner here at Ultimter. Uh I help lead some of our AI investments uh in companies like OpenAI and uh Glean and Parlo and Expo. Um, and excited to be here. Thanks for having me.

Yeah, you also did uh Reddus, right? Is that correct?

Back in the day.

I love that company. I'm such a fan. I I met uh maybe the co-founder, maybe just one of the the the creator of the original project like the the the the engineer behind it and uh he taught me a lot about uh about software development which was very helpful earlier in my career. Anyway, how are you thinking about the state of consumer AI? Who's winning? What are the narratives that are being discounted because of the vibe war and uh just all the different takes that surface every single day on the internet?

Yeah, look, I I I'll start at the top. You know, history has shown us that these consumer markets have been winner take most. Sometimes win or take all honestly, right? You look at the biggest technology market started the internet. We got Google out of there for search 90% plus market share, three and a half, four trillion in market cap.

Yeah. Mobile the next big super cycle led to the birth of Apple you know three and a half four trillion in market cap another uh winner take most dynamic take social the next one meta one and a half two trillion in market cap across WhatsApp Facebook Instagram they've got majority market share and so you know keep going down this list right hailing food delivery you know I think consumer AI uh I suspect will be similar uh and I think that's what's going on with Chad GPT you know it got lightning in a bottle 3 years ago

uh it continues to they continue to work hard, ship features, improved models, memory, voice, image, now Sora you saw yesterday

and each of that is adding to the S-curve and you know lo and behold Chad GPT is uh is is doing pretty well uh despite the vibes.

Yeah. Yeah. No, totally. Uh I heard a a a stat where my jaw hit the floor which is that Bing is profitable and growing. It's much smaller than Google, but it's a very interesting thing that even in something that feels like a winner take all market, the second third player can still be like reasonable businesses where Sacha Nadella doesn't need to say, "Oh, we got to shut this down. It's not working." It's like it is working. It's not winning, but it's working and it adds value to the ecosystem. And so I think that there's these there's these interesting. Has anyone ever tried to clock what what the actual enterprise value of Bing would be today? If

I think it's in the billions. I think it's definitely in the billions.

Definitely. I mean, search search is a valuable thing for agents more than humans, right? Because the Google API is no longer available as an API. So, so Bing's API is actually very valuable, maybe even more than the Bing consumer product.

Yeah. And I mean, uh, Google for a long time has always said, you know, competition is only a click away. Sundar Pay said this and it's interesting how uh how sticky consumer habits can be in a world where in tech we often optimize for the benchmarks and the latest features and I'm jump I got every app on my phone I I will switch dayto day I will export I'll jump through all these hoops to use the latest and greatest for two weeks and then go back to the old thing but I when I when I just talk to random people in the world it's like they're like oh I the Funny one is like uh Jordy was talking to somebody who was like he was like do you use any of these AI tools? And the guy was like no not really. And then he was like have you ever heard of Chad GPT? He's like yeah I use that every day. And it's like it's not even like he doesn't even think of it as AI because the the news when the news talks about AI it's like a data center or like the robot and and like he just thinks about like oh yeah that's the that's the text box that answers my questions

that I can do stuff with.

Yeah. Yeah. Yeah.

Yeah. how to make any decisions outside of chat GPT. Like if I'm buying something, you know, I got a button dedicated to it on my phone. Like it's a habit now.

Yeah, exactly. It's a habit and that's powerful.

Uh what's your framework for trying to understand the present through history? How much do you look towards the dynamics that were unfolding during the dot cycle, different consumer internet players? How much does that

like matter to you? How much have do you obsess over that versus obsessing over the data today and trying to predict the future?

Yeah, look, I think um in in the moment uh if you if you're if you're sort of living day-to-day and observing what's happening in the news, it it feels a little bit like there's four seasons to the year. It's called OpenAI, Google, Anthropic, and and China. Maybe those are the four seasons. And depending on what season you're in, like that's the news cycle, that's the vibe cycle. Yeah,

but I think to your question, you know, I try to zoom out of that and you know, obviously we run a very concentrated investment strategy. Our growth fund has only four investments, right? And so when you when you're when you're when you're running that concentrated and and and you know, putting your money where your mouth is you you have to zoom out and

history certainly rhymes in this case, you know, one of the things that I had written about, I'd say two two three years ago was look, there's two big conglomerates in technology, Meta and Google. Meta's got three billion users in their with phones in their pocket across WhatsApp, Facebook, Instagram, and and and Alphabet's got that with Google and Gmail and and and YouTube. And I said, "Hey, these guys are most likely to deliver the AI that everybody's going to use." I was wrong.

That's not how it happened.

Yeah.

Chad GPT turns out got magic in a bottle uh and continues to grind up. Now just under a billion, I think 900 million weekly active users reported uh over a billion month monthly active. And honestly, that was a surprise and we corrected that mistake uh uh uh two years ago and you know now the largest investment in in ultimate history.

Mhm.

How do you think about pattern matching against previous capital wars? I was I was loosely thinking about theft versus Uber fight that happened and I don't know it's always hard to overfit on previous cycles but I'm wondering how you process that capital fight. Uh I don't not I'm not exactly sure where you were in your investing career at that time, but you clearly live through it. So uh I was wondering about um

lessons from that era and just and maybe if there's a different scenario where capital fights can result in igopies. That's the interesting narrative that I feel like is might be playing out at least in some subm markets. I I was a I was a young engineer at Palantra at the time and I remember the um Uber rides in SF and Palo Alto here, what we call the Shire uh got so cheap you were almost better off taking them uh than walking.

Yeah.

And and you know I think it is not too dissimilar right now, right? Like I'm sure you guys are on chat GPT and like like you said, it's a habit

and

we are certainly getting more than 20 bucks of value on this per month, right? May maybe hundreds for some people it's thousands. the the folks who are who are coding maybe tens of thousands and I think we are in this in this era folks that the price discrimination has not stuck

right it's it's still 20 bucks a month for basically everybody

right um and you know actually I'm going to talk to Nick Turley head of Chad GBD later today about all of that and this is one of my questions is like h how do we make sure we're you know for folks who are getting thousands or tens of thousand dollar value out of this product how do we make sure we capture that what I call expand the ceiling on the top.

But then also, you know, there's the vast majority, the billions of users, the Joe Schmo, they don't want Chad GB to book their flights for them because they're not booking 10 flights a month. They're booking 10 flights a year.

Yeah.

They they're deliberate about it. And so for them, how do we get them on an easy ramp? How do we lower the

lower the floor? And and maybe ads is the way to go.

Yeah.

Um um and you know, the Meta properties have done a great job. They've got three four billion users. They monetize at about $60 per user per year globally.

Uh Alphabet has a similar number. It's about 34 billion active users. Monetize at about $70 a user per year globally. And you know, OpenAI is at you know, we we we're at about you know, 10 $15 per per user per year across uh about a billion users. And both have got to go up over time.

Yeah. How do you think about squaring this idea that uh at least among the mag seven? Uh maybe you take out Tesla and Nvidia, but basically every company competes with everyone and everything. It feels like uh I've always I was always laughing about six out of the seven mag seven CEOs own or control social media properties whether you include LinkedIn, Twitch for Amazon, X for Elon, uh iMessage for Apple is sort of a social network now. Um and and and you know YouTube directly competes with Instagram. the shorts product is the same. Uh, and it feels like over time some of these develop oligopoly, some of them, you know, the monopoly sticks. But, um, how much do you think we can learn from the way the previous era of of the the the mag seven boom played out in terms of how the leading labs might compete over the next decade?

100%. I mean, it's happening uh right right now. I mean, you saw what happened and you know, they call it

you get cloudcoded. It's a cursor got cloudcoded. I I I think it's the biggest narrative violation. I was with Michael Troll over the weekend. They're crushing it.

Oh, yeah.

They're absolutely crushing it. And and it's it's it's what you said, you know, it's, you know, maybe to give a historical example.

Sure.

Um, data bricks and and snowflake, they were both built on the top of the three hyperscalers, AWS, GCP, and Azure. While them having their oneP products, Redshift,

Yeah. Big Query and with Fabric, right? And and it's a little bit of what's going on right now with Cursor. You've got Enthropic and OpenAI who've got

their models that that Cursor buys and they've got um their one pie products called Codex and Cloud Code that compete with Cursor.

Yep.

And you know, look, if you're a if you're a developer, if you're a customer, an enterprise customer who cares about the resilience and the and the and the multi- modality and the option of having this choice, you'll probably pick something horizontal across them.

Mhm. And so I think I I I think as you said, everybody's competing with everybody and um uh what a time to be what a time to be observing this, but I think the fog of war will will get clearer um um as we go into the second inning and the third inning.

Yeah. You mentioned that uh flight booking might not be the major driver of uh consumer growth this year. Maybe I I I sort of agree with that, but I'm wondering like uh I was talking to um Olivia Moore at Andre yesterday about the the the value that might come from uh integrating image, video generation, and reasoning models and deep research models all into one consumer interface. So you have this idea for this like essentially vertical video reel and you can actually like like plan out generate images and and wind up with a much more fullfeatured product. What what else are you tracking in terms of like consumer use cases that you think might unlock this year?

Yeah, look I think the um you know we we're uh uh investors in in a couple of these inf infense platforms. So we see the traffic of of of of AI tokens. I would say the two biggest sources of uh the uses for TI tokens is coding and video.

Yeah.

Right. And then there's like a long tale of customer service and and and legal and office of the CFO and and and so on and obviously chat is is you know I forgot to say chat but chat's one of the big ones. Uh, I think for the average consumer, um, the beauty of having it all in one place and and why I I I had wish Sora would have been in Chad GPD since day one is just so high. You know, I I'm not a professional video editor or image editor. I make that one card on my kid's birthday and that might be it. Maybe two, the second one on my anniversary someday. And so, you know, I think I just I would just like it to be there.

Yeah.

But but but and that's the vast majority of Joe Schmo. Now obviously the 1% you know like Dan our producer he's going to make that 30 times a day and he might you know go find the best and breed player but I think for vast majority having the one throat to choke is is is the way to go.

How do you think uh marketing and customer acquisition will change in the future with regard to consumer AI? Uh we're seeing Super Bowl ads now. Uh I've haven't seen a ton of direct response but maybe I've just missed that. Um maybe Meta's like, "Hey, we're not going to let someone uh run these ads." But um uh

in some ways, I'm sure Meta is funding their AI strategy through how much other AI companies are spending on the

Yeah, certainly on the longtail, you know, you see a lot of ads for like point solution AI apps for, you know, try on clothes digitally or create a custom avatar and you see those apps sort of climb up the app charts on the back of meta ads. uh what what do you think the shape of like customer acquisition looks like over the next couple years?

It it's it is such a good question because you know as I said there's four seasons to the year but turns out in those four seasons like 16 mini seasons where one of these app will go through like a marketing surge.

Some people might even say 365 mini seasons if you're running a daily technology and business news show.

We're we're fully employed over here.

Exactly. Exactly. And so, you know, we see we observe these spikes and so there'll be a new app, you know, I mean, not to dunk on Deep Seek, but like Deep Seek had a huge spike start of last year.

Um, and it was so hard to explain like what I don't know if anybody uses it anymore. U

I just I just assume that was bought. They just bought they just bought it their way up the chart.

I know. I don't know. I think I think I I downloaded the app. A lot of people were interested.

You downloaded it cuz you saw it on the top of the charts. No, no, no. I saw it on Twitter and people were saying like this is the first way that you can see the the new user interface for what watching an LLM reason and and you could see the token streaming in. you can see the internal reasoning logic and then you get the output and and that was a remarkable UI pattern that then was quickly ported back to chatbt and the chatbt client base and customer base and user base was very happy about that and they still are because you see the little like thinking about this okay I'm going to search this and everyone's happy

yeah one one thing that's surprising looking at the I the the app store charts right now

you have chatbt claude Gemini and 123 But then there's no other new apps

in the whole top 25.

And it's so interesting thinking if you could rewind 2 3 years ago and you knew how you knew how how how much acceleration how much progress we were going to see in AI you would assume that there would be like

five six seven maybe more of these like new entrance. But I think it's, you know, it just, you know, even even with the technological shift, it's just it's just proving that like it is so hard to

actually break out in consumer. And in some ways, it's going to

the the tools, the AI tools just make it harder to actually break out and chart because you're not competing with two or three other companies. You're maybe competing with two or 300 at some point.

The flywheel is too good. It's too efficient. The more users you have, the more feedback you have. you know, the thumbs up, thumbs down at the bottom of your chat GPD prompt

and they make the next

answer better. And it's just that flywheel is so has gotten so good. It's a little bit like what happened in search.

Sure.

After a certain point, it's just so hard to deliver a better search than Google. Yeah.

Um and which is why these consumer markets, as as we discussed, I think are going to be winner take most.

Have have you guys heard the story of how Google built their uh spellch checker autocomplete? It's a it's I think I've told this, but they basically they just look through the records and they said if someone spells a word that we don't know, just look at whatever they whatever they searched next because they will likely go do the work for you and actually figure out the correct word that they were thinking of and then so they just had this massive data set and so they just had the best autocomplete and that stuff compounds all the time. The the leaving the thumbs up, thumbs down, that's the version of like tipping in the consumer AI era. Like if you're not doing that at home, if you're not if you're not tipping chat with some feedback, you got to get the the flywheel going. This is important.

That's exactly right.

Okay, prediction time. Do you think any Neolabs will try and jump into consumer or are they more interested in playing in like the more oligopolistic B2B point solution enterprise where you could potentially

intelligence as a service?

Look, I'm a I'm a forever optimist on startups. Uh so so so I hope they take a crack at it and at the very minimum we will learn something that that one of the big platforms will get like what you just told us about Deep Seek. I thought that was a good feature and and we got it on the main app. So at the very minimum we will have that happen. Uh and at the very maximum we'll have a we'll have a we'll have a new product and Chad Gibbly will be the Yahoo. But I but you know the obviously that's a harder harder hill to climb.

Yeah. What about uh I've been I I've been talking to a few of investors about um where they sit versus where the timeline sits where they're thinking in much longer timelines than what X is going back and forth on who's the hot uh company of the day, the seasons as you put it. Uh what do you think about this idea of tall poppy syndrome that that uh the tech community seems to like to punch upwards, not punch downwards. And so no one wants to beat up on a company that launched and then never really got to escape velocity and is sort of like languishing. But if you're at the top, it's game on. We're coming for you and we are going to find every problem and make it very well known and it's over. It's over. What do you think about that?

Well, when you're the underdog, you got to pick a fight with a winner. You got to bring the fight to you.

This is alpha,

you know, not not the other way, you know. So, I think I think that's what's going on. People want to pick a fight with a winner.

Yeah. Yeah. That happens. That happens a ton.

Bring the game to you.

Bring the game to you. Um, what what else are you tracking in the AI buildout? What else is important in understanding like can consumer AI continue to ramp? You mentioned ads, you mentioned continued retention and growth, but uh more deeper in the supply chain. How are you feeling? We just looked at Oracle earnings. It seemed very positive. But what are what are you reading?

You know the number one thing at this game like you know 3 years in that that we obsess about is durability.

Yeah.

Um durability is the number one question and we have this like panel of things that we look at. I've been publishing some of that

is hey is this thing you know users show up that is step number one.

Users stay there. That's step number two. M

and the way we measure that is daily active users uh divided by monthly active users and you know I was fully expecting that to tell us a lot and it does actually the dispersion of the leading AI apps the three that you mentioned that are leading you know chat GPT Gemini and claude is actually quite wide

chat GPT's daily active usage over monthly active usage is about 45 50%.

Gemini is at 22%.

Wow. Meaning that the number of times people are showing up in a month is like twice as much more on Chad GPT, which means it's

so they're still like retain they're still retained on Gemini. It's just a much less valuable user because they're not using the app as much.

That's right. That's right. The usage is far far less frequent on Gemini.

And then, you know, the third thing that we look at is, hey, has this turned into a habit or is it still a curiosity? And that that we look at you know using the standard retention chart which is over 12 months how many users retained and um you know the the dispersion is quite wide there as well.

Um in fact the only three apps that show what we call a smile curve meaning that users come back over time after having churned.

Um uh the only one of them in AI is is is chat GPT. The other two are uh Chrome and WhatsApp. meaning that the the the the utility is so high that even if you made the mistake of turning you you find your way back.

I mean that happened to me not not with Chrome but with uh WhatsApp. I had it for one group of friends dropped off then eventually there's a new group chat and you got to become a DU.

Chrome Chrome seems like a better kind of comp for LLM because it's more of a single player experience whereas like WhatsApp I'm I'm the same way. I'm not a heavy WhatsApp user, but then every once in a while I'm like, "Wow, there's two people I message on WhatsApp. I haven't gotten back to them or whatever." So, I'll come back.

But, but Chrome is probably more relevant to

It's a great point. You know, it's a great point because you you know, WhatsApp has network effects. You know, if my friends uh Jordy and John are on on on WhatsApp, I'm more likely to be there.

Chat Guby doesn't have that. If you guys are on Chat Guby, I'm not more likely to be there.

Yeah.

The same thing with Chrome.

Yeah. Exactly.

Yeah. And and the other effect that is that that I that I think about a lot is, you know, the dopamine, you know, doom scrolling through cat videos like Tik Tok, Instagram all have this dopamine, this like addictive effect. That's why they're, you know, three four billion users. Tragically doesn't have that.

Yeah.

It's it's you're there for for for some kind of a knowledge workflow.

Yeah. They also have uh the the Tik Tok I I haven't had Tik Tok installed for years. Uh, but occasionally people will send me a Tik Tok and I will try and open it and it'll just ask me to install the app. It won't even play for me in the web browser and it's very difficult. But that is another viral flywheel that people often will generate something but they'll send me a screenshot from their chatbt uh results. It's pretty rare that someone actually sends me a a full link to their deep research report. I think that might happen more. I' I've said for a long time like I want to just have a feed of Tyler whatever Tyler Cowan is deep researching. I want to just read through the results of that because I think that would be interesting um without him needing to like publish it and then everyone be like this is AI generated. It's like no I want to I want to be able to follow people that are doing interesting research and then when they when they do some interesting research they can just share that on the network. I think that could be cool as a feed but again very knowledge based not nearly as uh brain rotty as Tik Tok. different different building on what you said here's here's a prediction I think I think the next phase of utility guys you know we at like a billion users on this consumer thing the next billion I think will come from these proactive proactive work like you described it's like hey I don't right now I've got to go log on to chat GPD or Gemini to like engage with it

but hey what if you like you you you you were like hey this is what Jordy said or this is what Tyler Cohen said you got to check it out

yep

you know I know you think about a lot about this company open. You think a lot about this company Revolute. This is what somebody smart said about it. Or hey, they got a UK banking license. You got to check it out and here's why it matters.

Yeah, Pulse was like a glimmer of that. But I felt like the daily functionality on Pulse was too much for me. What I'm actually looking for in a world that has slightly different economics, slightly different business incentives is something more like I like movies. I like a certain type of movie. I want you to actually get through to me when there's a new Christopher Nolan movie or a new Denal New movie and I want you to tell me about it, but only when that happens. And I don't want a weekly newsletter of all the of all the movies.

Yes.

And that doesn't make any business sense. Like you would never be like I'm starting a Substack.

I'll post it whenever something interesting happens. It's like it's got to be daily. It got to be weekly. We have a daily show. Um we couldn't just do just emergency pods. Like that doesn't make sense. But you might appreciate Yeah, you might appreciate this cuz because you brought up Reddus. You know, these weekly reports are like

they were like cron jobs back in the day.

Exactly.

Who knew cron jobs would be so valuable?

But I want the I want the cron job to run and then the AI agent to filter and be like, does this actually pass the quality bar? If not, let's just not let's just not annoy this guy and then if it is it if we got something good going on, let's let him know. Uh so tons of interesting things. What a fun time to be a product manager at OpenAI. It's got to be a great time. Uh but uh say hello to the head of

Ch. What's the bar for for adding a new investment to your guys' four portfolio fund?

It's uh it's it's it's not a low bar. Um I I'll say that, but um I'll keep you posted. I'm working on one trying to get add one and we'll see if we get it through.

Yeah. Uh but but but but maybe maybe to maybe to maybe to actually answer your question, I'd say that one of our beliefs is that the rate of change is so fast and so high that you know honestly it feels like there's a dozen startups that get quite endangered turn into like endangered species with every product release, right? with from from from one of the what we call private max 7 open anthropic uh leading that charge and and so the idea is to just like stay concentrated in the ones that we know are working uh and and you know we obviously we have an early stage fund that does earlier earlier stage investments but on the growth state side we we run pretty concentrated

uh the bar to get something added would be we have belief that this is a company that has a good uh line of sight to uh free cash flow over time and that is hard to say for for for for many late stage businesses uh right now but the ones that we the ones we feel so are in there

that's great

concerning concerning that that that you're not seeing the line of sight

yeah I think I think I think I I think about the private markets in three innings right now guys it's like you've got the early state stuff that's like you know sub a billion market cap right the second call it stage is the 1 to 10 billion market cap And then the third one is the the the tend to call it wherever you're staying private forever. This is these are this is what I call like quasi publiclix. You could go public. You just choose to stay private. I would put open anthropic revolute ramp in this bucket.

Very high quality companies that we'd love to own in our growth fund.

Sure.

The second bucket which is the 1 to 10 I think is the part where there is most dispersion and is is probably the place I see the most fog of war and hardest to figure out uh whether they they have the durability. Um and you know the early stage business will always be there. There'll always be great founders starting. I know you guys had uh I think you call it the Gupta nator.

I'm glad you caught that.

Crush.

Yeah, crush.

What do you last question then then we'll let you go. uh how do how are you processing the disconnect between public SAS apocalypse people concerned around SAS as a business model and then what's happening in the actual early stage with the there's so many companies that may position themselves as something other than SAS but functionally they are SAS like you could just you could just take if they say agents you could just replace it with workflows and it would sound exactly like a SAS business and it's this like very interesting time where you just have all this funding activity early early and then public investors are saying, "Hey, we want to put a huge discount on all on all these businesses because we just don't know what is going to happen in in 10 years." Uh do do you find that a bit strange?

Look, look, I I have tremendous respect for the public markets. They are a very efficient mechanism for repricing. Uh but I think they're not hyperefficient in in times like now when the when the change is so fast. And so I think it's a there's a there's a spectrum, you know. Uh I'll give you I'll ask you a provocative question. Why couldn't you vip code Chad GPT?

You could.

And by the way, when you vip code and come back,

we did we launched Claude with ads.

Claude with

Claude.

Oh, you Oh, wow. We clawedwith ads.com was up for one day during the Super Bowl and it gave you intelligence too cheap to meter sponsored by all of our sponsors and it served you an ad before delivering a query from Opus46 or something like that. It was a lot of fun.

Yeah.

But yes, of course it was not a real product and it would never have legs.

But I think that's the state of a lot of a lot of vibe coding right now. You know, you go build cloudwiths.com, you try it out for a day and then you're like, "Oh man, that real cloud was was a lot better."

Yeah. And then and then by the way, go buy some software in the public markets after that because you realize the staying power of some of that software.

I wouldn't say that for everybody, right? Like there's a lot of stuff that that that we have on our list that is truly endangered um in the public market. So I think it's it's it's pretty case by case, but some of that's pretty good.

Yeah, that makes a ton of sense. Well, thank you so much for taking the time to come and chat with us.

Have a great rest of your day. Safe travels. We'll talk to you soon.

Cheers.

Goodbye.

Let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agents to deploy web app servers, databases, and more. While Railway automatically takes care of scaling, monitoring, and