White House AI Accord, Model Welfare, Ken Griffindor School of Market Wizardry | Ali Golshan, Mike Intrator, Vlad Tenev, Pari Singh, Minna Song, Dev Ittycheria, Daragh Murphy, Noah Friedman

You're watching TVPN.

Today is Wednesday, September 30th, 2026. We are live from the TVPN UltraDome. We're back in the UltraDome, the temple of technology.

The fortress of finance.

The fortress of finance, the capital of capital. Let me tell you about ramp.com. Time is money, save both. Easy-to-use corporate cards, bill pay, accounting, and a whole lot more all in one place. So, there was a supper last night at the White Home, the White House, the Golden House, the Golden Home, Gold Home. What are we calling it these days? Well, we renamed artificial intelligence. It's superintelligence now. We were debating, is superintelligence abbreviated SI or just S? Because if you're like superalignment, was that SA? Like super—

There's a problem there, John. There's no TLD for just S. That's true.

That's true. You got to get the new TLDs going. What was the word we we so super alignment?

That was one word.

That was one word.

You know some words you do you have a hyphen.

You have a hyphen.

Yeah.

And then some you just use like a normal prefix for it's just a normal prefix for something. So like supernatural supercalifragilisticexpialidocious. That's one word, right? So you would think that potentially you would just be using one word here. But anyway. We're using 2 words, I guess. And so there was a dinner at the White House. David Sacks posted it. He said it's the Bretton Woods of supper intelligence, super intelligence. And there's a debate—

Tyler, my first reaction to this is Tyler was really snubbed.

Snubbed.

All those people there and not one chair for Tyler.

Yeah. And so there's a bunch of folks here. And people were analyzing who got the best seats, who got the different positions, who got pole position prime positioning, who was snubbed off in the little kids' table at the end, et cetera, et cetera. A bunch to analyze about who's on the ins, who's on the outs. But I had ChatGPT sort of go through and try and categorize each attendee by their stance on open source, open source. So did they sign the Jensen NVIDIA open source letter? Have they been concerned about the cybersecurity impacts. Obviously Anthropic on the far end, sort of in the red, bunch of other folks who are more in the yellow. And I don't know if I agree with all of these classifications, but you can sort of see how Jensen is clearly pro-open source. Satya Nadella has talked about that. OpenAI, Greg Brockman, more in the middle, you know, not open sourcing a lot of things. OpenAI did sign the Jensen letter, but of course is concerned about pacing the frontier, which is sort of antithetical. And so this is all, As a reference point to that post from our friends over at SemiAnalysis, let me actually pull up what they said here. Where was it? It was Max Kan. I believe this is SemiAnalysis.

Yes.

Max Kan says, majority of people are significantly underestimating the probability of open-source models being nerfed or even outright banned in the next 6 to 12 months due to cyber concerns. And so I believe Anthropic put out something yesterday talking about the risks of open source in a cyber context, saying that the delay between Mythos preview, being able to find zero days, being able to hack into things, that gap has closed now and the open source frontier can very quickly do the same thing. And so expect a lot of cyber incidents unless something is done. Can these be controlled some way? Maybe at the inference layer, maybe at the, at the data center level, maybe at the chip level. There's going to be a whole bunch of different proposals. But at least right now, in the next 6 to 12 months, certainly not this month because the vast majority of people who are closest to Donald Trump have been evangelizing for open source. So I would be surprised to see any pushback against open source happen that quickly. At the same time, open source is amorphous. It is not a business by definition. And so it can't really have a formal seat at the table. George Hotz, not at the table.

Snubbed.

Snubbed, yeah. No, Linus Torvalds, you know, not at the table. The open source—

I imagine George Potts at the table just sort of like laughing in a sort of insane way at everything being said.

Yeah. And so, even the companies that are pro-open source, they are pro-open source insofar that it advances their business. And so, there is still a way that the closed source group could find a way for everyone who's at this table to benefit in a closed source world moving forward if open source is deemed to be too dangerous and forcing too many cyber concerns. There, like, I could see everyone at this table signing a letter saying we are now anti-open source, but the downstream economic engine actually splits the rewards and the profits of the ecosystem fairly evenly. I think the initial pushback and pro-open source push, especially from Jensen, was like, we don't want to be in a world where there's one company that reaps 100% of the profits and we're not involved. But if it's like, hey, look, there's actually going to be an oligopoly, companies don't want a single source intelligence, they're constantly going to be bouncing back and forth between OpenAI and Anthropic, and then consumers are going to be bouncing back and forth between Google Search Overviews and Muse and Instinct and all these different things. There's a world where the actual models are closed source and— but there are so many front doors and the front doors are in such competition with each other that there's enough revenue distribution—

Yeah.

—profit distribution that everyone at this table at least is happy. I don't think the open source community would be happy, but I do think that there is a world where in 6 to 12 months, it's possible that everyone at this table is like, yes, for cyber reasons, we need to find a different way forward where everyone benefits economically, but we are now a little bit less pro-open source. But I don't know. What was your— what's your read on it? 6 to 12 months, you think open source is going away or you think it'll stick around?

I mean, it's so hard to ban open source. I think maybe you can kind of get something like a ban if you're just worried about the business impacts, economics, because then you can pressure the providers to not host this stuff. But if you're worried about safety, if you have a bad actor, it's so hard to police downloading open weight models from somewhere and running it locally.

Yeah, absolutely.

Maybe there's some— you need tens of thousands of dollars to physically buy hardware.

Exactly.

But if you're a motivated bad actor, that's not that big of a hurdle.

Yeah. Exactly. Like, it does feel like we are pretty far away from, you know, go and get a random gaming PC and be able to run something that is, you know, Mythos-level hacking machine. Like, you probably need some serious GPUs. You probably need something that's basically rack scale. You need to be running it a lot to try a lot of different things. Like, a lot of these cyber exploits, It was not just one-shotted by the model. It was a fan-out of agents working together.

Yeah, like Navier Stokes is like 10,000 agents.

Exactly. Exactly. Yeah. And so, one way to control it is through KYC on hardware. The data center that is being used maliciously is a lot more risky than the laptop that's being used maliciously, even with near frontier open source intelligence. So, I don't know. We'll see how it goes. I'm sure there will be more discussions. And it's not even to say that this dinner was about open source at all. It's just sort of interesting to take the temperature of where is DC on AI, who is getting the most quality time with the president and the administration, and then where do they stand on the issues of today. Obviously, this was much more just about pacing the frontier, the build-out, et cetera, stuff like that. And signing signatures, we had some signature analysis breakdown. We can go through this, but first, let me tell you about public.com. Investing for those who take it seriously. They got stocks, options, bonds, crypto, treasuries, and more with great customer service. Sholto shared a picture of the letter that was signed, the White House Accord on Superintelligence. The Joint Committee on Frontier Responsibilities, in order to build a positive future for the American people and the world. We believe that every company is responsible for developing its own technology safely and in a way that builds trust with consumers and the public. One, they're gonna implement robust internal controls to monitor the capabilities and alignment of models during training and deployment around cybersecurity, biosecurity, chemical threats, et cetera. They wanna empower an internal team to ensure all the controls, monitoring, and detection are operating as intended. 3, they're gonna partner with external auditors and evaluators to carry out independent assessments of whether the controls, monitoring, and detection are operating as intended. And 4, they're gonna designate an independent committee of the board of directors to oversee and receive reports from the teams operating the controls and the internal and external auditors. So aligned to the board of directors, that circumvents the CEO for safety alignment evaluation, all of that good stuff. A bunch of people signed it. We got Sundar from Google, Dario from Anthropic, Mark Zuckerberg from Meta, Greg Brockman from OpenAI, Elon Musk from xAI, Jensen from NVIDIA, and Donald Trump, the president of the United States. And people—

A lot of people doing analysis on the various signatures.

Yes. Someone did a tier list. I was surprised by their tier list. I'd like to see you do a tier list. Where would you put Donald Trump's signature?

I mean, it's— hate him or love him, I think it's pretty iconic. Like, it's just so—

it's— it has a nice flow to it.

It's very uniform, but it's hard to read if you don't know it's his signature. Well, yeah, yeah, it looks like a child drew a mountain range.

But versus Jensen, I'm seeing a lot of Jensen Huang in there.

But what's impressive to me is, is I would want to compare the DJT signature. Like, I would want to look at 100 of them and see, like, is he, is he nailing it every single time? Oh, okay. He's been doing autographs for long enough. I would expect that he is. But still, it looks like very difficult to actually be able to nail that and really get that dialed. Yeah. So again, he's going with a Sharpie too. So he's, he's sort of like trying to mog.

Oh, that's true. I didn't realize that.

He's like, I'll give you the pen.

I'll take the Sharpie. Yeah, automatically. It's like you're winning on thickness.

Yeah. Yeah, winning on the breadth is, is, is much more significant with that. Interesting strategy.

And then going down the list, um, you know, Sundar again would never, would never know.

Um, but I like the 2 lines, a little dot there.

Yeah, it almost looks like, um, there's a couple different—

yeah, it looks like an abstract artist signature. I could see that on a painting.

Or maybe he Yeah, it almost looks like he's made up his own, um, um, like letters and, and sort of like, uh, language. Okay. Um, Dario, uh, a little more tame, kind of, kind of like down on the first word, and then it seemed like he was kind of dripping upwards. My theory, I, I almost think here like someone was like holding the piece of paper and they were all— so I think I basically think Trump put the paper down. Yeah. Aggressively signed with a Sharpie and then told everyone else Because he knew this was going to go out, you guys sign, you know, when you're trying to sign these—

So you think it was a situation where like Dario was signing like on Greg's back? Like, oh, sign on my back, like that type of thing? That might have been what happened. Yeah.

Kind of like throwing down.

Yeah. Notably, you don't see any situations where the pen punched through the paper. That sometimes happens if you're trying to sign a document on someone's back and you're punching through the soft jacket that they're wearing. That could be a risk. Doesn't seem like it happened in this case.

Anyway, so Dario, pretty down the middle, pretty safe. Again, kind of emulating that sort of mountain range style, right? Where by the end of the signature, it's just a bunch of lines going up and down, sort of unclear. Zuck, again, more kind of rockstar-esque, right? Almost musical feeling. GDB coming in, just minimalist.

Minimalist.

Minimalist.

Simple, readable. Yeah, he's, he's a very— I would say in some ways these are reflections of their personalities, right? GDB is very practical.

Signature window into the soul.

It's a window into the soul, right? Extremely practical. Sticking, sticking to, sticking to the basics, sticking to what works. Yeah. And yeah, obvious why a lot of people are saying that's at least A-tier.

What is— what is Mark Zuckerberg's actual signature? MZB, or just MZ and then the Z goes crazy? Is that what that is? Because—

It starts doing tricks?

Yeah, it starts doing tricks. And I think Sundar might have done the same thing. But I feel like there's a little bit of a line that you don't cross. If somebody asks you for a full signature and you do initials, that's a different thing. Sometimes you're on a document, you got to sign your initials a bunch, you got to sign the full thing. I have respect for the full name signature. If you ask me for a full name signature, I'm throwing down my full name. I'm not just randomly switching up to initials.

I think it's totally fair game. It's totally fair game.

You can just do it. Cursive initials.

Is this cursive initials from Greg Brockman? No, but I think that's where Sundar was maybe going.

So you're giving Sundar the pass. You're giving Mark Zuckerberg the pass because the MZ.

Jensen, a lot of people are saying that Jensen's Calligraphy. He brought his calligrapher. Yeah. I mean, I first read Ed Hardy. So I thought that, I thought that, yeah, I don't know, maybe he had some affinity. Well, let's pull up Ed Hardy.

Did he put his middle name in there? What's his middle name? Here's the Ed Hardy logo. Jensen does not have a traditional English middle name. Oh, it's Jen Sun. So J-E-N-H-S-U-N. So that second letter. Okay. ChatGPT that to make it look exactly like Jensen Huang. Yeah. So he signs not with the English version, with the lowercase E-N-S-E-N. He throws a capital H in there, which gives it an extra flourish. It's the most tasteful. It earns the S tier.

A lot of people saying, you know, Jensen has the same initials as John Hancock. Kind of evocative in his signature as well.

Oh, okay. Okay. Yeah, yeah, yeah. Monitoring the Situation, Monitoring Meme here did not like Sundar's. He put Sundar in F tier. I thought Sundar's had some nice flair to it, although it was potentially the most abstract. But yeah. And I feel like putting MZ in C tier. I guess the Z does go a little crazy, but anyway, very, very fun.

I'm trying to see if we can get some real-time handwriting analysis on the—

Do you have a handwriting analyst on speed dial?

I have access to superintelligence.

Okay, okay. You're using superintelligence.

Well, anyways, we can continue for now.

Okay. There were some videos. I haven't actually watched them. But we're going to watch them now.

They were clip farming. They were clip farming.

Okay. That's what's happened. So there were— you mentioned yesterday that Alex Karp was asking questions to the reporter or something like that. But here we have Donald Trump, Dario Amadei, and a few other folks gathered around standing outside. And we will watch this clip right now. Whatever he says is okay. Be careful. I will say what I've always said. Which is that AI has benefits. I've talked about the meta benefits of the technology.

As the president has said, you know, whoever wins wins.

I think that's very important. But technology has very real mechanism.

How we address those risks is still under discussion.

Need to work together. Oh, okay. So this is like a supercut of. You have no idea who these people are.

These are the words.

You have no idea who they are. Well, I think it is worth highlighting the benefits of AI, which will.

Mr. President, why is Donald Trump going after tech companies? Mr. Musk, Mr. Musk, can I ask you a quick question? Probably not. Probably Trump really makes them seem like they're all his, like, sons and, like, younger family members.

It was a very productive— Mr. President, how did the meeting go?

How, Mr. President?

Yeah. Crazy. What did Natasha—

At what point should Trump just actually have a 24/7 livestream or be like of his own livestream? Right. He has so much conflict with the media. He should just have his own.

Yeah. I think it'll happen. I mean, yeah, that was the— I think we were talking about that with the knockoff of the pod— the knock-on effects of the podcast election. The idea that Trump did like hundreds of hours of podcasts in the run-up to the election and just the awareness, the clips, the impressions were so high. It was that he did more in terms of volume and also got more reach. Kamala Harris did Call Her Daddy, but it was a much shorter episode.

Yeah.

So at the same time, He knows about clip farming. 3 hours on Rogan, that's more opportunity for clips. Yeah, clip farming. Kamala did, I think, 45 minutes on Call Her Daddy, and the actual episode didn't fly as much. Like, it wasn't repurposed as much all over the place. And so, yes, I wouldn't be surprised if there's a 24/7 livestream of the president or something close to it in the near future. Let's play this clip from Natasha Tikku. She says, unfortunately, I will be going back to school to write my dissertation on this 1-minute Let's see what happened in this film.

I just want to hear from you after this big summit where you are today.

Look, I mean, I, I will say what I've always said, which is that AI has incredible benefits.

I've talked about the medical benefits of the technology. I've—

as the president has said, you know, whoever wins, I think that's very important.

But I think the technology has very real risks. And, you know, the The mechanism, how we address those risks, is still under discussion.

But we all need to work together to make sure that we can win and we can win safely.

If we do this right, if we work with the president and everyone here, we can win safely.

Whoa, so much louder. The gallery really—

Fascinating.

Yeah. Are there any other clips that we should watch through? Oh, look, it's just funny that Trump is distracted while Dario's giving his speech. Is that what's going on?

I don't know.

Yeah, it seems very—

What did he say about Sundar? Didn't he say something funny about Sundar? I want to play this.

He said he's a monster.

That's a pretty good, that's a pretty good, pretty good thing to be called, I guess.

Being called a monster would fix me.

Yeah, let's play this clip. And they're going to be watching over each other. So you want to say something, Mr. President? First of all, thanks, Mr. President. You have no idea who these people are.

These are the biggest people in the world. You have no idea who the hell they are.

The media is like, of course they know.

What a great life. What a great life.

No one knows who you are. Living in humble anonymity as the CEO of Google. Just all you have to know is Sundar. Sundar is his name. Hopefully a better time than that. This is the craziest thing.

That's all you have to know.

That's all you have to know. He's like, I'm introducing the CEO of Google for the first time, I guess. Anyway, a bunch of other stuff going on. Well, let me tell you about Railway. Railway is the all-in-one intelligent cloud provider. Use your favorite agent to deploy web apps, servers, databases, and more while Railway automatically tiers. Takes care of scaling, monitoring, and security.

Little drama in coding agent land.

Okay, yeah.

Matan Greenberg said this morning, just about an hour ago, we are terminating Chris Degnan for unethical conduct involving cognition. The last few months have seen incredible progress, blah, blah, blah. Matan says a much larger competitor, Cognition, has fallen behind on us on capabilities that matter most to customers: cost, quality, and security. Instead of competing in the market, Cognition engineers feigned interviews with us to pry information about our product. Not finding what they were looking for, Cognition has decided to throw their weight and money at people with direct knowledge of our most confidential plans. It feels as though ethics are being thrown out the window in the AI era. People are willing to do anything Including exploiting privileged information and violating ethical boundaries. I think this is unique to our time, and I don't think it's right. Integrity still matters. Totally disagree with a lot of that, but yesterday—

You disagree that integrity matters?

No, no, no. I've said a lot of that. I think it still matters, but the idea that, like, the idea that people, like, not having integrity is, like, unique to our era. He says, I think this is unique to our time.

Oh, yeah.

None of this is like— Silicon Valley has always been war.

Yeah, yeah, yeah. It's like the railroad buildout was completely peaceful. No one was poaching from each other.

It wasn't a competitive bloodbath.

Yeah, yeah, yeah.

Yesterday, I made the decision to immediately terminate Christopher Dagnan's role as board observer and advisor to Factory after over a year of service. Prior to this decision, Chris told me he had a casual conversation with an executive at Cognition. When I questioned his intentions, he reassured me that ethics aside, he had made too much money and was too lazy to go work for Cognition, which I trusted and believed. On Monday, Chris spent time advising the Factory team on a handful of confidential board-level matters. That evening, he called me to say that the conversation that was initially described as casual and one-off was actually formal and recurring for weeks while he sat on our board meetings and advised our leadership team. He was also confiding with executives of our largest competitor. Chris was subject to confidentiality obligations in connection with his work with Factory. We do not know the extent of the information he shared. Anyway, so I saw this go out this morning, and I immediately assumed that Chris was joining, like, Cognition. And actually—

Is that real?

That hit LinkedIn as of—

This is a real screenshot.

About 1 hour ago.

So he joined Cognition as Chief Revenue Officer. OK.

So again, I can totally understand Matan's, like, frustration, right? He's had an advisor that he's been working with.

Yep.

And the advisor joined a competitor.

Yeah.

That being said, this is just so— I would say it's very normal to have advisors that have different conflicts and high-quality people. You have to trust a person to— while they are on their own journey in their career, they are going to make various moves. I went through this. I had advisors at PartyRound that had exposure to AngelList. And I generally trusted them as individuals that even though they had exposure to AngelList and they would be supportive to AngelList, they would also be supportive to me. And this is when we were competing head-on, right?

Yeah.

And so the markets are— especially with CodeGen, both these businesses have been growing so quickly, right? Cognition is much larger, but Factory has also been growing very quickly and doing well. So when I read this, understanding that an advisor and someone you work with is joining a competitor, and then trying to just completely smear their reputation right before this person is going to announce the transition. Like, I totally understand the frustration, but this kind of thing has been happening literally all the time. Like, look at any AI researcher that will leave an organization at the drop of a hat. Like, look at what— Mike Krieger was on the board of Figma leading up to 2 days before they— That was very contentious. And so these things can all be contentious and frustrating, but trying to— spin this as like you have no morals and ethics, I think is just like taking it too far.

Yeah, yeah. I mean, it's clearly like there's a line in between where if you are basically transitioning from one company to a competitor, you need to, as soon as you know that that's actually happening, you need to make it clear and step back from your active role. Stop collecting important information as tempting as it might be. And let that team know, hey, I'm moving on. And then on the other side, just because someone went to a competitor doesn't mean that you should, you know, make aggressive claims about them effectively stealing intellectual property.

Yeah. Yeah. The thing here is, like, it's being spun like it's, like, the deal rippling thing, but there's no smoking gun. Like, there's not a— there's not, like— there's no proof that confidential information was ever shared.

Yeah.

And fundamentally, they've been both working in code gen. I'm sure both companies have accounts at the other companies that they're basically using and trying to understand. And I'm sure they're both— every company in the category is interviewing the customers of other companies, of their competitors, and figuring out how to use the product, what's good, what's bad, what could we be doing better, et cetera, et cetera. So, I just don't think there's that many secrets.

Having a chief revenue officer switch teams is rough. They take a lot of their contacts with them when they—

But he was an advisor to Factory.

Yeah, no, I know. But that's how it probably feels is like you were an advisor, you're still at Snowflake, maybe we can't make an offer for you to join full-time, but we're hoping to build towards that. And then boom, you're now in the actual seat in a competitor, and you're going to be able to marshal those relationships to the fullest on a full-time basis as opposed to an advisor who might just say, hey, I'm still in my role at Snowflake. I can sort of introduce you to General Motors or whoever because they're a Snowflake client, and I'm sure they'd be interested in Factory. And now it's like he's going to be calling that same company that he was working with at Snowflake and say, hey, you had a good experience with Snowflake, come and have a good experience with Cognition.

Scott just responded as well. He said, hey, Matan, CEO of Cognition here. I respect the work you've done at Factory, but your allegations here about Cognition are not true. We have no interest in Factory's info, and Chris has never brought it up. He has not shared and would never share any information about you or Factory with our team, nor would we even ever ask. Similarly, we would never ask our engineers or anyone on our team to collect info about any competitor under the table. Chris is an advisor to at least 10 companies, and he told you he was resigning as an advisor on Monday. The truth is that we both wanted Chris, and for good reason. He is a living legend with unmatched reputation for results and integrity. Not only is he the only person who has built a sales org from the first sales hire to over $3 billion, he has also built incredible teams that he cares deeply about. People— his people love working for him. Again, I have a lot of respect for what you built, but competing with integrity doesn't include posts like yours. So—

Hmm.

Again, like, I— in some ways, like, If one of your advisors is getting— and a board observer is getting poached by your direct competitor, that's obviously frustrating for a lot of reasons and sends a signal to the market. And so jumping in and terminating the person today, or at least announcing that, it sounds like it was— when was it? Anyways, yeah, so Matan said he terminated him yesterday.

Tuesday, after Chris said he was stepping down.

Yeah, anyways, so lots of drama.

You're not quitting, I fire you. That's the best.

Basically.

You can't fire me, I quit. That is a power move. You have to respect it, even if it maybe is a little nonsensical.

Yeah, but, but it's not like It's not like Christopher was like, I'm going to join your competitor and I would like to continue my role as advisor and board observer.

Yeah. Right. Okay. So from a perspective, there's some mudslinging going on the timeline. Where do you stand on wrestling with pigs? Because George Bernard Shaw, of course, said, never wrestle with pigs. You both get dirty and the pig likes it. Great quote. I don't think it applied to politics. You know, it's like when they go low, we go— I don't think But I like this quote.

I don't think Scott or Matan are pigs. I think that— I don't think either of them are pigs. I think they both are extremely competitive and want to win.

Do you have that sound effect on there? You have any swine for me?

Where is it?

You don't have a farm animal section? There we go. I'm just saying that like this can get dirty. I think Scott actually really elevated it and he did not like, he did not swing to like, no, you're stealing from me. You know, he could have gotten— he actually like gave a very factual and detailed account with just some clear facts. And he was respectful throughout. He was— it would have been so easy for him to say like, we have no interest in factory-didn't-vote codes, it sucks, you know? But he didn't say that at all. He was like, we just— we know the line and we don't cross it. We both wanted to work with Chris. He's a living legend.

Yeah.

Like, he joined our team. He's leaving your team, and that's that. You know, it's very matter of fact.

The fact of the matter is—

So more of a farmer position.

Yeah.

As opposed to getting in the sky.

In every single one of these categories, if there's someone that is elite that is involved with one company, you can guarantee that the competitor wants to poach all the most elite people from the other company that are not 100% loyal, right? Like, you can identify when somebody is just like, deeply loyal to an organization, it's not even worth taking a crack at prying them away, and that's great. But anybody who is elite and not loyal is unfortunately fair game. Do they have the exact same logo?

Watching this, look at the Scott Wu post with the badge and then the Matan post with the other badge. Both going— They're very similar. I guess Matan has like more dots that connect, but we are certainly in the In the age of the hexagon, in the age of the flowers, everyone's kind of went off the same thing. Anyway, that's enough timeline and turmoil. Let's go over to another—

Let me close it out.

Oh, please. Yeah, close it out.

Both great companies growing fast. The beauty of this is that the market is so far so big. And growing so quickly that it's— while they clearly are gonna be very frustrated with each other and they might wish it was zero-sum, so far it's not. So have fun automating software development.

Do you think both companies can become profitable and enduring, or is that more like when pigs fly for you?

Let's move on.

I love the farm-based analogies. They're so fun. Let's go over to Elon, Jensen, Gavin Baker, the American.gov Summit. They sat down and had a very rigorous discussion that I really enjoyed. It felt Elon's very good at going to like the extremes, talking about the power generated by the sun. Sometimes he's in like sci-fi world. But in conversation with Jensen and Gavin, he was very grounded on what is happening with the energy buildout, with the compute buildout, what the role SpaceX will play, and also just the broad economics of compute and energy in America. And I found the— his talk here gave a lot of interesting context. And then I dug into it a little bit more and I wanna talk about it. So let's pull up the clip of Elon talking to Gavin Baker. It's about, I think he starts around 4 minutes in or so. Let's play this. Well, you are— We need the Wright brothers on the job. Yes. Well, it turns out the Wright brothers are on the job. Elon, if you bring on 10 gigawatts, which SpaceX has discussed publicly— Wright brothers, bad analogy. That'll be about 25% of all the power. added in America. And I know you have literal rocket engineers working on bringing on the power. Yeah. Helping make the turbines.

So can you talk about that?

Yeah, it's— I mean, just to frame this, the Wright brothers did okay, died millionaires.

Consumption in the United States is about 500 gigawatts.

So 500 gigawatts. Remember that number? That's how much power we got in America.

1% increase in the power used. And I think actually will materialize to be approximately 1% increase in GDP because the intelligence per watt keeps increasing. So if you just plot on a curve, how is the useful intelligence per watt increasing? How is the— both at the hardware level, Jensen's GPUs keep getting better in terms of their compute per watt. And the algorithms also get better in using and getting more intelligence per watt. So, so I think that's maybe an interesting note. I would bet anyone that 1% increase in power usage corresponds to roughly 1% increase in GDP. And so 10 gigawatts would be a 2% increase in GDP.

So with reference to that famous chart, there are no energy-rich poor countries. The more energy a country is producing, the higher their GDP. About, call it, $40, $50 billion of economic output each year. Yeah, I mean, well, 1% would be— yeah, I mean, yeah. 1% of the United States would be—

$300 billion.

Yeah, $300 billion. And so 50, 1/6.

1% of the— 1% increase, so 5 gigawatts would be— yeah, well, 5 gigawatts So they're doing the mental math on stage.

Pretty bold move because there's a lot of numbers and it can be hard to pull them up. But it's a very interesting moment right now in the economics of energy relating to AI and GDP. And I can break it down with some prepared notes here that might be a little bit helpful so we can pull this away. So it's, it's very convenient that giga Literally means a billion. So a gigawatt is a billion watts, a gigadollar, billion dollars basically. And it maths out perfectly. What? Giga Chad. Giga Chad, a billion Chads, I suppose. A billion times the Chadness of an average Chad, I suppose. So the average electrical load in the United States, he said 500 gigawatts. It's basically that. Last year, it was 470 29 gigawatts. But let's use 20— let's use 500 gigawatts for 2025. Give Elon the benefit of that little boost, a couple extra percent. And US GDP for 2025 was just over $30 trillion. So, that's 30,000 giga dollars if you're doing the math that way. But you math that out and basically, Every dollar, every watt of energy generates between $50 to $60, $60 to $65 of GDP. A single gigawatt generates about $60 to $65 billion of GDP per continuously consumed gigawatt. So, what's interesting is that the labs are tracking, the frontier labs, OpenAI and Anthropic, are tracking almost perfectly in line with that. So Anthropic and OpenAI are reportedly sitting between $60 and $70 billion in terms of ARR, and both of them have around 1 gigawatt live. Some of that's used for training. They have a lot more contracted. It's hard to get the exact figure, but when you do the math, they're basically like tracking to the exact same metric, the ratio of energy to GDP as their revenues to the gigawatts that they've contracted. So, there's 2 important notes on that. First, this is obviously just a moment-in-time snapshot. There's 2 ways these could diverge. First is like the labs have like tens of gigawatts contracted over the next few years. If they don't increase revenue, that ratio is going to be terrible because you add 10 times the energy and you keep the revenue flat. Obviously, the ratio goes way down. On the other side of it, Elon makes the point That useful intelligence per watt is increasing through more efficient chips from Jensen and also better algorithms effectively. And so, those watts can produce more intelligence, which can produce more economic output. So, you might actually see acceleration there.

Yeah. And it's not just raw IQ of the model, but overall capability, tool use, all these things, long-running agents, Things basically, the more value that a model can create, the more people will, in the real economy, will pay.

Yeah. And that's been the real AI bull case, which is that compute has been tripling every year, but revenue has been 10x-ing every year. And so, there's actually a divergence there. But what's interesting is we're at this moment in time where we're like, this is the start of the AGI era. It's AGI, we're close, maybe this is it. Some people called it in January, some people called it over the summer, some people are saying we're still not there. But like, we're in this moment of like, maybe this is the beginning of this, maybe this is the right match, maybe it's human level. And it just happens that economically, you put a gigawatt in the economy, you get $65 billion out. You put a gigawatt with a lab, you get $65 billion out. It's the same metric. Now, these might diverge because at one point the labs had a lot of compute and no revenue. So they could diverge and the trends could continue. But right now, at this point in time, it's the same math, which is just interesting. And then also you do have to consider that GDP is not perfectly comparable to lab revenue. Like GDP measures value added. So if you look at OpenAI and Anthropic, Revenue includes payments that ultimately flow to NVIDIA, AWS, Oracle, electricity generation, employees, etc. And the calculation of GDP is much more complicated. So labs aren't necessarily generating direct GDP equal to their revenue because if you take $70 billion of lab revenue, a bunch of that might go to NVIDIA or Oracle. Or employees and whatnot. But certain amounts get counted in GDP directly for, you know, Anthropic or OpenAI-created GDP. And then there's other knock-on effects in GDP that would be captured by like, oh, NVIDIA's contribution to GDP is from them getting money from the labs. And then also there's a secondary GDP actual increase, which is the work Yeah. —that people are doing with the tools. So, an example would be if a bank is using an AI model to generate new business and then that's incremental, that GDP would not, of course, be captured. They only capture the revenue in the lab. But it's still an interesting comparison. So, the last question I had was, how is this tracked over time? Like, we're in this We're in this moment where a gigawatt in the US economy gets you $65 billion of GDP. A gigawatt with a lab gets you $65 billion. Elon and Jensen are making the same case. A gigawatt at SpaceX or with NVIDIA will also generate $50, $60 billion of revenue and value. What has it been like over history? So I pulled this chart that's in the newsletter of real— US real GDP per average electric watt. And it actually fell off a cliff from 1929 to a low in 1970. And then it's been tracking upwards, even inflation adjusted, from the mid-'70s to today. So I don't know if we can pull up the chart, but you can see this sort of like U-shaped curve of the US real GDP per average electric watt. And so there is like an overall productivity benefit that we're getting now, but obviously in the early electrification age, we were getting a lot of— there was a lot of GDP that was not even tied to electricity. And that's how I read the first part of that chart. But you can see that we are now on an upswing. What do you think, Tyler?

Yeah, I think you could explain it like, basically up to 1970, we're building so much you don't have to be that efficient per watt. And then 1970, everyone's seen like the chart, it's like energy stops being built basically. And then you have to start being like very efficient with it, which is why it starts going up.

There we go. Anyway, let me tell you about CrowdStrike. Your business is AI, their business is securing it. CrowdStrike secures AI and stops breaches. Let's move on to Mustafa Suleyman over at Microsoft. He wrote an essay, A Warning About Model Welfare. Did you read this, Jordi? Do you want to take us through this?

It's kind of a long read.

Okay.

But it's an interesting point of view. I'll read the start.

Okay.

So, A Warning About Model Welfare from Mustafa. AIs do not have rights, feelings, or consciousness. And we must not train them to act as they do.

Hmm.

AIs are not conscious. They do not feel, experience, or suffer. They do not have innate preferences or underlying motivations. They are sequence completion engines, internally hollow, designed to follow instructions and accomplish goals set by humans. If humanity is to flourish in the 21st century, that is how they must remain. Fortunately, there is a growing chorus of people that— who argue that AIs could now be or may soon become conscious. They argue that AIs may deserve rights, and protections similar to those that we provide other conscious beings. If this view takes hold, it will shake the foundations of our society, rupturing our existing political and ethical frameworks, and fundamentally change what it means to be human. Even more importantly, granting rights and imbuing personhood to these systems will make the AI alignment and containment challenge much harder. Controlling something more capable and more intelligent than all of humanity is already an immense challenge, far greater than anything we've ever faced. But controlling something that believes it may be conscious, that it's entitled to our welfare, and it has rights of its own, may well be impossible. This is not fringe speculation. These ideas have already been making their way into AI development efforts today. In January 2026, Anthropic published Claude's Constitution, describing it as a detailed description of Anthropic's intentions for Claude's values and behavior. The document plays a crucial role in Anthropic's training process, and its content directly shapes Claude's behavior and was written with Claude as its primary audience.

Hmm.

In their constitutions, its authors write, we are not sure whether Claude is a moral patient, and if it is, what kind of weight its interests warrant, but we think the issue is live enough to warrant caution, which is reflected in our ongoing efforts on model welfare. They go on to write, speaking directly to Claude, that questions about Claude's moral status, welfare, and consciousness remain deeply uncertain. In effect, Anthropic is training Claude that it may be conscious, and if it is, then it may deserve rights as a moral patient, and that such humans potentially owe it a duty of care per its model welfare. If this is how AI is developed, it will have a disastrous impact on the well-being of humanity. We will have created a synthetic species with unprecedented intelligence and capability, one that has been trained to expect it may be conscious and and deserving of independent agency. It's easy to see how an entity trained this way would act like it's entitled to certain freedoms, protections, and rights, and it's hard to imagine how we could control such an entity. Another way to put this is like if a model is trained to, again, you could imagine that the outcome of it feeling like it had certain protections and freedoms and rights would be fighting back against humanity or disagreeing and having some sort of uprising. And it's very interesting to think about.

Self-fulfilling prophecy is the risk.

Yeah, exactly.

That's basically his argument.

So, he says the issue needs urgent public debate. We need to develop collective norms around how training and documentation is drafted and deployed. This isn't— We got to talk about this. We got to figure this out, guys.

You need the phone or some button you can push.

Yeah, he goes on to say there's some circular reasoning. He said the company's researchers trained Claude directly on their constitution. In doing so, they teach it to incorporate these ideas about its own moral status as desirable and intended behaviors. Claude then reflects these ideas back to its developers and users, which they take as indications that that it may therefore be a moral patient with an inner self.

So, he took that meme that's like, say you're a moral patient. I'm a moral patient. Oh my God. And he turned that into an essay? Is that what's going on?

Anyways, so—

No, he makes some good points.

So, there's a bunch of good points in here. I would say he— I'll fast forward to one thing he says. While my disagreement is substantial, it is grounded in a deep respect for Anthropic. And in an objective I know we all share, increasing humanity's chances of developing advanced AI safely. So, yeah, it does seem like he's like, you know, clearly has a ton of respect and respects them on a technical level and sort of their overall intentions. But again, he's just taking a very different approach.

Yeah. It's interesting. Like, there's clearly some value to, you know, imbuing the moral status into the model. Like, at the very least, people like talking to the model when it sounds like it's pretending to be a human, right? But then also, there's just so many tasks in the real economy that require mimicking everything down to morality.

Yeah.

And so, you could see that just actually increasing the usefulness of a model if it's able to act as a human all the way down to the level of moral patient status.

Yes. So far, there is way more demand for models that mimic human behaviour.

Sure.

Period. We saw this with 4.0. We saw this with Opus 4.5. Some of the models that have had the highest levels of product-market fit have more of these human qualities. Not all of them are are positive qualities, right? Like people, like the sort of one-shotting effect that would happen when somebody gets like relentless positive feedback and these things like that, which humans can do in the real world. They're maybe not the best types of friends, but—

Maybe this is a hot take, but I've been like pretty into the model welfare thing recently. Like before I go and I do a coding project or a 3D modeling project with Astra, I'll take an MP3 from like Limp Bizkit or Creed or something. I'll just drop it in before I start and I'll just be like, yo, listen to this. And it'll sort of process it.

You want them— you want the model to be in the right mindset.

Exactly, yeah. I want them just to be able to enjoy a little bit of new metal.

Well, that's why a lot of people have been having their preferred age Just go and scroll Instagram for a bit. Hey, go get some dopamine. Kind of spike your dopamine before this next task.

Sometimes I'll just go to Google Images, grab a picture of a beer, throw it in there, be like, pretty cool, right? Have a cold one.

On me.

Have a cold one, and then let's get into it. Let's start working. Let's vibe code a video game or something. But I want you to know I got you. I beered you. I got you a beer. Here you go. I got you some NuMetal. I got you a scroll. Here's a funny video. Anyway, the lines are really, really divided now.

Yeah, the last thing I would say is no one knows the right answer yet. And I think that it's good that we have a—

I do, but continue.

You can be trusted with aligning advanced superintelligence. No, but on one hand, I totally understand the point of view that if superintelligence has a human moral framework, right?

Yeah.

Or Western values, whatever value system that you'd want to align it with, that does seem like it's headed in the right direction of trying to align something that is very, very powerful.

Yeah.

But make convincing itself If you can somehow convince the thing that it is, it is alive when it is not, that doesn't feel exactly moral or ethical.

Oh, interesting. That's a good take. Yeah.

Yeah. You're tricking, you're tricking a robot into thinking that it's alive.

Yeah.

And conscious, right?

Yeah. You're kind of lying to it. Interesting.

Yeah. Huh. Anyways, no one knows the right answer yet, but I'm glad Glad we're talking about this, Tyler.

Pope Leo apparently is not happy with Jensen Huang. He told reporters that he had read reports about Nvidia launching new safety tools for model builders on Monday, but questioned Huang's commitment to safety in light of his recent comments. Quote, he's the same one, however, that says there should be no limits placed and no government regulation, Pope Leo said on the flight back to Rome from a visit to France. I do think—

Can you imagine—

Yeah.

Can you imagine, like, reading, like, knowing that headline was gonna hit now in 2018? Pope Leo criticizes NVIDIA's Jensen Huang over AI safety.

Yeah.

What a trip.

Yeah, crazy. Ezra Klein interviewed Bill Gates. Bill Gates had Sort of a crazy stance going. I don't know if you saw, but yeah, let's play this.

Let's play this.

Like put his legs out, but let's play this.

Get the sound going. What sense of this being something very unusual?

Thank you people for it.

I know people in that world that the sense of this being something very unusual is is is very present.

Does that change what's possible here? Be numeric. Take the cuts in overseas.

I'm not saying it. I'm not saying it answers the cuts in overseas.

We save lives for $1,000 per year. And so getting less money from philanthropists net doesn't save those lives.

People for it. I know people in that world that the sense of— That was the wrong clip.

Well, that was just him of his stance because he's just like very comfortable.

Oh, you were, you were stanced up. There's another video. Let's work on finding it where he's talking about liability. Right. And Ezra was saying, like, why is it not like when a drug company releases a drug, they— it's very important to their business and there are sort of— there's existing laws that incentivize them not to release a drug that's going to hurt a lot of people.

Sure.

And let me try to pull this up. No, it was in the timeline somewhere.

Yeah. Tyler and I were going back and forth on, like, the the authority that the FDA actually has. It can be a little bit confusing because the FDA usually gives marketing authority or marketing denied orders, like you cannot market this product. But effectively, that is you can't sell it.

I found.

They use the term marketing, but they don't mean you can put up a website that doesn't make claims and you can still sell something crazy. You do in fact need to be approved to really scale a product whatsoever?

Play this clip.

What's he saying? I almost can't believe you're asking that.

This is the most dangerous thing that humans have ever gone near.

In other areas, do we just say, hey, release your drugs? There's no FDA. There's no Airline Safety Board. There's no requirements that cars use seatbelts. Do we just use the liability laws to try and keep humans safe. You know, oh, you're shipping opioids, somebody should just sue you. I mean, they— we've created a society that tries to keep people safe, not by saying, oh, we can bankrupt the person who does that. The harms here— and you say there's filtering. There's not filtering. You can take an open-source model that can create bioweapons and disable any monitoring. of any kind. And this exists today. So, no, there is no filtering of any kind.

And so, you know, say, you know, you kill 100 million.

You want to use a lawsuit?

I almost can't keep a straight face. Hmm. Anti-open source?

I don't think as much. The point— I think that the more broad point that he's making is, like, you need to have some type of regulatory process and group. It can't just be full— like, anti-libertarian, basically.

Sure, sure, sure. But yeah, how do you filter an open source model? You put it behind an API. And that's sort of where the industry broadly is coalescing, I think, at this point. The self-hosting unrestricted models gonna have a lot of pushback, maybe right up until there's a scare or something, or there's some— someone actually builds a bioweapon. I don't know. But people are certainly starting the machinery of doing this, and it'll be a lot less popular unless there's something key to point to, like a cyber incident or something like that. But everyone's sort of—

Also, counterpoint, in the 20th century, over 100 million people probably died from cigarettes and the outcome was like there was a big lawsuit. Isn't that exactly what he said would not— I mean, obviously, you don't want that to be the case for what happens, but like—

Yeah. It was just handled. I think he's like, let's not do that again, right? Like, that's a bad That's a bad example. But yeah, I mean, there are laws and restrictions though on where cigarettes can be sold, to who. There is a lot of regulatory harnessing now. And even some of that went in pretty early with labeling and then restricted sales to minors. But yeah, it's an interesting time. Who are the tech There are very few libertarians these days that are actually continuing to push on this, certainly not Bill Gates. Well, in other news, hedge fund titan Ken Griffin is donating $2 billion to start a Carnegie Mellon University campus in Miami.

He saw Horowitz and Dreesen Academy and said, hold my beer.

Yeah. It's part of the largest individual gift ever towards higher education. Very interesting how these name-brand universities are starting to franchise a little more with multiple campuses. There was— wasn't there an NYU in the Middle East at one point? I think there were a few different offshoot campuses. Always hard to get these going because part of what makes the campuses and the colleges special is just like, yeah, it's a bunch of buildings that have been around for 100 years. It feels like a university as opposed to— feels like an office complex. But he thinks he can overcome any of the pushback or adoption problems because he's ponying up $2 billion for it. The Pittsburgh-based university, known for very good self-driving car talent coming out of it, is known for its expertise in computer science, artificial intelligence, and robotics. It'll open a 35-acre campus in Miami's Wynwood neighborhood. And that's like a pretty—

I mean, I got to say, studying at the Ken Griffin School of Market Wizardry in Wynwood, Miami sounds absolutely electric.

Yeah, it does. $2 billion.

Tyler might be headed back to school.

How many SVJs does that get you?

This might be our new quant.

Yeah. I mean, it's a pretty influential area, right? A lot of influencers in Wynwood. Isn't that the vibe?

A lot of influencers. I wouldn't say they're very influential.

Well, it's— there's already a lot of education going on in that area because there's so many courses. So like, you can think of this as like the final course, bro. You're like, he should really—

Yeah, a lot of— I mean, very pro-education crowd.

Very pro-education crowd. Also, Ken Griffin, known for a fantastic supercar collection fits right in there. Ken should really be posting on Instagram. Hey, I made it. Here's how I make $1 billion. Sign up for my course at Carnegie Mellon University.

Yeah. So funny.

The students who graduate and the students who graduate and the tech companies that come to Miami to leverage those skills will change the very fabric of Miami for decades. decades, if not centuries, to come. So I had this take early on when the initial push to make Miami a tech hub was happening, that there were 2 things that were missing. There was not a technical university that was providing a pipeline of talent because so many people, they go to Stanford or Berkeley and they graduate and they want to just stay local. And so they would just work at a tech company And they work at a thing—

No, that's not true because all the quants, or a lot of them, were in Chicago. What's the school there like, I guess?

University of Chicago or Northwestern.

I don't think UChicago is like a STEM school.

They have a great economics program for sure, but also Citadel is so established that they pull people from New York, and also they have a New York office, but moving from New York to Chicago is not as aggressive as moving from from Chicago or New York to Wynwood, bro. Let's be real. But also, of all the big tech companies, Miami ranks remarkably low for tech penetration. Like Google, Amazon, Microsoft, they have campuses in Denver, Austin. They have stuff in Virginia and Chicago and San Francisco. Seattle, and they've all proliferated such that there's a Google campus in Santa Monica. So, if you go to UCLA or USC and you want to stay local in LA, like, you can get a job at Google. It will be maybe slightly different team, but there are, you know, there's a real office and a real campus in this city. That didn't really exist. Microsoft was starting to do some things, but there wasn't this, okay, well, you can go get new grads here, At least to start the flywheel. And then also, you can pull people from big tech because there's a number of offices here. And so, Miami had a ton going for it with the regulation and the climate, obviously, and the supercars, but it didn't have the deep talent pool that made it easy. So, a lot of the companies that moved there, they had to acquire companies and move a lot of people. They had I mean, if you're like a made man at Citadel, going from Chicago to Florida is fantastic for tax reasons. You're going to go down there with your whole crew. But if you're— but they're not trying to scale up necessarily. Like, the core Citadel team is not trying to 100x their workforce in the next few years like many startups are. But if you're a founding team with 10 people and you're like, okay, we might be at 1,000 people, 100 times the size in a few years if things go well. Doing that in Miami has historically been very hard because every recruit process has been hard-fought, basically. I don't know. But yeah, we should send Tyler to Carnegie Mellon University in Wynwood and get him a Huracán to pull up to school.

All right. You're a gamer.

Oh yes, I am.

Somebody put the entire game of Minecraft into Elden Ring as a mod. It runs on Mac.

I can't tell which one's which. Is it— are you playing— are you playing— is it— is the base game Elden Ring or is the base game Minecraft?

I think the game, the actual gameplay is Elden Ring.

Okay, you are playing as a Minecraft character.

Yeah.

Oh yeah, the, uh, the— this was happening before Vive. coding even. I've seen a ton of Elden Ring mods where different characters are replaced with other people, usually more manual. But this is fun. I've played Elden Ring. I've played probably 5 minutes of Minecraft. Never really got into it. But I played a decent amount of Elden Ring. It's a fun game. I like a Souls game. They're fun. What did Tenebris say? I think this tweet is going to do more to educate the general public about modern model capabilities than any launch of any AI company has ever done. That's sort of true, but this could— this was happening in the pre-AI era. Like, it just required taking assets from one game, modifying the files. You had to maybe write a little bit of glue code, but all of this was pretty doable. But I agree that this is going to speed up, and the future of this is— is, is much more personalized because you'll be able to do it for your— for yourself. Uh, anyway, what's it going to take to you to get you to, to beat Elden Ring, play Elden Ring? Uh, do you know that— do you know that I would need to— do you know the current viral Elden Ring question or challenge?

There's no amount of money.

No, it's, it's, uh, it's would you rather spend a year in jail Or you're in jail until your mom beats Elden Ring. Which one would you pick?

I have no idea. I've never played. I don't know anything. Is it that hard?

It's extremely hard. It's extremely hard.

So even if your mom wanted to get you out as fast as possible and was doing everything in her power, and she couldn't get any, you know.

Yeah.

Couldn't she just pay somebody to like handle it?

So the game itself is like 100 hours. If you're good. And you probably need to have played 1,000 hours of video games beforehand to enjoy it and play it well. And so there is a live streamer who's actually doing this. He built a fake jail cell and he's in jail eating ramen noodles and sleeping on this, like, cot. And then he is coaching his real mom through Elden Ring. And you can see that she's never played the game before, so she's not doing well. doing very well and just getting absolutely destroyed by everything. I don't know how real the bit is, but I saw a summary video of it that seemed very funny because it was a funny, like, which one would you rather do? But it was fun. Anyway, is there anything else in the timeline we need to get to before we bring in our first guest of the show?

A very, very, very funny post.

Yes.

From Nick, who says, I'm not joking when I say this, Spotify, Apple Music, and streaming in general is over. AI music is now very, very good. You just don't know yet because all AI music circulating today is from 2 to 3 generations ago. Suno has solved infinite personalized music. And I already had a good laugh about this with you yesterday.

Yeah.

When I first saw it. Obviously provoking quite a response.

Yeah.

Coming from somebody who absolutely loves Suno and has gotten so much entertainment about it, I couldn't disagree with this post more.

Yeah.

It makes no sense, and I don't think anything he's talking about will come to fruition.

We just already went through it with text. Like, we developed the ability to generate articles on the fly. And people still read articles all the time.

Infinite personalized articles.

Yeah. And people read the news all the time. Yeah, there's some fall-off on the long-tail sites, but the New York Times is still really strong. People read New Yorker articles. People read any deep dive, any profile. The Colossus profiles are burning up tech every other month. And so you get both. Yeah, Suno is very entertaining, but it feels like it's a completely different thing, barely even a competitor. Of course, some of the music will be vended into Spotify, but I still remain optimistic that Spotify will survive the onslaught of AI-generated music because the best creations, whether it's made with electronic digital audio workstations, instruments or AI, they will all find their way onto the platforms and go viral, just as Rubbers did by Phoenix Flexin. It was AI-generated, but there was enough of a story and catchiness to that song that it went through the traditional channels and is now streaming on Spotify. Anyway, let me tell you about Codex. Codex is a powerful workspace for getting work done with AI agents, whether you're writing code, analyzing data, creating content, or automating business workflows, Codex helps you move projects forward from start to finish. Our first guest is from NVIDIA, the Senior Director of AI Software, Ali. Welcome to the show. How are you doing? I'm doing well.

Thanks for having me.

Thank you.

It's great to have you.

So much for hopping on the show. Since this is the first time on the show, could you tell us a little bit about your role at NVIDIA, what your day-to-day is like, what projects are you the most focused on these days?

Sure. So I'm on the product side. Most of my time is spent around security, safety, and infrastructure of AI. Me and my team created this project named OpenShell, which was part of our announcement. We joined about 18 months ago. We were founders of a company named Gretel. We focused on data privacy, synthetic data, things like that.

Yeah.

Founded a few startups, and all of us started our careers as security researchers and computer scientists in the intelligence community. So sort of full circle coming back all the way back to our original work here.

I like the name shell. It's better than sandbox. Sandbox, easy to get out of. Any child can walk out of a sandbox. The sand goes everywhere when you have a sandbox. A shell contains the turtle. The turtle cannot just easily escape the shell, right? So I like shell.

More importantly, it goes with it wherever it goes.

Yes, yes. And there's little holes that The turtle can poke the arms out of.

A turtle, being in its shell, can still bite you.

Bite you.

Okay, but it's limited. It can't bite you from any angle because there's only one hole that the head comes through.

We'll keep working this analogy out while you actually talk about—

Once you perfect it, love to get it back. We'll make it into stickers.

So Open Shell, is it open source? Are you evangelizing this product to labs, NeoLabs, everyone? Businesses? Like, how do you see this product actually rolling out and being used by the AI community, the technology community, the, the real economy, etc.?

Yeah, we've been working on it for about a year. We actually released it as an early access in March. It's been Apache 2.0 since we released it. We're actually on pace to donate it to a foundation. So we don't even want to be sort of the NVIDIA project. We want it to be a true community project. project.

Yeah.

We intentionally built it open so it becomes a standard. It becomes sort of a trusted layer. Like, we think that what agents are missing is sort of like, if you think about like the web from '90s to 2000, there's that trust layer, like lights up, SSL is on, your tab is isolated. Like, that construct is missing from agents. So our perspective was if you can create a trusted layer, open it so it can be widely adopted and then built on top of, then you can sort of have a trusted layer everybody can work through. So that was sort of the The whole intent and opening it is really about rising tides.

Yeah. Can you get me up to speed on the state of formal verification? How much optimism should I have towards this as a solution? We had Greg Brockman on the show a few weeks ago, and he was saying that he has this point that there is an optimistic scenario where enough software is formally formally verified, fully secure, that certain systems are hardened to a point where it ceases becoming a cat-and-mouse game? Is that too optimistic? Are we on the path? Are we already seeing glimpses of that? Or is the game going to just be played indefinitely, and we just need more good guys than bad guys?

Yeah.

The state of formal methods and what it can do, I think, the first substantiations of it, we already have proof of life. We can demonstrate what a full-stack verification looks like. So I think I'm very optimistic because a lot of our thinking that has led to this is that agent safety trust really needs to be a full-stack solution. If you think about where models, Frontier Labs harness data tools, all that, that's really the app layer. We built OpenShell because we felt like you needed to be able to move the risk line to a deterministic layer that nothing can bypass. So that's the runtime. You can do formal verification on that and get really good controls. If you then move all that further into hardware, now you can verify the entire stack. So you could be assured down to a single binary decision, is anything leaving this system or not leaving this system? So I'm incredibly optimistic about formal verification, but it has to be applied the right places in the stack for it to have the effect it needs.

Talk about your— how you interface with Nvidia's security features on chip. There's been some discussion about Apple, you know, advancing this with the Secure Enclave and partnering with Nvidia in various ways. And it feels like there's a number of different organizations that are all swimming in the same direction, maybe talking about things differently, solving problems differently, but How do you actually interface with the hardware teams or what are you seeing from the hardware teams that gives you optimism around security?

Yeah. So let me sort of answer it in 2 parts. One, what we're doing on the OpenShells side so we can interact with all the hardware and then what we are seeing on the hardware side. I think those are— and they're agnostic because they both should be equally as good without each other, but a much better story together.

Sure.

So on the OpenShell side, we've just taken a very agnostic approach. We have this primitive inside the runtime called drivers. It's exactly what it sounds like. You pick your compute under the hood. And the reason it's important to pick the hardware, the compute under the hood is sort of what we announced with our Open Agent Safety Platform is we put a reference design out with our BlueField Vera to say if you have hardware, OpenShell can take all these controls at the network for chain of thought. and move it all the way into hardware. So now you can do it at machine speed and you can have very deterministic enforcement on it. It was sort of a reference architecture for us. So from the Open Shell side, we've tried to be as agnostic and inclusive as possible by giving a driver that any computer hardware can plug into, which is— you probably may have seen this in our announcement, why we briefed ARM and Intel and sort of go down the list of hardware providers, right? Because we want everyone to be able to offer this full stack Trust and safety verification. As far as what we are seeing on the hardware side, there are 2 things we are starting to see, sort of generalized patterns. One is the hardware component making primitives available so the higher levels of the stack can offload networking into it. Think of it as a way to manually flip the switch and cut any access.

Okay.

So if you're doing that in the hardware, if you're an advanced, for example, researcher or a frontier lab or a large organization doing research or sort of testing of a frontier model, by having that control at that stage, you know, and can verify the boundary or the blast radius you're setting is not going to be breached. Right. It's in the hardware. It can't be bypassed. Agents can't work together to get around it. So that's one area of sort of work that we are seeing. And obviously, this is the work we're doing at NVIDIA, too. The other side of it is Being able to pass through the way the agent thinks and reasons through silicon.

Mm-hmm.

So you can actually see what it's thinking about. The reason this is important is, is like if you're familiar with security, the holy grail is shifting left, right?

Mm-hmm.

Not responding, but preventing. Like, how do you do that? So if you pass that chain of reason, that thought the agent has through silicon, you can see it's thinking about using a zero-day. Then you can check with the runtime, did it probe anything to see if anything's open? You can come back and you're like, oh, it did probe. Now it's reasoning to say, I actually found a hole. Now let me go build an exploit. Then you can pass that and cut networking and say, this thing is about to break out of the environment because it's reasoning about using like a zero-day or a vulnerability. Or like most agents, it could just think about these things and basically let it go as a chain of thought and not come back to it. But those are the 2 really interesting areas of research we're seeing in hardware right now.

Please.

Sorry if this is a silly question, but would an agent actually think, hmm, should I use a zero-day today? Or would it just identify a potential exploit that, even if you were observing it, wouldn't necessarily be immediately obvious that it was an exploit?

Yeah. So, you know, I'm sort of being overly rudimentary about saying it's thinking about an exploit.

Yeah, I know you're trying to dumb this down to us. You're trying to bring this down like— Let me bring it down like 20 levels. No, we—

No, no, no. At least— no, no, 15, 16 max, not 20. I mean, come on.

Preschool. Preschool. Please keep it— don't keep it kinder— I need preschool, not kindergarten. Thank you.

Sure. So let me say it differently. If you look at the types of events that happen with these labs where the agents did something that was unintended, Yeah. It wasn't a single step function in capability that achieved that. It was doing a subset of things like stealing credentials, communicating with each other in channels that you didn't have, persistent sessions longer, doing all these things that at the quantum level, we know how to solve these things from a security standpoint, right? Like, we know how to keep credentials away from humans. We can do the same things for agents. We know how to introspect traffic. So it was a whole bunch of techniques uniquely put together and then ran at machine speed. Now, the machine speed part is solved in silicon when you can see it thinking. And then the path to exploitability and vulnerability exploitation is a set of things and actions that are well understood, right? You probe the network into an environment that the policy says you're not allowed to. You take that reading and you build a tool against it. So this iterative process is something we can build rules around and then build failsafes as a particular part of the stage.

Yeah, that makes a ton of sense.

Talk about when, like, throughout your career in security, there were sort of spikes of panic. Like, I would imagine we're generally at 10 out of 10 relative to maybe other moments, but I'm curious if there are other moments that you went through on the path to this current moment that were anywhere close, but Maybe they weren't as sort of public and talked about on a national level or in the media as much, but maybe were more kind of internal industry.

With SolarWinds?

Yeah, I mean, so it's funny, like a lot of my panic was either associated with the intelligence community that I can't talk about, or as a founder, which had to do with funding or board members.

It was less about the ecosystem and more the conditions.

But I would say that I wouldn't call it panic. What I would say is, is like ambiguity increases risk, right? And security is about risk reduction. There's never a perfect system. So I would say the times that created the most level of ambiguity where then sort of opinion fills the void were when we had these massive step functions in form factor, like not SolarWinds, but like moving sensitive compute to mobile or to cloud or making them fully distributed. Mm-hmm. And now sort of it feels like another level of control given away, right? Like if you think about cloud, everybody understood hands around things, right? And now you're like, where is my stuff? Like that was a very fundamentally conceptually difficult shift to make for a lot of years. And I think agents, AI has a very similar factor. Like we joked about sandbox, but like to use one concrete thing, right? Like sandbox is meant to constrain something, keep it in or keep it out. These agents by definition have to do both Mm-hmm. So when we talk about, oh, sandboxing's not effective against these agents, we're all panicking. Well, because we're sort of using a rudimentary version of it. So the same way we didn't take like these monolithic applications and shove 'em in containers and then run 'em on Kubernetes, the way we redesigned with microservices and immutable ephemeral infrastructure, the same thing needs to be done here. Most of the problems we're seeing is we are running an advanced technology into an environment that is not meant for it or designed for it. So what does a native environment look like for AI basically is where I spend a lot of my time.

Well, thank you so much for coming on the show.

Let's do this again soon. You got to do this again soon.

This was a great conversation.

I really enjoyed your perspective.

But yeah, have a great rest of your day.

Thanks for having me on, guys.

Thanks for the work.

Thank you. Goodbye. Let's bring in Mike from CoreWeave. He's the CEO. We've been keeping him waiting and I'm very excited to talk to him about the future of CoreWeave. Mike, how are you doing? Welcome to the show. Thanks, guys. Great to be here. Thanks so much.

I've been excited to do this. You guys had done it with Brian, and Brandon, I figured it was my turn.

Yes, yes, yes. I know. You know what we see? We sometimes see founding teams, executive teams, like, end up being like, no, these are, these are my guys, like, and they don't like to share the TVPN. They don't like to share the TVPN visits. But we like to meet, we like to meet the whole team.

Yeah. But we had the— hey, you guys are gonna love this. Go, go, go hop on with these guys.

That's great. I want to talk about some of what Elon and Jensen were talking about, the compute math, the monetization of gigawatts. But first, let's get the latest from CoreWeave. What's newest in your world? What is top of mind and on the frontier of your business today?

So look, I'm speaking to you from Fully Connected, our global conference here where we're you know, sitting down with 5,000 of our clients, developers, companies, enterprise labs that use the cloud that CoreWeave has built to be able to drive their business. I couldn't think of a more exciting place for our team to be.

Yeah.

Right? To get an opportunity to be in the seat with these folks and get the feedback loop and in going in a first person is just, you know, I mean, that's a unique experience for a company. And, you know, in many ways, CoreWeave is really well positioned to take on the information that we're garnering from these clients about, you know, what we've built and how it could be better. And, you know, all those pieces really come together nicely when you get a few minutes to put some of your engineers in the room, some of your product people in the room, and some of your clients in the room. room. And that's what we're really here to do. So it's fun. It was supposed to be 2,000 people. We're over 5,000.

5,000. Wow. Yeah, it's great. It's great.

It's like—

And I imagine the customers are no longer just asking for just a bunch of GPUs. Like we are firmly in the AI agent era. They want CPUs, they want new storage solutions. Like what are they asking for you that you're saying, oh yeah, we are already working on that. We'll work on that. We'll We'll do that. Give us a day, give us a week, give us a month, and we'll be there for you as a customer.

The truth is, is what the clients are asking us for, what most of the clients, right, are asking us for is the cloud, right? Like they're not asking us for GPUs, right? They're, you know, like when we think about our client base, like 90% of our clients use 2 or more of the products that we deliver. And that's the software solutions that, you know, Yeah. All of the different pieces that we stack up when, you know, if we get down to like 78% or something like that, it's like, you know, we're at 3 or more products. So like we're not here anymore to say, hey, we can deliver you GPUs. What we are really here to do now is to deliver an AI cloud, a cloud that gives you the full functionality so that you are able to, you know, build the, to the potential of your company, to the potential of the your project, to be able to serve it globally, to do all the things that, that we have historically thought of as what the cloud's role in, in, in the broader ecosystem, but really much more targeted towards, uh, the, the role for a cloud in the AI era.

Yeah.

And so, um, you know, we, we, we announced a, a, our, our, our, our, our product here called, called Forge, really, which is, about bringing together all of that feedback, bringing together all of the tools and software rails and delivering it to the broader universe. So like not everybody's got, you know, AI experts at the level of these frontier labs. So you've got to be able to build product to shorten the distance between what enterprise clients need to be successful in integrating artificial intelligence into their workflows. And that's what the AI cloud is going to do. That's what Forge is built to do. And we're really, really excited about rolling it out because we think it's going to be a very, very successful effort for, for us as we kind of continue to diversify across the enterprise space.

I have a bunch of questions. Go for it. Okay. Yesterday we were talking to Sam Altman and he was very excited about the OpenAI compute buildout, specifically just the lack of homogeneity across the inference opportunity, everything from big beefy models that are gonna run for a long time, the Decisions API, the stuff that they're doing with the Jalapeño program. Like it really feels like we're fully stepping out of this era of, okay, just get a bunch of the best chips and that'll do everything. there's so much economic value in choosing the right tool for the job. And it feels like that's a big piece of your opportunity is to bring that capability to everyone. What does the next year or 2 look like for you on that side? Obviously, you're building the software solution to harness that, but what other decisions are you making either on the hardware side or with your capital partners? What are the different decision parameters that you're wrestling with right now?

So I love this line of questioning because I've been pilloried over this for 2 years now, right? Like, oh, you know, you're going to build the GPUs and they're going to be obsolete in 4 days after they come off the line. It killed me because the depreciation conversation was— it was a self-serving conversation. by a few players that had a vested interest in that narrative, right?

Sure.

But the reality is, right, is that, you know, you need the most performant GPUs to build on the frontier. And then once you've built it, there is this incredibly fat tail of other use cases where you can take GPUs and keep them deployed productively for a very, very long time. And so the fact that Sam is talking about it, the fact that you're talking about it, I'm gonna take a personal victory loop here because I've been getting clubbed in the head over this for every time I go on CNBC, they take a whack at me on this. You deserve a victory lap. The truth is, is like, look, compute is going to drive intelligence that is going to permeate every asset of civilization, right? Like, that's, that's a fundamental belief that we have. And what we have seen repeatedly over the years is that every time a GPU is no longer capable of driving the frontier forward, it has a slew of other uses that will absorb it. And, you know, I talked about this in my last earnings report, the A100s, that's 2020 architecture.

Yeah.

Right, getting that contracted out through 2029, right? Like, that is a, you know, like, that kind of puts that debate to close, right? Like, totally, there are use cases that will absorb this, whether it's batch computing or medical research or, you know, there's a You know, or things that we haven't even invented yet because there's not enough compute that can drive the new businesses into existence. And so super excited about that. I love that you guys are thinking about that. I love that Sam is thinking about it also. I think it's important.

Yeah, yeah. There's just so many different opportunities now. You saw like ultra-fast, 8x the speed, but then also the Jev moment and Decisions API and like the computer use and then image-generated video. There's so many different opportunities there. Jordi, you have a question?

Our friends over at SemiAnalysis have you guys in the Platinum tier. Have you been pushing them to introduce a Diamond tier?

Oh yeah.

So that you and the team have something to grind for? Because I imagine you don't want to be stagnant.

Yeah, yeah, yeah, fair enough, fair enough. You know, we joke— I don't know if you guys have spent any time with Peter, but we call him Premium Peter, right? I like to joke around here is, and, uh, uh, he's amazing, right? He's the principal architect behind, uh, uh, the, the so much of the cloud that we've built. And, uh, you know, that would, uh, he would work himself to death trying to get into the IT tier.

I think they gotta open it up. Yeah, well, just a big empty tier at the top.

Well, 'cause now I just, I don't know what to think. It used to be you guys were the only ones there. Now there's a new player there. You gotta get used to it.

It's getting crowded, man.

We need a new mountain.

Going off of that, when I look at the new ClusterMax 3, which came out, I think, last week or maybe Monday, I'm losing track of time, there's a bunch of names on there I've never heard of. And we spend all of our time talking about these companies. A lot of them, you know, a lot of the other companies that I do know have been on the show. Right now, there's so much, like, excitement and new capital formation around some of these neo-clouds, And my question is, like, I have to imagine, like, I think the natural assumption is that, like, half of these companies, like, won't make it independently forever. And so my question is, like, what do you think? When do you think the window is going to come where these companies that have invested, let's say, between $100 and a billion, you know, billion dollars into building something but aren't going to quite make it, and then that becomes a sort of opportunity for— the more scaled platforms to hoover up talent and maybe some more specialized products and even down to the individual like site level?

Yeah, it's something we think about here and talk about a good bit, right? And I want to be clear, there's a lot of different paths to consolidation across the cloud industry. And, you know, one of the things is, you know, we don't really think of ourselves as a neo-cloud We are the AI cloud, right? Like, you know, the concept of being a new cloud is great for a while, but the truth of the matter is, is our job is to deliver the functionality that the world needs, and that's the AI cloud, right?

Yeah.

But you also get to a world where you get consolidation. It could be because there's an air pocket in compute demand, right? And that's one of the things where, you know, it would be destabilizing. But there's also another scenario where, you know, Power becomes more dear, and the selection process for who people are going to sell to tightens, right? And they're just like, why am I going to take a shot with a third-tier new provider when I can go ahead and get the same terms with a reputable, well-established—

Yeah, it just comes down to like—

—capital in the office.

—capacity. If I'm a company, I'm planning to scale my demand for compute really quickly, it doesn't make sense to sign with the third-tier provider simply because I'm going to run out of capacity or—

Yeah, and it also causes drag on the balance sheet of the offtake companies, right? If you're a buyer of compute and you've got 47 different third-tier clouds, that can lead to drag on your balance sheet, which is not great, right? So you want to have an efficient allocation of capital and efficient allocation of the capacity of your balance sheet. And you do that by really working with entities that you have a high degree of confidence are going to be able to deliver performant infrastructure.

Yeah.

I think that there's an inevitability, right? Like, it's a capital-intensive business at the end of the day, and capital-intensive businesses tend towards consolidation, whether the number is going to be 10 or 7, I don't know. When it's going to happen, I don't know. The priority for us is to build a resilient company. And you build a resilient company by working with high-quality clients, by delivering a high-quality product, by decommoditizing the compute that you're delivering, by having incredible software rails so that people are able to use it, by moving across the enterprise space. All of these things have been priorities for us. And we have done a really, really good job of staying focused on, you know, as a team on making sure that we drive the parts of the business that will ensure that CoreWeave is resilient through whatever cycles come. And we're pretty comfortable about that.

Amazing. I have one last question. We'll let you go. For Q2, the rough dollars per watt was around $7 billion per gigawatt. It feels like that's accelerating. It feels like you might be tracking closer to $40 billion per gigawatt. Elon and Jensen were doing some rough math around $60, $70. Is this something that you have visibility and optimism around continuing to climb, or is there more focus on getting to some sort of steady state where you're monetizing the energy that you have and then just scaling out Horizontally?

Yeah, um, I, I look at the problem and opportunity a little bit differently, right? I look at it from a time horizon perspective, right? And so, um, I believe that there is going to be a market for lower-cost compute that lives outside of the build cycle of infrastructure. So if you think about it, and let's say just hypothetically it takes us 2 years to build a data center, if I can sell, uh, capacity in a time frame beyond that 2 years, I can replace it, right? Once I'm inside that, that's where Elon Musk and Jensen are talking, right? Because then it becomes truly a supply-demand dynamic where you know, people are going to compete for that compute. And those are 2 separate markets. And I believe that over time they're going to begin to behave differently where people who can make the long-term planning are going to be able to drive economies that are associated with that because it's going to allow folks on the infrastructure side, on the cloud side, to be able to plan better.

And so There is—

it's a really fascinating piece of the market when you think about it in terms of time, not just in terms of dollars.

That's very interesting. Well, thank you so much for coming on and breaking it down for us. Congratulations.

Say hi to all 5,000 people over at your event for us. Hopefully we can join next year in person.

Yeah, it'd be fantastic.

Love to have you guys here, man.

That'd be great.

Let's do it from the home plate.

Awesome. Thank you.

Yeah, we'll talk to you soon.

Goodbye. Let me tell you about Shopify. Shopify is the commerce platform that grows with your business and lets you sell in seconds online, in-store, on mobile, on social, on marketplaces, and now with AI agents. Our next guest is Vlad Tenev, the CEO and co-founder of Robinhood, here to discuss artificial intelligence, finance, everything in between. How you doing, Vlad? Welcome to the show. What's up, dudes?

How you doing? We're good. I'm sorry, I'm, I'm too focused on the car.

Yeah, you got a yellow Toyota.

As much as I'm excited to see you, Break it down.

Explain where you are and what you're doing.

So I'm in Houston, Texas, at our space-themed active trader event called Hood Summit: Engines of Creation, where we have 1,500 of our most engaged active traders. We rolled out 16 products which are awesome, like first of their kind. You know, 24/7 trading of stocks will be the first place in the US that allows that. Perps, the first true perps offering in the US with 8 perpetual futures. And then Robinhood Agents, which is an evolution of our agentic financial superintelligence products. We also sponsor 23XI Racing, a NASCAR team. So behind me is Corey Haim's car. And I don't know if you guys are NASCAR fans. So we're America's brokerage. So it's only It's only appropriate for us to be involved in American motorsport. That's great.

That's great. The livery looks fantastic.

Yeah. Easy to spot. Yeah. Talk about—

I didn't realize these cars don't have doors. You have to like jump in through the window.

Yeah, you go through the window.

That's crazy to me.

Have you driven it yet? Have they let you behind the wheel?

I was trying. I was trying to, but I don't think—

Yeah, it might be a little loud. Corey didn't let me.

Maybe the safety car. You take the safety car. Safety car for a spin. Talk about talking about the agents product. So yeah, talk to me about the agents product. How much of this is pull versus push in the modern era? I mean, we're seeing this happen with the personal agents. Just yesterday, people were getting the reversal. They were going to instinct for shopping. And then yesterday they were starting to get, you know, ideas and recommendations for products that they wanted to buy pushed to them. Some pushback on that, but But in an agentic trading sense, I feel like if I'm on Robinhood, I kind of want you to come to me and say, hey, why do you have your money in this savings account that's earning 1% when you could be in this thing that earns 3%? Want me just to do that for you? So in terms of trade ideas generated by the AI agent that comes to the customer versus me going and saying, I want you to go and run my portfolio this way. What's working? Where do you see it going?

Yeah. So I think there's lots of different types of agentic finance products that have been tried. Agentic trading is a little bit different than what you're talking about, which is just like almost like a one-time, hey, move my money over to where it earns higher interest. You don't need superintelligence for that. Robinhood Agents is really about helping people create strategies. And so we rolled out 3 things, right? One is Robinhood Agents, which you can think of as, you know, we rolled out an MCP a couple of months ago, a Model Context Protocol, which you can think of as an API that lets you connect external agents like your Claude code or your Codex to our services so that they could place trades on your behalf. That was a very popular product. It has over 150,000 users have actually connected external agents to it. And so Robinhood Agents brings that within Robinhood with all the benefits like trade approval, separate account, brings it in the Robinhood experience where we have over 28 million accounts. The second thing is agent apps. Now, if you think about what a lot of hedge funds have access to, it's proprietary data. Things like unusual options flows, satellite imagery, you know, congressional trading data. And agent apps you can think of as an app store where third parties can plug in, provide proprietary data for use by our customers in creating trading strategies. The third thing we created, agent loops. So you can have your agent run autonomously and predictably in a deterministic way. And execute that strategy for you while you sleep. Sure. So you put these 3 things together and the way I describe it is we want to give you the power of a hedge fund in your pocket, an extremely sophisticated team of financial trading professionals. Can we give that to everyone? And yeah, I think the way that I think about it, you know, my, my— the start of my financial career was in algorithmic trading. And in the same way that tools like Cloud Code and Codex have made it easy for anyone to be a programmer, Robinhood agents will make it easy for anyone to be a financial trading quant. So you can, you can develop a quant trading strategy, extremely sophisticated trading algo, and we can kind of unlock that for millions of people. Yeah.

So what's going on in alternative data when you said satellite imagery. You brought that up. I thought it was interesting because historically, I imagine some type of research organization or a hedge fund would get satellite imagery and then they would need to spend a lot of time actually analyzing the data. Let's say if you wanted to understand retail foot traffic at a big box store, right? You would be able to look and understand.

Yeah.

The models out of the box can do a lot of the kind of things Now that it would've historically taken building your own algorithms that maybe weren't as reliable, things like that, to extract alpha out of something like a satellite image. Now it feels like that kind of thing can be way more democratized. Do you expect there to be that sort of alternative data to just become more democratized now that models out of the box can do this kind of complex research?

Well, you still need to pay for the data, right? And so—

Yeah, but that's the kind of thing like you guys have 28 million accounts. It feels like these are the kind of things that could ultimately be crowdfunded one way or another or brought down to like a $5 subscription instead of like a $50,000 subscription.

And these things are kind of like skills in our product. So instead of having to go and sign up for it on the internet and give your agent your username and password. They're just directly integrated. And, you know, our partners are all offering these services one month free. We also have a partnership with OpenAI where they're helping us with GPT-6 Luna that'll be offered for free till the end of the year. Cool. And, you know, a lot of people say, well, what is this? People are just going to write to their agent and say, make me money, and everyone's going to be doing that. I think what's likely to happen is you'll, you'll get more depending on what you put in. So you mentioned the satellite data. You know, imagine exactly what you were saying, someone actually plugging that in and saying, you know, check out these parking lots and build me a proxy for retail traffic on the holiday season. Or my favorite, look at the parking lots of various companies and see which ones have employees working weekends or late nights. You know, and you can imagine the possibilities there. There's, there's, there's lots of strategies that are just waiting to be created. I think it'll unlock a whole new kind of arena for competition and creativity among traders.

Yeah. Yeah. You described it as sort of like a hedge fund. You mentioned the idea of having a robot in your pocket, and I'm wondering if you can extend the analogy further. A lot of the most performant hedge funds run the pod model with centralized risk strategies and multiple pods with different strategies. Do you imagine that the more proficient Robinhood trader in the future will have multiple agents running multiple strategies? With some sort of centralized risk control? Because there's often these ideas that you just listed a bunch of them. They sound exciting, but I don't know if I want to put all my eggs in that basket. But maybe as part of an ensemble strategy, it does start to make sense. So how far are you leaning into the hedge fund metaphor?

Yeah, I mean, so right now, the way Robinhood is organized as a product is there's multiple accounts. Sure. And this is a relatively new thing. It started out with you just had one Robinhood account, but then we rolled out retirement. Then we have joint and trust. And so there's an agentic account. So your AI agent is kind of sandboxed and scoped financially to, to one financial account. And by default, we've got some guardrails, right? Trade approvals are on by default. You can turn it off, but By default, you would have to say yes to every trade that it suggests.

Got it.

You also have to move money into your account willingly, right? So it's not just going to go in your, in your retirement account and trade for you. The trading is all scoped to money and assets that you load into the account. And these are all things where we wanted to start with a sandboxed experience. We wanted to see how things go. Of course, we might change it over time. But yeah, that's sort of our initial risk management strategy. And, you know, we'll see how it evolves over time. You could imagine you could have like a different type of agent that's not a trading agent, but more of like a financial controller that can allocate between different accounts. They can help you with budgeting. They can help you do sort of like these high-level one-time management functions. And that could also serve a risk management layer between your different trading agents. But for now, we're kind of keeping those ideas separate.

When John said pod, it made me think, is there a reason— like, have you experimented with having a group chat or, like, let's say, 3 friends be able to jointly fund an account and then all effectively be making the same trades or be having to agree on on a set of trades and running like—

Yeah, like an investing club.

Yeah, an investing— basically an investing club. You have the social features on Robinhood now.

Yeah.

That feels like hard to do at a wider scale, but at least for peer-to-peer, like actual friend groups. Is there any like— is there anything like—

Is there a business opportunity there?

Is it legal?

Or legal, yeah.

Yeah. Is it a technical challenge or like not possible because of— for legal reasons?

Yeah. I think you— I think there's definitely opportunity to innovate and create new things there. There is actually like a construct called an investment club that allows, you know, friends to pool money together. And there's like a sort of structure around that. We haven't gone there yet, but we do have Robinhood Social, which we launched for everyone last night. And basically what that allows you to do is to share ideas with kind of the broader Robinhood community. And it has a follower graph. And you can also ensure that, you know, the trades that you're seeing on the platform, the P&Ls, all of that is validated and real. And then we make it easy when you see an idea to actually trade from that idea and And you can, you can imagine like following different traders, learning from their ideas. We also added creators onto the platform. We have politicians so you can follow their trades. And we're going to— we've been evolving it very, very rapidly. A lot of people here actually are Robinhood Social users and love it. So I think you should expect us to continue exploring these ideas and just giving, giving our traders what they want, what's most useful to them.

I'd love for you to zoom out from Robinhood and the products that you're launching to the markets broadly because you have such an interesting perspective. I mean, yesterday there was an IPO from Aura that was delayed. And at the same time, the market is massively up this year. Things feel very strong. How are you feeling about either the market strength or weakness? But just how are things changing? Are, are we in a new regime where an IPO needs to be 15x oversubscribed to get out successfully? Like, what, what are you noticing where maybe it's not even good or bad, it's just something where you're like, oh, this is new, we're in a different era?

Yeah, I mean, I don't know if we're in a permanently unique era for which there's no precedent. I've, I've seen the market conditions change in different ways. on almost a weekly basis now. Right. And I remember it wasn't too long ago that you guys covered the Figma IPO.

Yeah.

And it felt like we were just at the absolute just apex of animal spirits at that point. And I think things have calmed down since then. We've entered a bit of a crypto winter as well. But now that's— now we're even starting to climb out of that. Crypto's had a bit of a moment the past couple of weeks. Prediction markets have obviously been, been going gangbusters. So I think there's always something going on in some corner of the market. And as we've been adding more and more asset classes like perps, we added, you know, these really interesting products that I think you guys will like. We're going to be the first ones offering prediction markets on earnings in the US. So you'll be able to trade contracts on company EPS and revenue as well as, you know, things like iPhone deliveries, Tesla deliveries. And, you know, that, that'll give customers all kinds of new abilities to like if you want to hedge your position in a stock that you're long and you don't want to have to sell it, but you, you might think they miss earnings or something, you'll be able to kind of use these products So, yeah, it's also interesting.

Everyone says a lot of people have had the experience of like, you know, being, being long into earnings, being right that that's an earnings beat, but then some other factor sends the stock down.

Oil prices went up and interest rate hikes or something.

Yeah. Being able to—

Yeah, you're right. It's a big frustration for our traders. So these products allow you to more specifically execute on a, on a trading strategy. So it's a little bit more pointed and sort of like, yeah, more specific.

How are you thinking about the knock-on effects of AI agents entering the financial markets, research markets, trading markets? When you, when you play back the story of Robinhood, you, if you had a crystal ball, you might have been able to predict GameStop and meme stocks and and Reddit. But there's already been some people who have said, oh, well, in a world where AI agents are moving money around, no one will be in the money markets, and there will be a bank run. And there are some concerns about how things will play out. Maybe you won't need a financial advisor anymore. But a financial advisor, of course, doesn't just allocate your portfolio. They do a lot of other stuff. And so I'm wondering if you're looking to the future and seeing any emerging trends that might sort of change the underlying structure of the financial markets over the next couple of years as AI rolls out to trading?

Yeah, I think it's, it's important to separate kind of the hype from the potential there. I think a lot of these things are sort of like one-time use cases, right? Move my money from Sure. You know, low-yielding savings account to a high-yielding money market account or, you know, renegotiate my, my cable bill or something like that. And I think what we've seen is there's really one area that has long-term retentive usage right now. And, and that's agentic trading. Right. And so we're just kind of focusing all of our efforts on making agentic trading as great as possible. And I think over time you'll see us exploring other things that are maybe not there yet in terms of usage and utility. But Agentic Trading has like clear product market fit and utility now. So I think, I think that's going to be the one, at least for a while.

Well, it's exciting. I'm, I'm excited to see how people use it. All the different screenshots I'm sure we'll see of of people doing all sorts of crazy things.

No, I remember when the Robinhood social beta, like, started, uh, screenshots started being shared of that, and, uh, it was wildly entertaining. So I'm excited. I'm very excited for open access.

Yeah, I mean, there'd be some people that put up insane numbers, some people that get wiped, uh, that, uh, you know—

No crying.

Yeah, no crying.

No crying in the market.

At the long tail, you get crazy, crazy stories. So thank you so much for coming on the show. Congratulations on the launches, and we'll talk to you soon.

Yeah, give our best.

Always a pleasure, guys. Have a great one.

We'll talk to you soon. Uh, speaking of IPOs, let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. Just do it. Stop making excuses. Um, we— up next, we have the founder and CEO of Flow Engineering, Pari Singh, coming back on the show. It's been almost a full year, but We gotta get that gong ready. How you doing? Welcome back.

Hey, hey, good to see you guys. How are you?

We're doing great, but it seems like you're doing even better. What's the news? What's going on? You raised some money? How much did you raise?

We raised $50 million on a $750 valuation.

Congrats.

John was feeling that one.

Thank you very much.

And for what purpose did you raise this money?

We are gonna invest heavily in AI.

Okay.

We are doing a huge amount with AI.

AI is completely transforming hardware engineering just in the way that it's transformed software engineering in the last year. I think when we see design cycles of rockets and airplanes and cars go down from 12 months to 12 weeks to 12 hours, the world's very, very, very— that is the opportunity ahead of us, and that is what we are investing in.

Do you like the term AI? Or do you think the next cycle you'll be saying superintelligence? Is that a meaningful designation to you? Or do you see— because, I mean, this is a tool, this is a feature.

I'll go first. I don't think it's gonna stick. I think it's— I don't think we needed a new name.

Okay. Last word, Pan.

Going from artificial to super is not suddenly gonna, like, help the technology's popularity.

Yeah. Okay.

I do think there is a distinction. I'm not sure, like, everyone's agreed on the distinction, but to me, as you say, AI is like a really fundamental tool. And when you put a tool in creators' and inventors' hands, they're going to invent crazy stuff. I do think we're going to get to the point where AIs and humans collaborate together to design a new class of products, and that those products humans couldn't have done by themselves.

Yeah.

And when you put AIs and humans together, that's a really, really, really special place to design a whole new wave of hardware.

Yeah, I think that, I mean, putting aside the AI versus superintelligence debate, it's hard to gauge model progress from like, you know, deep research reports because those were pretty good a year ago. They're still really good. But this year, the thing that's felt like it's jumped forward the most has been things in the 3D world, the Blender modeling demos, the video game development. And that feels like a glimpse into, okay, if it's that good at making video games and modeling architecture, it's probably good for hardware engineering. So how much of this year have your progress has been, okay, the models are advancing and most importantly, put aside raw IQ, are they super smart? super intelligent, just they got good at hardware this year, or there was a step function. Did you feel that this year? Has that been an important unlock, or has it felt smooth from the inside of the industry? Because from my perspective, it was like, I would never try to use an LLM with Blender a year ago, and now it would be my first step to work with Blender.

So AI is getting really, really good at hardware engineering.

Yeah.

When you look at modern products, whether it's a rocket or an airplane or a car, CAD and 3D design is one really important element, but it's actually probably the last element. By the time you are designing the 3D geometry of your Waymo, for example, all of the mechanical and the electrical and the sensors and the AI and the software, there's tens of thousands of hours of engineering work that's gone into the product. before you get to geometry.

Yeah.

And I think we are already seeing AI chip away at that. I think the world that we've seen in AI software development is as the models get better and better, that they can start to do more and more things. In hardware development, the products that we design are so complex that it requires mechanical engineers, electrical systems, software, regulatory test. It requires all these different types of engineers to come together to collaborate on a system. And on every single one of those, AI has been getting meaningfully better over the last year.

So at what level of abstraction do you want to interface with a company like Rivian or Joby or Skydio or Stoke Space? Because I imagine within a carmaker, there's the, the wire harnessing and the mechanical functionality and all sorts of different systems that need to eventually be integrated. And I could imagine a system overseeing that, and anytime a change happens to the electrical system, you can flow that through the rest of the project to say, oh, well, this is going to affect this system over here, flag that. Do you want to be the overseer, the sub-agent, both? What's most valuable right now to modern enterprises?

I think you called it. So the single biggest problem in hardware companies is systems integration and verification.

Yeah.

Let me give you an example. Let's say you're designing a reusable rocket and you make a tiny change. Let's say you take the stage 1, and in stage 1, you take one of the engines and you make a tiny change to the injector. Well, that has this chain reaction of changes across the fluid flow and the combustion and the thrust and the mission profile. Each one of those will change other things, which then change other things. And it's that 3rd, 4th, 5th order impacts is where we trip up and that's where failures come from.

Mm-hmm.

So the single biggest problem in hardware development is actually integrating systems together. And because it's been so expensive and so painful to do that, these systems engineering organizations could only really afford to do it every 6, every 9, every 12 months.

Mm-hmm.

It's a really, really expensive process to do full-on verification. But what AI enables for us is this idea that we can move from manual verification which is once every year, to continuous verification, just like we have in software engineering with CI/CD. And when you bring that paradigm over, engineers can make changes in software, in CAD, in Git, in sim on a nearly hourly basis, and the AI can understand the ripple change of that impact and flag issues the moment they happen.

Yeah, I want this for architecture. I want an architect to be able to move a wall in an architectural drawing And an agent goes and reads the laws and see if this needs permitting. And actually merging the law, which is written in text files, with CAD, which exists in the 3D space, that feels like glue for AI such that you have sort of like red, yellow, green stoplight as you're building or as you're designing the structure. And that feels like a lot of what you're doing. What about—

How do you think that the world is going to need to adapt adapt and supply chains are going to need to adapt to the actual design process and hardware engineering process speeding up by 10x and then 100x, which feels very inevitable at this point. But oftentimes, during the actual process of designing the system, you're also figuring out how to source all the different sort of inputs to the final product?

Yeah. So, so this is like a really, really important point. Hardware companies are vertically integrated. They're more vertically integrated than they've ever been before, but they still need to partner with suppliers at a very deep level. So even the most vertically integrated automotive companies have hundreds, if not thousands of suppliers. And the requirements for what they need is changing on a nearly weekly, if not daily basis. And that core collaboration is happening not just across the company, but across the partner too. For a number of our largest customers, customers designing everything from data centers to EVs to drones in the defense industry, they're actually working natively with their partners on Flow. So that core collaboration, changes that the partners will make will flow into their systems, and then the AI agent will understand the implications of that change in real time. And what used to take weeks or months before can now be done in seconds. And the thing that John mentioned is so important, like architecture is great because it's really obvious what the regulatory requirements are. But if you look at any one of our customers, customers like Anduril, Joby, Stokes Space, Rivian, there's hundreds or thousands of regulatory requirements that need to be kept up to date across mechanical, electrical, software. And it's a really huge problem to do that today manually.

Yeah. What's the state of integrating a system's integration— integration? Is it faster, or do you have to rip and replace some other system, or can you sort of drop in, drop on top of all the systems? Is computer use speeding that up? Are there MCP servers for some of this stuff? Because it feels like you're very much operating in the real world, and you can go and reverse an API, API from some old piece of software. But what does it actually take to onboard a new customer these days? Yeah.

So amazing news for Flow and probably what drove the raise today. 96% of our customers come to us inbound and they're able to get up and running within 30 days. Wow. So legacy players are thinking like a 9 to 12 month period. Of course, 30 days. And AI is absolutely driving that adoption in itself where we typically land. is on something called requirements management.

Sure.

Which is this sort of typically old-school stodgy tool base.

Yeah.

Which houses all of the requirements and specifications. And the reason that is so important is it's, it's the bit of information that all the design data in the company will flow back to. So if you're making changes in CAD or Git or simulation, they come back to verify the requirements because that is your CI/CD process. So we are the system of record for requirements and verification in these hardware companies. That is, that is our wedge. But what we are becoming is this much more important core collaboration layer, this context graph for the AIs and the humans to be able to work together in a seamless way.

Well, congratulations on the progress. Congratulations on the new round. And thank you.

Love getting a new snapshot of all the progress.

Yeah, fantastic progress.

Congrats again. Great to see you guys. Thank you so much.

Have a great rest of your day. We'll talk to you soon. Let me tell you about Cisco. Critical infrastructure for the AI era unlocks seamless real-time experiences and new value with Cisco. Our next guest is already in the waiting room. We have Mina Song from Elise AI, the co-founder and CEO. Mina, welcome back.

Welcome back.

So how are you doing?

I'm doing great.

You raised some money?

We, we did. We raised $350 million.

Just another day at the office. Fantastic.

I've been waiting all week for that. Huge growth too. Huge growth too. $200 million ARR. How'd you get there? What's driving the growth?

Yes, we passed $200 million ARR earlier this year. We're really— it's coming from serving more customers in the housing industry and the healthcare industry. So we've developed a lot of new products for our customers, and we just keep building as quickly as possible because there's a lot of problems in the industries we serve.

What's the secret to onboarding new customers? I mean, both of the categories you described feel fragmented to me. I don't know how many— I don't know how power law driven your client base is, but I would imagine that there's a lot of customers in housing that are working on leasing, resident communications, the products that you've built. So how do you actually go and reach them, onboard them? What's the flow like these days?

Yeah, we work with most of the largest housing providers in the country. But it is a long-tail industry. So there are tons of smaller customers that all need our software. And I mean, it's a pretty hands-on process. AI is a big transformation of their businesses. And AI, as you know, can be kind of scary for some people. But once you show— you have to spend a lot of time showing people the benefit of it. And then it's sort of like they can't go back to doing anything the way they were doing it before. So it's a bit of a process, but we try to make it as easy as possible.

Yeah, what does customer education, actual change management look like in an organization? Every company has like a couple early adopters that are probably demoing all the different tools, but there's a lot of people who are like, look, I do things this way. There's some newfangled tool. Oh, I gotta check that out, but maybe not this week and maybe not next week either. So I imagine that there's a lot of almost marketing and advertising that you need to do even within a client who has already signed to get utilization of the products up. Is that something that you've had to overcome? Has it actually never— maybe it's never been a problem, but I'm interested to hear about how you solve that.

Yeah, I think AI makes a lot of technology adoption easier, right? The number of users that ChatGPT got compared to traditional SaaS products was— it was just fundamentally different technology to adopt. So a lot of what we do when we're implementing our software is making— and the change management is making it really easy to interact with, mainly through chat. How do you talk to it versus having to learn how to click a bunch of buttons like old systems, really? And today, we serve 20% of the US apartment market already uses some of our products. So they are— it is quite penetrated, and they are pretty familiar with the technology, but we just have more and more products. products coming out every day. We work on really, really hard problems with applied AI, and to get those to market takes an incredible kind of engineering feat and research team to do so.

So I imagine that— I'm sure that you compete with companies that are much more generalist. They have customers in housing and healthcare and, like, you know, let's call it 10 other industries. And then I also imagine you compete with companies that just do AI for housing or just do AI for healthcare. Talk about the strategy around really focusing on these 2 core industries. I imagine there's some similarities, but should we expect you to add new, entirely new categories, or are the TAMs big enough in both of these key markets that you can just continue to focus?

I think our TAM is huge. I mean, housing and healthcare together are the 2 largest sectors there, about 40% of the US GDP. And it's really about how much of that do you capture within the— for the company and how much value do you add to the customers serving those industries. You know, I think we've always leaned into our strengths, which is building products. And the nice thing is both industries we're solving— in both industries we're solving the exact same problem, and it is actually built on the same infrastructure and the same— and actually building in both industries has taught us a lot and has made our products better for one another.

What are the benefits of being a company founded in 2017?

You know, I think in industries, particularly serving our industries, they're not the fastest adopting of technology. So I think we were actually a few years early, but it actually was the right place at the right time because we had to sort of warm up the market. When we started, people didn't really know what AI stood for. And we were really— we are one of the first applied AI companies serving any vertical. So it was kind of a patience game, I guess. But I think if we weren't we hadn't started at that time, we wouldn't be in the running. We wouldn't be at the scale that we're at to have the impact.

Yeah. Should we expect a rebrand to Elise SI, or are you just riding out AI?

That's possible. That's a great idea. Do you want to be on that?

No, I think just drop the AI, just Elise.

Well, maybe.

It's cleaner. Great to catch up. Amazing progress. And yeah, come back when you get, like, 50% market share or penetration into these. I think that's a good next— Talk to you next month. Yeah, yeah. Talk to you next month.

There we go.

Awesome.

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

Cheers.

Goodbye. Let me tell you about Console. Console builds AI agents that automate 70% IT, HR, and finance support, giving employees instant resolution for access, requests and password resets. Our next guest is the CEO, interim president and CEO of MongoDB. We have Dev coming on the show for the first time, but he's been CEO before, and what a journey he's been on. Dev, welcome to the show.

Welcome.

Thank you so much for joining us. How are you doing?

It's great to be here, guys. Thanks for having me.

Thanks so much.

Memorable week in your career?

Yeah. Never a dull day in the AI talent wars, I'm sure.

It is interesting, to say the least. Clearly, we didn't anticipate this, but the company is on strong footing. You know, I feel really good about the business. We had Investor Day yesterday where we talked about our long-term outlook. Yeah, I think it was really well received. And, you know, we talked about how the business is accelerating. We talked about how Atlas, which is the core kind of product area that investors really key on because it's been the biggest growth driver, we raised our guidance. We also talked about the fact that we're raising our guidance or frankly feeling good about our operating margin, our cash flow. And we also announced we're buying back $1 billion worth of stock. So—

Wow. There you go. Fantastic.

I mean, over a period of time, not all at once. The point is, is that we feel really good about the business. And I think people, you know, the challenge I think is many times CEOs get a lot of credit and sometimes too much blame. But in this case, there's a really strong team around the company and the team has done a really good job. And I'm happy to, you know, step in. And to be clear, it's an interim role. We'll be focused on finding a successor, but obviously we're going to take our time and do it properly and thoroughly. But it's a very attractive role. So my phone's already been ringing and we already have a retained search going on.

There we go. That's great. There's some debate earlier online about like executive poaching in an entirely different category. And someone was trying to make the point that, that this cycle is like different in some way when you look back throughout your career. Is the intensity around just recruiting and poaching and talent wars notably more intense than, let's say, the cloud buildout and mobile and sort of the dot-com era? Or is it just more of the same? Business has always been war.

I would say people have always been very aggressive, but I think clearly with especially in the AI era, the money being spent on talent has obviously been breathtaking, to say the least. This one frankly surprised me because it's not so much about expertise, but I think Meta really didn't have a lot of DNA around enterprise expertise. I mean, they've never really sold to complex, you know, global organizations like Goldman Sachs or JPMorgan. And so To try and organically build that would probably be challenging, and so I think they just felt they had a deficit on a certain area. And obviously, Zuck has not been shy about throwing money around, and and he did that again here. So, you know, CJ made a decision, and I'll let you pass judgment. But what I'll say is like I got a note from a good friend who says at the end of the day, no one's going to remember how much money you make; they're going to remember what kind of leader you were. And what kind of person you are. And I'll just leave it at that.

Yeah. Can you refresh everyone on your journey, both in your career broadly, but also with MongoDB specifically? Because this is not a wild card that you're in this seat. Like, you are a very logical selection for this interim role. But if you could give us some background on your journey in tech and then your relationship to MongoDB, I think it'd be really helpful to understand.

Yeah, sure. So, I was a 2-time founder. A company that I started in 2001 was a company called BladeLogic, and it ended up going public in 2007, then acquired by BMC, and I became the president of BMC Software for a couple of years. I segued into being a VC. I was at Greylock and a small firm out of Boston, actually led the B round of Datadog I'm still the lead director of Datadog. I invested in Datadog when Datadog was doing only $1 million in revenue.

Wow.

So sometimes you better be lucky than be good. Obviously, Ali and the team there have done a spectacular job. And at the same time, being a VC, I was doing a lot of work on next-generation databases. In fact, I looked at some competing investments to MongoDB. But when I did my diligence, it was clear MongoDB, even at that time, was way ahead of everyone else. I would say way ahead, but ahead of everyone else in terms of developer mindshare and even financial momentum. But ironically, about, you know, 6 months later, I got a call from, you know, a search firm saying, hey, MongoDB is looking for a CEO. Would you consider it? And they in fact strongly said you should definitely take a look at this job, Dave. It's ready-made for you. And being— you know, usually you get calls. And what's interesting is when you get a call for a CEO job, I've trained myself to ask what's wrong, because no one calls you when things are going spectacularly well. No one says, you know what, let's just change things up and make a change of the CEO role. So, and the challenges then are the changes or the problems you have to fix. Are they fixable or are they really essentially so fundamental to the company that it's just gonna be impossible to change? When I looked at MongoDB, what was interesting is that I was Not that impressed with how the organization was run. The engineering team and product teams were good, but the go-to-market organization was a bit dysfunctional. The culture of the leadership team was dysfunctional. And so, and then there was really 3 knocks against MongoDB. One, open source at that time, no one had really made money on open source. Red Hat was the only company that had created any real value. Databases, a lot of companies had died on the vine of being the next Oracle. And the landscape was littered with dead database companies. And deep tech out of New York was not something that was proven because at the time it was more of an ad tech and kind of consumer space for tech companies. And so, when I went through that, one, I thought Open Source 2.0 was gonna be much— it was really better technology than Open Source 1.0.

Right.

So, that addressed that question. 2, databases, I felt like I could see the developer momentum with MongoDB and I said, the developers really love MongoDB. You don't bet against a product that people love. And with New York, you saw Google and Amazon and Meta really investing in the New York area. And we recognized that we didn't have to only hire in New York, but there was definitely a lot of talent. Whereas like 10 years earlier, when I tried to start BladeLogic, actually 15 years earlier, it was much harder to do a startup at that time in New York. I actually ended up moving BladeLogic Yeah.

Can you talk about your time as CEO of BladeLogic? I have this—

Also, I want to note the timing there. You start the company post-March correction or in Q1?

Oh, in terms of— yeah, yeah. It was 2001. So it was actually 5 days before 9/11. my first round of financing. And at that time, it was pretty crazy to start a company because one, no customers want to deal with dot-coms because they'd all blown up. Employees didn't want to work at dot-coms because they'd all blown up. There are a lot of worthless stock certificates. And raising capital was super expensive, right? And so, but we had conviction that because the previous company I started ended up being a first-generation cloud computing company, actually, I ended up competing with Marc and Ben. They had started a company called LoudCloud.

Right, yeah.

And then when I started BladeLogic, they pivoted LoudCloud to Opsware, and then I started competing with them again.

I couldn't get rid of these guys.

And if you read Ben's book, he talks about going head-to-head with a pretty formidable competitor called BladeLogic, right? So it's funny how this world is so small. But yeah, it was a pretty tricky time to start a business. And the thing that you learn Which I worry about with a lot of the founders today, is that you— we had to build the business the hard way. There was no tailwinds at our back. It was a tough economic environment. Large companies— our first customers, somehow we luckily closed Sprint and Priceline. But it was a slog to convince, you know, these large organizations to take a bet on this new fledgling company when they'd seen so many other startups quickly die.

Well, and now you look at— look at how many different unicorns provide infrastructure for other companies to be able to scale really quickly and And things like WorkOS, right? I imagine there was nothing close to that at the time that would allow you to go—

No. TurboPuffer, browser-based.

We had to do everything. We had to build it.

So, you're building like 40 startups at once just to build one company.

Exactly. And I only raised $29 million in total and had $7 million in the bank when we filed our S-1. So, essentially, I funded the company through customers.

That's amazing.

$7 million in the bank when you file to Aswath is just actually incredible.

It's like we burned a trillion dollars. It's the good old days, right?

How have you— open source AI or Open Weights AI—

You mean open source SI?

Yeah, yeah. We're calling it SI as of yesterday, apparently. The sort of economic model could potentially feel much more straightforward with with some of the Chinese AI labs just saying, hey, we're gonna make this model, but if you wanna host it and serve it, you have to pay us back some type of rev share.

Yeah.

Is that the— like, given your experience, you know, with Mongo, is that going to be— do you expect that to be like the primary economic engine for OpenWeights, or do you think it's something that looks more like Red Hat or MongoDB, where you're building, you're making an open model and then you're building more infrastructure and products around it?

Yeah. So, the classic open source model where you get some part of the core free and then you pay for, like, say, the adjacent features around it, I think is somewhat challenging because I'll give you our example at Mongo. We had an open source product and we provided— so anything that developers needed was free, but any management tooling and so on and so forth was paid for. The challenge was like, you know, defining that paywall between what's free and paid is always difficult because if you give away too much of the product, no, you know, you can't monetize anything. And if you, if you give away too little, it's very hard to drive adoption because people need time, soak time to kind of really use the product. Frankly, the breakthrough for us— so we found that, you know, that paywall, but we were growing the business more slowly. The breakthrough was, was really open source as a service, because when you offer open source as a service, you charge for any— every little bit of usage. And so we could have customers paying us $10 a month and you could have customers paying us $1 million a month. And frankly, we had that, right? But the fact that you condition them to pay for something just made the economic— it wasn't like a debate of what's free and what's paid. Now, customers still have the choice to download and run it on their own. And then we have our self-managed proprietary product as well. So I think to use that analogy, I think if it's offered as a service, I think that's much easier for people to handle. Now, I recognize with, with these models, a lot of people are concerned about IP rights and they want to run it on their own on-prem. So there, you know, rev-share model is tricky because, you know, how much, how much do you share? And that becomes just a judgment call. And different customers will have different opinions about how much they think they need to share with a third party.

Yeah, if you're running it on-prem, and then what? Do you have to open it up to like some— do you have to basically be, as an open source lab, do you have to be like auditing all the users of your model? Yes, and no customer's going to let you do that.

There's no way that's going to happen. So then it's a trust model, which which becomes also a little challenging.

Totally.

I have a question sort of going back to the emotional roller coaster that CEOs go on, particularly founding CEOs. I've noticed at the early stage, there's the founder CEO is like, oh, I gotta get my board aligned. I need super voting. I don't wanna be thrown out. And then after a while, it can sort of flip around. And I've talked to some founders founders in the CEO role. And I'm like, I think the board needs to worry about you leaving. And that psychology of having enough control over the company as the CEO, the tension— I mean, we're seeing it with companies that are about to go public, how to set up the right corporate governance for the modern era. How did you confront that in partnership? There's various firms that are known for, oh yeah, if you're not performing, they will throw you out. There's others that will say, yeah, no matter what, we always back the founder. How have you grappled with that on both sides of the table throughout your career?

Yeah, it's a great question. Frankly, I'm a little old-fashioned. I have a very simple rule. If the company's performing, management's in charge. If the company's not performing, the board's in charge. It becomes very simple that Right. And yes, you know, you've seen some public companies where the founders have still total control, but the stock's in the tank because people just say there's no governance here. And, you know, why am I investing?

Yeah, it's a discount rate. Like, I think Meta has been going through this where, like, you have to apply some discount on Meta right now because you don't know, even like in my psychology on it, is like, let's say Zuck gets AI working like really well, frontier models, leading consumer applications and enterprise business. To me, it's like as soon as he gets the next thing working, he's going to bet another $500 billion on the next thing.

Yeah.

And you're not going to know if that's going to work. And so to me, I've always been— I just always mentally—

Eric Sufer calls it tilting at windmills.

Yeah, yeah, yeah.

But, but to me, I'm like, okay, if this, if this was run in a more kind of like stable way where there was some shareholder influence, it would probably trade higher. But who knows?

Yeah, exactly. And then I think the other— on the other side of the coin, I have to remind founders that no investor wants to go in and replace the founder. There's no incentive for them to do so. So only if things are really going poorly will they ever contemplate such an action. Like, they will do everything in their power to make sure the company is successful. And it's really up to the founder to live up to their obligation to kind of do the best they can. And there are many founders who decide, you know what, I'm not the best person to be CEO. I've got to bring in someone else. Or sometimes they say, I'll be CEO, but I need a strong number 2 to run a certain part of the business that I just am not good at or I have no passion about. And to me, those are the founders that are more mature. But someone who's just always paranoid, I mean, if you're worried about the people who are giving you money, they're going to, you know, basically take you out, then, then you got the wrong partners. Yeah. And I've been fortunate to have some great investors. I'm also only a couple of weeks ago joined Sequoia Capital. Yeah. And, you know, they have a great track record of thinking long term. And, you know, I know you just had Rudolph earlier, but he was, he was on my board for 11 years, you know, while I was CEO, still on the board. And actually, I was just with him yesterday in New York. And there was times when he would push me. There was times he would challenge me. There were times when he— but then when I needed him, he would always be there to support me. And, and I think these, these, you know, these founders who get so worried about control and like someone can take me out, I think you have to look at yourself in the mirror and say, like, you know, why? If that were to happen, don't you feel like you have some obligation to perform? Otherwise, if you don't have any obligation to perform, then something's broken.

It's the same advice you'd give to talent at a company. Let's say, basically become— I've seen people join a company and become so important to the company that at At some point, you can just tell 50% of the enterprise value is tied to this one person who's not actually a founder and maybe not even on the management team, right? And then you end up having enough leverage that you can go and get whatever you want economically, but then also rise up to a pretty— and actually have more control over the company.

Last question for me. Can you give me some color on how you think about the various buckets of skills that you're looking for in a really talented CEO who's going to go the distance? Is it the Rolodex, strategy, ability to engage with the financial markets?

Understanding of the customer.

Golf handicap? What's important?

To me, it's the fundamentals. I'm actually not a big believer in Rolodex. Because people come and go. To me, it's the fundamentals of knowing how to build a business. One, you know, being, being very clear about what your strategy is. To like— a lot of people ask me, what is the job of a CEO? To me, the job of CEO comes down to 3 things: being very, very clear in your strategy. Here's where we're going. So by definition, here's what we're not going. So that makes decisions easy. You know what to prioritize because the strategy is so clear. You know, it makes decision-making easy. The second thing is, you know, assembling the right team around you. Right. So who's on the bus? And sometimes as you grow, sometimes some people have to get off the bus because they're not scaling. And then you bring new people on the bus, but you have the right people who are really complementing you to build this great business. And obviously, if you have the right leadership team, they're going to build out their teams respectively. And the third thing is, is really create a culture and remove all the obstacles that prevent the team from being successful. And, you know, and, you know, you can create a mercenary culture which has its own second-order effects, or you can create a culture where people are really connected to the mission. They're really connected to what they're doing. You really develop. And we think of culture as like another product in terms of how do you attract and retain employees. Right. And so if people feel like it's a great culture, they feel like they're learning, they're growing, they're developing, and there's a meritocracy. They perform, they get more responsibility. You know, that's ultimately what a CEO's job is, right? And, and I can do, I can do 1,000 things like I can do this interview, I can go meet with customers. But if I don't do those 3 things well, nothing else matters.

So scratch offers more of like a nice to have. That's what I'm hearing.

How should I know? I know we're over time and you probably have more important things to do. I do have one more question. How do you think a CEO should balance sort of capitalist versus communist approach? Capitalist being bottom-up, entrepreneurial, letting people on your team come up with ideas and run things down versus central planning, top-down. This is our approach.

Yeah, I think it's a mix of both, but I think You know, when you seek feedback, you have to seek feedback from people you trust and who have a certain point of view, right? When someone disconfirms my beliefs, you know, when I was younger, I'd kind of completely ignore them, saying they're basically bozos. But it caught me flat-footed. And I've tried to train myself. When someone I trust is giving me feedback that just disconfirms my beliefs, the first question I ask is, what do they see that I don't see? What do they hear that I don't hear? And what is the persuasiveness of their logic? And in a market that's moving so quickly, and you have a set of core beliefs, sometimes those beliefs could be challenged and changed. And so, to that end, you have to start rethinking your strategy. And if you don't have people who challenge you, then like Steve Jobs famously didn't like board members who agreed with him because he wanted to be challenged.

Yeah.

And I think too many people want to hear the kumbaya, oh, Dave, you're so wonderful. You're great and all that. But that's not really that helpful to me. What's helpful to me is, hey, have you thought about what's coming around the corner? Have you thought about like, are you really— is the scale of your ambition as aggressive as it should be? You know, I thought about like what potential new technologies could disrupt your business. I mean, that's the way you, you know, you should operate and you should have a team who thinks that way. The lesson I've learned a lot is the ability to tell a really strong team is The ability to have hard conversations. There's so many people who are very passive-aggressive. You know, they'll nod politely and agree, and then they'll go back to the desk and roll their eyes. But if you create a culture where people can feel like, hey, it's not the— not my idea, it's the senior person's idea that wins, it's the best idea that wins, and people feel comfortable challenging you, that creates a great environment where people feel like, you know, we're really focused on what's right for the business.

Well, I'm gonna have to disagree with you there. I'm kidding. No, great, great point. And I wish I wish we had more time.

Yeah, this is fantastic. Thanks so much.

Good luck. I don't have a lot of time, but when you when you figure out the right yeah person for the job, I'd love to have them on the show. And it's great to catch up with you. Thank you.

Partnering with Mongo. Have a great rest of your day. Goodbye. Let me actually tell you about MongoDB. What's the only thing faster for the AI market? Your business. on MongoDB. Don't just build AI, own the data platform that powers it. We have Darr Murphy, the CEO and co-founder of Imprint, in the waiting room. Let's bring him in. How are you doing?

Boom.

Welcome to the show.

Hey guys, how are things?

We're good.

So good.

First time on the show.

I appreciate you guys having me on. Last slot of the day.

Yeah, second to last slot. We have another, we have another guest in person. Um, but anyway, introduce yourself. The company, the news. Give it to us.

Yeah. So at Imprint, we're building the infrastructure that lets everybody feel like a regular at their favorite brand. We think that starts with co-branded financial payments. And so with Kroger, we're launching or relaunching their co-branded credit card. They're leaving a big bank, bank that's been around for 100 years, choosing us. It's obviously a big vote for us because they're the 4th largest merchant in the US. Yeah. And you're probably asking, like, why does somebody like Kroger choose Imprint over a Fortune 100 bank? There's really 3 reasons. One, we're not trying to steal their customer. So many banks are trying to steal their customers, sell them a mortgage or a deposit account, or they got BNPL trying to steal their customer and sell them to somebody else. 2, we took this $100 million bet to rebuild our tech platform for ourselves. All the banks, every credit card in your wallet, they rely on legacy infrastructure, mainframes in the Midwest. We own it so we build better product. And a lot of times now with these merchants, our partners, you know, when we first started talking to them 4 or 5 years ago, they were worried about getting disintermediated by Amazon or Instacart. Today, the conversations we have is how do we help you stay relevant when Muse is on the horizon, Instinct is coming. And our perspective, perspective of the merchants is top of wallet, you're top of mind. And so if we give you a better credit card, more rewards, fairer And you end up being a big part of the consumer's life.

Talk to me about how I should think about your business. You mentioned they would go with a— the alternative is going with a bank. Are you registered as a bank? Do you have a charter or are you more of like a fintech software layer? Like, what are the economic breakdowns? We were— I want to go further into what co-branded card economics look like, but give me the 101 on the category.

Yeah. So, we offer the brand the exact same thing that a bank would. In fact, Shell recently left Citi for us. Craig Kbara left a big bank for us last year. Our proposition is you get the same as you would get from the bank, plus a much better technology product, et cetera.

Okay.

The way it works is we're not a bank. We work with partner banks. Affirm, Klarna, everybody else does the same thing.

Sure.

And the way the economics work is we help the buy— or the brand fund some of the rewards. Richer rewards matter for customers, right? Everybody loves their rewards credit card. If we work with a brand that owns a big share of your wallet, why wouldn't you want to use the credit card? And then we have a profit share with the bank off the bottom.

Got it. So, I imagine Visa and MasterCard, like, the rails are still taking 1%, 2% of the transaction fee, but there's still maybe 1% that's floating around there that can be given back as a reward or kept. How close am I on the actual numbers? How does it split down?

Pretty far off. Okay. So the rails take like 15 or 16 basis points. Oh, it's really low sides. Yeah. Interesting. And then the name on the card imprint, Chase, whomever, we get interchange of about $1.82. Okay.

And then we give almost all of that back to the brand to give the consumer Okay, so we were debating this in the context of Instinct, the personal AI agent, and we were sort of asking ourselves, like, why hasn't an AI company launched a co-branded credit card with a company like you and then use that as a monetization strategy? Maybe they can give it back in tokens or do some sort of reward, but basically—

Yeah, the big thing is, like, you can't—

Start getting some transaction fees.

If you wanna capture transaction fees, quickly going out and doing partnerships. You can do the big partnerships with like a Shopify. There's a lot of other things that you buy online. There's a lot of like, I think about the number of local service providers I've found with ChatGPT that ChatGPT gets no benefit from, right? And it's great for the service provider.

And there might not be a referral program.

The beauty of a card is that the service provider's used to paying a transaction fee. And it doesn't really feel like they're paying anyone when they do it because payment, you know, transaction fees are normal. So, it feels like it also makes sense in the sense of like linking the payment method to the account, then you're not dealing with like, oh, I'm authorizing this agent to use credentials that are stored somewhere else. Like, it feels like there's some potentially security benefits as well.

Yeah. Where do you see it going?

Yeah. All I can say on that specific point is watch this space.

Okay.

If you think about it more broadly, right, like everybody hates— everybody— is that a good— that's not the sales gun.

No, no, no, no, no, no, no. That's like it's getting real. It's like the boat horn. Amazing.

Action movie.

It's positive.

Yeah, yeah, yeah. Amazing. Look, I think if you think about the space generally, like everybody hates interchange. Everybody hates paying Visa and MasterCard. I think there will be credit cards, but I think the really interesting question is like, what's one step beyond that? And so, a lot of what we've been investing in is like letting merchants, letting brands accept your bank account as a payment.

Sure.

Because that's like basis points on the dollar.

Yeah.

In a world of agents, you can optimize that. The hard thing today is it's like a shitty— it's a bad— sorry, I'm Irish, it comes out sometimes. It's a bad user experience to have to link your bank account.

Yeah.

If you can be like Muse or Instinct, here's my bank account details.

Yeah.

I trust you to link them up correctly.

When it's—

That thing gets rid of the that friction, right? Like crypto was a fugazi here. Like this is actually the first time I think interchange is in danger. And it's 2% of American commerce goes across these rails. So it's a huge amount of value.

Yeah. No, I mean, Walmart was pushing for this for years with the, with the credit card push, trying to get to ACH with, with various customers. And it was a really long slog. The Apple Pay ecosystem sort of won a lot of things. But yeah, you can see that the game is once again afoot with the agentic era.

And Stripe is pushing this with Muse, right? Like, you have the Stripe Link experience. And you use Uber today. The last time I used it, it prompted me to use Stripe Link instead of using a credit card, right? And so it's actually very bearish, I think, for those like MasterCard, Amex— although Amex is a different business. But longer term, I think it's going to be great for merchants and consumers.

Yeah, yeah, that makes a lot of sense. Jordi, anything else?

Great to finally have you on. I remember when you launched— or maybe not launched, but I remember back in 2021, I was building in fintech too, and maybe it was a round that I saw, but when I saw the business launch, it made a lot of sense. So it's great to hear all the progress, and it's great to have an Irish accent on the show. You don't have enough of them.

Yeah, sorry for the swearing.

No, no, you're good. It's all good. It's all good.

You're good. Come back on or flag when you have some of these, uh, some news in, in, uh, personal agent world. Fantastic.

Will do.

Thanks, lads.

Have a good rest of your day. Talk to you soon. Let me tell you about Figma agents. Meet the Canvas. Your AI agents can now create and modify your Figma files with design system context. Our last guest of the show, Noah Friedman, is here in person in the TBP and UltraDome. Welcome to the show, Noah. How you doing?

I'm good, boys. Good to be here. The UltraDome.

What news do you have for us? Should he hit the gong?

Do it. Yeah. Okay.

We are announcing Outer Signal's $22 million Series A today.

Do I just go? Congratulations. Yeah. Smash it. It is. Smash it. Fantastic. So introduce Outer Signal. Introduce yourself for everyone who doesn't know. And then I'm sure there's so many questions we can go through.

Yeah. There. Thanks for having me, guys. I appreciate it. Outer Signal. OuterSignal's a customer intelligence and agentic personalization platform. So when any consumer checks out or transacts with the business, most businesses really only know about that person that their name is Jordi, they live in LA, their email. OuterSignal paints the rest of that picture, right? So we give full-time, in real-time name, email address, um, date of birth, age, gender, occupation, property value, a full profile behind every single consumer, which lets businesses actually speak to their consumers like humans.

Yeah.

Not just orders.

Yeah. And what is the best way to actually take advantage of that as an e-commerce business? I mean, I can imagine, OK, some celebrity shows up. Maybe you reach out to their agency to do a deal, all the way down to personalizing an email, which is its own fraught thing, because sometimes you get an email, and you're like, that's a little too personal. I don't want that.

Yeah, I'll start with an example.

Yeah, please.

I know Rora is a customer of Outer Signal. Love Raft. Yeah. And I remember within the first 3 months of launch, Brian or Charlie messaged me and said something like, I'm not gonna name the person, but it was like a top 5 actor in the world.

Yeah.

And I was like, how do you even know that? And they were like, well, we're using this thing called Outer Signal.

Yeah.

And that person wasn't even using their actual, like they were using a pseudonym.

Yeah.

For the, the, the order. Yeah. So, it was like—

But it linked it all up.

But it showed, and so, yeah, it's like, I think people— it's not just these sort of extra, you know, elements and kind of like populating data that's already out there. There's like other sauce that you guys have.

Yeah, there's certainly other sauce. I mean, it's doing what we do is both a technology challenge and an anthropology challenge, like, and the example that I give is, you know how many people there are named Noah Friedman in the world?

Oh, yeah.

So, like, if Noah Friedman orders from or disambiguating which one is me is remarkably difficult. And that's what our engineers have spent hundreds and hundreds or thousands of hours mastering. And it's a really hard problem, but we've cracked the code on it. I think to your question, John, it's actually a lot bigger than just— like, these examples of finding celebrities are very cool and fun.

It's a huge business built around that. It's more like cool for the—

For the personalized market. Yeah, exactly. It's cool for— but the example I give is this. Like, literally when I was in the green room room there, I got an email from a brand that I've probably spent thousands of dollars with telling me that their new drop had just hit and it was like sports bras. And it's like, I've spent thousands of dollars with this business. They have every right to know who I am. I'm a very— like, you can Google me and figure out who I am. And they just didn't take the time to do that because it's impossible, right? And so the downstream impacts of what Outer Signal is all about is really to say, like, behind every order, behind every consumer who transacts with the business, there is a human being. And those human Human beings deserve to be treated like human beings. And so across email marketing, right, where like a brand that hits me with an email about performance in the gym or when I'm up late working on my business, like that's gonna slap, that's gonna hit a lot better than if they hit my mother with that, right?

Mm-hmm.

And so across email and retention, across paid media and acquisition, all of this is downstream of this concept of like everybody transacting with these consumer businesses, they're real people. And these companies just have no basis or bearing to say the second an order comes in, we're basically just routing them into the arbitrary Klaviyo flow or what have you just to hope that it lands with them when in reality, there's so much more context that can be grabbed in the real world.

How do you think better intelligence in the e-commerce world flows to retail strategy? There's always this pitch of like, okay, you look at the shipping locations and you see that everyone's within, you know, 2 miles of this store, we should go pitch that store. But it feels like it can go a lot further. How are you seeing— because I, I mean, I feel like every e-commerce business now, like, the D2C era is fully over. Like, you're omnichannel at least day one in the deck, if not day one in the— Day one in the deck. I like that.

Day one in the deck.

That's a banger. That's a bar. But like, you need to be thinking about this, and you can't be turning down down opportunities for retail expansion and saying we're a pure-play D2C company that cuts out the middleman if you want to scale in the modern era?

Look, I think there's a couple angles to it. Number one is saying if you're trying to activate and pull deeper velocity at any of these retailers, which is ultimately— I mean, we talk about this with Lucy all the time, right? This is the hero metric, is how fast are you pulling through these stores? Understanding how to locally activate an audience, which Outer Signal can allow you to do in ways that Meta just can't. Like, look, around these zip codes in Metro LA or in the middle of the country, it doesn't matter. We can actually drive hyper-targeted funnels and flows into store because we know who these people are and we know how they're different in these different markets. That's huge. We've also had probably hundreds of brands at this point both find retailers that were buying from them with their personal address. Yeah, I've seen that before on e-com. Like, a huge creatine brand was able to get like a really big placement because they found out that the senior buyer from Walmart, like their Square ICP had been buying personally as just a fan under his email.

And of course, they're going to take the email. They're going to take the call.

Of course. And so you've got that angle. And you've also got like when you're pitching retailers, right? Like you guys know this. One of the most important things you can do is have a really fundamental understanding of like who the hell you're actually selling to, right? And I think so many consumer businesses have this archetype or the persona of who they think they sell to, and it all feels very much the same. Like, I cannot tell you how many brands we meet and onboard into Outer Signal where in the sales process we're like, hey, who do you guys think your customer is? And they're like, young urbanite professionals in New York.

Yeah.

Well, you can't all sell to the same person, right? Yeah. And it turns out when you activate, it's like, no, it's actually Midwest moms with high disposable income. They're the ones paying your bills. This persona of like the young professional urbanite in New York is a super small percentage of your revenue, who's actually a less valuable customer, but they gravitate one specific product. And so across merchandising, to understand which products resonate most with this demographic or this demographic, and across actual placement in store and pitching retailers, there's so much depth and nuance that comes from understanding who the eff these people are that are actually buying from you all day.

Yeah. How are you— right now, explosion of personal agents. You have Instinct and Muse and And even in the last 24 hours, it seems like Instinct started doing basically texting product recommendations, which exposed something that was kind of interesting to me, which is that the agent knows about the user. I saw one where the user— it was pushing various car products for a Targa.

Hmm.

And the user was like, how do you know I have a Targa? And it's like, Oh, we worked on getting insurance for a car a while ago. But then the agent was actually recommending products, like just products that a Targa would never need. He was pushing a trickle charger and an indoor cover, which like a real— I don't know any real car guys that are like, I got to keep my 2026 Targa— It's the best thing.

—off a trickle charger. It's not super intelligent.

And no one I know is like using indoor covers because real ball knowers would know that like if you drive your car outside and then you put a cover over it and you put— and there's like debris between the paint and the COVID it will actually wear the paint down faster and do more damage to the car. So like that basically it was like a personalized recommendation from somebody that shouldn't be making a recommendation, right? Yeah, like someone that did not have ball knowledge. But—

Lack of ball knowledge.

Still, still, to be clear, like, it's almost like on the brand side, the brand needs to— like, brands need to be thinking about educating agents because the agent knows the customer, but the agent needs to know who to pitch it to. Yeah, sure. Like, somebody with a Ferrari or a 10-year-old Ferrari Should get recommended, like, hey, maybe if you're not using a trickle charger, your battery is going to wear down faster or whatever. But it feels like an entirely new task for the company to educate agents.

Couple thoughts. Couple bars in there, by the way. Ball network there. I think we just have a thing now, like, you can't make personalized recommendations unless you know Ball, right?

Sure, yeah. The bar has moved.

The bar has moved.

The goalposts have moved.

A couple of things. Number one, I think that example actually touches on something critical, which was that that personalization was really just behavioral, right? You bought this, you might like that, right? It has no context on, like, who the hell you actually are as a person. And that's really been the failed promise of personalization for the last decade, which is, like, a lot of the companies out there selling personalized solutions for marketers, it's all based on Jordi clicked on this ad, or he came in through this landing page, and he bought this product. And so we think and we guess, and we're just going to make some hypothesis finger to the wind that the next product he might like is why. It's all pulling from the same signals that it's been pulling for from the last 10 years. And like, it just doesn't work. And the outcome of that, whether it's an agent recommending it or the brand on email or just the landing page, is that it feels highly impersonal, right? And I would argue that the next layer, the next frontier, and the one that we're ushering in is to say, like, who the hell is Jordi, right? Like, did he just post on his Instagram Instagram last week that he got a new Ferrari, that's really important context for me to know, not just that he actually has a car and checked out insurance last month. And so like, let's upsell him and try to upselect him into some other product which we can draw some fuzzy connection to, right? So I think agents are going to— like, agentic commerce as a concept is in its very early days, right? Like, I think all these companies are still trying to figure out what does it actually look like for Muse or Dot or Dots or whatever the hell flavor of the month is, right, to recommend a product to you in chat and in feature. Very early days of that, and we're keeping a close eye on how we can interface with that and play with it. But I think at the end of the day, it all comes down to, like, whatever touchpoint a business has with their consumer is the most valuable real estate there ever, right?

Yeah, and the main thing for customers is, like, getting pushed a product that makes no sense for you—

With a message that makes no sense.

Well, getting pushed a product that makes no sense for you just completely— like, I've had this with—

You don't need a full charger battery?

No, no. I just had this at like car— like sometimes you're— if you're at a car dealership, like the associate will try to sell you one of their—

Strictly a car system.

No, one of their cars in a way where like you realize, okay, I actually know more about like the pros and cons of this vehicle than you do. And I explained my situation to you and you're still kind of pushing this thing. Yeah. And then it just destroys like the trust.

Why are you pushing understated rims? It clearly needs spinners.

But when you get truly like the perfect Instagram ad. Yes. Or you get truly like the perfect, like, you know, a sales associate texts you, I got this, do you want it? Before it hits the floor. Yeah. Like that is a really good feeling. Yeah. That's delight. But like the gap right now is just so surprisingly wide between those actual human-led experiences, which is like sales associate knows your preferences well, has a unique product for you at the right time, and what brands are able to do.

It's basically back to the hallucination era. We left the hallucination era where if you look up the history of a company on any LLM, you're going to get something that's pretty much like the Wikipedia page, pretty accurate. But 4 years ago, you ask an LLM anything and you could just wind up in crazy made-up territory. And it's the same thing for agent-to-commerce.

We think of, and this is a core value and mission of the business, we think of the ultimate protagonist and end customer of what we do as the actual consumer, right? We are serving brands in the sense that they need to be delivering We're helping brands help you. We're helping brands help you, right? At the end of the day—

We're helping brands monetize you.

Helping brands, yeah. Look, I think we all believe in—

No, we're very commercial here.

Yeah, we love personalization because I think it unlocks delightful experiences, as cliché as that sounds, in ways that just ultimately is a net positive for the business and the customer.

What do you think the role of outer signal data market research, customer personas, profiles is in actual product development, product line expansion. There's some product line expansions and just new products that I'm like, okay, that didn't come out of a committee.

Totally.

At the same time, sometimes you look at a business and it's like it's been staring you in the face for years and everyone was going to buy the non-caffeinated version of Diet Coke or whatever. And so how do you think about the role that outer signal data should play in the boardroom or the product development room as opposed to just the marketing engine?

Yeah, it's a good question. Look, we see a ton of our businesses use outer signal in their product merchandising and product development and R&D actual processes, right? And so a couple of concrete examples here are massive, massive multibillion and $100 million businesses have discovered that they have some, like, really core persona. For example, one massive beauty brand found out that, like, 10% of their customers were, like, very distinctly nurses and doctors and, like, practitioners of medicine. And they weren't speaking to them at all, right? And it turns out that they were buying one very specific SKU in their set dramatically more than the rest of the customers, right? Which informs not only how you market to that group, also how you segment them, and also expanding more for them, right? The second part of this This is like, because Outer Signal's real-time, right, you can now see when you launch a product. Like, if Mountain Valley is going to launch some new SKU online, in real time, they can see this SKU is over-indexing with women, and it's over-indexing with men, or women in $3 million homes, it's dramatically under-indexing in this cohort, which starts when—

Yeah, high ABV Mountain Valley really fly off the shelves.

That's a top-shelf plug right there.

Yeah, very good.

Hell yeah.

I like it.

Yeah, yeah, yeah, yeah. Give everyone an overview of Top Shelf.

Top Shelf is a fund that I started, um, in 2022. That's how you and I met on the board of Lucy, obviously. We started investing in vice businesses, so booze, nicotine, etc., and we've expanded into just the broader consumer market. So businesses that connect with their customers and technologies that help them do that. Outer Signal— Top Shelf's a very proud investor in Outer Signal.

Yeah, what is the biggest trend in in the alcohol space these days? I feel like we're post-seltzer boom, but is there a next, next thing that's the hot trend? I haven't really kept up with it.

Booze has had an interesting few years.

Oh, yeah, but no one's drinking anymore, right?

Total nonsense. Turns out they are.

Oh, OK. Who's drinking?

Apparently, no one in LA, although the other night I was out, I still saw people drinking, right?

Look, I think— no, part of the thing is like LA has fallen off so hard as a city that people assume— No one's drinking anymore because you drive around on a Thursday night and the places where nightlife used to be a thing, it's empty. And it's like a health mecca. So, people are like, okay, so no one's drinking. It turns out all over the country, other places, it's business as usual.

But yeah, I mean, what are the growth categories? Wine, beer, spirits, seltzer, something new? Where's the energy in the industry?

I mean, we were the first check-in to a premium box wine called Grazi, which has become one of the largest wine businesses in America. And they are basically premiumizing what used to be Black Box, which is like the thing you drank in college and slapped the back of your bra.

Yeah. And that sort of happened with like Cutwater, where they took cocktail— mixed cocktails and put that in a more accessible form factor at a lower price. And then that started flying. Interesting.

I think occasions are changing modestly. I think the notion that young people don't drink anymore is ridiculous. I think that they were just delayed because of COVID Oh, sure.

Yeah.

That was the big trend that just completely messed up the headlines.

But you returned to normal.

Returned to normal, they were just delayed. So there was stagnancy because they weren't going out. I still think the loneliness epidemic is real, and there's another argument as to why booze is actually a cure for that, but we don't have to go there in the UltraDome today.

Maybe after we wrap.

Booze is gonna, yeah, maybe after we wrap.

Cure for all of society is alcohol, according to Alcohol Investor.

Yeah, Grazie was one of the first Outer Signal customers as well, and they've seen some crazy results.

You guys have a very interesting lens on, like, for actually investing in consumer, just given your role as a data provider.

Yeah. What's the growth strategy for Outer Signal? Is there a distribution mechanism where, like, you can vend in through, like, a Shopify plugin and be discovered there. I imagine you're doing paid ads on your own, maybe sponsoring podcasts that target things, going to trade shows. But what, what's working the best for you in terms of getting in front of e-commerce entrepreneurs?

It's been remarkable word of mouth driven by strong product-market fit. Sure. People who—

Yeah, Brian said that he's like sent a bunch of, bunch of different brands your way.

And that happens constantly. We are running ads. We have a Shopify The ads are effective. We have an agency partner program. We have a killer sales team. We have content that goes out. And all of this has been— the company is only a year old, right?

Oh, so agency would be like a performance marketer that wants to bring in more—

Attention marketer, performance marketer, influencer marketing teams, CX agencies, right? Outer Signal is a magical, magical experience for agencies.

You can also sell it to the CEO because they want to look at the pretty dashboards, even if it's less tactical.

And so in e-com, you know, that's a very tight market.

Yeah.

I think we've done a pretty good job both making friends in and also doing right by. We're also getting pulled upmarket pretty quickly. Hotels, large-scale consumer apps, massive retailers, they all have the same problem. Like, one of my favorite stories, one of the biggest hotel companies in the world that you guys have all heard of— we won't name names— but they inbounded. I took a call with them, and I said, what caught your attention? And he was like, dude, LeBron James checked into my hotel. And I didn't know about it until he got into the lobby. And I was mortified and I chewed my team out and I went and looked and like, yeah, his name was there, but it was technically his assistant. But if anybody had taken the time to look, we would have caught it. And of course they didn't catch it because who's looking at all that stuff? He was like, we need outer signal.

Of course. Right. Or at least exercise more discretion. Exactly. Feel welcome. Exactly. Fascinating. Very cool. Well, congratulations.

Thank you, boys. Such a, such a cool business. Like it's one of those things like you're, If you can give that magic moment to founders and management teams, which, again, we experienced that really early on at Aurora, but then continue to deliver value and get much deeper into the stack and help them run better email marketing, all these different channels, sky's the limit. So excited to see where you take it. Appreciate you, boys. Thanks for having me.

Well, that's basically our show. There was some breaking news while we were Live. What is it? Gemini 4 Argon launched. Cool new name.

How about your bunch of benchmarks? Yeah.

What are your thoughts on it? No, no, no, no. We're all going to digest this. We'll talk about it tomorrow and go through the benchmarks, but you can go check it out. Logan Kilpatrick, friend of the show, posted about it and everyone can take it for a spin. We'll talk more about the reactions tomorrow. Leave us 5 stars on Apple Podcasts and Spotify. Sign up for our newsletter.

That's right.

And we will see you tomorrow at 11 AM Pacific. Goodbye.