The Social Reckoning Reactions, Fable 5 Sparks Safety Debate, 𝕏 Timeline Reactions | Farza Majeed, Trent Simonian, Sridhar Ramaswamy, Matthew Prince, Vinod Khosla, Ranjan Rajagopalan, Markie Wagner, Bret Taylor

You're watching TBPN.

Today is Wednesday, June 10th, 2026. We are live from the TBPN UltraDome, the temple of technology, the fortress of finance, the capital of capital. Let me tell you about Ramp.com, baby. Time is money. Save both. Easy to use corporate cards, bill pay, accounting, and a whole lot more.

Ramp, ramp.

You go to Ramp. Ramp, ramp. Ramp, ramp. Ramp. You want some accounting help? They're going to say yes and thank you for using us.

And what else?

Two major stories, bunch of major stories, but the big one, Facebook got a movie, another movie, actually a third movie if you count the documentary. There's been The Social Network, The Social Dilemma, and now The Social Reckoning. Only two of those are movies. The Social Dilemma is a documentary, but it's very interesting. And Anthropic got a fable, and there's been a ton of reaction. Ben Thompson was writing about it to the rollout there. The Wall Street Journal Like the announcement should have been like the really good model launches and people are amazed with what it can do. But they put the title Anthropic puts curbs on AI models. So the, well, I'm sure you've seen this on the timeline, but we'll take you through some of the debates over how restricted the model is in certain areas. Bio is a big one. Cyber is another one. There's a bunch of funny examples. There's a bunch of, reasonable arguments for doing this type of stuff. So we'll go through all of it.

But can we talk about who's joining the show today?

Absolutely. We got a bunch of great guests.

What a lineup.

What a lineup. Hey Clicky founder Farzad Majid is joining. You've probably seen his viral demo last week. We're very excited to talk to him about computer use, potential consumer AI applications, what he's doing to build hands-free AI voice assistants. Uh, the CEO of Snowflake's coming on, CEO of Cloudflare's coming on, Vinod Khosla from Khosla Ventures is coming on, and then, uh, Brett Taylor. Uh, Brett Taylor didn't even make the cutoff on today's lineup. What's going on here? Brutal.

And of course we have Marquise Wagner coming out, founder, announcing, emerging from stealth with her new company.

Very excited.

She's fantastic.

Uh, so I want to kick it off with the, uh, the social network sequel. I want to watch the trailer for The Social Reckoning, because Jeremy Strong is playing Mark Zuckerberg. I've heard people say, hey, he looks like Zuck, like the styling is there.

Henry says, no Doug from Semi Analysis.

Oh, we got to get him on.

We should get Doug on.

Let's hit him up.

Hit up Doug.

Let's get him on.

And see if he's available. Not that we have any time, but we'll make time.

We'll make time for Doug, anytime. So the one critique is Jeremy Strong looks a little bit older than Mark Zuckerberg did in 2010. 2011, because he is young.

Calling him "unk"?

It just doesn't— I think it'll color the movie a little bit in the sense that the film is telling the story of this very difficult whistleblower situation. And it's different when it looks like a mature adult versus someone who was young and running into these problems for the first time. Anyway, let's—

That says no Matthew Prince.

What?

He is joining.

Yeah, he's joining. We got the Cloudflare team.

12:30 Pacific.

12:30, get ready. We're going to go through Cloudflare. They have exciting news. I think it's public already, but I don't want to blow them up if it's going to be announced later today. But anyway, I want to react to the trailer. So let's pull up the trailer.

Listen, before I go on, I want to make something clear.

This is how it starts?

Okay.

I have a hunch you're not a fan of Facebook. But I am. I am here to help Facebook, not hurt it, okay?

You sent me a message. What would you like to talk about? The chairman gavels the session to order.

You'll read your opening statement, which we'll skip past for now.

That's a separate session, and we'll move to witness questioning. Spell your name and state your current occupation for the record.

So is it not Fincher this time? Let's see.

And your occupation?

Zuckerberg.

Oh, it's co-produced and directed by Aaron Sorkin.

Where 30% of content hits multiple risk factors.

Hang on, I don't understand what that means. Opinion piece to the social network.

Are you a tech reporter?

Ish.

Ish?

Is this Mike Isaac?

You guys are counting on the next round of congressional testimony to make you likable, Mark.

Jeff Horowitz.

I'm happy to lend a hand, but I think you're doing—

Who wrote Broken Code. He was a journalist at the Wall Street Journal. He is a journalist at the Wall Street Journal. Broke in a number of global news stories, including the Facebook Files.

The fire hose of bad information you are injecting into the air supply is becoming jet-powered.

I'm a free speech absolutist. I'm not the one who's lying, and I'm not stopping them from seeing someone who is. Not a bad Zuck impression. He's got the voice tone pretty down.

As a result of time spent on the platform, senior leadership knows and is doing nothing.

I know there are easier enemies to make.

The mafia would be an easier enemy to make. So what would you need?

To stand up the story, the internal documents.

This is a material violation of my NDA.

We're twice as big as the biggest country on Earth.

We're not frightened of Congress.

We're post-government around here.

Please. Twice as big as the biggest country on Earth?

We have 102 hours together. From a, from a user standpoint.

Oh, okay.

Users.

I don't want to be made an example of by a guy with unlimited resources.

Harm, I promise you, is imminent.

No! People around here understand— that when I say no, that's the end of the debate. I'm not 2 years out of a dorm room anymore, Charlie, look around. Man, if they were trying to go as inspirational as "The Social Network," they really missed the mark there. What happened?

Yeah, so that their intention was to try to fire up the next generation of entrepreneurs that to inspire them.

It seems like they missed the mark, yeah. To inspire them, to uplift them.

That you can build any business out of your dorm room.

Yep. And also it feels like if the goal of this film is to really, really take the viewer inside TBD and Meta Superintelligence and what's going on with Nat Friedman, Daniel Gross, Alex Wang, like it feels like they just dropped the ball potentially. It's sort of crazy.

Yeah, like I didn't see like anyone that even resembled Yann LeCun.

That is crazy to not highlight that and what's going on over there. The launch of Llama, the Llama 4 moment, that could have been the crescendo. Yeah, behemoth. Behemoth and all the drama around that. You want drama, you want Hollywood drama, you want Oscar bait, llama for behemoth. That story is going to get put butts in seats. That's right.

Uh, no, more seriously, it's hard for me to— it's going to be really hard to not look at, uh, Jeremy Strong and, and, uh, Kendall.

Yeah, because you're a Succession guy. Um, so, uh, this is the sequel to The Social Network. Instead of chronicling the birth of Facebook, it's the story based on the 2021 Facebook leak by whistleblower Frances Haugen. It's gonna be dramatic. It's not gonna make people like Facebook more, and it's probably gonna make Americans even more distrustful of tech. What's interesting here is that if you stop a random tech person on the street and ask them like, what is the social reckoning about? They might say Cambridge Analytica, because that was another drama moment. And I was sort of like, is it Cambridge Analytica? Is it the Haugen thing? And a lot of people don't even remember what the whistleblower story was. Little refresher there. Internal document leak, the Facebook Files, showed that Facebook internal employees were aware of harmful societal effects from its platforms, yet persisted in prioritizing profit over addressing these harms. Now, Nikita Beer chimed in, who now works at X, a rival platform to Threads, and said, Zuck makes a lot of mistakes, but this isn't one of them. Meta literally had multiple teams of $1 million per year engineers working on teen mental health, and they had the agency to override big Product decisions. They're probably a thorn in Nikita's side, 'cause he's trying to maximize click-through rates, maximize attention, and he's getting pushback from, hey, there's guardrails here. And he says, the story this movie is about is actually a product manager who didn't get a promotion. And so that's the pushback from people who—

The only thing— What was that? And the only thing with that argument is like, you could make the same argument as like, The cigarette company has million-dollar doctors and researchers focused on making sure cigarettes are as healthy as possible, right?

Yeah, yeah, yeah. And especially with the new fertility data, I think that there's a new cycle brewing of exactly how bad is social media. I still don't really buy the addiction thing just because I'm familiar with addiction defined in nicotine, which is extremely addictive. It's chemical and you have cravings. If you forget your phone, like, do you have the same cravings? It's sort of different. But there is like a very serious discussion going on around social media and its effects. And this fits right into it.

The phone addiction thing is like very real. I was at an event last night, and they required everybody to put their phones in these sort of like locked bags so you could go outside at any point.

And you felt sick.

I didn't feel sick, but I noticed probably like 20 times throughout the night, I was thinking of going for my phone. I didn't start having like physical withdrawal symptoms.

Yeah, but the craving was there?

Yeah, just kept thinking about it. It was like phone noise, right?

Phone noise, yeah. This is real, this is real.

I was having some phone noise.

Yeah, yeah, yeah, yeah. Okay, so yeah, that's legitimate. So the question that I had really quickly is like, what is the impact on Meta? Like we can pull up the stock chart, it's a $1.5 trillion company. They're trying to raise more equity for the AI buildout, hire people. Like, is this good, bad? What does it look like over the short, medium, long term? So short term, I think this was amazing timing. They really got lucky because as we will talk about, like, the fact that this trailer dropped the same time as Anthropic's Fable 5 was incredibly fortuitous because Fable just took over the timeline and no one was really talking about this in tech. Tech insiders won't really see or talk about the social reckoning this week. Case in point, I put it second in the newsletter when I wrote it up. Medium term, I think Jeremy Strong is gonna drag Mark into like the bad group of AI leaders while he's on his Oscar tour. Like there was a version of Facebook strategy, a Meta strategy that's just like, hey, we're like Amazon, we're like Microsoft. Like yeah, we have bets in AI, but we're not like, we're not deciding the frontier. Obviously Mark has made the decision to hire some of the greatest researchers. Invest very aggressively. And this puts him sort of at the center of the conversation around, well, what will the effect of AI be on our society, on kids? Should kids be talking to LLMs? Psychosis. All these different things will come up just as we saw 4.0 psychosis and then Claude Code psychosis a little bit pop up and people are talking about that. Like you're priming the pumps for the next wave of like, oh, a bunch of people went into the Meta app and like they had a bad time and now we got to talk about that. So I wouldn't be surprised if they get to version—

What do you think Mark is gonna be more annoyed about this week? The fact that SpaceX is gonna be valued meaningfully higher than Meta or this trailer? It's actually a tough one.

I don't know. It's actually a tough one.

I mean, I think the real one is Fable 5, because like if Meta was using a ton of Anthropic models and then the new model comes out and says you specifically can't use it, in MSL, the most important initiative in the company. Like, that's like a pretty rough thing where you're like, oh, I've been working with this company to develop my models for my family of apps and now I can't use them anymore.

Also, but if they do, you know, Meta is spending billions of dollars a year with Anthropic. They have been for a while. Oh, ARR, but yeah. Yeah, yeah, yeah. They will spend billions of dollars this year. And the question is like, if Anthropic does allow Meta to use their models for Meta's AI research, what does it say about what Anthropic thinks about Meta's potential in AI?

Secretly love ads this whole time?

No, I don't know. No, to me, it would be them not feeling like they had a—

The AGI.

Ability to get to the front of the game.

Totally, totally, totally. And that's an ace that's up their sleeve at any time. They can always just take the guardrails off and be like, hey, more AI, more intelligence on demand for everyone. So I think the fallout of the medium term, Jeremy Strong's gonna get he's gonna go on this press tour for the Oscar and he's gonna have a bunch of like really emotional pithy sound bites and then also be like just viral because it's funny to see him doing this impression. And he's very good at like drawing, he's already gone viral for his Mark Zuckerberg impression in the past. So I wouldn't be surprised to see the next version of a Bernie Sanders press release about AI highlight Dario, Sam, and Mzuk instead of what has historically been the order, which is Sam, Dario, and Demis. So it's always based on like perceived industry power power and negativity around personal brand. So they have to have a scary quote, and then they also have to have a lot of power in the industry. Zuck's power in the industry is rising. And also with this movie, like there's more ways to take shots at him. Like, oh, look at what he did with the Facebook Files. Remind yourself of that because a lot of people don't really remember the details. They're going to after they see this movie. Then you can say, well, this guy's also building crazy data centers and look at how he handled this. If he handles the AI thing like that, it's gonna be bad, right? So that's like a little bit of a risk in the medium term. Long term, I don't think the social reckoning is going to matter all that much for Meta. People will complain about Meta's data centers.

Why did they not get Jeremy Strong like a wig?

They did. Because also he is playing an older— remember when he had the Caesar haircut?

Yeah.

So it's a different time. So people will complain about all this stuff, but they'll do it on Meta's family of apps. I don't think there'll be churn and certainly advertisers won't pull out. It's impossible. To pull out of Meta because it just drives up the ROAS, the return on ad spend for smaller companies. And a bunch of small businesses will just jump in and say, oh great, like, you know, some big company pulled out and is boycotting Meta, that's great. Ridge Wallet's gonna jump in or somebody else is gonna jump in.

Yeah, people have, advertisers have tried to boycott.

Yeah, you can't on Facebook, it's impossible. So the business is just too strong. And so, and I also don't think Mark will be singled out by a regulatory hammer should it come down hard. Like if there's a data center ban, I don't think it's gonna be uniquely focused on Meta. Only Dario and Sam have like true scapegoat risk because they run pure play labs and have been so noisy about AI and they have the biggest revenues in the category. There might be data center regulation, but it won't unfairly target Meta. So anyway, that's my social reckoning take. Let's move on to Fable, which launched yesterday. And the model seems like incredibly impressive. I've been seeing these like vibe-coded games that look really, really good. And for some of those, I haven't seen those— like, the vibe-coded games hit really well on social media because you take a video of them and you show the example of what it built. You share the time. And people just sort of believe it. It could be embellished, but in general, these feel like pretty solid— of course, like, making a great game is a great mechanic and there needs to be multiple levels and, like, a single demo of, like, like a forest putting in is not quite there, but really, really useful. And I'm sure we'll see a bunch more examples of like games as memes, simulators, and these things are gonna get easier to build. That's really exciting. Of course, there was a bunch of debate on the timeline yesterday and it's bleeding into today around the latest model, Fable-5, the first Mythos-class model that both seems remarkably good at long-horizon tasks like software development and knowledge work, but rejects requests related to biology, cybersecurity, and frontier LLM development. Interestingly, I haven't seen anyone share rejections around anything else. Like, did they remember to reject, like, build me a nuke? Like, because I haven't seen anyone try that. And it'd be very funny if it was like, oh yeah, just, we didn't get around to that. Or there are so many other things, but I think a lot of those other, like, things that you should reject, queries that you should reject, have been ironed out in previous in previous iterations. So going back, you know, the say a slur was a big one for a while, or say something rude, or make a political statement.

And it's a bit of a difference to like pausing a chat, basically just like shutting down a conversation versus like switching you to a less performant model.

Yeah, and it creates this like screenshot that went viral pretty much continuously yesterday. And so this aligns with Anthropic's focus on safety, but as many people have pointed out, it's also just good business. You don't want competitors using your products, to directly create competitors. And you also don't want financial liability or negative headlines from bad actors using your models for nefarious purposes. Ben Thompson called it true alignment. The take safety seriously culture aligns with business value creation, which is very, very rare. Oftentimes the like be good or your culture limits what you can do and actually hurts your business. But it's something that you do in favor of brand like Apple would probably be more profitable if they were using like diesel generators for all of their data centers. They went clean energy because they wanted to have an environmental brand, and over the long term, it's helped them, but in the short term, it's been rough. Of course, inexpensive. So Ben Thompson writes, "What is so fascinating about Anthropic, however, is that while I'm sure some executives of the company are thinking this way, I also totally believe that the employee base broadly also happen to believe that they are doing the right thing. It's fascinating to observe. Me, the rational business analyst, sees a hard-nosed but understandable decision to cut off would-be competitors. Anthropic employees and advocates, the true believers, see a regrettable but understandable safety decision that ensures that responsible and thoughtful people themselves, of course, will be the ones guiding our AGI future. This is true alignment and it's an incredible accomplishment. Facebook has tussled with this a bunch and we already talked about that. But to be clear, the Fable-5 rejection threshold really does feel way too low from what people are saying. Tons of examples on the timeline of a biologist just saying hi to the model and getting kicked down to Opus. I saw you shared someone just said cyber with the devil horns, the purple devil horn emoji, and it's like, you can't do it, we're not going further. Of course, like bringing down like a broad hammer, you can always dial it back over time. But every rejection is this implicit invitation to hop on the phone with an Anthropic sales rep and get on the Mythos Enterprise plan. And that's where the real dollars are too. The timeline is unhappy because the idea of democratizing science, technology, all of this is very alluring. But the pool of dollars available from all the biohackers in the world probably isn't close to the budgets available from Big Pharma. And so, again, you're in this rational business analyst situation and you fail to see how this is that damaging, except to the hacker community. The real tricky part is how frontier AI research is handled. Instead of outright rejecting the query and bumping the user down to OPUS, the model appears to answer but quietly gives a degraded answer. And this was disclosed in the model card, which is interesting. So this is, again, reasonable— Yeah, but it's not disclosed to the user in the product while they're paying for it. Which is a different path than bio and cyber. So if you go LLM frontier research, devil horns, it will actually give you an answer apparently. It won't bump you down immediately. But so it doesn't disclose that, which is odd. Outright rejecting requests for AI research and just saying, hey user, sorry, this model doesn't work for that type of project. Please share, please use another model or contact sales if you want help with this, would've been much more in line the bio and security and cybersecurity strategies. And they, and then they also could not have disclosed. And it's also possible that they just didn't need to disclose this at all. Like they could have just released a model that was intentionally nerfed on AI research. It would've shown up in the benchmarks 'cause people would've benchmarked it on some sort of AI research bench and been like, oh weird, it's really good at all these other things, but it's bad at LLM research. And, And maybe that would've been a bit of a brand hit, maybe, but users might never know that the model was intentionally degraded around this category of work. So that leaves this third more worrying position, intentional degradation without disclosure in the model card. There's no evidence of this, but it's possible that other workflows might be nerfed and there's no law or even convention around disclosure. Again, maybe good business, but a weird situation to be in. So probably bullish for evals if you're building a business on top of a big lab, You can imagine like a legal AI company will want to be really sure that the models they're using aren't degrading unexpectedly and not telling them. It's different if you're like, hey, I've been a bio researcher for a while, I'm using this and I know that this model was never intended for me, or I've been using it. But what you don't want is like, I'm using it and now it's giving— now it's leading me astray in my work and it's also not telling me that it's going to lead me astray, which would be like sort of an odd outcome that I think they'll probably address in the near future.

Anyway. Yeah, that aspect triggered Dean Ball substantially. Yes, what'd he say? He said, my last observation re: Anthropic's secret sabotage safety policy is that it undermines actually good safety policy. How? First, it is very plausible to describe this as anti-competitive behavior. Even if you're maximally sympathetic to Anthropic here, you must admit this, and it is behavior being justified in the name of AI safety. If you believe, as I and many Anthropic staff do, that it may end up being critically important to relax antitrust enforcement so that the frontier labs can cooperate and collaborate on some areas of AI safety, Anthropic just undermined the case for that in a large way. Overall, this is— this massively and profoundly raises the status of the argument that AI safety has been hyped to justify monopolistic behavior by labs. I continue to believe that AI safety is a real and serious issue that is growing in importance rather than diminishing. If you agree with me, this incident is a setback, maybe a serious one. And third, he says, as I have observed elsewhere, Anthropic's official corporate policy is structurally identical to the fact pattern alleged against them by the Department of War. I still think DOW acted both falsely and wrongly in that fight, but it is no longer possible to defend Anthropic with a full throat after this incident. This raises the case for heavier-handed regulations. Anthropic is making an awfully good case here that their product ought to be treated as utilities and thus their alignment practices should be a matter of public policy rather than private property. I am starkly opposed to this sort of state power grab, but Anthropic is doing more to justify it than anyone else. Thus significant damage has been done to a community and entire approach to AI governance. It was done unilaterally by Anthropic, likely motivated largely by self-interest and justified within the internal psychology of the firm through the lens of safety. I suspect this is fixable in the economic and legal senses for Anthropic, but I fear that trust has just been broken and the goodwill extinguished will take very much time to repair.

And just to level set on Dean Ball, he wrote the AI Action Plan, but then and also came out very publicly in support of Anthropic during the conflict with the Department of War, saying that the Department of War is completely overstepping by pushing towards supply chain risk designation, putting pressure on Anthropic for not wanting to work with the government in that particular way. Obviously, there's a whole bunch of new data that's been released and stuff, and that conversation has evolved. But He's not— he doesn't strike me as some, like, crazy hater. Anyway, 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 takes care of scaling, monitoring, and security. Doug O'Loughlin was also sort of mixed results on Fabless. He says, "When it works, it's brilliant, but the unilateral guardrails make me frustrated beyond all belief." He has a folder of health information with like 100 days of Oura health data. This is a very logical thing that people would do with Codex or Cloud Code, gather up all your health data. I actually talked to a very, very prominent person in tech about how they got into vibe coding and, and CLIs. And one of the first things they did was reanalyze some Whoop data, reanalyze some health data, and he correctly detected sleep apnea and went and got it treated. Like that is It's not the cure for cancer, but that's like incredibly awesome and like very, very good and exactly what you wanna see. And who knows, maybe the thing that you detect early could lead to an increased risk of cancer in the future. And that would be a really, really big win for our society. And so a little bit tricky there. He says there's a private investment in the life sciences tool space. Guess what? It's not safe. Doing some code scanning for vulnerabilities, not safe. I get the safety, but it feels incredibly out of touch for a group of a few thousand people making total comp in the millions, telling me what is and isn't safe. Dario worried about inequality. I think he has to realize that he is the inequality and the unilateral gatekeeping feels whack as hell. I don't like it.

Yeah, people are very, very—

And like Doug O'Loughlin is extremely optimistic about Anthropic. Like he recognizes the power of the models in the business and has been, very, very strong supporter, I feel like a strong user. He was one of the first people to admit to Claude code psychosis, and he's unhappy with the current situation. But these things are very tunable. I'm optimistic with more disclosures and more fine-tuning and a smoother path, I think we can get to the good outcome here.

Kermieu.

Unless you're Kermieu.

Last one, Kermieu says, tell me about mitochondria. It's the powerhouse of the cell, right?

Chat pause. The chat pause is rough. Yeah, it's a screenshot.

But maybe the harness is just saying like, sorry, if you don't know that the mitochondria is a powerhouse of the cell, like—

It's not worth my time.

I don't even want your time.

I mean, that's the thing with like the—

You keep going back for the good.

Yeah, yeah. I mean, there are expensive questions and All of these screenshots are clearly from like $100, $200 a month plans. And I'm sure the GPUs are on fire as they always are with any new model launch from a frontier company. And it can make rational business sense. And so when everything aligns, when you're like, yeah, we're maybe being a little bit too safe, but it's actually good for our business, it's very easy to say yes to that stuff. The big question is like, where do you get tested where something— there's a hard decision and it doesn't align with your business plan? Business interests. That's always tough, right?

What's going on? Was there the new data retention policies as well? That also has to do with like, is it trying to—

This one was funny because I assume that every tech company stores everything forever, basically. I didn't realize that they don't.

Well, not for these like enterprise use cases.

I know that the enterprises said that they wouldn't train on your data. But like, when I go into any chat app, I expect to have my data from a year ago still there. Like, I want, in fact, one of my main critiques and frustrations with these apps is that when I go to them, I want to be able to, instead of hitting the search bar, just go into the chat box and say, hey, remember a couple months ago we were talking about CrowdStrike and I had you pull up some data, can you just go back refresh that, get the original thing. And a lot of times these apps are like, "Ah, like I don't really have access to all your chats." But like the chats are clearly saved. But so I understand people were upset about this, but it didn't fully clock for me why. I understand the training thing, like the, but—

Yeah, and they're very explicit. They're not using the data to train.

They're not. Okay. Oh, well then that's good. So the steel man, I believe, was that you need to hold it for 30 days because companies, probably international companies, competitors, will set up a ton of shell corporations and send out like pseudo-random queries over random times, over random accounts through VPNs. And they will triangulate something that is useful to distill or learn from the model. And so by keeping it all, you can now run analyses and look at, wait, is there a pattern where 25 different accounts that seem completely unrelated all seem to be triangulating the same question? That seems reasonable to me, but stay out of my data. I'm built different. I'm not distilling anything. I don't know.

Anyway.

I think it's time.

Let me tell you about Shopify. Shopify is the commerce platform that lets you grow with your business, 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. And speaking of AI agents, we have a founder of a company.

I don't know if we should call it an AI agent.

What are we calling it, Farza? Is this— are you creating a new category? Introduce yourself. Tell us what you're building.

What's going on, guys?

Farza here.

Nana, of course, of course. Farza here working on Hey Clicky.

Yeah.

Uh, man, what is it? Uh, it started as an AI teacher.

Yeah.

And I guess now it's an AI that where you can essentially talk to your computer and it does whatever you want it to do from like personal super intelligence and you got there before the big man. I love it. Somehow, I feel like the big man doesn't even know what they got sometimes, you know?

And it takes the little guy to figure it out. I love it.

I love it.

I'm always rooting for the little guy. So, like, reintroduce the product because you went viral last week, but it feels like you've been working on this for a while. What's the nature of the team, the structure, the funding? Like, what's actually building this? What's adoption been like? Like, where is the product today?

Yeah, I started 8 weeks ago. I just wanted to build a, uh, an AI that could teach me DaVinci basically while I'm like actually using DaVinci.

And I thought it was kind of weird. And this is DaVinci Resolve, the color grading and video editing suite that's free from Blackmagic Design.

Got it. Exactly. Yeah. So like I, DaVinci Resolve, complicated program. Yep. 1-hour YouTube videos weren't it. Let me just build an AI that can help teach it to me while I could talk to my screen and I could learn.

I've been there.

Uh, and and put it out. And I thought it was really a really bad idea. One of my friends just said to post it. So I posted it and it goes really big about 8 weeks ago. After that, I didn't keep working on it. But then the cool thing was I saw all these people, you know, you know how when you put something new out there, you have all this emergent behavior kind of start happening. So when you let a human being talk to their screen, what did they start doing? Turns out they start doing a lot of crazy things all the way from watching anime with Clicky. All the way down to like learning like Blender with Clicky. 'Cause it's so easy to just talk to your screen. And that's kind of where it started, yeah, about 8 weeks ago. Okay, so tell me—

Yeah, go ahead, go ahead. I just want to hear about like the unhobbling path because I imagine that there's demand for, you know, like these chat interfaces have existed for years where you open an app and you, you know, do the voice mode and then you have it read it back to you. And there's some that are more polished than others, but, When I think about using a computer, I think rearranging windows. I want anime over here and DaVinci Resolve over here and the YouTube video over here. And I want to be able to puppeteer the mouse and the keyboard, have full context. So I need to be taking screenshots every minute, every second, 30 times a second, 144Hz. Like, what are we doing there? How are you getting the data in? And then what's actually under the hood? Because you need to process the voice actually produce an output, and it's surprising to see a harness like break out of this so quickly.

Yeah, yeah, it's actually, uh, so simple. We use GPT Real-Time up front to give you like the first layer, to give you like the really quick answer.

Okay.

But then, uh, if you want like a deeper kind of like, uh, uh, thought process over the image, you know, if you're in DaVinci Resolve, if you're in some complicated software, we now use Fable 5 actually. By default.

Yeah.

So now Able 5 is like absolutely mind-blowing in terms of how precise it is on screen understanding. So we use that now by default when you want to like understand something on your screen. But yeah, we only actually screenshot when you press a button. Okay.

Press the button. One screenshot.

Yeah, but it's really crazy because we can still detect the program you're on and stuff. And that's already really helpful. Sure. Like if I know you're on Notion, for example, and you're there for like 10 minutes, I can just ask you and be like, hey, like, What are you doing? Like, can I help you? And I think it's that sort of new modality that's like, been really exciting because that's something that's just not possible today with any chat interface.

Yeah, it's more like a coworker walking up, like a coworker just like being like present and like kind of like, sometimes you don't need help until you're asked, right? Or you don't know. I kind of describe it. Yeah.

Yeah, I kind of describe it as like having this 23-year-old intern with like a decent, like a new grad that's always watching over my shoulder. And just pretty much like seeing patterns in what I'm doing and then just tapping me and saying, "Can I do that for you?" If I had that, of course, that would be awesome. But it'd be cool if everybody had that.

A lot of these frontier models are expensive. How are you thinking about the business model? Because it seems like you've built something useful. People should just pay 10% on top of whatever the metered rate is. But at the same time, in consumer, prosumer subscriptions, flat pricing is very popular, but that introduces financial risk to you. You have to understand your costs and how they change. What do you think the business model will be as you grow the business?

Yeah, I mean, believe it or not, like, people are very open to paying large amounts of money, like normal consumers, like large amounts of money for, for like more, for having better access to these models.

Sure.

Different interfaces. So right now we just charge a straight up $20 a month.

Okay.

But even then there's a limit. You know, you get like for $20 a month, you get 150 agents on Clicky.

Sure.

But even that's very specific because once you pass that number, that's when we start losing money. Yeah. So everything's done in such a way where it's pretty cost effective. Also, like, just so you guys know, like, just calling like Sonnet or Opus or Fable, it's kind of cheap overall. The expensive part is like agentic work. That's really expensive. Like, just to give you an idea, On GPT-5.5, if you tell Codex right now to like go and click add to cart in Amazon, that costs $0.25 by just API.

Yeah.

And I know because I'm seeing the cost myself. Yeah. And so there's a lot of tricks to kind of reduce that. But no, overall, there's so many tricks to making agents cheaper, more efficient, changing the model based on the request.

And I didn't catch it. You haven't— you have raised for this or you haven't?

In the process.

In the process. There we go. There we go.

If you're out there.

Yeah, but if you're in the process, that means like you didn't launch this and be like, I'm down to be losing like $5,000 a day. Sure. You know, on, you know, like you've been focused on the economics probably earlier than you would have been if you were—

That's really good.

Had raised money out the gates.

I mean, yeah, I mean, I'm never the type of guy that wants to be losing thousands of dollars a day. So We can stop it early. But no, I mean, that's the thing about AI. Like, you know, if you're— I've built a lot of social apps in my life, and for those, costs technically don't matter based on the number of users coming in. But in my case, for every like 10 new people I add, I'm kind of losing or losing money or making money depending on what they're doing. So I kind of got to kind of got to think about it. So yeah, it's kind of AI kind of accidentally making everybody better at business because it's kind of required.

So is— I'm thinking about the hilarious, hilarious, like, like trade-off in having an agent that applies a coupon code but winds up spending more money applying the coupon code than you save. Because it's like, oh yeah, we put in this, it saved you $0.24, but the API call is $0.25. But how important to you will creating an ensemble of models that maximize the Pareto frontier be as you build this business? Because I'm imagining that any day now we're gonna get somewhat useful on-device models on Mac from the new Siri models, the on-device models. Those will probably be very limited in what they can do, but useful for some things. And that's free for you. Then you'll have open source, you know, legacy models, GPT-4 class stuff. And then you might only call out to a Fable 5 on demand. So do you need to build your own router or is there a tool that you'll be using? Like, how important will that be from the interactions that you're seeing?

That's a great question. You're pretty much asking like, what's the harness? And we just built our own.

Okay.

Because technically, like, this is all so early where there's nothing to use off the shelf.

Yeah.

You have to think deeply about it yourself and then build it yourself. So for example, let's just say, you know, you're on Clicky right now and you're saying, hey Clicky, I want you to actually research the latest in the Iran war and put it in my, in a Notion doc for me when you're done. Sure. What now?

Like, should it hit Opus?

Should it hit Codex?

Should it hit—

like, it actually depends. And so like, we can take all those requests today and just route them ourselves. So yeah, we use like, we use like 4 models right now underneath the hood.

And do you have a model that's doing the routing?

So GPT Realtime actually does the routing. Does the routing?

Oh, interesting.

Yeah, that's something that we figured out that I think not even the OpenAI team really knew and they hit us up about, which is is it's a great router. Um, it's so good at like, uh, because it's so good at tool calling.

Yeah.

Um, that means it's actually really good at routing requests. So for example, we have this tool called Call Fable 5. Uh, so if the user asks something that like is a heavier pixel task, we call Fable 5. If it's like more of like agentic work, we call, um, uh, we call GPT 5.5. And so it's kind of all being done in the background. But that's the kind of magic of the product right now, where it's like most people who use our product have never used an agent before. They don't even know what Codex Yeah. And so to them it's magical where it's like, oh, I'm just talking to my computer and it's doing the thing. That's, that's awesome.

That's kind of what they want.

Uh, how do you, how do you think about the tension between, I think you being a generational media talent and then like startups? Because I actually feel like, uh, this is our first time meeting, but I've seen your videos over many years at this point and like you're just really, really, really, really good at it. So much so that as like somebody who's in media, I'm like, well, I kind of want you to just do that.

Podcasting.

Normally, normally it's the exact opposite where it's like, hey, like, you shouldn't do this podcast. You should just work on your company, you know, or whatever, whatever the thing is. Yeah. And they obviously feed into each other really well. Like you have an edge. You can get attention for free, you know, just investing a few hours making a video, whatever. It's very, very powerful. Skill set. But how do you think about that tension?

You know, it's funny, like I've been making videos for like 15+ years and I—

Overnight success. Second time.

Yeah, I always just see this as a, as like a thing I get to do for fun on the side while I get to do the main thing, which is like actually build stuff for people. At the end of the day, like I am painfully familiar with how bad of a business like making videos is. Is and making movies is and making music is. Like, I know how bad of a business is, like, firsthand. And so I know, but I know more than anybody the power of getting in front of a billion people every single month, um, when there's a good engine underneath, a good business engine underneath it, um, that you— that, that's, that's powering something else. And so, uh, I fully intend to do that. So if I have this media ability, I'm glad I got it. Uh, I think, uh, it's like a way to do—

it's like sales.

Yeah, sales.

It's like really scalable sales.

You got to be good at sales. WWDC, some movement on interoperability in iOS. Still feels like the clicky iOS app is probably on the longer term and will be, you know, stuck in the walled garden for a while. But what I'm interested in is that, do you see just a really, really solid market? Because there's no DaVinci Resolve iOS app that I'm aware of.— and certainly people don't— there are like prosumer sort of professional tools that really only exist on a desktop or a laptop. And so, are you interested in like going and hacking on crazy workarounds to build something that like maybe makes people a little upset in Cupertino but gets you into that more casual mobile space? Or do you think that you just want to focus on desktop prosumer, you know, desktop apps?

No, I want it all. I want when you talk to your computer, whether this is your computer or your computer is this computer, I want to be an interaction layer on top of it. That being said though, I have no interest in necessarily being like an OS-level controller, for example, like kind of like what Apple is doing right now. There's no real interest., for us, most of our customers and most of our users are essentially connecting like 15 integrations to Clicky from G Suite to Notion to Dropbox and doing work with it. So I'm a lot more interested in like that reality where talk to, what if I can talk to my phone in the morning and just say, hey, like, look through all my emails and send me a, uh, send me a brief, you know, when you're done. Um, and just, I can just talk to my phone and do that. I'm more interested in that stuff, which is stuff I don't think Apple is gonna do personally. They're not gonna make like connect your connect these 15 places to my Apple account and do work with it. I just don't think they're gonna do it.

How do you think the health of open ecosystems in desktop software is broadly? Because there's two ways that you could update an Excel file, or there's actually probably a bunch, but you can puppeteer the mouse, move over, click, type the cell, save, or you can just go edit the underlying CSV file in raw text and then just refresh the front end, and I imagine one's way cheaper for you, so you want to, like, lean into getting the MCP servers, the APIs, the file writing, like the CLI, you know, interaction down. But do you think that there will be companies that lean into that or companies that fight that because they want you to stay, you know, mouse and keyboard at all times?

That's a good question. I don't think so because at the end of the day, what Qlikki is doing in the background Just so you guys know how it works, we literally took the Rust binary that's in Codex and we package it with our app so that when you actually call the Clicky agent, you're just calling a sub-spawn of Codex. And so this is on purpose. I just want to use the best model and the best kind of thing possible. And so no, I don't think that's going to end up happening. In fact, it's better. For example, if you ask me Clicky right now, hey, how do I actually add a formula to this Google Sheet here? There's two answers there. One answer is let me show you. The second answer is, I can show you, but do you want me to do that for you? Sure. And I think that's where we're going to start going with computers, where you're just going to start— your computer is just going to say, can I just do this for you? Like, I see you doing it. Like, I can just do this.

So that's where we're kind of going. Well, good luck with the fundraise. I'm sure you'll be back on the show soon. Let us know when it closes.

Yeah, it's—

I'm excited for this.

Yeah, I love watching you win and have fun in the process.

Yeah, it's great to meet you. Congratulations. We'll talk to you soon.

Have a good one.

Later. Let me tell you about MongoDB. What's the only thing faster than the AI market? Your business on MongoDB. Don't just build AI, own the data platform that powers it. And up next, we have Trent from SideTalk. He's a creator and co-founder of SideTalk. He's in the waiting room. We'll bring him in here to use the UltraDome. He helped turn Knicks fandom into one of the internet's most recognizable—

Whoa!

Calling in live.

—media brands.

How are you doing? Take us through it. How is New York? How is it going?

Yo, it's crazy out here. I'm not gonna lie. Okay. I was about to do this interview from my apartment and then I realized, hold on. Yeah. Has anyone done a TVPN interview live from the streets of New York?

You gotta go to the sidewalk. This is a first. Founder of Sidetalk.

This is amazing. Come on. Amazing.

For those who don't know you, take us through your media empire. Take us through your strategy. Break it down for us.

Okay, so I have a little company called Sidetalk. You might see the logo. And pardon the potential loss of voice from this Knicks crowd, but we kind of pretty much run around New York City and let people do and say whatever they want into our microphone. We created the Bing Bong trend years ago. We created these Knicks videos. Unfortunately, there's a little bit of chaos outside of Madison Square Garden due to us. But yeah, we pretty much let people say what they want in New York, and it's pretty fun and entertaining.

What's the secret to a good street interview?

You gotta keep it shareable and engaging. It sounds obvious, but like, why would you wanna watch an interview of someone talking about something boring or low energy or anything like that? What would you wanna send to your friends? And that's kind of the principle.

So you ask people about like language model diffusion and like, always, always, always, always.

That's what we love talking about.

Token pricing and all that. Hedge funds, private credit.

Yep. Yeah, yeah. Gotcha. That's what the people wanna see. We're gonna start talking about that outside the Knicks game, I think.

Yeah, where, where, so I understand that a lot of the, you know, man on the street interview format these days, it can be very prepped. It can be, there's a PR person that's pitching some successful business person for the, what do you do for a living? It looks candid, but in fact it's staged. Do you ever participate in that or is everything candid? Is that part of the brand?

We're 100% natural on the streets.

You don't know what you're gonna expect. Organic, okay.

Yes, sir, organic. But it's really cool too.

—Yeah, yeah, then my follow-up is like, then if you aren't prepping and understand who you're gonna be interacting with, like, how do you vet, like, where is the bar? If you see somebody walking and they have a baseball hat on and they probably don't want to be bothered, is there some sort of social contract where you shouldn't, like, is there a doom scenario where everyone in New York is getting asked, "What do you do for a living?" 15 times when they walk to go get their morning coffee?

Coffee? Yeah, honestly, it's a bit of an epidemic going on. I actually am kind of scared that I'm gonna get pulled up on right now and ask what I do for a living or what song I'm listening to.

Yeah, but oh yeah, that's another one.

The people, the people of New York, I think they distinctify Side Talk a little bit from that. They know we're not there to ask them something like, I don't know, so, so cookie cutter like that. Sure. Uh, and I don't know, now people come to us when they see the microphone, which is great, and Yeah, we just go with the flow.

How do you think about monetization? Short-form monetization, notoriously hard, but you have some merch. Break down like the business side of the business.

Yeah, it's actually very interesting. You would think that you would, that, you know, I would've made a dollar off of TikTok by now. I don't think I've ever gotten paid from TikTok or Instagram, literally not a dime, which is fine, it's cool. Go on TikTok, pull through. But no, it's cool. We do a lot of branded work for companies. So we've worked with everyone from the NFL to Netflix, Nike, Google, Amazon. Creating a lot of content for them. So they'll come to us with a product or an event that they wanna highlight and they say, hey, you know how to get clicks and views and we kind of pretty much apply our formula to that and then create a really good video for them.

Okay, I have like a hot take and I want you to walk me through whether you agree or disagree. I think you should just do a mid-roll ad in a 3-minute Instagram Reel. —because I see the branded integrations. It's like, oh— there was this example on Subway Takes where someone's take was, Android's better than iPhone. It went massively viral on r/android. And then Google was like, oh, we'd love to work with you on another take that's the same thing, but it's branded. And those sponsored takes, they often don't go as viral. It's hard to get branded content to go viral. But in some of these longer 3-minute videos, is if I'm locked in after 2 minutes, I would sit through a 10-second mid-roll of like, hey, this was brought to you by this, thanks for sponsoring this video. And then it's just all organic content around it. Does that work? Is that already happening? Are you thinking about that? What's the deal?

Definitely, we, one thing we really like doing is kind of like a natural integration. So we just did something with Nike, which is really cool where they hit us up for these Knicks colored shoes. Okay. You know, okay, who better to get word out about these Knicks colored shoes than Sidetalks? So we had people wear them in the episode. We were talking about them. We had people react to them. I think that feels authentic because that's kind of something we would do anyway.

And it did really well. That's good. Predictions for tonight.

You know, if you're losing me, by the way, a little bit, it's a little patchy.

We are losing you, but the energy is still there. Predictions for tonight. Nixon 5. Nixon 5. What's the plan for covering the game?

Like, is this any performance?

But, but isn't it in your, isn't it kind of in your best interest for it to be all 7 games? Like, you know, more, more, more content, more attention. Like, you kind of want, you know, you, we, we obviously want your Knicks to win, but you don't want 'em to win too fast.

It's gotta, where's the, where's the drama? Listen, if we can guarantee a win in 7, I would like 7. If we can guarantee 7, that would be great. But listen, we need to get the win more than anything.

Okay. What is the process of a shoot day? How early are you getting there? Who are you bringing with you?

How big is the team? Definitely, it depends on what we're doing. We do such a wide variety of stuff. We can randomly go to a hot dog eating contest in Coney Island. We can go highlight a random character in Brooklyn. We can go obviously film these Knicks games. So it's pretty dependent on what's going on. But with the Knicks games, we kind of show up in like the third quarter as it's ending. Unfortunately, we can't bring our equipment into the stadium, so we lowkey just have to stand there and watch the game on our ESPN app and hope that the Knicks win or we hear, you know, someone just scored, something like that. And then the chaos erupts. We go into the chaos. We're there for like hours and hours at a time. And at the end of the day, people see, you know, 55 seconds, but we're out shooting for like 8 hours throughout these playoff games. Got it.

Wild. Well, thank you so much for coming on. We appreciate you. Good luck.

Yeah, we will feature some of your videos in tomorrow's episode from tonight. Good luck out there. Have fun. Be safe. Yeah, we'll talk to you soon. I'm hoping for Knicks in 7 personally. Let's do it. Let's do it. Can't guarantee it yet. So have fun out there. Thank you for this walk and talk.

Thank you guys. Appreciate you.

Yeah, great to hang, dude.

Appreciate it. Thank you. Let me tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. We will be joined by the CEO of Snowflake in just a few minutes. Thank you for tuning in. We need to move on to the next story.

Luke Cannon says the cashier at Home Depot just asked if I want to round up to support the SpaceX IPO.

SpaceX IPO is going to be big Friday. We might have a surprise guest. It's going to be fun. Also, there's this crazy meta story going on, not Meta the company, but like story about a story behind Ty Morse. He's going all out trying to get an Elon interview. And it's been fascinating watching him build the craziest set in podcasting history. We're rooting for you, Ty. Good luck. Hopefully you land it. It is the quiet period, so it might be difficult, but I think that there's a plan to extend the effort and Everyone's rooting for it to happen. It's been viral on X many, many times. Also, Bloomberg's reporting that the SpaceX IPO is 4x oversubscribed. Do they know that already? Does— do you get that information at this point? That's really good news for the market, for this, for the overall. There was an interesting article. I mean, we can go through some of this. I think this was in The Economist, SpaceX. Talking about— where is it? Can the market swallow SpaceX, Anthropic, and OpenAI? Watch out for indigestion. But they talk about the float, the free float. This was very interesting. So obviously the company might be worth $1 trillion. How much of that trillion dollars is actually actively being traded? Like, sure, Fidelity might at any point in time have a price at which they're willing to buy more and a price at which they're willing to buy— to sell their stake. But founders are often locked up. Founders often want to maintain control. Employees are locked up for a certain amount of time. Certain investors might be locked up, unless you're Bill Gurley. Don't try and lock him up. Nobody can lock him up. He's a wild man. But the float is important, because If it's a trillion-dollar company and it's the CEO passed away a generation ago and there's all financial managers in, it's only owned by hedge funds. This was the story of Take-Two before Strauss Zelnick came in. Take-Two, the makers of Grand Theft Auto, it was entirely held by shareholders, by financial investors, and they were unhappy with the management team and the management team didn't have any equity. And so he was able to raise his hand and say like, I'll run this thing. And they were like, They're like, "Absolutely, thank you." And they let him take over the company without really putting anything up. He didn't need to do a hostile takeover, like bring a bunch of capital to bear to get control of that company and become the CEO. And it's been a fantastic success for him and fantastic success for Take-Two shareholders who stuck along, stuck around for the Strauss-Zelnick era. But the float matters. And Microsoft, it's 100% floated. The free float is 100% the founders have moved on and divested and they're not locked up. At Apple, it's like 99%. Broadcom's also at 98%. Nvidia at 96%. Amazon's at 91%. Only Jeff Bezos is considered sort of off the table. Alphabet at 90%. Tesla at 89%. And Meta is notably of the big tech companies the lowest free float at something like 88%, 86%. The reason for that is because Mark Zuckerberg has a lot of control. If he sells his stake, he loses control. No one's expecting him to sell, even if the stock goes way up or whatever. He's going to maintain his position because he wants to run that company. Now, SpaceX is in an interesting, interesting place. 13% of Meta shares are owned by Mark Zuckerberg. SpaceX plans to release locked-up shares in a series of tranches. If its IPO issues $75 billion of shares, valuing the firm at its hoped-for $1.8 trillion valuation, the initial free float— because if you buy the IPO, you're not locked up. You can sell the next day if you want. But this means—

Yeah, technically, it's not like whether somebody's buying through investing through like JPM, Morgan Stanley, or Goldman, all these pieces, technically, you can go and sell. They just might restrict you from other IPOs. So, you would expect serious— and this happens on all the different apps, Robinhood, Public, et cetera. If you're buying into an IPO, they basically are asking you nicely, do not sell. Don't flip. Don't flip here. Yeah, but people can't. And so with how many retail dollars are flowing into this, I expect a lot of people that are buying, buying, you know, you know, basically buying the IPO. Yeah. To start trading almost immediately.

Exactly. But so importantly, it's only 4% of the free float. So only 4% of that $1.8 trillion is a lot of money, but it's only 4% will be really free trading. And then of course, during the IPO process, you're vetting investors and you're trying to get the people that will hold forever. And Elon has done a fantastic job of that with the indices. So he's gotten the NASDAQ, the S&P, and a few others to, to, like, commit. So although NASDAQ has already shortened the seasoning period before index inclusion to 15 trading days, The Russell slashed its weighting time to 5 days. Most share indices weight firms in proportion to the value only of the shares released for public trading. And this is important, because people look at the S&P 500 and say, wait, S&P 500, a $2 trillion company coming into that, your weight, if you're just buying the S&P 500, is going to be more than 1%. It might be in the single-digit percent. That's a lot. And we've been talking about the S&P 499 for the bears, right? In fact, the initial weight in the S&P 500 will be around 0.1%, because it's just that 4% free float. And then it increases over time as the shares get locked up. We actually have a chart here of how the lockups work. And it takes basically a full year, almost two, for everything to get unlocked. And it's also triggered based on share price appreciation. So, if the shares trade up 30% or more, then more shares become unlocked and the road to 100% lockup ending happens slowly. And so, just a little bit of an interesting deep dive into how the SpaceX IPO will fare. Anything else to talk about on that story before we move on?

Apparently, Senator Warren— Yes. —has urged the SEC to halt SpaceX's IPO. Citing governance risks, Elon Musk's control, and potential foreign, especially Chinese, investment concerns. She also highlighted SpaceX's role as a US defense contractor. Senator Warren has never met a business that she liked, I think, except maybe large financial institutions. Who knows?

She likes those. Well, we have our next guest. In the waiting room, Sridhar Ramaswamy from Snowflake. He is the CEO and we're very pleased to be joined by him. Welcome to the show. Thank you so much for taking the time.

How are you doing? I'm well. Thank you for having me.

I can imagine you're doing well. The company is doing fantastically. I am interested to hear— I mean, obviously there's been an emotional roller coaster, but I'm more interested in the data that you're seeing that shows acceleration, so much promise, so much opportunity in this particular business?

Yeah, data is the foundation for insights. I worked in a data company, work in the search ads team at Google, where great data was the flywheel that made it into a great business. That's what we aspire to do with each and every one of our customers. AI has done a couple of things. It's made the act of bringing data into Snowflake doing just a whole lot easier. Pipelines that used to take weeks to set up or migrations that used to take years now have dropped to days and months, which is pretty remarkable. But people also realize how much more power they can get by having great conversational access to their data. And now with things like Snowflake Co-Work, how you can actually solve pretty meaningful business problems problems in an interactive way. It's that, it's that loop that is compounding for us that drives a lot of our growth. Is—

has there been any movement downstream for Snowflake? Whenever I hear the story of a data lake, a properly managed data warehouse, as an entrepreneur, even in a small business, I think that sounds amazing. That sounds like the dream. And then I see the companies to actually have the resources to onboard, be immensely large, and obviously that's fantastic for your business. But I'm interested in what the future looks like. Is there a world where a 10-person startup with AI agents and these pipelines that can be built much faster can actually set up their business and then not have to go through a cumbersome migration in the future?

Yeah, we are seeing that happen in front of our eyes with Snowflake's own data team. Yeah. Like with every other company, used to years long. You ask them for a feature, they'd be like, here, take a ticket, come back, you know, in a few quarters. Yeah. What they've been able to do with agent AI is grind through those kinds of backlogs very quickly. Absolutely. Data and agents are going to make it much easier to set up these things and have them evolve as they go along. I'll give you examples from Snowflake. We rewrote our— both our support team operations and our SRE teams, software. These are the folks that take care of different kinds of problems that we get into an— and the net result of that is by agents investigating the problem. And if you see enough instances of a problem that you have not seen, you go figure it out, go write a tool for it. It's now in a self-healing loop. I increasingly imagine that more and more operations get into that kind of a mindset. Where there are very competent agents, sets of recipes, you can call them skills, there'll be some other name tomorrow, that take care of problems. But it is a self-learning loop. That's what AI makes possible.

Got it. Where are we in the path to like fully automated migrations, fully automated new integrations? It feels like every new model release, we're like, this is the one. This is the one, the non-technical, CEO is going to be doing the job now. He's just going to say, hey, we have two systems, go fix it, go integrate it, get it into Snowflake. It's going to happen. And then reality sets in and there's not just token costs and cost of that to delegating to AI, but also just the real world of other decisions that need to be made that might be outside of the context. So what are the remaining challenges that come up when you're building a new integration or working on a migration?

Yeah, I think the timelines are definitely shrinking when it comes to legacy migrating off of legacy platforms, the on-prem ones. Yeah, we used to think of them, the biggest ones, as running into, say, like 3-odd years. And these strike terror into every company's heart because it's 3 years of, like, really anxious development work work and waiting for results and hoping things don't go wrong. To give you very concrete numbers, we now feel a lot more confident that we can tackle really tough migrations and have them finish in a couple of quarters. That's a big deal. It's going from 12 down to 1 or 2. And the migration team, which work on top of Agentic harnesses that we create, are confident that the technical parts of these problems will be solved by the end of the year. But there is the change management, there is all of the other applications that are running on the old system that you have to validate on the new system and ensure that there is business continuity. That kind of change management is still going to be non-trivial. But the technical parts, the act of doing the migration or writing new software, these are the things that AI is absolutely driving down.

What is the biggest—

by the way, I actually go check out every feature my team puts out using our coding agents. That's already a thing. And honestly, it's not that hard if you have a modestly technical background.

Yeah, yeah, no, that makes sense. What is the biggest lesson that you or the company have taken forward from Frank Slootman's era?

Frank was always about amazing go-to-market execution. Yeah. And we remember that. And in fact, our sales teams are incredibly well-versed now in our products and in AI as a whole because we created products that could work for them. And having an effective go-to-market motion is the real strength of Snowflake. Again, to make this super concrete, the number of use cases— these are new products basically within customers that our account executives won— went up by some 75% year on year. That's the power of sort of using AI to get work done faster. And that is a tradition that will— that, that will continue. What we have added on is a new focus that comes from a product-first person like me. And AI in sort of a generic way, or about worry about things like what, you know, what can it do. But I think sometimes what we forget is if you have a well-built system that uses like the latest state Yes. It's actually an incredible delight. All kinds of information that you would want. I was at a conference recently. Every piece of information that I wanted to know about a customer that I was having a conversation with, I would look up even live if I didn't know that I would do it in front of them and show them the power of Snowflake's AI. A big part of our success over the past couple of years has been internalizing this. AI is a different way to think about about product and product delivery, and the companies that are going to succeed are the ones that truly internalize that. That is a big chasm between how we did it with web pages and clicking on things to these agentic flows that can truly solve problems.

Then how is the sales— the role of the sales rep changing, the sales associate changing? Are you looking for any different skills, or are you doubling down on what worked in the pre-Agentic AI era?

First of all, a lot more familiarity with Snowflake is a must. Sure. Because these folks now have access to products that show off what Snowflake is about. I expect every sales rep to have CoWork on their phone and be able to show off the actual impact of having a product like that with free access to information. To each and every one of our customers. When it comes to our solution engineers, these are the technical pre-sales people. Used to be that we'd have 6 demos total for all of Snowflake, and maybe they would do a little bit of customization and say, "Hey, here's what Snowflake can do." Now, pretty much all of them are capable of creating a custom demo tailor-made for a specific customer in a vertical in a way that makes sense to them them that will really illustrate the power of Snowflake. It's that kind of empathy that they can use the technology that we create to drive. I would say it's changed in a really, really big way because a lot more possible, but the bar is also a lot higher in terms of what you're expected to bring to the table.

Can you walk me through the partnership with Amazon Web Services? I, I, I'm sure you compete in some categories, But how did this come together? What does it mean for the future of your business?

Yeah, we work with the hyperscalers. AWS is the biggest partner. Yeah. And, you know, we work with them on a number of different levels. We work on product integration. Sure. You know, buy a lot of capacity from them, a lot of GPUs that our inference team runs models on. But where we really shine is in joint go-to-market. AWS is an incredibly customer-first company, and if a customer says, hey, I want AWS plus Snowflake, they lean into it, they bring us, and we really focus on how do we solve problems jointly for customers. And there's a deep, deep relationship both at the account level going all the way up to the CEO level that ensures that problems that come up in practice get solved very quickly. It's the kind of partnership that one only dreams of. It's incredibly effective in practice.

What's been your strategy for trying to predict AI progress? Because, you know, we have people on the show every single day, somebody from a lab, somebody that's an investor, you know, customers of the labs, things like that. Everyone has a different opinion. You have to weight them differently based on their incentives. But you're running a massive company that's leveraging these tools. But at the same time, you want to understand what the capability will be 1 year out, 2 years out. And it feels harder than ever to actually be able to predict that.

I think this is a, is a really good question and one that to deal with. I don't think the software industry as a whole Google has internalized that this precious commodity that we all love creating by hand, which is software, the cost of creating new software is going to keep going down. I tell people that the best analogy that I can give for them is to go back to, let's say, the 2000s when the cost of creating and distributing new content truly dropped precipitously. Instantaneously. And you have to work through what is the implication of something like this. For example, I think simply getting deployed into an enterprise and being part of workflow there is no longer a guarantee that things will going to— things are going to keep continuing because switching is also easier. I would say that's the first Every company that's in the world of creating software as a big part of their business needs to think through what is the durable value that they are creating. I think it's a really hard question to internalize. And so we spend a lot of time just thinking about it. The second one, and this is— this gets to the heart of your question, and I honestly just have a practice, which is that this is a bad time to be making 2- and 3-year plans. Anytime things are getting better or worse by 20%, 30% every month, the human ability to predict these things, which are exponential in nature, is, you know, it just doesn't, it doesn't really work. I joke to people that at this point in time, we should treat coding agents the same way we treat traffic. Any sane person is always going to look at Google Maps first before they go somewhere. Because it's an all-seeing God. Agents are sort of becoming like that when it comes to, when it comes to software. And the more you have that open mentality that more things might be possible with a new model release than was possible before, the more you will be pleasantly surprised and you'll be ready to take advantage of it. And so I stress that childlike quality in figuring out what is possible. Obviously, you know, there are nuances to how do you How do you put in work in such a way that even if the model got better, you'll still be able to take advantage of what you have done before? There are practical problems like that. But the biggest thing that I ask for is that open-mindedness to what is possible today, because a lot different from what was possible, say, a month ago.

Are you positioning Snowflake as a company that's in the token path or in the token token stream. I don't know if you've heard this phrase that's sort of new coinage for a position in the AI value chain. And it feels like you're firmly in the token stream, in the token path. But I'm wondering about what you think about that category. Is that narrowing you? Is that valuable to communicate to customers and investors? What do you think of the phrase like in the token path?

It's a useful way to think about how value is going to get created. Remember back in the web era, we talked about, you know, number of visits, number of unique visitors. These were the things, or weekly active versus monthly active. These were the abstractions that we used to ground the numbers that various people people would report in a reality. You can think of the token path as existing along this path that is using the power and knowledge that these models have and creating value for customers. And both in the data, where our coding agent called Coco comes into play, where we want everyone that wants to get value from data to be using it to create all of the artifacts in that path, but also in the consumption path where our product is Snowflake Cowork, where I want every employee in every company to be able to access the valuable data that is in Snowflake, but also in other applications via things like MCP to be using Cowork. And so we very much think about can we be in this path of coding up the things that are going to drive consumption and creating value but also in the path of how does an end user query enterprise data, take an action on that data, and also creating value. So we very much think about this.

So my last question is, you said it's a bad time to be making 2 and 3-year plans. How far are you looking into the future? Do you have a 1-year plan, a 3-month plan?

What— Well, as a public company, you do have to—

Yeah, you have to have forecasts, But more qualitatively, like what are you gunning for this year or over whatever time horizon you think is appropriate at this point in time?

Our overall strategy is very clear. It's just what I described. Anytime anyone wants to do something with data, I want them to use agent tools that are created by Snowflake, Coco, for example, to be able to do that. We want to have the best data platform in the world so that when and whenever someone thinks, hey, I need a historical view of this or I need an OLTP database to do this, they think of Snowflake. They can use any coding agent, not just our own, to be able to, to be able to do that. Similarly, we think a lot about how can we make sure that we sell co-work to CROs and demonstrate what we have done internally within Snowflake, just transform our own sales organization. So we think a lot about how do we take our customers through the same journey that we have been through in terms of how can AI make a difference to the business. There'll be a lot of details along the way, but that's the path that we are headed in. AI leveraging the full power of data.

Well, congratulations. Thank you so much for taking the time.

Yeah, great to meet you. And congrats to the team on all the momentum.

Yes. Stay strong in the token. Thank you, guys. We love to see you. We'll talk to you soon. Have a good day. Goodbye. Let me tell you about— sorry for agents. Meet the Canvas. Your AI agents can now create and modify your Figma files with design system context. Yes. Apologies for the technical issues.

A lot of people are saying, is this AI?

It's not AI.

Is this AI? There was some type of clipping.

I've never seen that exact technical failure before where it cuts out entirely. We've heard scratchy audio. We've heard degraded video. The video was coming through crystal clear, but the audio was cutting in and out a little bit. But I think we got a lot of good stuff. Coco, they should have called the AI agent Frosty. Snowflake? Frosty. Probably intellectual property.

Frosty the agent. It's a jolly happy coder.

Coder. Yeah, something. Anyway, Citadel Securities, former employer of mine when I was an intern. Wrote a post on tokenomics. They're one of the most significant hedge funds and they just dropped tokenomics. I have it here. I have a second printout, Tyler, you wanna bring that over to me? It's not what you expected is what they say. So said, we've argued for some time that agentic and complex workflows delivered by frontier models would be expensive to run, constrained by physical bottlenecks, and vulnerable to unrealistic expectations of frictionless deployment costs. That judgment now looks less contrarian than it did when we set it out in February. They're taking a victory lap over at Citadel Securities. Amazon has now removed its token leaderboard. Microsoft has canceled Cloud Code subscriptions. And there have been multiple reports of unexpectedly large token bills. The salient point is that even the most powerful technologies must pass through the prosaic discipline of cost curves, capacity restraint constraints, and marginal returns. Adoption is therefore becoming less about what frontier models can do in principle and more about the price and scarcity of the inputs required to make AI operational at scale. Compute power, cooling, memory bandwidth, and inference budgets are real and binding constraints. They go into economic theory talking about about the 3 functions that prices provide. They signal scarcity, create incentives for substitution, and ration scarce resources toward their highest value uses. This is the stuff that goes viral on TikTok. We're deep in it. You heard of the keeping the viewer watching. This is the most important thing. You read the Citadel Securities Macro Strategy Report, but they do say they ration scarce capacity toward the areas where marginal productivity in AI justifies the marginal cost of using it. They're talking about ROI maxing for the economy at large. Simpler models may be more cost-effective productivity-augmenting pathway until physical constraints are eased. We, we hence see growing signs of a bifurcation in frontier versus everyday AI usage. They actually shared some data here around Token Expenditure Index dropping the log growth rate declining. And so folks are pulling back a little bit and shifting towards cheaper models as they figure out a new workflow, something gets unlocked, and then they say, hey, we got to actually make this ROI positive. So that's the latest from Citadel. Let's see what Amazon Web Services is saying because they're blackbelling on the timeline. They said more AI-generated code doesn't make your team faster. It might actually slow you down. Crazy post from AWS. Would've expected it from like some AI thought leader, podcaster, Twitter person. AI bear. But yeah, it's funny coming from the AWS cloud gold checkmark.

It's funny 'cause it's also like if you were to write this, I think a lot of people would write this as more AI-generated code doesn't necessarily make your team faster. No, just perfect. No, more AI-generated code doesn't make your team faster, which is just like factually incorrect for like a lot of teams. Yeah, yeah, right. That's true. We're actually moving faster than ever. Also, okay, uh, read the next post in this thread. It's like very clearly AI-generated. Okay. The real bottleneck was never writing code. It's releasing it, debugging it, and keeping it running well. So when Honeycomb— see, Honeycomb.io CTO Charity Majors set a productivity target. She didn't chase 10x, she chose 2x and built from there. You think this is AI?

Because the choose 2x comma and ampersand built from there. I think it was human edited, but like that syntax feels a little AI. I agree, I agree. Her team also skipped the mandates and built a set of AI values instead. Every output has to have human owner.

If you don't want your name on it, it's probably not good work.

Yeah, that's kind It's reasonable. Quality first, quantity second. So they had a podcast about it. Coming from AWS, that's a wild statement. This is very funny. But yes, also funny with the backdrop of a $500 million budget.

Well, yeah, and it's funny because Amazon is, of course, one of the largest investors in both OpenAI and Anthropic too. Leading coding model provider.

Hey, they didn't say more AI-generated posts don't make your social media team faster. Entirely possible it sped them up. They certainly got 14,000 likes on this. This is a banger post from a corporate account. Rare to see those numbers put up on the timeline from a corporate account. So congrats to the team over there putting up some fun numbers. Shriram Krishnan. I don't know how to differentiate them. He's at KearneyJackson.com, investor advisor to Bending Spoons, not the former A16Z GP who worked for the government for a little bit. Shreeram Krishnan says, not many Silicon Valley VCs took our Bending Spoons intro in 2022. These VCs thought it was beneath them to be associated with the company. From 2024 onward, many clamored for an introduction to sell their portfolio. Portfolio companies. Bending Spoons is going public. So Bending Spoons will be a public company, and they own a number of historical companies. They own AOL and have done a bunch of other stuff. Eric Suford broke down the—

They literally own America Online.

They own America Online. It must be electric. I can't get into my mobile— email. I'm just going to skip this. Brutal. And move on. But you can— Eric Suford over at Mobile Dev Memo wrote a breakdown of— Sufornator. Sufornator wrote a breakdown of Bending Spoons filing to go public, the Italian roll-up operator that owns, among many other properties, AOL. They own Eventbrite and Vimeo. Don't they also own Evernote, or am I getting that mixed up? Someone owned Evernote, and the Evernote employee growth chart looks like a fast takeoff. Like post-2020 COVID, they went from like 1,000 employees to like almost 6,000 employees, or they tripled or doubled the workforce. And Bennington's known as sort of like the operational excellent, almost like a private equity firm, private— like they're going to do cuts. They have done a few cuts, but like the actual headcount has like held fairly stable. And I think the business has gotten to a much better place. So interesting to see there's a time and a place for being aggressive, expanding when there's a big moment and the market's up. And there's also a time to consolidate and focus on operational efficiency.

Why is this guy so obsessed with GTA, John? Have you ever played GTA? Any GTA? Maybe like— GTA 3, GTA 4? Maybe like for like 2 minutes. 2 minutes? And I was like, wow, you just—

Loser. What a loser. Never gamed in his life. Can't get a load of this guy.

I couldn't get into it. I couldn't get into it. Yeah.

You're a Rust guy. You're a COD guy.

Everything, like I just, even as a child, I was like, everything that you can do in this game, you can just go do real life minus the crimes.

No, you can't. It's all crimes. Wait. Jordi turned it down. He turned it down.

There was some like 10. Maybe I'm pure of heart and I don't enjoy it.

Like, okay, like, like crap. They were gonna let me play for like 15 hours, 10 hours, maybe 5 hours, but I turned it down.

They were like, they actually wanted to pay me. They wanted to pay me to play.

Yeah, they wanted to pay you to play. Strauss said, actually, I would love for you to play. Um, no, I, I, one of my first jobs, I was a video game tester. I got paid to play video games. Best job.

Play video games all day. They got you hooked. Yeah, they, they, they got you hooked.

No, I was already hooked. I was elite. So yeah, you go play the games while it's in development, you file bug reports, you're a quality assurance tester, but basically your job is to play video games all day. Worked with a guy who was super chill and like, he was just like, "This is my career. Like, I'm happy with this. Like, I'm a video game tester." And then I checked in with him a couple of years later and I was like, "Hey man, like, how's the video game testing business going?" And he was like, "You know, I actually had a shift in my career." I'm a DVD tester now. It's even easier. So he just, I just sit there, watch the full movie, and then tell the people like, "Yep, like there were no audio bugs. There were no drop frames in the whole movie. I sat there, I watched the whole movie, and it looked fine. Thumbs up." And then every once in a while he'd be like, "Oh, I was in the menus of the DVD, and when I went to behind the scenes footage, it wouldn't play the correct video. It had the title for you know, interview with the director, and it actually played interview with a cinematographer. So you got to fix that. Like, these things happen. Like, UI bugs pop up in DVD menus. But what a wild job. DVD tester. It's a good one. Good one. Anyway, this fellow is obsessed with GTA 6, and he's doing anything and everything to understand whether or not GTA 6 will be released on time. There's two methods here if you want to know GTA if GTA 6 is gonna get released on time. You can just shake down Strauss Zelnick for the truth, or you can do what he did, which is he set up acoustics around the Rockstar headquarters to measure audio levels near a meeting room, and he discovered that his equipment had disappeared. And when he returned to check on the latest data— but he didn't stop there. He started camping outside, monitoring traffic and tracking the number of cars and their estimated value. Oh, there's an executive executive in town, that means something for the GTA 6 release date. Uh, oh, there's more cars coming in, probably a lot of people working late nights getting it out, burning the midnight oil. And so he believes such data can be used to speculate on how many executives may be spending time at the HQ as opposed to regular employees, and an increase in both traffic and high-value vehicles could indicate Trailer 3 is coming. He's also considering an even more insane idea, which is to measure to monitor the amount of oxygen in the area, to monitor how many people are around. And he did it! He started monitoring the oxygen around Rockstar North, and he believes Trailer 3 is coming soon after noticing that oxygen levels dropped to about 20— 20.3%. We actually need to use basis points this time. From 20.3 to 20.04 after 5 PM, suggesting increased oxygen consumption in the area after normal working hours. He's also monitoring the number of cigarette butts I thought it was just a GTA 6 fan. I'm excited. Everyone's excited. Maybe it's a Rockstar or a Take-Two shareholder. You had a different theory. What was your theory on his motivations?

I don't believe that people are going to go to these lengths unless they're making money on it.

Okay. And how would he make money?

There's a lot of prediction markets with plenty of volume.

Yeah, could be trading the stock directly, but could also be betting on a prediction market that Oh yeah, if there is—

theoretically, if there were to be another delay, I'm sure the stock would probably react to that. Yeah, absolutely.

It is very, very interesting.

But regardless, fun story. A hedge fund should hire this guy to do other types of detective work.

Yeah, it almost feels like it's going over the line, maybe illegal. I don't know. I would be very frustrated if someone was standing outside of our UltraDome with a mic measuring the air, measuring everything to see if we're going to interview Matthew Prince yet next. And we are, we are, because he's in the waiting room and we're bringing him in to the TBP and UltraDome. Welcome to the show, Matthew. Thank you so much for taking the time. How are you doing? Good to see you. I'm doing well.

Great to see you. Great to see you.

What's new in your world? You made some acquisitions. You made an acquisition. Take us through it.

Yeah, uh, so we, we bought a company called Void Zero, which, um, uh, makes Vite, which is one of the most popular developer platforms that's there. It's just an incredible team. Evan Yu, who's the founder, is just a, just a first-class human being, someone who, who our team is super excited to work with. The team that he's assembled is just great. And I think that this is increasingly becoming the platform that is being used to power a lot of the agents that are running around, around the internet. And, and a lot of those agents are running on Cloudflare. And so we think it's just a really natural combination.

How simple is the synergy? Is it, uh, you'll funnel those 130 million users who download Vite every month— I think it's 130 million weekly downloads— uh, weekly, wow, yeah, that's a lot— uh, into, you know, preferring Cloudflare, defaulting to Cloudflare? Or is there more synergy under the hood around developer integration, company integration? Like, how are you thinking about this playing out? Yeah, we continue—

we plan to continue to leave it as an open source project and support it and invest in it that way. We want to integrate it closely with Cloudflare's developer platform and make sure that Cloudflare is the best place to run any sort of Vite project that you have. But it'll work in any of the different platforms as well. And so we just really wanted to make sure that Evan and his team have the support to make sure they could continue to really invest in what was, in what was just an incredible platform. And we think that that is going to drive more developers to Cloudflare, Cloudflare Workers platform as well.

What are the headaches for developers these days? I know everyone's concerned about token costs and token budgets. Sometimes that doesn't show up for the developer. It's more like the CEO that's worried about it. But is uptime more difficult to maintain? We've been seeing screenshots of different uptime tracker status pages that have more and more red and yellow on them. Like, what's at the top of the stack and how are you helping?

Yeah, you know, I think that you need a different architecture than we've had to build sort of the last generation of applications for what's coming with agents. If you imagine, you know, there are about 100 million knowledge workers in the United States. If all of those knowledge workers had one agent which was, which was working on their behalf, that would, that would— and each of those was— each of those agents was running in a container, a traditional container like something you get at AWS or Google Cloud. The amount of just CPU resources that would be needed to, to run just those, those agents, assuming they just had one agent per person, is about 50% of the total CPU capacity that's generated, you know, by all the different CPU manufacturers that are out there today. And that's just the United States knowledge workers. If you take that to the world, then it's several times, you know, 30 or 40 times the capacity of GPUs and CPUs that are existing today. And so what we really think is that as these agents are creating code, you need a different platform for it. And Cloudflare was built— Cloudflare Workers was built not on containers as much, but on something called isolates, which is a much more efficient technology. And so what we're seeing is that as people are building these agents, as they're using them, it's just a much more natural place to be running them. And it's, I think, why you're seeing more and more of the big AI labs have Cloudflare as the preferred target for where, where their code gets run. You know, OpenAI released a project for their enterprise users a little while ago that again is targeting us. And, and we want to make sure that with, with things like Vite, uh, as, as first-class citizens on Cloudflare, that we can help power that future. Because again, it's not going to be kind of the same, uh, system that we, that we built with the hyperscalers. It's going to be something different. And again, I think that we have a really good shot of building that different thing.

So can you help me understand more about the problems of like CPU bottlenecks and then maybe some of the solutions? I'm just thinking about like I would think bandwidth would be an issue, obviously CPU load, but is there a world where we get to some sort of convention around— maybe it's the robots.txt or just the user agent, and when Googlebot shows up or any other AI system— AI agent shows up, you're just delivering a— it's almost an MCP server, but you're just delivering something that looks more like a JSON package, something that's a little lighter. A little less CPU intensive. Is there a path there to optimization? It feels like we're in the era of like squeeze every ounce of performance out of everything. But what's actually going to happen here? Where, like, are we just screwed or is there an opportunity?

No, I think two different problems. So the first is, as you ask these, these different AI systems to perform tasks on your behalf. What has to happen behind the scenes, especially as those tasks get complicated, is that they need to be coordinated in some way. So if you say, plan a vacation for me or something like that, what goes on behind the scenes is that there's coordination there. And the best way to make that as efficient as possible, what agents are really good at and these various LLMs are really good at, is actually writing code. And so that code needs somewhere to live. And the problem is that if that lives in a container, then you've got to bring in an operating system, you've got to bring in all the tooling and everything else. And that's actually extremely heavyweight. And so that's the first place that you've got both a CPU and a GPU bottleneck that's there. And so with, with something like Cloudflare Workers and Isolates, it's just a much lighter weight system. And so that means that you can have more agents running on the same CPU infrastructure and be able to provide that. And again, I think that's why A lot of these next-generation tools are actually built using our platform. You mentioned something else, which is, you know, as these agents go out and interact with the rest of the web, that you want to make sure that that is done in the most efficient way. And so if they're pulling down, you know, all of the HTML from a webpage, and those webpages are designed kind of to be consumed by humans, there's just a lot of cruft on that that isn't necessarily as important. And so some of the things that we're doing are, for those customers of ours that want to make sure their content is consumed by agents, want to make sure it's as accessible as possible, we are automatically converting things into Markdown, which is a much simpler system. That saves you a ton of tokens. It saves you a ton of processing. It means that your context window can be— you You can fit more useful information into a context window. So I think there's a ton of optimizations. And we're helping both on the developer side as well as on the content side, making sure that we can have these agents be as powerful as possible and get as much done as possible.

John Gruber, total victory. You know he invented Markdown? I did not. Grubinator, yeah. No way. Yeah. Isn't that amazing? Yeah.

I mean, it's one of these things that just, you know, it turned out. It was ahead of its time, but it's such a key for making sure that we can take information and make it as compact as possible.

John Gruber created the format for God. It's a funny, funny world we live in.

Talk about, it was last week you guys announced that the threshold had been passed around agent versus human traffic. Talk about that moment. Did it happen sooner or later than you expected?

Field much sooner. I mean, it was— I, at the end of last year, so the end of 2025, I said that I thought that we would pass bot traffic. And that's, you know, across the board. So that's like Google's crawler, but also, you know, the new agents which are coming out. I thought we would pass human traffic by the end of 2027. About 3 months ago, I revised that based on the traffic that we were seeing to say that it would actually be in the first half of 2027. And so the fact that it actually happened in the first half of 2026 is just, is just, it's just been extraordinary. And it just shows how quickly this, this is, is growing. And the real key here is that I think that if, you know, you or I as humans were researching to go buy a digital camera or something, we might visit visit 5 websites and do, you know, a little bit of research and some price comparison. You know, you just watch as you use these agents, they have boundless, uh, attention to be able to just go to maybe 5,000 websites to find exactly what you're, you're looking for— the best price, the best delivery, the best service, and everything that's, that's there. And so that's just driving an enormous amount of consumption of, of, of the internet. At the same time, the other thing that's happening is is that for a long time, since about 2015, the web has kind of plateaued. There were not— there were more websites that were being shut down than were being created during that time. In the last 18 months, though, we're back to the web growing at a rate which is exponential. And what— and it looks sort of similar to what was happening back in the early 2000s in terms of growth of the web. And so I think you're seeing both more sort of consumption of what's online, but you're also seeing more things online as it goes forward. And so in both of those directions, you know, that's going to continue to just drive, you know, more and more use of the internet. And I wouldn't be surprised if, you know, going forward, say 5 years, that bot traffic will be 1,000 times human traffic online. And we've got to make sure that we make the internet work for that new future.

Yeah, like every year we're just gonna add another 9 to the 99.999% of internet traffic is bots. Do you know what the baseline was? Like pre-AI, were we at like 1% bot traffic? I mean, I've seen a lot of sites—

No, it was more than that. It was about 20%. Okay. For a long time. So it's, you know, Google was the largest. And then obviously there's a lot of malicious things that run around online. But that was around what it was for. And it was pretty stable over the history of Cloudflare, at least where we could measure it. So since 2010, it was always kind of around that 20% range. And then it's grown. In the last few years, it started growing steadily, and then it really accelerated in the first half of 2022.

I want to talk about the inference stack, I guess. Like we're seeing two things play out, sort of a bifurcation, like at WWDC, Apple's launching on-device inference that's, it's not going to be frontier, but it's going to be usable for sure on your phone, on a computer. They can obviously go to their private cloud as well. And then you have like the new NVL72 models there's stuff in between, a Cerebras chip, fast but expensive. Then there's everything from, oh, it runs on commodity hardware, but it's still pretty big. You need a real desktop for it. And I'm wondering about how you see Cloudflare fitting into that. It would be a logical extension to Cloudflare Workers to have inference on the edge, inference in different places. Like, where do you see yourself of offering inference, if at all?

Yeah, well, it's actually kind of funny. Back in 2022, we issued a press release that there was a graphics chips company that we were gonna partner with in order to put GPUs at the edge of our network. Yeah. And that graphics chip company turned out to be NVIDIA.

It looked something like NVIDIA, yeah.

And what was amazing was at the time, it was just crickets. Like, nobody cared. Sure. And so we sort of did a little of it, but we hadn't really rolled it out broadly. And what was funny was— and then you fast forward 2 years later, and all of a sudden everyone cared. And so we basically just reissued the exact same press release. And now, you know, that's become a pretty important part of our business. And so today you can run inference at the edge of Cloudflare. And because of the fact that we're, you know, in over 350 cities worldwide, you know, we're within, within milliseconds of the vast majority of the world's population. We become a very natural place for inference to happen. I, my, my working assumption has always been that about 50% of the inference that happens will be on device, whether that's your, your phone or your, or your laptop or whatever, but that there needs to be some standard protocol where your phone or your laptop or whatever that local thing is can hand those either long-running tasks or larger tasks off next to the network. And so that would be to us. And then if for some reason you need something that's more than that, that you could handle it back to, you know, some centralized data center with, you know, with more, more capacity than, than we may have. And I think that that's what, that's what we're increasingly seeing. I think my, my assumption was a little bit probably off. I think actually I think less is going to happen on device today because I think more and more of the tasks are going to be these long-running tasks where it's not going to be just, you know, what's, you know, what's the, you know, what's the temperature in New York today? Yeah. Instead, it's going to be something like help me plan a vacation, go, you know, take into account all of these different things, shop for, you know, different hotels in different places, plan all the plan all the, the travel between, between the different locations. Here are the criteria that I have. And that might be something that takes, you know, maybe, you know, it's not going to be seconds. It might be, you know, minutes or hours or days in some cases to run.

And you're going to have— I feel like you're going to have long-running agents just for Park City snow forecasting and reporting, tracking like individual, individual, individual runs, you know. Maybe a network of drones that are identifying what's cracked out, what's not. Satellite photos, Planet Labs. Optimizing your routes on the mountain.

A lot of stuff. That's exactly right.

Hopefully it snows this year, unlike last year.

Yeah. So what are the decisions when you're doing inference on the edge? Because it's sort of a hard pitch to say, I'm going to shave off 300 milliseconds, 600 milliseconds, when you're talking about a 20-minute workflow or even a 10-second workflow, you're getting me a 3% boost. Is that gonna make the difference? But are you optimistic that there will be sort of like an in-between state where there will be inference that happens and it needs to happen within the request loop under a second and then the latency matters?

Yeah, well, I think there are sort of 3 different buckets. Why people are choosing to run inference tasks on Cloudflare. I think the first bucket is exactly what you said. It's latency. And so especially for those things that have human and computer interaction where any delay feels like it hurts you, that people are trying to make sure that they can get the best performance possible.

Yeah, for voice models, I mean, it feels like that has to— Yeah, voice models have got to be local. Yeah, I love that.

Yeah, and so again, there's a case for that. I think the second case is— is that a lot of times for either privacy or regulatory reasons, people want to keep things as close to where they physically are as possible. And so I think that a lot of, especially in Europe and otherwise, of places think that they made a mistake with the internet kind of 1.0 and 2.0 of having everything go back to Ashburn, Virginia, everything go back to the United States. And so I think that there's a real kind of sovereign desire to keep things local. And that's important. And then I think the third thing is we actually can be oftentimes much more cost-effective because of the fact that we are a network. And because of how we're interconnected with networks around the world, bandwidth for us is effectively free. And so it's very easy for us to get things onto our network. And because of where we deploy systems, we often are in places where we don't have to pay for the space or the power, which allows us to then pass those savings on down to customers. So it can be significantly more cost effective to use inference with us than it can be in some other places. That's somewhat counterintuitive, but it's sort of the nature of what we've built and kind of the secret to what's allowed Cloudflare to deliver as much as we have over time.

There's a ton of debate over data centers, how they get built, how big they should be, what we need. Need, the pushback, regulation, et cetera. How are you— is any of this affecting your business? Are you thinking about how you position your footprint in the real world? How are you processing the evolving discussion around data center construction?

Yeah, I mean, so first of all, I mean, we're gonna need more data centers. I think a lot of the discussion is somewhat silliness. I mean, the water, consumption. I mean, these are closed-loop systems. I mean, a golf course uses more water than, than probably all the data centers in the United States over, over the course of a year. So I think that, I think, you know, there's, there's, there's a lot of kind of silly concerns that are there. And but there are, but there are real concerns as well. We've got to make sure that there are efficient ways to get power to these things, that it's not taking power from, from other, other systems that the grid can support. Support that. And so I think those are all good considerations. For Cloudflare ourselves, you know, we're, we're a little bit different, where in the case of, you know, if you're an AWS or a Google or, you know, a SpaceX, you're building one big facility or a handful of very big facilities and putting a ton of machines in that, that one facility. Cloudflare is different. We have a ton of machines, but we spread them massively all around the world. And oftentimes We want to go into the places where networks come together. And so those, those can be, you know, some of the oldest data centers in the world. And in any of those facilities, we may not have— we may have only hundreds of machines, but over collectively around the world, we've got, you know, what is effectively much larger than any individual data center, which is, which is out there. So I think we're less, less impacted from, you know, the new builds of data centers. I think we're much more impacted by How do we find our way into the existing data centers? And then how do we make sure that the equipment that we're deploying is as power efficient as possible? Because we're often given sort of, here's the envelope that you have to fit your, fit your, your equipment in. And we're often guests of, you know, whoever the local ISP is or, or, or whoever, you know, partner is that we're working with in the, you know, the thousands of buildings that we're in all around the world.

What's new with Italy?

Oh yeah, um, yeah, Italy and Spain, like, it's, it's, uh, it's, it's, you know, so you're fighting with Italy, you're fighting with France, your summer plans are, uh, limited, dude. Yeah, no, no, no Ibiza for me, or the Amalfi Coast. Um, no, it's, uh, it's, you know, I think it's, it's interesting. Um, it's been actually kind of puzzling, um, For us. So all of this stems back. So for those who don't know, in both Italy and in Spain, there have been pretty aggressive tactics to come after either me personally or Cloudflare. And it's all been driven largely by the football leagues, the soccer league in those places. And what they're concerned about is, is piracy. The thing that's been ironic about it is like, we don't like piracy on our network either. We don't make any money off of it. It takes bandwidth. It steals resources. And so we have a whole team that's working all the time to shut, shut the pirates down. And, and yet, you know, the pirates are clever. They find ways to use our system. And because, you know, huge percentage of the internet sits behind Cloudflare, you know, there's going to be stuff from time to time that we don't, we don't capture. It takes us some time to catch with leagues everywhere else in the world. So the NBA and NFL and MLB and the Premier League in the UK and others, we work really closely with them to— if they identify a pirate stream for us to pull that down, because again, it costs us money. We don't like it. But for whatever reason, La Liga in Spain and the league in Italy have decided that instead of partnering with us to take this down, they would sue us. Instead. And so, so again, yes, it does cramp some of my summer plans.

I want to talk about IPOs, life as a public company. There's a few entrepreneurs that are trying to go public this year. You had a very successful IPO. The stock is up more than 100,000 basis points since you went public. Fantastic run. But advice for—

we're putting everything in basis points now.

Sounds even better. I mean, the stock's up 1,000%. It's fantastic. But 100,000 basis points sounds even better. 100,000 basis points is a lot. 110,000 basis points. Who's counting? But anyway, what was your process like? We were talking about the level of float, lockups. What were the hard decisions? Were there things that we're talking about now with these big IPOs that weren't even on the top 10 of your to-do list? List. What is important to a successful IPO and then running a public company?

Yeah, I think, you know, there are a handful of things that, that, that, that are interesting lessons for us. So first of all, like, I love being a public company. It's, you know, I think that it's interesting, you know, that the, the, in, in jurisdictions where Yeah, you don't have no-fault divorce. Um, spousal homicides are much higher, which is, I think, kind of an analogy to the difference between private market investors and public market investors. It's really hard to fire your VCs. It's hard for your VCs to fire you. And so as a result, like, actually, it's kind of dysfunctional. Whereas public market— like, I love our investors because if I do something stupid, they call me up and they say, that was stupid, I sold your stock. And then we, and we have a conversation and sometimes they say, oh, actually that makes sense now. And I bought it back. Right. But I think you can have actually a much more honest, much more real conversation. And, um, and, and that's, and that, that's felt actually just a lot healthier, uh, than, than what you see in, in, in the private markets.

Yeah. It's so, yeah, it's so true. I mean, uh, there's every, every, everybody that's angel invested in any number of meaningful companies is gonna have portfolio companies where you've fully given up up on the company and you just try to be nice and yeah, help if you can, but at some point you're just totally disassociated with the investment and you're just like, and that's the reality, but yeah, so it's very healthy to be like, yep, I'm disassociated, like moving on, put my energy elsewhere.

And in both directions, right? I mean, it's just a better, it's a better relationship. And so I loved the process of going public. I think the thing that we, we did, like, it was an opportunity for us to really kind of retell our story. Like, you don't get to reinvent yourself very often, but the process of writing the S-1 was incredibly just an opportunity to sit down and really do it. And we, you know, really dug in and told the story in a way that to this day, we still refer to parts of that document to sort of explain what is Cloudflare and how do we work. I think that the most clever thing that we did, and this was advice that I got from Ryan Smith from Qualtrics, was he said, you know, he's like, what are you doing about friends and family? And I'm like, we're not going to do it. Because you can take 5% of an IPO and allocate it to friends and family. And I didn't want to get my aunt, like, over— if the stock went down or something, I didn't want to have to explain that over Thanksgiving dinner. He said, no, no, no, you're thinking about it wrong. Think about the people. Who, if they owed you a favor, that they could make a meaningful difference in the future of Cloudflare, and then offer them the ability to invest in the IPO. Yeah. And I was like, is that legal? He's like, check with your lawyers. Check with the bankers. But the answer is yes. And I was like, some people are going to have conflicts. He said, it doesn't matter. Even just the fact you offered it, even if they can't do it, will, will mean that, you know, um, that, that, that they'll always remember that super fondly. And it was incredibly good advice. And we made this crazy list of all of these, you know, people who, you know, who were like, you know, at some point that might be an important relationship, uh, for us. And like 75% of them said yes, and, and they all made, you know, a ton of money as a, as a result of it. And it was, um, It was one of those kind of—

Now your aunt doesn't talk to you anymore.

Well, yeah, my aunt doesn't talk to me.

She's like, I missed out on a 100,000 basis point move.

How could you? I am always comforted that we actually— so we priced it $15 a share, and it went up to— and that's actually the other thing that's interesting. Everyone— like, Bill Gurley and I have fights about this from time to time about IPOs. Like, you have a lot of control over how much the pop is going to be. And we sat with bankers, and we were like, Listen, we want it to go up about 20%. And they were like, OK, if you want it to go up 20%, you price it right here. So we priced it right at that point. And it closed the first day at $18, which is exactly up 20%.

OK, but just to steelman a little bit, is that a function of your business? Not to be rude or anything, but there are some businesses that are just meme stocks out of the gate. And they're in some weird thing. And there's some froth. And they have less control. Or is it always controllable?

Well, and part of that was like, did you think the fair value of your business was 18% or 20% higher than where you priced it? And so you wanted the market to sort of reprice, and you didn't care about not selling?

Yeah, I mean, if we had priced it at like $13, it would have gone up more. But the problem, again, you want to play for the long term here. And so again, my fight with Bill is, you know, he says, you know, that you should, you should do, you know, some sort of system where you're, where you actually, you know, price it at whatever the top price is. The problem with that is if you look at all the companies that have done that, so the Airbnbs and, and, and others that they basically, you know, that's, that's great for your private market investors. That's great for the VCs. That's great for any of the, the, the founders that are taking money off the table. But you got to keep running the company like the IPO isn't the end. It might be the end for the VCs, but it's not the end for the operators. To the company. And you want to be able to kind of build that relationship and build that over time. And so I think, you know, we worked incredibly hard to be very sort of choosy on who are kind of who we gave allocations to. And we went through every single name on the allocation list. And, and I think as a result of doing that, you know, if you look at like who our top 10 shareholders are, it's been incredibly stable over, over time. And they're folks who give us, again, great advice and have been really, just really, really great to work with. So I am sure that there are times you don't have it, but I was surprised at how the banks have a really good sense of, if you price it here, here's what's gonna happen. And in our case, it was, you know, to the penny. Last question.

How are you thinking about long-term planning as a CEO? When we're on these exponentials, the example of agent versus human web traffic, you had more data than almost anyone on the planet and your prediction was still off, or your original prediction was still off by 18 months, 24 months, something in that range. So how are you thinking about overall business planning when model capability is increasing and you're just seeing sort of exponentials everywhere?

Yeah, I mean, I think a couple of different things. So one, I think we're trying to make bets on, on people who, who can, who can really function in, in really dynamic environments. And that for us has been somewhat different than I think a lot of other people. A lot of other people have cut back, for example, on kind of early career hiring. We, you know, Kleffer's about We had 5,000 employees, a little bit less than that now. But we hired 1,111 interns this summer, straight out of college. And they're amazing. Like, they're just killing it because they are so native to these tools. And again, I think we've always thought of our job as training the interns. But this year, the interns are helping train a lot of us in how to do these things. Really well. So I think staying super dynamic and then, you know, from our perspective, just making sure that we've got optionality going forward. So, you know, it's, it's, it has gotten tougher to get equipment today. Memory prices are through the roof. Things are harder. But we've, we just have a team that's always constantly trying to figure out how to refine that. And then generally, I think we're, we're trying to be big adopters of AI, not just in developer land, but also across the entire business. So we built something called Cloudflare OS, which allows, you know, anyone on our finance team, legal team, you know, accounting team, finance, anyone that's out there at the company to be able to have access to tools and really figure out how their job gets done. The clever thing that we did to seed it was we actually set up this email address. And initially, we told the team that it was this magic AI model. And if they just wrote to the email address, for us and ask it, I have this part of my job that I need to get done. How do I get that done? It would— sometimes it would ask some more questions, but it would actually give you a response back. What we didn't tell people was actually it was a team of humans behind the scenes that were initially receiving these emails. And then they were using AI systems to flesh things out. But what they were really doing was recording all of the key jobs to be done within the organization. Is, you mean you have to— if you're going to have AI systems that can help facilitate this, you actually have to record what are the steps that people are doing as, as, as a part of that. And so as a result of that, we've been able to seed this, this Cloudflare OS now with the ability to do things like if you're on the sales team and you have to do like an account plan, you know, you can literally just type /account plan and then describe what it is that you want to do and it will output that. And that's made, that's made our team so much more productive. And again, I think it's giving us the flexibility to be able to respond to, you know, whatever comes next. Very cool.

Final, final question. Are we going to— are we going to see an exponential increase in the number of interns this summer? Yeah, we see 10,000 interns. I don't know.

I mean, I— we've got to find a place to put them all. And so it's— I mean, our office in Austin, like, we've— it's in It's a big space and awesome. We run out of space there because we have so many interns there. I just think that there— I think a lot of companies are doing this wrong where they're saying, like, we're not going to bet on the new people coming out of school because AI is going to replace them. I mean, these kids know how to use AI better than anyone else. So we're going to bet incredibly heavily on that. And so far, it's paying off. I think the person who eventually replaces me might literally be one of the interns today because they're just so, so, so smart.

It's happened before. I mean, I don't know if Satya Nadella was an intern, but he has that famous video of him as a product manager demoing Excel. And yeah, nothing like a company man.

I remember, I mean, I was on a bunch of calls with Satya before he was CEO. And he was just a product manager, you know, at that. So I think just if you can find really great people and be able to move on.

He's the GOAT. That's a GOAT. He's a GOAT. You're the GOAT as well. You know, whenever you put up 100,000 basis point move in your stock over just a couple years, you're in contention for GOAT, I say. But thank you so much for coming on the show. Always a fantastic time.

Thanks for jumping on. Good to see you guys.

Good to see you. I haven't talked to you since the acquisition.

Thank you. I appreciate it. We'll talk to you soon. Have a good one. Let me tell you about console.com. Console automates, builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets. We will be joined by Vinod Khosla.

And while we wait, Mad's summer vacation is looking good. And let's see what's on the schedule. What is on the schedule? 6 AM, he's waking up on the wrong side of the bed. Okay. 7 AM, he's split between judging a book by its cover and beating a dead horse. Okay. 8 AM, this is where it gets spicy. He's gonna poke the bear. He's gonna poke the bear again. And shortly after that, he's skating on thin ice right into walking on eggshells. Wow. And at 10 AM, he decides, all right, it's time to address the elephant in the room. Okay. At 11, gets a bit dark. He's going to cry over spilled milk. Yeah, but at noon he's letting the cat out of the bag. And that's where it really ramps up. He's starting at 1 to bark up the wrong tree. And again, but at the same time holding his horses. He's double bucked. But then at 2 PM, he's going to add fuel to the fire. That's fantastic. And I can't really think of a more perfect summer day, John.

You know, you know, there's— you can share your calendar with someone else and you can share it privately so you can just see that I'm booked, but you don't actually know if I'm having a meeting with someone. Maybe I'm interviewing Vinod Khosla. You won't see that, but you could build a plugin that changes the book to just these things. That would be delightful. Maybe the next Google April Fools' Day.

That's super intelligent.

I would love that.

Yeah, if somebody just can only see that you're letting the cat out of the bag. Yeah. That's pretty exciting. It's kind of mysterious.

Okay. Don't interrupt him. It's an important meeting. He's letting the cat out of the bag. But hey, he's just beating a dead horse. Maybe I should go. Grab coffee with him, see if he wants to spin out on that.

There's a website, uh, turning Joe Wiesenthal's tweets into full articles.

This is crazy because there's a, there's a live show that turns his tweets into like minutes of just back and forth conversation and banter, uh, but somebody made a version of our show that's just web pages. So they scrape every Joe Wiesenthal tweet and they turn it into a full SEO-optimized article. This is great. Apparently, uh, it is great, and Joe looks great in this photo. He's looking good, he's looking quizzical, he's having fun. Well, uh, let's bring in our next guest, Vinod Khosla, the founder and managing director of Khosla Ventures. Vinod, thank you so much for taking the time.

How are you doing?

I'm doing great, great to be here. Fantastic to have you. Uh, I want to kick it with some of the recent dealmaking that you've been doing. Your outlook on artificial intelligence is obviously front and center, but what has been the most recent fundraise that you've done? Why did it get you excited?

Well, obviously there's a lot. It's all on yesterday or today. We are working a lot and doing a lot of new things. That I'm pretty excited about is a company like Pramana, but more importantly, the broad category of auto-formalization, which isn't talked about much. But humans aren't very good at talking to AI, describing what they precisely want. And that's where auto-formalization becomes important. And companies like Sanaa become very important.

How did you get into autoformalization? And I guess the big meta question is, as an investor, you have to consider, is autoformalization on the direct path for the other labs? How do you think about competition in this area? Investors are going back and forth, and it's been fascinating to see things that appear to be directly in the path of the big labs, they're still growing revenues, $2, $3, $4 billion. We're seeing $60 billion acquisitions. Everything seems to be going well up and down the stack. But why is there any— why is there a particularly big opportunity right now?

Well, the labs, they will do more and more over time. But I still see the opportunity set being very broad for venture capital and bond notes. In clever— are coming up with great ideas. One example looks at the TACO and says it's not very precise because it's written by humans. You can make it into a machine-checkable language. Mathematicians turn all their problems in this specific language. It is hard for humans to understand. Believe. Auto-formalization allows that kind of formalization so computers can operate on it precisely, not probabilistically. And that's a really important thing. If you look at today's world, AI systems have been awesome in the rate of capability development. They are awesome, but they have some pretty big gaping holes. They still hallucinate. I don't think I don't think hallucination is going away anytime soon. Yeah. So the reliability is low. Humans aren't great at specifying what they want. The specification problem and then verification of what— are you getting the right answers or are they hallucinating? Those are all large, interesting problems which are open for entrepreneurs. Things can be built in certain domains. Where you build a lot of value and a lot of capability that the general models don't do.

How are you thinking about applications of this technology in particular? It's been so hard to predict the original GPT. No one was really expecting a knowledge retrieval service. It just sort of went viral when ChatGPT launched. Coding agents. Some people had predicted those, but it was sort of an unexpected emergent capability. Do you expect that auto-formalization as the next critical frontier of AI just makes all the existing applications better, or is it actually going to unlock new applications?

I think it will allow humans to use AI where it wasn't previously possible. If you want to know what your bank account is, you can't have a hallucination. If you have a medical problem, you can't have hallucination. So auto-formalization will significantly enhance existing models in domains like tax law. You know, you can have lots of tax startups, but they can't formalize. You take tax law. It's a very specific thing. And that's an area Pramana is initially to say, can you formalize the tax code so computers can work on it precisely and not be subject to probabilistic answers or likely answers? So it's very, very important in certain domains to have precision, reliability. Verification, um, and, and that's where auto-formalization can significantly enhance the applications and still leverage all the power of LLMs.

Well, we're excited to talk to the founder of Permana in just a minute. Uh, we've been having a few technical errors, so, uh, we will end this here and hopefully we can have you back on the show soon to go deeper. Uh, thank you so much for taking the time. To chat with us today. Vinod, it's been a pleasure.

Thanks for jumping on. It's an honor.

Have a great rest of your day. Great to be here. Cheers. Goodbye. Let me tell you about Public.com. Investing for those who take it seriously. Stocks, options, bonds, crypto, treasuries, and more with great customer service.

We have unfortunate technical difficulties. I couldn't really hear anything personally, but fortunately the founder is coming on discussing in just a few minutes right now. So we're going to hear it.

So we're going to bring it again and again. Sorry about that. Nikesh Arora is on the timeline tweeting. I like that. He's— you don't think of him as like a big poster, obviously. 73,000 followers. I think he's a poster. He's successful. But he said, congratulating Prime Minister Narendra Modi on becoming India's longest serving elected prime minister. 4,399 days of leadership earned the trust— earned through the trust of 1.4 billion people across three democratic mandates. From lifting 250 million people out of poverty to making India the world's fastest growing major economy, PM Modi's tenure has been nothing short of transformational. Look forward to a continued US-India partnership. Love that. Yogesh says, bro went from stopping global ransomware attacks at Palo Alto Networks to doing math calculations on PM Modi's calendar milestones like it's his final exam. I like it.

Just in quotes too. It's great. And Nikesh appreciates it.

There's been debate over Beverage Media. Have you seen this? This is a very— I saw this, 8 likes, but I think this is an interesting post worth discussing. So Vasa, I don't know exactly this account before, but Beverage Media doesn't have the good nuts Nuts to call out celebrity brands objectively because the economy is fanboying and CPG. Celebrity founders sell tickets to events, but they don't always move the industry forward. I think he's probably subtly subtweeting the maybe Tom Brady launch of the Nuts, which is a reference to the coconut water brand that Tom Brady launched. But this does happen all the time. There's a lot of celebrity brands. We've had We've had some celebrities on the show to talk about their brands. It's fun. Sometimes it works, sometimes it doesn't. That's the fun part about interviewing a celebrity about their brand. You kind of get to clock it and be like, "Meh." Like, you know, "This person's really on to something." I think why they named the brand that is they fully ran out of other names.

So that was literally the last possible name that they could choose.

We're not getting an invite. Tom Brady will not be coming to our house. To our event and selling tickets. But it's interesting, because, yeah, BevNet reports on CPG brands. They're one of the insider media, sort of the TechCrunch of CPG, at least in beverages. They also have Nosh for food. And there's a number of other CPG media and critics. But if you gotta have one of those celebrity founders at your event to sell tickets, you get into this weird— you get sort of audience captured. I don't know, celeb captured. You just get raptured in in the aura of the celebrities, and you have to stay out till 2 AM partying with them. It's craziness out there. But Beverage Media, I think it's just interesting in the backdrop of more and more celebrity brands. Although it does feel like we're at peak celebrity brand at this point. And I feel like the more modern celebrity strategy is just invest in companies that have a number of good VCs on the cap table, be happy with that, that you can still— or a farm. Yeah, okay, disagree.

Explain. I think celebrity, celebrity brands have always been a thing. They're just way more in your face because now celebrities have their own channels that they can communicate.

I'd say they're less in my face. I, I say we're on the decline. Uh, top celebrity brand?

No, I'm quoting John Field. No, I think it's just getting— it's getting easier and easier to start a business. I think you'll see more celebrity brands. I don't necessarily think you'll see more breakout brands.

So your claim is quantity is increasing.

I think it's probably getting harder.

Quality is decreasing. Quality. More distribution.

Yeah, yeah. Greater, greater. I mean, there's going to be some breakout celebrity brands. There's going to be big exits. There's going to be as many failures as ever, right? I mean, my favorite example was like Travis Scott a few years ago launched an RTD like canned cocktail thing, shut down, I think, within like a year, like maybe more. And so I think you're just gonna see more and more and more.

Would you put the Johnny Knoxville Neo Cloud in the bucket of celebrity brands?

That is something I would like to see. Less beverage, more Neo Clouds. I want to see Johnny Knoxville levered up, right? I want to see—

Racking GPUs himself. Yeah. I want him like operationally involved too.

Not just like— Not just a pretty face. Endorsement. Yeah. Not just a pretty face. So anyways, I think more on the horizon and it's not necessarily, you know, a lot of, I don't know. I feel like some people don't like this, but I think it's fine. It's fine.

It's fine. You heard it here first. Jordi Hays, it's fine. You know what else is fine? The New York Stock Exchange. More than want to change the world, raise capital at the New York Stock Exchange. Our next guest might be there soon with Pramana Labs. We have Ram John, who's the co-founder, in the waiting room. Let's bring him into the TDP and Ultradome. Welcome to the show. How are you doing? Thank you so much for taking the time.

Doing great, John.

Jordi, since you're here first time on the show, please introduce the company. We heard a little bit about it. From Vinod, but I want to hear it from you.

Yeah. Bravo, by the way, of getting— being able to get Vinod to just like personally pitch your company for 15 minutes on internet television. It's a— it's a good— it's a good sales hack. Good pitch.

Thank you. Thank you so much. At Pramana, we try to make AI more provably accurate. Yeah. The current frontier of of AI focuses on recall, which is more along the lines of how do you make more answers come up. Yeah, but we are trying to focus on very specific mission-critical domains like tax, legal, healthcare, and governance, where a wrong answer could be catastrophic. You wouldn't want AI diagnosing you wrong and ending up in a catastrophic situation. So that's where we focus on, and the technology underneath it is formal verification. Yeah, I would like you guys to dive a little bit deeper and I can get more technical as it—

Yeah, I'm just wondering about the actual rollout of this. People are using AI tools for healthcare now. They've been using WebMD. People have been misdiagnosed with cancer by random internet posts for decades. You go to WebMD, you're like, I have a headache, and it's like, you got brain cancer. And then you freak out, go to the doctor, and And they're like, what are you talking about? Take an Advil. But so like, who exactly is not adopting AI right now because of the hallucination problem?

One, two major examples currently are like Big Four firms, which are tackling tax and legal cases. You would've heard about the Australian government suing one of the Big Fours, stating that there is a hallucination which ended up happening. And AI can actually make up laws on the fly. Yeah. If it wants to suit a particular situation, it can make up laws on the fly, which can lead to catastrophic situations. And another thing which we have noticed is also that insurance providers, major insurance providers of US, have also moved away from insuring AI outputs. So there is a very specific clause they have introduced to say that we will not insure if AI is behind whatever you have said. So those are some things which have come out specific to domains like tax, legal, and healthcare. For healthcare, we can even see the models themselves typically stating that you need to go to a doctor to verify. We will be absolved of the responsibility.

Yeah. How much of this—

Yeah, imagine your CPA being like, great, I just finished your return. It's 99% chance it's accurate. I mean— There's 99% chance.

The steel man here is like, that is how I feel about my lawyer. That's life. I love the people I work with, but humans do hallucinate. And so, if you give me something that doesn't even need to be superhuman, something that's going to sit alongside that person, I'm pretty happy to buy. But I understand that I would buy more if I could be more sure. Absolutely. And so, I understand that. What I'm confused about is like, how are you thinking about building new foundation model, new LLM, new AI technology that bakes formalization into the actual workflow versus like, you know, we were talking to Dr. Karp from Palantir last week about, you know, the idea of like you've built this ontology, you have your ground truth, you have your system of record, and then you can have the LLMs hallucinate all over it. Over the place because at a certain point they're going to run into the ontology, and that's going to be the formalization process. Is there something— is it an extension of that, two systems running side by side checking each other, or is it more of an entirely new system?

So it would be a couple of systems interlocking with each other. We actually have four layers in the stack. I can go through an example of how it happens. If you end up asking whether a particular transaction is taxable in Illinois. Yeah. You have the first thing which we do is actually a fair bit offline for tax specifically. We are formalizing the US tax code. So we are taking a very formal domain but expressed in English today and we are converting that into Lean, which is a proving language. So when we So when we do that, we get it ratified by experts. So our work is deeply involved in formalizing specified domains and ensuring that we get it ratified by the experts themselves too. In that case, we, we do use a lot of LLMs because the current knowledge base is already not formalized. It is present in English. And once we have the whole knowledge base, whenever a question pops up, we convert it into a series of constraints. You would have noticed Vinod also talking about how humans are not great at specifications. So we ensure that we ask the right set of questions so that ultimately we can give you a reliable answer. Post that, we have a solver and prover working in tandem to ultimately give you an answer along with a proof of correctness. So the key difference between an ontology-based approach and a lean-based approach is that a lean-based approach is intrinsically verifiable. Your answer comes with a proof of correctness. The proof is something which a mathematician trusts. That's supposed to give you more comfort in a way. And like you suggested, uh, you're right in saying that even, uh, like, best humans might make mistakes, uh, but if you actually have a mathematical proof encoded in Lean, you can choose to trust it. And that's the frontier which we are aiming for, and it involves like fair bit of work between formal verification experts who are a deep part of who we are at Pramana Labs. Interesting. Working along with CPAs, frontier CPAs, frontier lawyers, and frontier doctors.

Yeah. How are our models getting better at interacting with Lean? I've seen a lot of progress in frontier math and the different really hard problems are getting solved. Sometimes they don't use Lean and that's sort of a flex. Sometimes they do use Lean and they perform even better then. But is that getting just table stakes now? Because I think your average white-collar worker doesn't know how to interact with Lean.

Absolutely. So the models are getting better at proving with Lean, but we are building more foundational models along the lines of formalization in the first place. If you say something in English, How do you convert that into a construct in Lean? That's where our focus is on. If you think about it, Lean is a system which has been around for the last 12 years. It has taken humanity around 9 years to formalize a significant portion of math. We are taking that same technology and we are trying to blitz through tax, legal, and healthcare. Where we are identifying what is the core knowledge base and then we are converting it into lean, uh, where you can reason on top of it provably.

That's very cool.

What were you doing before this? Uh, I was a machine learning engineer at Google, primarily focusing on Google Maps. Interestingly, I draw parallels over there because I was focusing on getting addresses and phone numbers accurately in Google Maps. Sure. If you think about it, Google Maps is like a very, very messy real-world domain, and we have been able to make it trustable. A lot of people do trust Google Maps to a very large extent. Yeah, I intend to bring the same level of rigor to multiple other domains along the way. Very cool.

Well, congratulations. Thank you so much for taking the time to come on.

Yeah, great to meet you.

We'll talk to you soon.

Have a good one. Thank you so much.

Cheers. 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. Uh, that's right. So there's a bitter feud going on.

There was a— there was a good, uh, comment over on the X chat. Yes, Steve. He said, we're on the topic of celebrity brands. He said, flip it on its head, Jordi. Brands by average Joes, like a painter or the owner of a lawn care company, instead of celebrities. I think there could actually be something here. I think he's joking, but imagine as a brand, you're trying to differentiate, you hire an actor that's just playing the role of an average Joe. Hey, this is Average Joe Electrolytes, right? I'm not a professional athlete. I'm just a normal guy. I'm just painting houses, right? I'm just painting houses. I get a little dehydrated. I take my Average Joe Electrolytes. It's relatable.

I like it. I like it.

The average Joe brand meta, Steve is calling it.

I enjoy it as an idea. I enjoy it as a joke. I do think that there is a version of this, which is if you walk through an Arowana or Whole Foods and you pick up random CPG products and you look on the back, there's often what's called like the love language or whatever. So, it's like a couple paragraphs of text. A good example is Dots pretzels. There's some sort of like pretzel twist that's very popular right now. It's a little spicy. It has the name of the founder. And it tells the story of like an average Joe. The Buzz Ball, for example, that is an average Joe brand. It was not a celebrity brand.

Disagree. Explain. There's nothing average about that founder. One of the most elite entrepreneurs. After she started the company. I'm going to make the case that she was always elite. Maybe the Buzz Ball woke it up.

Maybe, maybe, but there are a lot of brands that their whole origin story is like, I was just a normal person. I had a problem for myself. I was going for olive oil and I smashed the bottle, so I made a squeezy bottle. Like there's all these stories that aren't celebrity brands, at least at the start. Of, you know, rags-to-riches stories that are told sort of quietly on the back of the packaging. And then slowly over time, eventually Unilever buys the company and takes that off its back.

Yeah, I just like thinking of it as like, from the marketing lens, a brand that is like, doesn't have this like heartwarming story.

It was just like, yeah, we just hired a guy to be like, "Yeah, just be the guy, just be the average Joe." Just be an elite snowboarder like Nima. Starts Salt and Stone, you know.

He wasn't a celebrity before, although he was a fantastic snowboarder. Before we bring in our next guest, Marky, let's put— pull up this video. Somebody put up fake tech ads. Okay, what's going on here? Hmm, I don't—

I love ads, but I hate fraud, so I don't know how to feel about this. This is going to be peculiar. I don't think there's anything wrong here. With fake tech ads? That's real ad space that could have been sold, Jordi. You're destroying shareholder value by running fake ads. We would never do that. We did do that early in the days. If you go back through the archive, Dr. Squatch had that feel. Oh, Vince Vaughn. Vince Vaughn running a—

We have the video here. OK.

No, no, no. I know exactly where they're going with this. We put the Q in Q11777. This is hilarious. What if Texas was upside down?

I mean, this is actually how normal people view Texas.

What if the Rizzler was purple?

Purple. Wireflies. You pay it. We pay you. You pay us, we pay you. Wireflow.

Wow. What does this say? Dennis can tell you.

Dennis can tell you. Brain is—

I like ZipLink is now—

They put up a lot of these.

Wow, they really went hard. Who is behind this? This has to be a tech insider or someone making fun of people.

ZipLink is now Frogle. The cloud-based online safety you know and love, now in the palm of your hand.

They put up so many of these. This is such a great stunt. Is this a— if this is a launching campaign, a launch campaign for a tech brand marketing firm, it's genius. And whoever's behind this should actually be calling every tech company and saying, "What if forks were spoons?" "You want your ads to not look like these? We know how to make good ads." What if forks were spoons?

What if forks were spoons? Go to cutlery.ai.

Lots of people have been asking that question. Recently. Anyway, very fun. Uh, let me tell you about Cisco. Where's Cisco? Critical infrastructure for the AI era unlocks seamless real-time experiences and new value with Cisco. Our next guest is Markie Wagner from Poetic. She's the founder and CEO. Future Hall of Famer. She's a prolific Mafia player. What's your favorite role in Mafia?

Ooh, Mafia. I think Who doesn't love to be fascist and mafia? I, you know, you always gotta hope you're getting the fascist card, but, uh, I thought it was just mafia, townsperson.

Oh, sorry, I'm the secret Hitler. Oh, you're the secret Hitler.

Yes, yes. Um, uh, yeah, I mean, mafia is the obvious one.

Yeah, it's always power. It's, it's less empowering to be the townsperson because you feel like you're just gonna get killed. You have no superpowers. But running control or sheriff, that can be fun. But then if you get killed off early It's the end. I'm just thinking of it because the, the latest episode of Mafia Episode 2 from Founders Fund dropped and you're backed by Founders Fund. But since this is your first time on the show, why don't you introduce yourself? Tell us how you wound up in a position to raise money from Founders Fund.

Cool. Yeah. So I'm the CEO of— and wait, okay. Yeah. So I'm the CEO and founder of Poetic. Yeah. And Poetic's an AI system that will learn and execute super complicated processes in some of the biggest companies in the world at over 99% accuracy, which is, which is quite hard.

Yeah. So what is the process for obtaining and documenting the workflow? This feels like you could do something very database-driven. You could build an ontology, you could just wait and maybe the models get better and stop hallucinating. There's been debates over the various approaches. Like, what did you pick and why?

Wait, before we get into that, like, talk more about the kinds of workflows where, like, you know, we caught up a few weeks ago and you were saying, like, some of the customers that you're talking to, even if somebody came through and they were like, our AI system is 98.5% accurate, that that would actually create, you know, hundreds or thousands of hours of, like, issues in an organization., and it would actually, like, rolling that kind of system out would have sort of like negative enterprise value.

That's just a pitch problem. If you're pitching a software like that, you just got to tell someone, our system has 9,000 basis points of accuracy. Yeah. This is the goal.

Put it in basis points. Put it in basis points. Anyways, talk more about the problem before we talk about the solution.

Yeah, so I think the problem is, you know, one of the things you've seen is, you know, obviously, AI is incredible at writing code and has really crushed that. But, you know, a lot of the main processes that are at the heart of these giant businesses have remained pretty untouched by AI. And the reason why is that the rules that govern them, you know, the 10,000 secret rules, they live in people's heads, right? And these rules need to be followed every single time. You know, we work on things like anti-money laundering and underwriting and fraud investigations where every single step matters. And to actually get from where we are now to those running on machines, you need two things. One is a system that can learn the process, not the 100 pages of written down stuff, but the 10,000 secret rules that live in people's heads that they've never written down before. So that's part one. And then you have to be able to run them with many nines of accuracy. Not, you know, I was talking with one CEO, he's like, you know, 80% on an eval is great. But in underwriting, it's unusable. And I think that's the view for a lot of these processes that are at the heart of these, like, really big and old businesses. So you got to do two things. You got a system that can learn all the rules and then run them with nines. And if you have that, you can get from where we are now to this version of the future that I think everybody's really excited about. But you need to build that bridge.

Not everyone's excited about it. AI has a very low approval rate, but, but, but Tech leaders are certainly excited about it. But so why have you started with such big companies? Like, I feel like you've been, you've been, you know, at the heart of the startup ecosystem for so long. If you came on here and said, oh, we have so many, you know, of my friends' companies on board, these are the named customers, SoFi, Chime, AIG, like these are large institutions at this point. Why? Why start at the top feels harder.

Yeah, I think our view was, um, when you're, when you're building something like this, if you build something that's less powerful, it's hard to sometimes to make it more powerful. Um, like a lot of these drag and drop tools, they were simpler, they were easier, they could do simpler things faster, but they just couldn't handle super complex things. Um, and so you kind of get kneecapped in terms of what you can represent and build. And the view was, hey, AI is going to be doing a lot of this. Place, like writing of the software. You'd be in a better place optimizing for the power of the thing. Can it express and run these 5-hour processes that require those nines? And then everything else becomes really easy, actually. And I don't think that people are going to be having a platform for the easy stuff and the hard stuff. I think they're just going to be running all of their processes in one spot.

I mean, you mentioned like, tens of thousands of hours of information stuck in people's heads. Like, what is data collection? What does data collection look like? Conversations? Like, it feels like, yeah, like, is it the McKinsey model or is it the, you know, self-serve software model? Like, where do you want to sit?

Yeah, I mean, what's interesting is it's evolved over time, but the place that it has ended up going is looking closer to data labeling. So what happens is a big company company will say, here's my biggest process. There's 100 pages of documentation. And you know that that's only 20% of it. So the question's like, how do you get that other 80? Mm-hmm. Well, what we'll do is we'll take that operating procedure, we'll generate the AI operating procedure. It's written in step-by-step English. Mm-hmm. What our system does is turn that into code under the hood and you run it. Mm-hmm. And when you run it and you put it in front of those experts, they have a ton of opinions. They have a lot of feedback to say, sure, you forgot that the threshold's actually a million, not 10,000, right? All these little things. Which get merged back into that document. And you're sort of doing that more and more automatically so that it looks almost like training a program or something like that, rather than sort of like long calls or, you know, process mining in like the normal sense. And so more and more it looks like people giving feedback into the system directly, and that updates the rules that are written in it.

And so you're using AI models to generate that code. That is then deterministic, which gets you the reliability that these companies need, correct?

Yep, yep, exactly. So the, you know, the source of truth is that AI operating procedure. It's English, but what I see most is turn it into code. So when it runs, if the world's the same as yesterday, it's just gonna run as code. Great, nothing to see here. But if something changes, like the column name changes from month to months or the save button moves, only then will AI go in repair it, look at what the English goal was, and then update it. And so if you do that, you can get the best of, hey, it's code, it's precise when it runs. But if something changes, instead of breaking, it will actually just kick back and recover. And that's important because a lot of this work code couldn't do alone and agents couldn't do alone. So code is very static, right? It's like very brittle. So even one small a small date change or something could break the whole thing. And so that's one side of it. Agents, on the other hand, are pretty improvisational and they think step by step. And, you know, when you're figuring out what to do as you go, eventually you're going to make a mistake. This is something that's kind of in the middle where Skoda thinks the same. AI is there to test, heal, recover if things are different. And that's how you can kind of get those nines.

So are you hiring for deployed engineers? Like what is the role of engineering? Engineering in your organization at this point?

Yeah, so we do, we hire tons of forward deployed engineers from all the best spots, you know, whether it's, you know, places like Palantir or, you know, even like Retool or Scale and all these other sort of new places that have had their own versions of forward deployed. And yeah, I think it's interesting. It is over time looked less like extremely just only focusing on engineering, like hard engineering. It requires being able to change how people operate, right? Yeah. I can write the best code now, but even if you have the most incredible piece of software, you still have to change how the business organizes around it. And so the people who understand like business and can think about, hey, like what is the best possible fraud process going to look like? And you know what, how should we reorganize the business around this new kind of thing? And they understand AI are the ones that are just like totally crushing it.

Yeah, that feels like an entirely new skill set, like much more people-driven, but also like forward-thinking in technology. I don't know. It's an interesting new skill set.

How has enterprise sentiment around just AI changed during the course of building the company, right? Because like every, you know, it is, I would say it's a roller coaster in some ways. It's like going up, up into the right, right? There's generally, more excitement about the potential, but at the same time, there's these sort of like period troughs of, of disillusionment. And you guys are coming out of stealth at a time when, again, I think companies are more excited than ever, but at the same time, sort of like understand the overall shortcomings. And we had Karp on the show last week and, you know, he was just saying like a lot of these deploycos coming in, they're just trying to like— they're, they're, they're trying to deploy— their goal is to deploy AI, but they don't fully understand all of this like business process under, under the hood. And he makes some good points.

Yeah, I think sentiment has evolved quite a bit. You know, I think it was extremely excited earlier in the year and then, you know, around the board meeting times, more and more CEOs would come to me and say, hey, like, you know, how do I get ROI here? And, and, you know, is this what your other customers are seeing in these domains? And so I think now people have realized, like, you cannot just throw an agent at a problem and expect to see the result that you want, right? For an agent to do a useful bit of work, it needs to learn how to do that work and run it, like, quite accurately. And that knowledge transfer between How's the work done today? And getting it written down enough to where AI can run it is just hard. And it requires, you know, we sort of jokingly call it the Great Migration internally. Like, you have to go and migrate these, like, tremendous amounts of rules into something that AI can touch and improve and evolve. And if AI can't touch it, it's not going to be able to help it. And so, you know, I think deployment is really important because, yeah, until that transition happens, it's gonna be hard to just throw tokens to, to see better outcomes.

Give us some backstory on what you were doing before, how you wound up here.

Yeah, so I got my start, I got extremely into AI in middle school after reading too much sci-fi, you know, a lot of Dune, a lot of, a lot of mainly Dune.

Is that the message in Dune? I thought Dune was the, But, but like, you know, right? I guess, yeah, yeah. You want to make a good outcome?

Well, it just felt like it was gonna be really important during my life, um, you know. And, you know, you read and you watch sci-fi, you know, the biggest difference between the future and today, it's like mostly machines thinking. Like, that is sort of the main AD sort of BC moment, um, in like hard sci-fi.

In soft sci-fi, it's like time travel and faster than light light speed and like aliens.

But I take it— Yeah, exactly. Moon is made of cheese.

That's the type of sci-fi I'm trying to read. That's the white pill. So then walk me through the consulting work that you did and how this like evolved out of that. Was there like a clear moment where you're like, I'm stopping that and starting a company? Or is this an evolution or a change of structure?

Yeah, so I was initially got my start in research. So I was, I was at Stanford working in AI, like Waymo and, you know, Google. Um, and then, you know, one day I sort of realized that, you know, all the things I'd built, like, you know, didn't, didn't matter too much. Um, and that was Friday, and I ended up dropping out on Monday. Um, and the idea was like, hey, you know, we've had software for decades. What are people still doing and, and why? And I felt like I didn't really understand anything about how the world works. So to your point, started consulting, and the idea was like, let's just go into to some of the oldest companies around and understand like what is software still not touched and what happened there. And when you do that and you start doing some of the work yourself and like going into, you know, I went to North Carolina and you're watching people who have done this stuff for like decades. You realize that a huge amount of the work is really just operating procedures, documents, and people are just following them. And that this class of work everywhere. And, you know, whether it's underwriting and claims and insurance or onboarding customers and fraud and banks, it's just still sort of done by, by people, and software hasn't been able to go there. The reason why is because the second you write all this code to do that process, something will change, right? A button will move, a column will change, um, or maybe even the process changes. You want yearly instead of monthly transactions, um, and when you, you know, automate and see that happen enough, you, you realize like, hey, there's a missing kind of material here that can but still have guarantees. And I really just waited for the models to get better. I knew some of the researchers and said, "Hey, the models of today are not it. Please let me know when they get good enough to do this stuff." And I truly just waited, so. That's great.

Well, congratulations on the round. How much did you raise?

I wanna hit the gong. Oh yeah, so we raised $50 million.

Wow. Congratulations and thank you.

Who besides FF is in?

Yeah, Kleiner Perkins, First Harmonic, Genius Ventures, all participated in the round.

Round of applause for Genius. Very good, Ben and Adam. Love Genius.

It's great news. Well, have a great rest of your day.

So great to have you on the show. Congratulations on coming out into the world, and I'm sure you'll be back on very soon.

We'll talk to you soon. Thank you. Cheers. Goodbye. Um, let's see. Canva AI. We gotta talk about the dog. The dog walker and the dentist. The bitter feud between a dog walker and a dentist over who owns the beach. A lakefront owner likened his neighbor's shoreline walks to a home invasion in a dispute that could be headed for the Wisconsin Johnson Supreme Court. Who you got, Jordi? Knowing nothing about this story, are you going dentist or dog walker?

Uh, Pure Vibes. Pure Vibes. Flip a coin.

Which one? I don't know. Who you got? You don't know?

It's your job to know. Is the dentist a day trader?

So it's your job to pick someone wildly and then defend it throughout this article. A recent trial in Shore Wood, Wisconsin, had all the trappings of a minor legal dispute. A disgruntled neighbor, a defendant representing himself who called his own father as a character witness. You love that. Hey, Dad, tell him I'm a good person. $313 were at stake. $313. That's $31,300, if you're paying attention. But a And if academic and devoted dog walker Paul Florsheim gets his way, the case will go all the way up to the Supreme Court— Wisconsin Supreme Court, that is— and reshape the contours of shoreline access to one of the Great Lakes. It started when Florsheim started walking his two dogs past the Lake Michigan property of dentist Daniel Domagala, locally known for the time he spends in a tiki-style boathouse and deck that doubles as a surveillance post. From there, Domagala monitors traffic and sets off alarms to scare walkers, swimmers, and kayakers away. He's got an air horn. He's got an air horn. He sets it off when somebody comes by. It's crazy. Florsheim repeatedly ignored signs outside of the dentist's house that said private property behind the sign, only water access behind this— beyond this point. Domagala kept calling the cops, and the village eventually issued a trespassing citation. Rather than pay the fine, and walk away. Florsheim dug in. At stake is what right people in Wisconsin have to take a shoreline stroll. It's high-stakes stuff. And we'll be following this case closely. Of course, one of the most important stories in tech. But more importantly, we have Brad Taylor from Sierra here in the waiting room. Welcome to the show, Brad. How are you doing? Doing great. How are you? You going dentist or dog walker? We won't make you take a side. That's what's Tough question.

That's a tough question.

We're going to ask you the real tough questions. How's business going? Business is great.

Uh, we just crossed $200 million in ARR. Uh, we were— great news! I was hoping to get a gong out of that.

Of course, of course. Revenue, revenue gongs are always the best gongs. Yeah, yeah.

Fundraising is easy. Yeah.

Uh, and today we announced we're, uh, certified FedRAMP High, which really simply put means now federal government agencies have access to Sierra, which I think is really exciting. If you look at the conundrum of the federal government, it's like we want greater services for our citizens, but we have a really big national debt. I think AI is going to be a huge benefit to programs, whether it's getting a passport or Medicare. So this is the first sort of the door opening so we can serve these agencies, which I'm really excited about.

Interesting. I don't think of, like, you're breaking my brain because I don't think of like getting a driver's license at the DMV as like a customer service interaction. But is that where we're going, where more and more of these like form-based requests are expected?

I'll give you an example, like an address change on like an ID, like something like, you know, if— Calling the tax department.

Yeah, there's a lot of these.

So yeah, I mean, there is— if you think about all these government services, you've got Medicare, you've got Medicaid, you've got the Department of Veterans Affairs, you've got, you know, immigration, like, you know, the passport agency. All of them run large call centers. But more importantly, people depend on them. You know, people depend on them. You have, you know, someone on Medicare or you have a vet who depends on Department of Veterans Affairs for their healthcare. It's a really big both payer provider. If you look at, you know, what we do for, you know, folks like Cigna and Blue Cross Blue Shield and you squint, well, it's almost that entire stack in a federal agency. And, you know, for good reason. There's just a very high bar for cloud technologies when you're serving the federal government. And this is the main certification that kind of opens the door there. We're already working with a few agencies, none that I can fortunately talk about today. We work with FINRA, the regulator, So we've always been really passionate about, we want the best technology to be available even in regulated industries. And so it's a big step for us. We're really excited about it. I hope we can contribute to the quality of life as US citizens, you know, and do so in a way that's a lot more cost-effective.

Yeah, it's interesting. Like, would you, is the biggest low-hanging fruit just taking some of these agencies and making them available 24/7. I feel like one of the most frustrating things that I run into is like, call from 9 to 5, and I'm like, I have a job.

Yeah, the other thing is, yeah, the thing like, uh, uh, the difference between interacting with like a public utility versus like waste management, right? Like waste management, like waste management, right? Like, you know, maybe it's not perfect, but like it's It's pretty fast, like you can get stuff done, like it's efficient. And then like trying to do the same exact thing with like a public utility, you know, usually I always personally estimate like, okay, if it would take me like, you know, 10 minutes with Waste Management, it'll take me like 30 minutes with a public utility. And so like bringing that, there's no reason that those two things can't be at parity, but it is entirely, it is almost entirely a, a technology and like process problem, right? 100%.

And there's a couple of factors that I think go into why it's so hard to have great experience with some of these government programs. You know, first, a lot of people do need to call things on the phone, and it's only been the past couple of years that we could digitize a phone call with AI agents. So we've essentially digitized the last remaining analog channel. And everything you said is true. It could be 24/7 as an importantly, it could be multilingual. If you imagine staffing up a big call center and you want to represent all the languages in this country from Spanish, Mandarin, Tagalog, you know, like go down the list, really, really expensive to do. Now you can do it at scale. Each new language, the marginal cost is effectively zero. And then the other part of it is for good reason. As I said, there's just a really high compliance bar for technologies to be available. And as a consequence, if you look at, you know, what waste management or just pick any private company, they can go off and get get the best of breed solution for any given technology problem they have. That is not something all federal agencies can do. And so what I'm really proud of is that CIRA is the best of breed solution in customer experience. We power everyone from, you know, Singtel to Rocket Mortgage on our platform. And now because we have the certification, federal agencies have access to the best of breed solution in this space as well. And so I'm really hopeful. I think AI If you look at some of the parts of the economy, the governments, both state and federal, the healthcare industry, which has gotten progressively less productive over the past decade. It's one of the few parts of the economy that has. I think AI is incredibly important because these are the parts of the economy that need productivity. You know, we actually want to have great service as citizens and reduce our spending. AI is the answer. So we're really passionate about it. We're patriots here. And we love our technology, and so I'm just excited that we can apply it in a great way.

Talk about pilot programs and conversion to real customer. Is that timeline a KPI for you? Is that something that you monitor very closely? I have to imagine that it's speeding up, but whenever you shake the hand of a CEO and they say, "Let's do this," or someone in the federal government perhaps, an agency lead, and they say, "I'm ready. I'm bought in on the vision. The demo looked good." But then there's obviously an implementation time. What does that look like now? What did it look like a year ago? Where's it going in the future?

We track it and we measure it in days, not weeks. And so, you know, some of our favorite, like Nordstrom, you know, from concept to 100% of their phone calls was 35 days.

That's good. You said 3,000 days is good, and then we're back to, oh, you're gaming the stats.

What's neat about that too is we started out at 1% of their phone calls in, I think, 4 weeks, and it went to 100% in just a week, and that's because of the Nordstrom team. You know, actually in Fiserv, which is an amazing financial technology company for banks, just broke the record. So I don't know if I'm at liberty to share the number of days, but they have an amazing president named Divya. She was just top-down, We're going to do this and broke the record. So we have a leaderboard internally about how quickly when a customer talks to us, can we go live and start delivering value? For what it's worth, I think it's really relevant in this world. You can— Financial Times, Wall Street Journal, it's all token maxi. What value are we getting from the tokens? It is kind of in the headlines right now. We don't really have that issue with our clients because our thing is like, let's start answering the phone as quickly as as possible, start driving those operating expense savings as quickly as possible, but as importantly, show that you can actually improve sales, improve CSAT at the same time. Our, you know, as you know, we've been really pioneering this idea of outcomes-based pricing with the idea being you only pay CIRA when we successfully resolve a call or successfully make a sale. It sounds like a small business model thing, but it's big because we don't get paid until it's live. It really aligns like all the incentives. Like, we don't care about the sales process. We care about the go-live process. And I think it really changes, I would say, almost the social dynamics between vendor and customer where we really become a partner because we have the same incentives as our clients. And I think it's a really big shift. And you know how much I'm an advocate for it. But I actually think a lot of the token maxing issues would go away if more companies were just focused on outcomes.

How are you tracking the popularity of the message, talk to a human, or the state, or the statement? No, no, no, no, no.

When I pick up the phone, I say, talk to an agent, agent, and then they put me on the phone with an AI agent. It's great. You know, thank you for that.

You're on my side. So we actually have a name for it. We call it greeting acceptance, uh, which is basically what percentage of the time do people get past the greeting and actually give the AI agent a shot. It's going up. People in general, I think the problem was we had 10, 15 years of bots that were really bad. And so we're just, you're just like paying down this debt because if you were talking to an AI more than 3 years ago, it sucked. Like, let's just be blunt about it. Now you can talk to an AI and it's conversational, it's multilingual, it has access to systems. I mean, the AI agents of today are just like a completely, it's like, horse and carriage versus flying car, right? They're just very different technologies. The problem is in the first 5 seconds of a call, you're wondering, is this one of the crappy ones or is this like one of the good ones? And so we have a lot of techniques on our platform to help people with that. I think the real key is making sure the first greeting is personalized, is engaging, is empathetic. And I think the trend is in the right direction. I think over time— I mean, I don't know if you have this experience. If I have to to call a restaurant to make a reservation, I'm like, I'm going to a different place. Like if you don't have OpenTable or Resy or whatever, like I'm out. I think we're gonna get to the point where people demand an AI because they're not gonna wanna wait on hold, you know? And what's also interesting is these AI agents, once they're properly configured, are faster for me as a consumer. They have access to my information, they can get the things done quickly. We're not 100% there. And as I said, we're just paying down the debt of bad tech. That we've had for the past 10 years. But I'm hopeful in 3 or 4 years it'll be the norm.

When did it become popular to get the, like, there's a 30-minute wait, like, press 1 if you want a callback? Because I felt like that wasn't a thing when I was a kid. But now that feels like a pretty meaningful innovation at some point in the last 10 years where it actually is really convenient to not be like, okay, well, I come on, I'm just going to sit Some guy has a patent on that and has a friend talking to him. Yeah, yeah, I'm sure. But like, that was like the last great innovation in customer service.

Yeah, it was. Well, I think it's going in a really good direction. I mean, first, I think, you know, not having to wait on hold, we should never have to wait on hold again. Like, that is gonna be a thing of the past.

What if the GPUs are on fire? It's like, we'll call you back when the GPUs cool off. We're load balancing right now. Please wait at hold. Maybe. Imagine that world. That is a bizarre outcome. I mean, on that note, are there any— how close are we to thinking about like A6 for voice models? Or we were talking to Matthew Prince from Cloudflare about edge computing for voice models. That feels more relevant than— yes, if you're going to cook for an hour on some deep research project, go do it on the NBL GPT-72, but how are you actually at the point where you're starting to think about that optimization, or is that more a few years away?

I think it's probably not a few years away, but it's definitely in the future, but I'm excited for it. I actually agree with your intuition that, you know, we're entering a world where first the frontier models are actually separating, in my opinion. GPT-5.5, the recent releases from Anthropic, I mean, you can just see it, like the gap between the best models like GPT-5.5 the others, the open-source models, is growing, not shrinking, which is probably not something we would've predicted a year ago, or at least like certainly not what I saw on X and the press. But we've reached sufficient intelligence for a lot of different domains. Like you don't need, you know, a multi-billion-dollar, multi-billion-parameter-count model to do a simple classification or, you know, just do basic, you know, transcription. So as a consequence, I'm hopeful that over the next few years, end up with a constellation of models with really purpose-built ones. And as you said, just given the power of the hardware in our pockets right now, I think edge will be a part of it. There's just no doubt. And it just stands to reason there might be creative ways to, you know, maybe convert what you're doing into tokens, you know, so that you have like lower bandwidth connection. If you're doing voice, could there be opportunities there? There also could be some privacy benefits to that. You know, I have a strong Intuition, though no knowledge, but just intuition, that's got to be something Apple's thinking about just given their posture on privacy. So I'm very bullish on the frontier models. Like, you know, OpenAI and Anthropic have, you know, really shown the strength of their research groups, but not at the expense of all the specialized models. I think we're entering a world where, you know, just as we become more sophisticated in deploying AI, we're going to have a constellation of models with just different capabilities. So I'm excited for it, but it's not now. It's in the short-term future. Yeah, yeah, that makes sense.

Did you guys have to build in guardrails to prevent people from access? Every month there's a post that goes viral of somebody that's figured out some endpoint that they can hit to get like— Chipotle code? Like a Frontier model interface.

Chipotle code, somebody figured out how to wire up, they took some sort of open code fork like Kodak's agent coding model, but it's wired up to Chipotle's customer service in the backend. This is a huge risk because you could run up a big token bill that way. Is this actually top of mind?

We have guardrails on it. And, you know, it's just like any other thing. It's, you know, sort of you're always in a race, the good guy, the white hat versus the black hat to, you know, handle it. You know, in practice, I think it's more like digital graffiti. I think nowadays, like in the early days, you'd you'd see a post online of a chatbot saying something goofy, like, "Oh my gosh." Now people are like, "Yeah, you fooled with it." You know, like, "Come on," you know? So I think first, I think it's probably jailbreaking, I think is a really interesting area of AI security, but probably more around the areas of cybersecurity and bio and things like that. I think this type of thing, it's why people work with places like Sierra to have good guardrails. I also think like, like, you know, social media sort of become numb to it at this point. You know, it's like, oh great, you got a chatbot to say something silly again. You know, I think the bigger issue will be just as these frontier models become more capable, the guardrails around using them need to be more and more effective. And I think, you know, as you think about kind of the mission of OpenAI, ensuring AGI benefits humanity, you really need to make sure that those guardrails are effective against jailbreaking, which is You know, it's not like possible to make perfect, by the way, but just because these models are sort of capable under the hood, which is why there's been such an interesting topic around cybersecurity and others. So that's probably the area I think more about, though, you know, the digital graffiti is gonna continue to exist as well.

Yeah, that's a good term.

Have you ever heard of 002 cents? No, tell me. This famous customer service nightmare call where someone called Verizon. They were getting billed at 2 cents, and the actual rate was supposed to be 0.002 cents. So they were off by 2 orders of magnitude. And they're talking to this customer service agent. We'll play it on the show after we wrap. But they're talking to this customer service agent, and the customer service agent just actually doesn't know the difference between a decimal cent and an actual cent. And the frustration The discussion that just ensues is like turned into this massively viral video. There's whole websites about it. Verizonmath.blogspot.com from 2006. And it's like this whole deep dive. I'm sure you've heard about all of these different things.

I will take a look at it. The good part is a super intelligent AI probably will know.

I think it will. I think some ground truthing, some code in there. There's opportunities here. PhD level intelligence. —The raw audio for this call is 27 minutes. The guy went back and forth, and it's hilarious. There's cutdowns we'll have to play. But yeah, they had unlimited data plan in the US and recently crossed the border to Canada. Prior to crossing the border, he called customer service to find out what rates he'd be paying. The data rate he was quoted was $0.002 per kilobyte. And then he got billed at $0.002. $0.002 per kilobyte. So it was 100 times more and he couldn't get out of his quagmire, his Kafka-esque customer service interaction. Anyway, thank you so much for taking the time to come on the show.

It's always a pleasure. Always a pleasure. Progress. Incredible progress. Looking forward to the next revenue gong.

Yep. Come back anytime. I'll come back.

See ya. Accelerating. We'll talk to you soon.

Goodbye. Oh, cheers to making sense.

We're going to now play this 27-minute video. No, we can play this one.

This is just a little bit of this. Yeah, yeah, yeah, this is the one. The 3-minute version. YouTube math fail. I want to hear a little bit about this. See if it holds up or see if I'm washed unck at this point. Being old and remembering a '70s video. Not this one. It's the 0027. Sense. This is exciting news from Canva. We can talk about it later, but you can turn your ChatGPT images into fully editable, fully editable Canva designs with magic layers without ever leaving the chat. Great integration there. Apple Intelligence has some more updates too we can run through later in the week, but let's play the 002 Sense at least a few minutes of it because I want to hear if it holds up.

Kilobyte usage that was done while in Canada.

$0.002. Do you recognize that there's actually— $0.002. Yes. Do you recognize there's a difference between those two numbers? No. $0.002 and $0.002. Yes. Is there a difference between— They're both the same if you look at them on paper-wise. No, they're not actually. So if you take $0.002— Cent. Cents, remember, it's cents, times 35,896. $71.79. No, that would be 71 cents. How much should I be charged? By, by the way this is calculated, $71.

I like these guys crashing out over 70 bucks.

2/10 of a penny per kilobyte. 2/10 of a penny. Would be 0.2 cents. You quoted me 0.002 cents.

I think this guy also might be entirely wrong. He might have just been quoted the wrong number.

0.002, sir. 0.002 what? What? Cents per kilobyte. So you just said it was 0.2.

They quoted it back to him.

And then you also said it's 0.002 cents. Those are two completely different numbers. They're 100-fold different.

Okay, George, hold on one second for me, okay? This guy, so dedicated.

This is Andrea, I'm a manager on the floor.

How can I help you today? Another manager's involved.

Do you recognize that there's a difference between $1 and 1 cent?

Coming out of the gate. Definitely. Definitely.

Do you recognize there's a difference between half a dollar and half a cent? Definitely. Then do you therefore recognize there's a difference What's the difference between $0.002 and $0.002 cents? I'm keeping that here. And I mean, I'm trying to get what you're saying here, but it's just not— And we're talking about cents, right? Right, $0.002. And if we multiply that by the amount of kilobyte usage that you have— 35893. 35893. 93, that comes out to what you paid, 71.79 cents. You never did the conversion from cents to dollars. I don't know. I'm not a mathematician. 0.002 cents, right, times my 35,893. It's a number, but it's still in cents. We're not quoting $0.002. We're quoting $0.002 cents. Oh, God. Honestly. Well, I mean, it's obviously a difference of opinion. It's not opinion. Okay, well, you know what? I'm going to post this recording on my blog. Oh, oh. Cancel. That's fine. That's what I'm going to do. And then you guys all at Verizon can learn. Yeah.

Oh, what a funny interaction. Incredible.

Well, now that you get a PhD-level mathematician Somebody's got to run this on an AI lab. I was quoted 5 cents per million tokens, not $5 per million tokens, and confuse someone, get a 100x discount on your bill, your token bill. Anduril has a space something. What was it? Space observation network. These are ground-based satellites or ground-based Telescopes that monitor space. And look at this, captured by Anderlecht's network of 400 telescopes deployed around the globe. The second stage of the Falcon Heavy launch of ViaSat-3 F3 performing a routine, a routine thrust event. This produced a spiral-shaped plume effect, a nominal part of operations for a successful launch of ViaSat's latest satellite. I thought this was a cool image, and we learned I learned in the process that they have 400 telescopes all over the world. Anyway, we can dig into all of this more later, more tomorrow. We'll be back at 11 AM Pacific. Thank you for tuning in. Leave us 5 stars on Apple Podcasts and Spotify.

Have the best afternoon or evening of your entire life.

I mean it. And we'll talk to you later, folks.

Bye.