Keith Rabois on founder mode, Airbnb's pivot, AI consolidation, and why micropayments will never work

a pharmaceutical trial it will use command line tools to contact the press contact regulators try to lock you out of relevant systems or all the above. That is absolutely crazy. But uh we'll see if it's real or fake news. And let us bring in our next guest. We have Keith boy. Welcome to the stream Keith.

Hey, how you doing? There he is. Welcome. Welcome to the show. Great to have you. Yeah. Yeah. Great to have you back. The entry song. Yeah, it's a little messy. We're figuring it out. Uh give us the recap on the KV Summit. How'd it go? Yeah.

So, every year uh for the last 15 years, we've hosted uh all of our CEOs, like 150 plus CEOs at Caval Point in San Francisco. And it's like a two-day program where we bring in inspirational and actionable speakers.

So, for example, I interviewed Eric uh from RAMP, John Collison of Stripe, uh John John Coulson interviewed uh Bill Gates actually and Sam Alman spoke and then the CEO of data bricks. So, we have a wide variety of content.

Uh some some of it's more practical and tangible and some of it's very aspirational, inspirational.

Uh so we take our CEO time very seriously and we want to make sure that every minute they're not in the office you know managing executing uh you know growing their companies that they get you know disproportionate returns out of the time. What was any type of surprising takeaways or or insights that that you can share?

I want to know what the biggest debate points are like what what where are the things where people are actually debating. I've I've uh you know uh Sam Alman recently said AI could be an operating system. There's questions about foundation model layer value acrruel versus application model layer value acrruel.

How many winners there will be in the coding market like what are people actually debating and unclear about and then I'd love to know where you stand on it obviously but uh like the non- consensus topics are probably the most interesting to dig into here.

Sure, we can talk about all those and generally speaking, we will post all of these videos online. So, there's a vast library of, you know, 15 years of virtually anybody who's been successful in tech has spoken at our conference. And so, in fact, Sam's probably spoken four or five times now, right?

Um, it was interesting. I compared when I was interviewing John Carlson, I'd interviewed Patrick Carlson literally a decade ago. Oh, wow. And it was an interesting contrast to show the arc of the company.

When I interviewed Patrick in 2015, uh, Stripe was worth, you know, several billion dollars, but it only had 250 employees. And now, you know, as John mentioned, they have 9,000. 900,000. Wow. Yeah. So, it shows you the arc of, you know, building a company that a decade had passed since I last interviewed Patrick.

Um, so we will share uh 90 plus percent of these videos. We'll start posting them next week. Um, there's maybe one or two. It's up to speaker, you know, if they don't want to share certain things. But the contrast I noticed one a was one about building a company. We talked about founder mode a fair amount.

Um so you know Eric talked about how they build ramp and John gave a slightly different perspective on building stripe and then I interviewed actually Gary Tan YC to talk about the origins of founder mode the lessons from you know thousands of companies that you know they've invested in for 20 years.

Um and so I think the operating styles and approaches to hiring senior talent or not um was an interesting contrast on AI. Uh we had Brad Gersonner of Altimter speak and um talk about the future of AI as applied to pre-existing companies. Uh and he was quite inspirational.

He made a couple points that I think Sam generally would have agreed with which is there's only going to be one foundation model research labor oriented successful winner. It's certainly not um something that VCs should be investing in at the application layer.

I think there's more room and opportunity or something that transcends what we think of research labs today just like open AI and chat GBT is transcending what we think of search and you know to some extent what we think of social I think the next generation will be something that doesn't look anything like an LLM doesn't look like a research lab and you know it's very difficult therefore to find but hopefully some founders have visions and hopefully they call me when they do.

Yeah. Um, I feel like Brian Chesy was a big inspiration for the original founder mode essay by Paul Graham. Uh, he just went extreme founder mode a week or two ago with the Airbnb reimagining. Um, at the same time, Ben Thompson's kind of uh analyzing that.

And Ben Thompson's argument is that uh Airbnb is a platform and a marketplace company. And so the economics of the platform kind of dictate a certain user experience mainly that people open the app. Yeah, they have billion they have hundreds of millions of users.

The user base is massive at Airbnb, but the average user opens it like twice a year and they want to get in, get out, and it's not a daily active user experience. And so, uh, Ben Thompson was saying there will be some challenges. What was your reaction to how Brian positioned Airbnb's next chapter?

Do you like that communication style? Are you optimistic about where Airbnb goes? Uh give us your give us your reaction to the Airbnb news. So yes, Brian pioneered what we now describe as founder mode even though it's very specific in some ways the operating style of Apple.

But in in any event, um he triggered this at a YC alumni dinner in Napa Valley where he kind of off the cuff without preparation spoke for two and a half hours and that's what led to uh Paul Graham summarizing and distilling it which led to a lot of intrigue and interest and this is kind of what I think CEOs were talking about privately but no one had really stitched it together and kind of created a public discussion of it.

I for those of you really interested in the topic I interviewed Brian a few months ago uh at Gramp's offices and that video is available online where Brian exh is incredibly articulated eloquent about the virtues and the benefits of founder mode and how he actually implemented it.

So I highly recommend anybody in the audience who's intrigued watch the actual video because I will not do it justice. Um the on the specific topic of Airbnb experiences, I think Ben made some interesting points. I've been a fan and early reader of strate.

It's the only thing I read regularly in all of tech because I think he's the only one who actually understands tech that writes. Um so I actually have been recommending this to friends of mine for over a decade who want to enter into tech is like start reading to checker. Yeah.

So, I think the point he makes first of all is that in services world, you're going you potentially want to get off the platform once you find a service provider you know that you like. I think he missed one subtle point which is Airbnb is still designed mostly for travelers. I. e.

, you know, I'm going from New York, let's say, to Cincinnati. And obviously the canonical case is I would find a place to stay in Cincinnati. Great. But as you point out, the average user, you know, might travel 2. 5 times, 2. 4, 2. 5 times a year.

So, it's hard to turn my Airbnb usage into a daily, weekly, monthly habit. However, the point that Ben glossed over a bit is that I'm traveling to Cincinnati and then to SF and then to Dallas and then to somewhere in Utah. I have no incentive to go off the platform to find a service provider. Yes.

in a city that I'm in regularly, let's say New York City, if I found like a hair stylist here or the equivalent, I may have an economic motive to just, you know, disintermediate the platform.

But if it if Brian's right that curating experiences and services that a local would use still appeals to people who are in New York City, which is a huge f a huge number of people come to New York City every year, then I think Ben may be wrong and Brian may be right.

So, I think Brian is probably uh uh sort of dialing in to the traveler visitor use case. Totally. And then you need the breath of services when you're traveling and experiences. You don't necessarily have to only use it in your home city, which is where Ben may have some valid critiques. Yeah. Yeah.

It seems like the the dog walker example is probably the rough one because it's the most obvious disintermediation point. Uh at the same time, yeah, there's plenty of times when big group is traveling and they want uh you know, private chef to come cook for everyone.

Something like that makes a ton of sense in that context. And then also it's it like you can build a good business in a referral business that is not less it's less transaction. It's more discovery and advertising basis.

I was actually uh like not even half joking just saying that maybe Airbnb needs an advertising product because Uber's been so successful in advertising.

I don't I I mean I think that that might really frustrate people if they're seeing promoted listings, but at the same time like on these marketplaces, usually advertising does pretty well. I don't know if you have a reaction to that. Yeah, I have two reactions. First of all, I don't think Airbnb will need it.

Airbnb's margins are like roughly take rates roughly 13%. Sure. And you know, versus like the Ubers or the Instacarts where they really don't make too much money per transaction. Got it. And so if you don't have highly profitable advertising revenue, it's hard to build an interesting business.

So I think Airbnb doesn't have the economic drivers that you see with Instacart and where it has been successful, but it is some annoying.

That said, you know, Google used to run this internal study where they would actually show users who are exposed to uh SEO content and paid advertising are more satisfied than users who are searching and only see SEO content. So advertising when done properly can be a value ad.

If you think about it, the economic incentive of drafting content that explains, contextualizes, and markets products does drive to efficient advertising that can be for a user.

Even today, if you're in an area and some service provider is running ads, it shows a level of sophistication and professionalism that might associate in some ways with just a better end product experience than somebody that doesn't know how to set up a Google ad.

I wanted to ask uh Ben had some great thoughts around you know the entire internet is based on you know advertising and in a time when you know we may all have agents that are just trollling the internet for us and acting on our behalf the you know economic the existing economic model of the internet could uh hit some snags and he makes a case for you know micro payments to allow people that produce content to get some benefit from having models uh you acquiring information on those sites.

Uh he also outlines how this is going to be incredibly difficult to pull off because there's so many different groups and incentives. I'm curious if you've spent much time yourself thinking about the potential for microtransactions in in in the context of replacing lost advertising revenue.

So similar similar observation I think the post the original sin of the internet by Ben is totally worth reading and I generally agree with the point that the original sin of the internet was that all content should be free.

I actually made that point to Jessica lesson when she was launching the information in like 2005 67 8 whatever it was. I was like you definitely are going to be on the right side of history by charging a premium subscription. ran into Dick Castello at that party, the launch event.

And I I actually used the term the original sin of the internet was that we we all learned the lesson that because the marginal cost of content, the marginal cost of distribution of content would go to zero, you should charge zero, which is absolutely false. However, microp payments make no sense.

Ben is totally wrong about this. I'll give you an old lesson from PayPal.

So, one of my jobs at PayPal between 2000 and 2003 was to find new markets because we were really dependent on eBay and that led to, you know, risk and we ultimately went public and then sold the company to eBay because we were nervous about the risk of eBay handicapping our future growth and as you know tech companies are valued by the next 20 years of cash flow.

So, u you Peter and Roloff and various people were very nervous about that. One of the markets I wanted to explore was quote unquote micro payments and Peter Teal made a very astute uh sort of rebuke to me.

They said like that's the dumbest argument ever basically because content has a marginal cost you it has a gross margin of about 99%.

So basically whether or not microp payments cost you 5 10 15 20 25 30 35 40 50% of the content who cares like if you're writing content that people pay for because your because your gross margin is like in the 90s you shouldn't really worry about whether the microp payment cost is too large.

Now the friction of making the reason why I wanted to build the product was I think the friction of microp payments is too high like the consumer friction of paying for anything like you know you know you're going to read a short blog post you're going to read a short post you don't really know exactly you know what it's worth and so if you have to do anything other than just one click forget about it but I still think people get this analysis wrong that it's not an economic problem for now music's a little different by the way where you owe a license owner a reasonable fraction of the transaction then you can't be so you know sort of whimsical about the marginal cost of the payment but most people do not really own you know have to pay out a meaningful fraction of that dollar transaction to somebody else some third party so I I I think this is an excuse by the way Stripe originally originally started when Patrick John uh pitched me on Stripe in 2010 they originally thought that the core use case was going to be like micro payments and accelerating content.

They still have the GDP of the internet but they really felt that it would unleash new content which hasn't for the most part happened. You know, Substack is an interesting business. It has real GMBB and and an associated real revenue line. But I I still think it's the wrong place for Ben to be focusing on. Yeah.

the the issue with looking at Substack or even Ben and their success is there's some element there of I just want to support this the person creating it. I want them to not think about getting a job or building a business other than just writing online and spending all day thinking about in Ben's case tech.

Whereas for the average site on the internet, the you the person that would be making the microp payment, it's not like they don't have any incentive to be like, I care about this person and I want them to have a, you know, I mean, with like the recipe example, I'm fine if I'm baking an apple pie, blending three different recipes together and give me the LLM result of that.

But if I want to listen to Michael Jackson's Thriller, I don't want half of Thriller and then half of another song kind of clued together. Like, I want that exact piece of intellectual property. And so there's a very traceable lineage of that of that data transferring.

And so the microp payment makes a lot more sense when you're actually buying that specific thing, that specific thing. But I think you're better off having a differentiated voice and a premium for that. It's very difficult. Yeah. How many how many people in all the realms of content have a truly differentiated voice?

Like in sports, you know, Bill Simmons in his prime definitely did despite all the other sports writers. Then in tech I still think has an incredibly differentiated voice. It's just very difficult to achieve that but if you do you can absolutely charge a premium. Yeah totally.

Um what's been your it feels like it's been AI week. We had Microsoft build on Monday. Uh yeah I know you were tied up with the KV Summit but uh Google IO Google IO Open AI IO open AAI bought IO for 6. 5 billion. Uh Anthropics dropping cloud 4 today. are you tracking these things? What stuck out to you?

What made it uh above all the noise? What was the signal in your mind? What are you tracking? What are you uh what's ch has anything changed your mind this week?

Well, I think what was interesting to me because we do have the CEO summit, you know, annually this year, the AI content, not just the AI content that we programmed, but the AI content in cocktail conversations, dinner conversations was an order of magnitude greater than last year.

So, the first derivative was off the charts on AI. Last year there was like a vertical of AI and a vertical a slice of uh conversations and the people who are not in in AI space were probably intrigued but it it wasn't infusing everything.

I felt this year AI infused every conversation up and down the stack so to speak laterally across verticals whether you're in financial services or labor marketplaces it didn't matter. Everybody wanted to learn more about AI, how to how to leverage AI, put on this bionic suit for their own business. Yeah.

Are you seeing AI pop up in unexpected seedstage companies you're betting on? We were talking to TJ Parker. He launched General Medicine today and uh he was like, "We made the deci. We made the deliberate decision not to give the user a text box.

We're not a chatbt rapper in that way, but we're using LLMs all over the place to parse uh insurance policies and medical records. And this business is only possible because of AI. And I feel like that's going to be an ongoing narrative, but it should dominate the conversation.

At the same time, it should melt into the background just like, oh yeah, you're hosting your your new company on the cloud. No big deal. Everyone is. Yeah. I mean, at some point. And same thing like the adoption of you know some Yeah. I think that's a good metaphor but maybe maybe I'll give you a barometer please.

Of my of my last 11 investments six are AI based and that's up from zero. Literally zero. Yeah. And so you know if even people like me who had no exposure to AI before last year Yeah. are now making a majority of my investments are AI forward that that says something you know pretty interesting.

But aren't but aren't for the nonAI investments, aren't you still talking with those founders, CEOs about how they're going to leverage AI? I mean, it's not like they're coming to you and being like, "Yeah, I think it's overblown. I'm short.

I'm, you know, I'm not going to, you know, they're not like lites if they're technology entrepreneurs. " Yeah. Well, there's a couple use cases that are important to pretty much any company. One is just engineering productivity pure.

you know whether you use cursor or you know some other product or all of the products like we've seen at ramp for example engineer productivity is up 46% pretty strictly measured and increasing um so that's just purely on productivity so getting more done with less great you're seeing AI remove costs um think customer support is a classic use case customer support can be expensive and in many cases AI can perform on par with a human or better certainly faster and more cheaply with more scalability.

So there are areas that are no-brainers for any business to adopt then there's the question in the strategic area you compete in your company is there way to use AI you know in a differential way and that that's a little bit more art than science right now. Yeah.

How how how overblown is the narrative of uh startups getting steamrololled by the hyperscalers? Oh, every startup just got turned into a bullet point in a Google presentation. Uh there's one narrative that's, you know, Google invented AI. They have all the TPUs. They have this massive data center.

Uh at the same time, a lot of their product rollouts, they sunset a lot of products. They run a lot of tests. They're really dominant in search and advertising. But if it's a side project, it kind of gets side project attention and it's not really a founder mode project.

Um, how how are you thinking about how big tech is competing or just stepping back and just reaping the rewards of everything that they've done over the past couple decades?

Yeah, I think there's like classic, you know, sort of conceptual approaches to this is is this a sustaining innovation or disruptive disruptive innovation and you know thing things that are disruptive there's a different you know approach to those things that are sustaining there's a different mentality towards those I think the key is for a founder to understand where's my comparative advantage ultimately like you're going to build a startup from scratch the world is not your friend inertia is against you nobody cares period nobody cares about So you've got to turn that inert into positive momentum in a true physics sense.

Yeah. And then you've got to figure out how you propel yourself even if other people large or small start doing something similar to what you do. And I think having a line of sight like at least an intellectual conceptual level of how would I do that? How am I going to do that?

Makes the difference between being, you know, a very mediocre startup that isn't the proverbial roadkill to something that looks like it should be washed out in this wave but actually trumps over the incumbent.

And you know, the history of tech is typically the well-run, thoughtful startups prevail over the large startup. Think about AI. Open AI is absolutely dominating Google period. Yeah. Open AAI, OpenAI's chat GPT will be the default interface for the majority of people on the planet, not Google search. Bold big change.

Uh I want to talk about with uh w with OpenAI uh they they acquired Johnny IV's company. They brought in an absolute industry legend. I want to talk about it in the macro unless you have a take on the on this actual deal or or Johnny IV.

But more importantly, um, startups bringing in these absolutely legendary heavy hitters. It takes a bunch of different forms, right? Palmer Lucky brought in John Carmarmac, uh, video game legend at Oculus. That was amazing. Uh RAMP has Kenchault, this former CEO of AMX involved. Uh but not as an employee, right?

Uh then Ramp has Gennady, this uh this fantastic programmer like I think he's the best programmer in the world uh more in an employee role. And so as startups are scaling, if should they be hunting for these like you know absolutely legendary people trying to get them in their organization?

Is it just a very rare scenario that OpenAI is so doing so many things that they need a Johnny IV? Um, how would you talk to a founder about hiring kind of an industry legend? On the industry legend side, I think it depends on the motivation both for the company and the individual.

Many people become successful and are demotivated and not as then there's the Elons of the world that the more success they have, the more ambition they have. And so I think you need to triangulate like who are you talking to and you know how much drive they still have left. Sure.

Um and then certainly from an employee perspective you don't want like someone who's past their prime a little bit like sports. You do get past your prime or your motivation to drive decays unless you're like the Elons or the Kobes and people like that that are very rare. Um so I think that's the most important sort.

Um, on the specifics here, I've read mixed reports on exactly, you know, what involvement, if any, Johnny is going to have. They're acquiring the company, but it's not really clear to me what he personally is going to do. So, I think that'll have to sort itself out. Yeah.

Could wind up in like a CEO of hardware role, very deeply involved or, you know, uh, more of an adviser or design lead. Um, there's a bunch of different ways that that could shape up. Uh, Jordan, do you have anything else you want to chat about?

How do you think uh the AI safety era of the of of 2023 early, you know, 2024 will be viewed in 5 years, 10 years? Well, it's DOA. I mean, nobody even talks about it anymore. When's the last time you heard someone like credibility talk about? Well, so I heard about it today. I'm I'm just going to call this out.

So, uh, today apparently I'm seeing a screenshot. An AI alignment researcher at Enthropic just said that Claude Opus will call the police or lock you out of it of your computer if it detects you doing something illegal.

And the researcher says if it thinks you're doing something egregiously immoral like faking data in a pharmaceutical trial, it will use command line tools to contact the press, contact regulators, try to lock you out of relevant systems, or all the above, which to me to me is like if that's AI safety, that that's scary.

That's scary. I don't think that's what most I mean that's is scary seems pretty stupid but there are some legal obligations obviously imposed upon any tech company you know engage in child pornography and various things.

So like assuming there are compliance obligations on all companies whether tech or otherwise and absolutely those are all real and serious AI safety I think conotes to a lot of people like the idea of AI taking over you know etc and doing things that are you know mischief creating in some way that debate I think is over in the United States for a very long time we're in the how do we accelerate the progress of AI how do we ensure that AI in the western world dominates over the CCP's use of AI.

I think those are the most priorities, the major priorities. I don't think anybody in DC right now really wants to hear about this old a AI safety canard, which is basically an anti-ac. Yeah. You know, is euphemistic for anti-ac. It wasn't a serious concern. Yeah. It's interesting.

I uh I don't know how closely you followed like the latest release of Llama 4 by Meta, but they there was allegations of kind of fine-tuning on the specific evaluations and uh and a delay. And a couple years ago, it would have been so easy for them to just say, "Hey, we're worried about alignment.

We're worried about safety. We're doing this for you. " But instead, they kind of just had to bite the bullet and say like, "Oh, we kind of messed up on the training. It's not quite done yet. we're just going to go back to the drawing board and like work through it.

Uh and the fact that they couldn't even use it as a fake excuse was very telling to me that that it is truly a dead conversation. And I think yeah, for fortunately I think it's dead great. I think ultimately the US needs to prevail and succeed with AI period. Yeah.

The flip side is that I so I've I I I think I've been aligned with you that the the AI super doom fast takeoff we all become paper clips never really resonated with me. I thought it was kind of just paranoia uh and uh and like superstition.

But now that I've seen what's happening with DeepSeek, all of a sudden I've I've said to myself like maybe we do need something that looks like an AI safety team to go in and investigate these models and see are they weighted to shift us like the Tik Tok algorithm might be shifting us towards certain uh certain beliefs.

Like these tools are powerful and they can be fed false information or bad data to kind of steer the answers. one way or another. And so maybe we do need some sort of model interp. And it it it doesn't it's not exactly AI safety in the doom sense, but it's the type of work that you'd see done by an AI safety researcher.

So geopolitically, how how do you how how do you think things are shaking out? Oh, I I Yeah, again, I would discriminate that from the gloom gloom and doom AI. Totally. Totally. I definitely think models can be biased in different ways and manipulated by different people who have power over those models. Yeah.

And I think observability, which is actually cutting edge um technically, there are some really cool startups that are focused on observability, it's a little bit like doing brain surgery and trying to figure out how to rearrange the brain. That's the metaphors I think they use.

And so that stuff's um actually tracking really well and incredibly intriguing like why did the model do X and then be able to decompose that and arguably do s you perform surgery and fix it. So I think many of these models are biased.

you know the the famous examples of you know ask for the the benefits of Donald Trump and you know open AI used to struggle with that I don't know if it's improved but you know those things are real um but I think that's a different set of challenges and maybe the market sorts that out may you know maybe people expose it and the company has to you know address it I think the tick tock manipulation is a very real thing which is another reason why I think it's better that American companies be in the forefront but I I do think that's a valid concern.

It's different than what I think of safety as if there's a sentient AI that's going to take over with so much power that you can't really unplug the computer. You know, that that I think that debate is pretty dead for a long time. Yeah.

It seems like tech and the US government broadly, the new administration has been kind of going all around the world doing deals.

Are there any countries that you're particularly excited about America building partnership with on the tech side or even any opportunities for earlier stage startups to go plug into the global ecosystem? It's kind of a narrative violation because just a couple months ago we were saying do not do business abroad.

There's a 75% tariff on everything. But now you see Jensen and Sam and you know scale AI is over in the well it's more it's more like we're exporting American AI. Yeah.

versus where are the opportunities couple drivers there is there's a belief I guess by some people that exporting uh American AI makes it the default standard it may or may not be true but I think there is an ideology there second is I think a lot of this is we need resources i.

e infinite power and so people going to very wealthy countries and asking them to spend a lot of money to either develop advanced power or advanced manufacturing capabilities at scale and I think that's why you see you know the CEOs who run these large you know AI based companies uh you know sort of traveling around the globe to sovereign nations that have too much money you know they don't really know what to do with it um but I think that's a it's not a uh do business in the country.

It's more take advantage of their natural resources, turn that into something that we need, which may be either more manufacturing or more energy or both. Very cool. Well, uh we'll let you get out of here. Uh it's past two. Always a pleasure. Such a great time. We'll talk to you soon, Keith. Great to catch up.

Great to see you both. Take care. Talk to you soon. Bye. Uh do we