Keith Rabois: AI app-layer companies must have positive gross margins, and the 'deal guy era' echoes the dot-com bubble

growth. And we have our first guest of the show. We have Keith Boy here in New York City coming on down. Look at this. Thank you, Keith. Good to see you. How you doing? [music]

Thank you so much. Thanks for having us. you are the mayor of New York now. Uh we have partnership with the New York Stock Exchange. It's a great setup. It looks great. Uh and obviously it gives us easy access to folks like you. So thanks so much for coming on down. Uh how have you been? What's new in your world?

Great. Pretty busy. You know, of her adventures pretty busy right now. People doing deals every day watching left and right. Crazy valuations.

Is that a green flag for you or is that a red flag?

It's hard to tell like and I think a lot of people are going to lose money. Hopefully we're not included. Um but the valuations have an extra zero.

Yes.

You know sometimes two

literally um before this recent like sort of trend I had never invested an entry price for like any normal company above an $86 million valuation.

Yeah.

And you know that whole world is completely different.

Uh so like everything seems to have

How many deals a year are you trying to get in under under 10?

Like I would imagine all No, I know. I know. Realistically like in the last year maybe like one one a year somebody you knew well so you can

my philosophy is bold early impactful bold early impactful that's the KV mantra you know I I love that

so early means first institutional capital

even today if you're right bold and early that's still maybe over $10 million it used to be you know 500k when I invested Airbnb it was 500k 3.5 million post. Wow.

Door Dash was something like 8 to 10 post.

Yeah.

So, you know, the world's very very different now. They're going to be epic companies. Like what is very clear is some of these companies are absolutely iconic and will be iconic for decades.

I doubt there's as many as people think. So, the argument goes that, you know, we've changed the physics and there's going to be 40, 50, 60, 70, 80, 100 billion dollar plus companies. I don't believe that stuff. Uh, certainly not on the consumer side. I don't believe it on the consumer side. Possibly an enterprise if you believe AI is going to transform every business everywhere possibly. But the consumer side, consumers are busy. They have 20, there's only 24 hours in a day. Tech doesn't change that yet.

And so every time someone adopts a new app, a new product as a consumer, they have to substitute from their friends, from their family, from church, from basketball.

And so the bar just keeps going up. So I don't really believe there's going to be too many breakthrough consumer companies. Chat GBT, OpenAI will be one. Not really sure who else.

And it just feels like we're in this era where if you do have a breakthrough, the big tech companies are in an offensive position. They're in wartime mode. Some of them like Meta are are are uh they're founder still. They're still in founder mode. And so you're going to have a pretty good copycat on your tail within 6 months of you going viral, within a year. That is an interesting dynamic. Typically the large companies we don't all make fun of you all entrepreneurs and we don't make fun of these old sluggish you know large companies with tens of thousand now they're our friends well they may not be your friends but they are fast and they are paying attention like they're usually sloppy and slow like I remember when we were doing PayPal eBay was like the most incompetent organization on the planet like Google was literally being built underneath them and eBay hired like management consultants at Bane and told them don't worry about Google like that was the classic like large tech company and admittedly bag was worse than most But like still now, you know, all the large tech companies except Apple seem to be paying attention. Um, you know, Apple is behind the eightball in AI.

Yeah. Unpack that a little bit because there's a lot of departures and, you know, you're wearing an Apple Watch today. Like, you're not giving it up anytime soon. I love my Okay. What's the nature of the risk? Because I don't see us actually going somewhere else because we're locked in, right?

No. I think, you know, building vertically integrated products. So to do hardware well you need hardware skills chain battery innovation

that's before you get to the software but does somebody figure out how to do an AI unique unique to AI custom device that is compelling possibly depends on what time frame we're talking next year unlikely

10 years I'd be one of that you know more more likely than not

so at some point Apple will be

Bernard Arno has a best quote here he's like in something like in 50 years will people we be drinking champagne. I think yes. In 50 years, will we we all be using iPhone? He's like, I'm not sure.

Yeah, I'll take the champagne or at least tequila. [laughter]

Um, but

how is your how is your uh kind of view on Google's uh like competitiveness in AI changed over time if if at all?

Well, I we had a debate several several years ago. Is it you know a sustaining innovation, you know, disruptive innovation? Um after chat GBT it's clearly more disruptive potentially. Um and I do think though they're they are paying attention. They are taking it seriously. That said Chad GBT is still the fastest growing product consumer product in the last like 15 years.

Um and I think if OpenAI plays its cards well and focuses on Chad GBT versus gets distracted as being a pure research lab. So there's probably this tension there. Yeah.

Research lab AGI blah blah comes at sacrificing the quality. It's not just that, but it's it's a let's launch a short form video app. Let's automate. Let's do all the science. Let's do they Sam I mean I think uh uh Sam's answer to Brad which kicked off like a broader discussion. He was saying we're fine. We're going to do consumer electronics and we're going to do science and I didn't I think a lot of people read into that and just say like hey Google does that. They lose a lot of money on it. It's not exactly and those are their bets. Google means money so that it's going to be clear so they can fund it. Um I think there's some things that Sam has to do. Hardware is probably one because the risk that somebody else puts AI together on a consumer device before he does

can rip them sort of out of

what do you think about what do you how are you processing Apple with Gemini as a risk to chatbt?

I don't I I

because to be cuz cuz like to be clear like Gemini has an amazing model. They're well be behind on a just a product uh product experience standpoint, but there are quite a lot of searches that I hit Chad GPT or Gemini with that like a Siri integrated with Gemini, I'd be going there right away if it was like

Yeah, it depends on the kind of query you're doing or prompt you're doing.

Yeah, but right now if I if I hit like for a fact that's not that far away that's not that far away from doing a a com like a a like

Sure. But then there's other kinds of prompts where you didn't used to maybe even use Google. For example, I asked to write a book outline for me proposal.

I wouldn't have thought that [laughter] you could you could probably one shot a book because somebody was outline was actually moderately creative. What I expected and this blew me away was that there was a spark of ingenuity to the proposal. It wasn't just like right down the middle. Um I tried it on Gemini.

Are you writing a book?

No, but maybe one day. You see the problem with long form content is apparently because of token ratio ration rationing rationing. Yeah.

If you try long form content they they'll only spend a certain amount of tokens on you so it degrades. So you'd have to write almost like a paragraph or a page at a time otherwise the quality will degrade. But the outline the proposal was actually quite good. I did try it on Gemini. It wasn't as good. So I think 5.1 5.1 thinking mode

actually still has a bit of a human personality and I don't see that when I try the same prompts in Gemini. So I think you could build an excellent cutting edge

person human personality product if you focus and then I think you do need to try a device just to make sure you know you're kind not blindsided by somebody else.

Some of the other stuff though may interfere. There's only so much time. There's only so much band. Sam only has so many hours in the day etc.

Um may interfere with success with CHBT. Chad GBT should be a $4 trillion business if you maximize it.

Yeah. Talk about uh hardware. It feels like we're in this weird era where everything is getting more capital intensive earlier and earlier there we

So this is a good question. I think there's some companies where that's true. Some types of markets, some types of products. I think other people are raising capital for no apparent reason. Like let's talk Harvey Legal AI. I have a competitor investment that doesn't need a lot of capital.

Which one are you in?

Uh it's called Spellbook.

Spellbook. They target somewhat different markets, but they directionally legal,

but there's no real reason to spend that money. Like like in other words, their compute costs can't be that high.

Sure. Because they should be able to pass that through as gross margin. Like they should have good gross margins on day one.

So when I invest in AI companies, so 2/3 or 3/4 somewhere in between of all my investments over the last 18 months are AI based.

Yes.

Most of the application layer but not all.

Yes.

Of the application layer ones. I expect them to have positive gross margins. Extremely positive gross margins. Actually, I'm like, you have to build a business, too.

And I don't believe in these excuses. Now, if you're building frontier models and cutting edge or you're going to do hardware, you're clearly going to spend money or infrastructure clearly going to lose money intentionally. And that's always been true in tech. But if you're building an application layer, I don't buy that margin upside down.

There was this hilarious uh expose on HackerNews about uh AI startups that aren't actually training their own models. They're just rappers. And they were breaking down the gross market. was like, "This is a VC's like like celebration because they were like

the only thing that was bad is they were kind of making the claim that they were like training models when they're not." But the the my takeaway was that like the rapper market is extremely healthy in certain in certain markets

because workflow like if it you have to productize things. Yeah.

Like if I'm working as an investment banker

Yeah.

or an accountant or a lawyer,

you it's not just the performance of the bottle that I care about. It's like how efficient, how improve, how how does it improve my life. Yes. And that's a function of how intuitive is the UI, how easy does it make the entire job to be done, so to speak, uh, be able to be accomplished. So, it's not just a model performance. I think when I'm looking for like a consu like an application layer, uh, CEO, I want them to really understand what their ability for, why, and what the P&L impact to their target customers are. Like in legal, like it's actually a complicated question because of the billable power. You don't necessarily want to make your lawyers more efficient if you're on a law firm.

We've been talking we've been talking about this so much. We had a I know a lawyer who switched firms because he knew he was with a big like probably top 30 law firm. He's like they are not going to adjust in time. Like there's no in they're like we like that our associates are just running up hours on tasks that could be oneshotted by AI.

I built 300 3,600 hours my last month as a lawyer. And and and what what the last month is Yeah. Next time someone tells you they're working too hard in tech. Although I literally build three 3,600 hours. That was not my average month, but still I hit over 3,000 a few months a year though. Usually today we'll do like a few thousand. The culture has degraded to like

Wait, but uh how do you build 3,000 in a year?

I wasn't anywhere near the top 5% for a year. Yeah. Yeah. Yeah. 3 Oh, sorry. Here it is. 360 hours in January 2000.

Okay. Okay, that makes sense because the top associates in S&T were over 3,000 for the year. Yeah, I'd be like 26 2700. So, I was a slacker.

Uh I mean, you're a buyer of legal products as well as an investor in legal AI products. Are you advising your uh your portfolio companies to push harder on their law firms to get

Well, they are. So, every founder I know actually uses chatbt for legal research. Then they call up their lawyer and their lawyer's giving them advice and they're literally sometimes they're actually screenshotting and saying what you're telling me is wrong.

Does CHBT need some like legal like clearly a lot of people are using it, but there was just a judgment today that was requiring CHBD to turn over a bunch of stuff.

So it's not technically not privileged. So it's complicated to use it. They could

we'll see how the law evolves, but it's probably the right decision as a legal matter. Um, but I do know I just had a conversation with a CEO of a more than hundred billion dollar company

and he has a legal team that spends $40 million a year and he authorized all his lawyers to use chatbt.

Wow.

He's just like you need to induct you need you need to adopt this tool efficient right of course. So if a lawyer technically uses chat GBT it might be privileged but if you as a consumer normal person input stuff like oh you know how do I hide the murder weapon that is definitely not privilege. for them. Now,

I I wonder I wonder if it uh there's certain businesses that uh start and uh build knowing that they're going to get sued a lot. I wonder if this levels the playing field for businesses like that that are kind of going after markets that don't want innovation and they can just say like, "Yeah, we would have needed a team of 20 lawyers back in the day, but actually now we can have like a few like really efficient."

Yeah. Well, there are startups in the legal space that basically are a substitute for a lawyer. And then there's the legal AI startups that make lawyers more efficient.

And which is uh which is Spellbook? Spellbook is making lawyers mostly making lawyers more successful for firms and now for companies. We have a lot of like uh actually fortuneous companies because you think about the incentives of a corporate legal department. They're different than the law firm frenziers like corporate legal

departments want to spend less and get faster responses. So it makes tons of sense and then you can customize to the culture of that company. Yeah. So like eBay uses Spellbook for example and they have their own preferences about how they want to practice law and we can implement that on software versus like Google's not a customer but Google might have different preferences.

Yeah. Yeah. Yeah. Yeah. We we've been seeing this new trend. I mean the the the lineage here in legal is clearspire and then atrium and and now people are trying it again. What do you think about folks who say uh I was on a particular track where I'd be at a firm now I'm going to start an AI powered firm. Should they be raising money or should they just go do that on their own?

Usually the way they start though is a human a ratio of humans and models. It's not like just all model, right? So you're going to have to pay lawyers for a while and like you want to take the ratio from 80% humans to 80% automation and like hopefully quickly. Atrium basically never did that and maybe they didn't have the technology to do it. They got hooked on the drug, I think. Yeah. Of just like

I knew some of the machine learning folks at Atrium at the time and they were like, "We're we're close to understanding documents with AI and but close doesn't pay the bills." You still need a human in the loop technically for most of these products right now. That's going to change and the ratio should be like one human for 99 customers or 99 documents or 9,000 documents and then the economics could work. But you I think you would need to raise venture money upfront also maybe for credibility. there's some signal you know like people are buying enterprise company enterprise all across the globe any vertical are buying AI products that's what's allowed these you know vertical application companies to hit you know 100 million of revenue quickly is there's top down demand at very large

for how do you how do you assess quality of revenue if a founder comes to you and says yeah I am raising at 100 million uh for my first round but I have a $5 million deal with some Fortune 500 company and so it's really not that crazy of a revenue mult But then you start wondering obviously is that going to stick or can they just bounce?

We ask for the contract actually because the terms may vary. We call the often ask to call the customer like why are you buying this? What do you hope it does you know and achieve? What are the business goals? How confident are you that you're achieving those business goals? So that's actually a pretty standard reference. Yeah. Does that mean that uh like the level of the level of due diligence required on these deals even if the early stage like this the the the round numbers it might still be a seed but it's a bigger number you're treating it with the due diligence that you would bring to an I think I think you do need to look at what's the source of the revenue what are the terms of that relationship are they a pilot permanent do they have cancellation rights and then you can look at the underlying metrics like we always used to which is what's the active usage like ultimately everybody in your organization is using the product even if you have contractual right to cancel, you're not canceling it. So like you can look at MAUs, DAUs, all those kind of things. And actually we do tend to look at that. I tend to personally look at the engagement metrics as much as the revenue metrics. You can also look at the margin metrics because ultimately like yeah, if you give away someone pays 5 million that cost you 10, it's a great deal. [laughter] Like that's not a long-term

business. I mean, if you're if you're a enterprise uh you know uh CIO or CTO, uh you're probably getting pitched some stuff where you're like, "Wait, I was going to buy those tokens anyway." Yeah, was going to be and I don't even have the infrastructure to run it. So, it's like totally I don't have to buy I don't have totally

there was a lot of bearishness around certain vertical specific AI tools and enterprise AI after that MIT report came out that nobody read the source material uh and and uh just anecdotally it feels incorrect. Uh but uh I think the the simplistic way to look at it is like will large companies spend more on AI in 2026 than 2025 and I think like you lean like

the answer is definitely yes I mean I think you could critique you're right nobody read the report they just write look at the highline you know X or something but I think it comes down to as talked about the quality of the implementation if you have the right people implementing it you're going to get high quality results if you just think the software is just going to show up out of the box it's going to work and transformation highly unlikely unless the product was designed to be that way. There are a few products that you see that you know are designed for a consumer to just like or user like an average employee to just use out of the box. That's pretty rare. Usually you have some implementation hurdle and I suspect a lot of these things are getting trapped in the implementation sort of you know uh box somewhere.

Yeah.

You guys have stayed out of the prediction market wars.

We have not invested in a prediction market. Is that was that uh like was there a moment where you wanted to and the dynamics of the round didn't work out or was that a conscious choice?

It's a it's a good question and I had the maybe I've been burned because I helped uh start and like seed invested way back in the day this thing called blueap which is an early prediction market. Um it's a really cool product Kevin Hartz Javin and I worked with this founder on back in the day like 2005 or six

from uh YouTube or PayPal.

Yeah. Yeah. Right. There's never any new There's never any new ideas.

We we could never get it to work. The hardest part that made it work wasn't like on the better side or the the wager side. It was actually getting enough bats.

And maybe now with AI, you can scalably create enough interesting bets. We had to do it by hand and we just never [clears throat] could get that treadmill working

or you maybe didn't do sports bet sports.

Well, we didn't do sports betting. It was illegal. So, yeah, obviously the on traction has been sports and then cleverly politics. We did have political bets there. Um and so anyway, maybe I was burned by that. Then there's another question which is I like the political markets. Yeah.

How you know I said we're bold early and impactful. You do have to start think about which market segments are actually positive impact in society. Like just giving average Americans more places to gamble certainly is not something I want to like invest in. Now there are markets that are illquid and they the byproduct of the liquidity is a good thing for society like politics. Yep. But I don't want too many people speculating like Americans don't have a lot of money.

They gamble, that's fine. They bet on sports now, which maybe is fine. They bet on crypto whether they call it that or not. So adding more speculation to the American, you know, consumer, I'm not sure is something I want to, you know, invest.

How much how much attention healthy American consumer? How much attention have you paid to just this tension between states clearly want to regulate this stuff and yet the CFTC has decided like hey these are event contracts like we're it's under our jurisdiction. How how have you kind of like watched that play out?

Well obviously the poly market guys have played their cards really well. Um controversial you know at the beginning probably very high risk. You know Alfred I saw Alfred Lynn talking about a little bit. I'm sure he probably opposed it, but the founder was pretty tenacious and it worked out. Or, you know, they all played the regulatory cards, you know, in interesting different ways. Um, so it's worked out well for them. Uh,

but the story is not over.

Well, I think they're in pretty good shape. Um, you know, they the biggest issue they had historically was more not more US citizens were using their products before they were supposed to. Yeah.

But it I think the statute of limitations might sort of be hitting on that.

Sure. Yeah. It's fascinating. Do you think this will be a political issue in the next cycle? Uh in terms of tech stuff that we've been monitoring, it feels like the uh the the AI data center energy use, water use, like those are the questions that people want to answer in DC. Job displacement stuff

displacement is obviously top.

That's a bigger question than than okay. Yeah. Like my, you know, my uncle, he was always a sports veteran, now he uses a different platform. Like uh it's kind of

I think unless people start losing a lot of money. So, you know, in sports betting, you definitely have this. You have the opposite of Wales people, you know, Wales the company, but like people lose massive amount of money and that becomes a really sad story for them, for their family, and it becomes a political issue. Then

if for some reason you get more more people betting, betting, wagering, whatever, predicting,

but they're not like massively exposed. I think that it won't be a political issue.

Yeah. Yeah. Um, we I mean, we've been tracking AI diffusion. We're obviously excited about a lot of the products getting in people's hands like like Spellbook, like law like law firm for like legal AI. Uh at the same time, we're just not seeing that much job displacement. Maybe it's coming, but it feels like there are if whatever's happening when somebody lays people off, it's probably not because they're just, oh, now it's just a prompt. Uh but but do you think tech needs to do a better job of telling that story or understanding how?

Absolutely. Tech is not I mean you know I was in a speech that President Trump gave and he says like I don't like the terms artificial intelligence. You guys need better branding

and nobody's come up with a better answer. I think someone should like artificial intelligence sounds scary. It sounds bad artificial when you say like you're having artificial sweeteners you know. So like I think you can rebrand it like that.

Yeah. Now it really is like a power up. Like if you're very creative,

credit to Sachi Nadella and and the Microsoft folks like like co-pilot feels like okay I have a helpful assistant that's alongside me. Does the co-pilot ever put the pilot out of business? Never.

Not yet. Not yet. Maybe you have a long tail kind of situational planes that makes it prudent. But I I don't think there's there's going to be job elimination. I think there's going to be job substitution.

Sure.

You know, we're going to we need more data centers. Well, to build data centers requires a lot of advanced plumbing,

a lot of advanced electricians. These people get paid a lot of money. Um, like Yeah. 3 to 5x easily the median income in the United States. So, we're going to be building a lot of infrastructure. Building infrastructure creates jobs. Are the people who, you know, are doing this now

going to have to substitute something else? Maybe.

But I don't I do think there is going to be some substitution for white collar work like accounting, lawyers, maybe do primary care doctors.

Yeah.

So, what advice would you give to um high school student?

I think you're still like read learn reading, writing and math very well.

Even though the computers can read, they can write and you can definitely do math.

You need to know the fundamentals. I mean like there's a research paper I I I retweeted this week about if you if you handw write your notes versus type your notes in class on any metric you match back

and where if they're in middle school, if they're in high school, should they be fighting to the nail to get into the Ivy League? Where are they going? Where are they learning?

Oh, that's a good question. Chill fellows seem to work out well. It does. I'm a fan. I'm not, you know, my kids are four. I I hope by the time they're better, they better not have to go to college by the time they [laughter] grow up. It' be really sad.

You're not saving for it.

I know.

Sorry. No college on AI. [laughter] Um, we'll get a little data center. What about what about for the the the next generation of entrepreneurs that are just getting to Silicon Valley right now and they're seeing a trend of what we call like the deal guy yuga, the deal guy era. This idea that uh it used to be you get to Silicon Valley and you're writing code and you're in and you're in the basement or you're in the garage. You write the code and then you put it on the internet and everyone shows up and your elegant algorithm just produces Google and you create a trillion dollars in value and it's because you're the PhD scientist, you're the mathematician that you get rewarded. And now it feels like you can maybe make it in Silicon Valley maybe not caring so much about the technical aspects but instead about the deals putting pieces together.

So the funny thing though is like people use Sam as an example of this but like he was running a research lab for years and and so like even though like

he was like he was a CS dropout. He was yeah he was also a CS drama and like definitely built like an app uh bubble yeah he built an app but now so the reason why it's scary and this is maybe betraying my age is back in the late bubble internet bubble 96 2000 there used to be all these what we used to call business dev people running around all the companies and all doing all these deals

and then the bubble collapsed

and every a lot of people ascribed the bubble and collapse to these business dev people.

Oh interesting.

So everybody stopped using the label. When was the last time you met a business dev person like never like for the last 25 years. BDR BDR rebranded. Well, Bisdev was like higher like partnerships.

I was like I come to you I say Keith I'm going to I'm going to give you $100 million of revenue. Funny enough I'm also going to give you invest $100 million in the company. I'm going to give you $100 million in revenue. That was the classic that was the classic example. So that's how I started my career in tech actually was a business head person. And if you look at my LinkedIn profile it's still there [laughter]

but it became like this you know sort of negative signal on your resume. those business have people running around and so everybody's doing that now that is a scary indication that maybe we're hitting you know something

yeah yeah so I mean would your advice to new entrepreneurs say ignore the noise ignore the frothy rounds uh retreat to what product market fit programming

I I think look ultimately deliver value I think it's actually really simple building a startup

it's like you create a value proposition

yes

and then you explain the value proposition to people

and then you actually yeah there's product there's market you know that's basically it like what is my value proposition to who like what do I do for you that's good for you it could be good for your business it can be good for your life and then can I articulate that in a way that causes you to interrupt your life and test it out and verify those claims and then rinse and repeat so you can create a value proposition through massive use of technology create it through design you can create it through the intersection of humans design and technology it doesn't actually matter but you have to have a very crisp succinct compelling value proposition. In business cases, you want to move the P&L of your customers. Like I had an enterprise company, maybe the only enterprise, true enterprise company I ever funded and joined the board of where our first customer is Walmart. Everybody's like, "No, how's Walmart going to be your first customer?" And they spent $5 million on us.

The reason why was we figured out there was one of the top three priorities for the company we could address. Our second customer was AT&T, another classic customer. spent 10 million or it was a telecom. It wasn't an ATD. It was a different telecom actually. Next we could our technology allowed them to understand their churn better. What's the only issue in the telecom business? Churn. Subscriber churn. That's the only thing that matters and like you know cuz they all compete on like churn.

Somebody else is going to offer.

Nobody had a technology that could literally take all their data and actually figure out why people were returning. So if you have a right the correct value proposition all the rules are changed. So just dial into the value proposition and then figuring out how to articulate it, which sometimes can be different for my post. Sometimes you have actually marketing fit and sometimes you actually have your product positioning fit and sometimes you actually have value proposition fit.

Sometimes you start with both, but sometimes you're bad at one and it kind of masks and is not clear that you're actually really good at

marketing market. Marketing market fit will mask the lack of product market fit for about a year.

The other one the other one doesn't get masked. It's hard sometimes hard to fix. Like you may go back to the product and think you don't have a value prop. You actually do. You just can't explain it.

Uh because you only get a certain amount of attention to explain. You got to distill it. Once you distill it to the right people, then it might take off without changing the product. But that's the art. True CRO is actually good at this art of triangulation.

Mhm.

Uh how do you uh there's been some kind of coalitions or or not really coalitions, but uh Walmart and Etsy have leaned into LLM commerce. They've partnered with OpenAI. Amazon and eBay have not yet. Shopify has. How do you who do you think is going to look smart in two years for like leaning into these platforms early?

They're leaning into learn. So there's a question, are they leaning in or are they leaning into learn? Leaning into learn is a great idea. Like if you're Shopify, of course, like will people use ChachiP in a way that leads to product discovery and product purchases?

We had a friend of ours has a Shopify brand. They do like 200 million of revenue. He says uh people that land on his site from chatbt convert at 12%.

Incredible.

That that I agree with. Sure that's true cuz they take they it's authoritative the recommendation the the signal it's better than a blue link.

However, the question is the volume and what's the increase in that volume like how much of today.

Yeah. So that's the question is what does that change and what triggers that change? What catalyzes it? But I mean you know we invested in a company called profound day. They're sponsor.

Yeah. So they allow you as a brand to track how various whether it's Gemini chatp etc are showcasing your products and brands. That's critical to the future.

But what consumers ultimately decide to do they actually start using chatp to recommend the next tennis shoe to them or not? I think in some vertical almost surely yes.

I want I want Christmas data so badly. I I I feel like this is the first year that we're really going to get an answer. At least the labs will now. I don't know if we'll figure it out, but uh about the agentic shopping, just the propensity.

So, let me give you an example of something they can solve. So, like think you're a guy and you want to buy a new blazer like I need a blazer so I don't look like ridiculous on like a nice Christmas

sweater. So, I got to figure out what what blazer is going to fit me off the rock. I don't have time to get a tailor.

Sure.

Almost an impossible task like today.

Yeah. Like like literally like you would actually even if you know your fashion you almost an impossible task like you might know your brands might

I can imagine though chatb kind of knows

you know what I buy how it fits what things I return what things I don't over time

and just says Keith go to Hugo Boss

y

38 or 48 or 42 will it fit you off the rack like that is something that doesn't exist online and there's a lot of latent demand for consumers not to go to retail like the reason why I go to retail typically is I either need something immediately or I need to try it on.

Yeah. Yeah. There's also this interesting thing where uh chat and basically every LLM has been only pull in the sense that you have to go and prompt it. It never just sends you a push notification and says here they they're experimenting with with pulse. So they'll say okay we know that you like venture capital news and you like learning about data center. They have depp prioritized it, but I still think that there's a glimmer of something valuable there where they could proactively send you results. It's very comput intensive and I don't know if they got the form factor right, but there's clearly there's clearly a glimmer of something.

There's a glimmer there. The question is going to be the quality, right? Like ultimately if Yeah, totally magical, but I'm going to

react to the first five they send me and either love it

or hate it and they're not going to get a second shot. Totally. Totally. Totally.

So, I think that's hard to do in a beta.

No, no, no. It's hard, but uh I think over time the scale like all the math and like the actual value we've seen it work with like some of the Instagram ads where you're like, "Wow, I didn't even know I wanted that. It's really cool. People are satisfied." I think that will come.

People are also more comfortable sharing data with CHP than any other tech product I've ever seen. So, I think like my sizing and things like that, you know, if they allow me, if I put that in memory, it makes it a hell of a lot easier.

So, so help me square that because I feel like that's true. I feel like people have these incredibly personal conversations with LLMs. They use them for all these funny things. They're obviously seeing AI search results. And yet AI as a whole feels like the most unpopular technology in decades. And

well, I think it may be a 1% issue too. Like there's a lot of people who are motivated for their own reasons, you know, to cause uh to create negative feedback on be [laughter] 5 years ago. There was a lot of interviews that happened where they were talking about job loss and and just like totally

well even Sam made the mistake of like talking about UBI. We need UBI like this is a mistake.

We need UBI because there's going to be job loss. Well, first of all there's not going to be job loss as we were talking about arguably but B like highlighting that a decade before any of that happens. Makes no sense whatsoever. And UBI by the way is a terrible solution. So [laughter] for all those reasons it created like more fear. And then the safety hoax people the people are always creating hoaxes about it's literally a hoax. Safety is a complete hoax. This is my challenge. Name a single person who believes we have an AI safety issue that's been right about any political issue for 50 years.

Every single person who's talking about safety has been wrong on the environment, has been wrong on equality and all that nonsense. So like I just don't believe any of this because the people are all always finding excuse for bureaucrats to interfere with progression.

I mean that that I I hear you. That's just like that's kind of sad for our tech community because uh

well, as you know, a lot of people a lot of people in tech though have very sad views. Yeah. [laughter] Yeah. But what I mean is like is like Mark Zuckerberg has deal has dealt with a lot of attacks on like social media is bad, right? But at least you got 5 years of people kind of liking posting Instagrams, you know, people kind of liked social media before they started critiquing it. AI day one immediately

which is critical like you know uh there's this old there's this old like line that's stuck in my brain from when uh Nuke Gingrich was speaker of the house. He said, you know, if the electric light had been invented today, it would absolutely be banned because it would have threatened the candle making industry. You know, the safety, you know, it actually was somewhat unsafe. Actually, thousand people died,

but we never would have allowed electricity and think about society without electricity.

Yeah, we have feedback loop on fear and and actually taking action and then monetizing that fear or one way or another.

And we saw this with nuclear power. you see if you see a a an AI like video uh that was generated and it's bad you're like I don't want to lose my job to this thing that I don't like you know and so I I I

well we have to remember people also there's a bell curve distribution of people's content let's take video

of all the humans assume there's no technology

what fraction of human generated video is great

or good even

it's probably a fairly small fraction

but then you see the AI you're like oh

you know this is terrible it's

the guy that made this said I'm going to lose my job and that I'm going to need UBI. Why don't we just stop all of this

alto together?

Yeah. Yeah. Yeah. It's a little bit like Paul Graham had this epic uh blog post about, you know, if someone um unfortunately like meets someone, you know, and sexually assaults them at a grocery store like Safeway. It's not like a front page news story for the New York Times that I mean this happens all across the country unfortunately. God forbid, you know, someone meets someone on Instagram in a DM. Front page news story, you know, something bad god awful happens. But it's the grocery, you know, versus the new technology. And so, you have to always remember that people are going to critique the new technology differently than what the real risk in the real world is. That's actually probably more material.

Uh, we have a couple minutes left. Uh, and we're here at NIC. What's your IPO kind of like outlook? How are you thinking about 2026? You think a lot

got some good ones coming up for you hopefully. Um, you know, ventures going out.

We've heard rumors about some other funds. We are definitely I can assure you we are not going to be we have never had the conversation about whether we should be a public company. We're in the craft business rumor swirl. IBO rumor swirl.

Absolutely not. We'll let someone else do that. But who else?

But um you know obviously we've been involved in really good companies. Ramp your sponsor obviously is phenomenal. Can be a public company whenever they choose to be. Um got a trade republic in Europe. Not as many people are paying attention to. Great future company. Um there's a bunch. like so I'm excited about the future.

Saw there was some news that SpaceX is maybe thinking about going out next year.

I don't have any inside information there. I mean I read I read the same thing. I think it would be great. You know it's 202 years or [laughter] something.

Yeah. A long time and also just a lot of Elon fans have been riding with Tesla for a long time and that's unlocked to invest private. They'll take it back public at some point that you can invest. through what what you're advising uh those portfolio founders without naming names, but uh uh on whether or not how they should be thinking about timing an IPO. You've obviously been along for the ride on many uh what are you counseling them on where when to do it?

Well, I'm always a fan of Go Public as early as possible. Okay. Yeah. So, I wrote a whole chapter in Eli Gil's book, High Growth Handbook that explains why.

One of the best books in industry.

It's a great book. I have two chapters. One about hiring and then one about going public. Sure. My general view is $50 million in revenue plus predictability. You should be a public company.

Wow.

Uh there there's so many benefits of being public. Um so I'm always counseling when you're as soon as you possibly can go early, not late.

What about what about burning money? I feel like there's another bar which says, hey, if you're a public company, you shouldn't need to raise money anymore. Has that changed?

Well, it depends on the company. Some businesses are profitable. Like Trade Republic wouldn't need to raise money, for example. Uh we actually pay taxes already. Yeah, [laughter] I've actually never had that happen to me. I've never been on a board of a company that's actually paying taxes cuz you know you to pay to pay corporate taxes, you actually have to offset loan and profit. So like we're such a great company. You burn through all your care.

Exactly. Wow. That's fantastic. Well, thank you so much. Is there anything else we should talk about being why why New York? Why are you here? Is it the ramp mafia? You want to make sure you pick off all

pick off all? So we actually looked at my geographic investments. So I mentioned AI is 2/3. New York is the number one geo has been the number one geo for me for five or six years. It's about 60% of all my investments. Application layer stuff, financial services, you know, like Imprint, RAMP, a lot of great companies. Basis, accounting AI, Robo, Investment Banking, AI, Traa is here, you know, RAM, they a great set of companies here. So, and then it's better for me to be on the East Coast for like personal reasons,

of course. Well, we appreciate you taking the time to come down. This is fantastic. I'm sure we'll catch up with you soon.

Thanks for the invite.

Have a good rest of your day. I'm going to tell you about Numel.

Compliance handled. I think I already told you about Numeral, so I'm going to tell you about Figma again.

Think bigger, build faster. Figma helps design and development teams build great products together. And our next guest is Lin Martin, the president of the New York.

While we wait, uh Dan Sroker's company, Limitless, has been acquired by Meta.

Oo, congratulations to him.

There's a video here that we cannot play, but Meta acquires AI wearable startup Limitless. Ooh, I we were talking about I feel like there's a lot of opportunity in the AI wearable space. There's a lot of cool stuff happening and like it just feels like fertile playground. I feel like it'd be a fun time to be a hardware hacker. We talked about that transcription hard piece of hardware that's making $250 million out of nowhere. Uh very interesting companies. There's obviously cool things to do. a lot of the a lot of the the you know the the the the a lot of the attention goes to the rabbit R1 the uh the friend pin the humane AI pin but there's cool stuff happening all over the industry and so uh yeah again voice recorder they had a hardware device

amazing

uh notable that yeah again Limitless I don't think was able to crack uh like real growth with the hardware but that company Plaude has and very so uh but again this is great outcome for for them uh I'm sure the whole Limitless seem as uh

and let me tell you about privy. Privy makes it easy to build on crypto rails, securely spam white label wallets, sign transactions, and integrate onchain infrastructure all through one simple API. Um what else is going on in the timeline, Jordy?

Let's see here. Uh

yeah, I I mentioned this briefly in the conversation with uh with Keith, but apparently OpenAI must turn over 20 million chat logs to PL

Judge Ona Wang has ruled.

Yeah.

Um

you have to send an email to a to a server that adds a Chat GPT response in with your lawyer CC. That's how you maintain confidentiality. Harvey who Keith also al also mentioned raised 160 million at a $8 billion valuation [laughter] this is why Keith was talking about 2% dilution round from A16C they 3x revenue this year to 150 million they raised 300 million from Sequoia at 3 billion in January 300 million at 5 billion from KOU and KP in June and then 160 at 8 from A16Z shout out to Spencer and the boys at at uh COU a little mark up there.

And I feel like I feel like law firms have got to be pretty discerning on making this purchase decision at this point. Like it's not it's not entirely exploratory budget anymore. And it's not viral like something that's like oh it's like a flash in the pan and then all of a sudden like they're not going to be able to monitor.

Yeah. My question is like if anybody that wants to say like oh this is like out of control it's like do you think they'll be able to get to like a billion dollars of revenue? legal industry is pretty big and and it's been traditionally pretty hard

and they actually they actually so so the labor displacement thing does feel real here because I do think a law firm would say hey currently we have 100 associates we can condense that down to about 20 and do the same amount of work by working with uh with something like Harvey and again the the vibe had had uh at least uh with a a lawyer buddy of mine had shifted drastically with Harvey a year ago he was saying uh we're like using it a little But now he's like it's oneshotting stuff that uh I didn't think it would.

He says I'm I'm oneshotted actually by

I'm one uh

well oneshot your data anal your data analysis with Julius AI the AI data analyst that works for you.

Uh this is

join millions who use Julius to connect their data ask questions.

We should have got Rahul here. You should have we should have told him to come here and be in the capital with us. Uh

we'll do that soon.

YC has a new company called the hog. Oh, wait. Really?

It's called the Palunteer for Go to Market. We were just We were just talking about

uh Yeah, we were talking about swine themed startup names. You have Pig. You have

apparently carried no interest. Watch their launch video twice. I evaluate software businesses for a living. Still