Third Party Evaluators, Fed Hikes Rates, Zuck Pushes Back On AI Slowdown | Jeremy Allaire, Tomasz Tunguz, William Layden, Justin Beroz, Eli Wachs, Sean McCarthy, Tom Mueller

You're watching TVPN.

Today's Wednesday, September 16th, 2026. We are live from the TVPN UltraDome, the temple of technology, the fortress of finance, the capital of capital. Let me tell you about ramp.com. Time is money. Save both. Easy to use corporate cards, bill pay, accounting, and a whole lot more all in one place. Jordi, is that even going to happen?

I don't think so.

I don't think that's going to happen.

We're running into problems.

Strip the gears off that one. That's rough. Well, we have a fantastic show for you today, folks. I made it to San Francisco and back since the last show. This is a new thing. I like this. Being able to get up from the show, go do something in San Francisco, get back. Very excited for today's show.

Why didn't you want to stay?

I don't know.

I like doing the show. It's pretty simple. Good to be here in the TVP and Ultradome. We have a great show. We have a bunch of great folks coming on the show. What was on my mind last night? I was listening to China Talk. Jordan Schneider was talking about evaluators, third-party evaluators, and I was noticing this discourse around like it feels like we funneled into like very clear camps super, super quickly, and it feels like I don't know too calcified for how fast it happened. Like the the Dario essay comes out, and you know he throws out meter, says he's bringing in meter. And the backlash is like immediate. The New York Post is crashing out saying like these are handpicked AI watchdogs. Martin Casado, Martin Casado over at a16z, he's pushing for the Department of Energy. He's like full nationalization now. And both of those have their advantages, disadvantages. They both do good work. They are the extremes. And so I was just sort of wondering a few things like, First is like, what other regulatory bodies can actually work? How do they work in other industries? It's very interesting because like the revolving door is something that's common like in financial regulation. You get people that work at banks and then they go and work at the regulator and then they go back and forth. I mean, this is the story of the like many people in the crypto industry where the regulators who are regulating it, they go back and forth. Like revolving doors exist, but I think the The pushback to meter is very much like the door is like too revolving or it's too close, I guess. But the more interesting question to me is—

Yeah, the defensive meter is that I don't know many groups that are qualified at all to even understand what's going on at the frontier, right? And so—

Is that true though? I see, that's the thing I disagree with.

I would say that there are not that many groups that have been this invested in understanding frontier model behavior for this long.

That's a separate thing.

It's just a small group right now.

That's a separate thing. So, I think there's 2 separate things. There's one which is like super forecasting, seeing the future, predicting what's going to happen. I think that's important. I think taking that seriously is important. But then there's the other side which is like doing the work, reading the logs and being like, This violated this rule. This hack happened. Here's how it happened. And I think that those are actually 2 separate disciplines, 2 separate jobs. And you can just tell the regulator if it's someone. You don't need to tell— like, if you hire someone and they are able to get up to speed on how these systems work and evaluate them and look at logs of different incidents, see what's happening, assess the risk level, assess the liabilities. If you can get those people up to speed and you can just be like, it is your mandate to take this seriously. And the example that I'm pulling from is like, you can be 22 years old, not graduate from college, go into the Navy, enlist in the Navy, not even in the officer training program, and in 18 months you can be responsible for the security of a nuclear power plant on a submarine. You do 6 months. I actually looked it up. Like, you do You do 6 months in Nuclear Field A School, about 6 months of Nuclear Power School, and then 6 months of hands-on prototype training before arriving in the fleet. Nuclear Power School covers math, nuclear physics, reactor principles, health physics, materials, thermodynamics, electrical systems, and reactor technology. There's no part of that that's like, you need to really, at a deep level, understand that nuclear annihilation is bad and could happen. You don't need to be able to forecast out, well, what happens if China and geopolitics and Iran gets the bomb, and then these people, and then Pakistan. Like, you don't have to understand that to just know, like, don't let it blow up. That's your job. Here's how you don't let that happen. We've created a plan for you. You're 22 years old, but we know that you can do this job and you are enlisted to do this job. And it doesn't matter if you think nuclear war is impossible, or it doesn't matter if you think it's going to happen tomorrow. Your P-doom on nuclear is completely irrelevant to you doing this job, right?

Yeah.

And I think that that's something that is maybe being missed here a little bit. There's like, There's like, well, you have to take it seriously. You have to have seen it coming. And I actually think that like super forecasting, super important, awesome, also just like fun read, really cool. And if you take it seriously, you make a lot of money, you work on interesting stuff, you get amazing technology. There's so many things that come downstream of that that are really, really positive and really important. But I don't necessarily know that it's actually a prerequisite. And then the other thing is— yeah, Tyler?

I mean, like METER, they're not the ones putting out those forecasts though. That's like— those are other groups.

Yeah, yeah.

If you look at like METER research, it's like, you know, very complicated benchmarks and like these kinds of things.

Totally, totally. Yeah. I'm sort of collapsing the METER criticism from like the New York Post perspective.

Yeah.

And I think that is—

Overall, I think generally, if people are going to talk like if there's a third-party regulator, it seems like it has to be an entirely net new group because I don't think anyone has Like, whether or not METER— METER could be operating and their actions could reflect that of a fully independent group, but the financing structure, the history of the various parties there just makes it so that people don't have any trust that they would act like that.

Sure. The interesting thing is that there is actually a distinction in the history of nuclear regulation, which is— there's a difference between advisors in the— before the NRC, it was the Atomic Energy Commission. And the AEC was created to oversee nuclear development. But— and many of the scientists who did forecast the importance of riskiness of the technology were involved. Interestingly, like Einstein writes this letter and says like nuclear war— he basically, you know, describes what nuclear annihilation could look like. There's a whole bunch of other scientists that actually run the numbers, and they're like, this is what could happen if all the nuclear bombs go off. They do all the calculations. There's some that go a little bit too far, but in general, the scientists were the ones who got to it earlier, but the scientists didn't actually wind up being the regulators. They wound up being the advisors. So Oppenheimer became the chair of the AEC's General Advisory Committee, which had enormous influence but didn't actually issue licenses. So, the actual work that was being done, it was like, okay, Oppenheimer in this advisory role says like, well, we need to have a security guard here with a gun that makes sure that no one can steal the nuclear material, right? But he's not the one doing it. He's not the one actually hiring that person. That's just like an engineer who's qualified for that job. And I think that actually takes a lot off of it because You could say, oh, there's all these conflicts of interest here or there or there, whatever. But if you just say, well, they're just putting out a proposal that then people are going to go and implement, but then the actual people that are doing the implementation are much less conflicted because they're just drawn from the broad pool of engineers and scientists and mathematicians and physicists and whoever else we have in America who can do this type of work, it gets a lot less complicated in my mind. There's a whole bunch of other interesting details from the AEC history. Enrico Fermi and Glenn Seaborg became committee members, with Seaborg later becoming the chairman of the AEC itself. Of course, Fermi from the Fermi paradox, famous scientist, very interested in forecasting, seeing the future, understanding the implications of things. But again, becomes committee members, not actual regulators, not actual licensors. And then, Edward Teller is sort of like the most accelerationist because he was advocating for the development of the hydrogen bomb. But he ultimately became the chairman of AEC's Reactor Safety Committee, talking about how to actually secure reactors in particular. But again, not actually on the team regulating— staffing working on this directly. And so, today, the regulators, like nuclear regulators, are— talented and hardworking, but these are not like the most elite jobs. Like, you can just be a nuclear engineer, mechanical engineer, material scientist, physicist, health physicist, geologists, probabilistic risk analysts, cybersecurity professionals. They hire for this emergency preparedness. They hire lawyers. They hire inspectors. Pay for some of these jobs ranges between $125,000 and $187,000. For many NRC technical staff, roles. So, like, once the machinery— the machinery did have to get sort of described by the scientists, but then the implementation of that machine, of that regulatory structure, is actually done by really hardworking, really talented Americans, but not head in the clouds, not thinking about the future in some bizarre way. That's enough to happen at the democratic level, and then it gets implemented. And I think that that might eliminate a little bit of this, like, oh, okay, well, like, this person who's really tied to you is now, like, inside actually the one with the keys, the one with the role overseeing you.

I don't know.

What do you think about this, Tyler? You have some pushback?

Yeah. I mean, like, I know that CAISI, the Center for AI Standards and Innovation, like, I think that they've had a hard time, like, staffing.

Getting funded, yeah.

Yeah. So it's like they're having a hard time, like, finding people who will come, like, from the labs. Maybe they should just broaden who they're looking for. But it seems like it's still like the ways that we detect if a model is safe or not are not like set in stone yet. Like, it's still like that's what the role of meter, that's kind of what they're doing right now.

Sure, sure, sure.

So I think it is still different than like, you know, there are predetermined accounting practices and you can just kind of check the box and like follow the rules. Yeah, it's still going to be implemented. Yeah, it's like a moving field, you know.

Yeah, yeah. And I mean, there is the question of, like, with particularly with, like, agent swarms perpetrating cybersecurity violations that don't directly cause economic harm. Like, I didn't tell it to hack you. It hacked you, but it didn't knock your payment system offline. So you didn't lose a dollar of revenue. It's, like, very hard for you to prove that I acted wrong and then also economic damages. There's a whole new level of, like, you know, tort battles that need to be battled out in the court of law to see, like, what exactly do I owe you? Because I shouldn't have done that, but what do I owe you? What's the damage? What's the problem there? And then you can go and say, okay, well, you know, how do we measure that? How do we prevent that? And how do we work through that? But the discourse is getting, like, more and more— more and more polarizing by the day. We'll see. I think there will be the big sit-down between the lab leaders. I wonder how important it will be to have Jensen with a beer alongside Sam, Dario, Elon with beers, because the clear proposal, there's gonna be beers involved. That's what we know from the interviews. Everyone's asking, why can't they just sit down and get beers?

6 months of this could have been on the Cheeky Point podcast.

It really should. That is sort of neutral ground too, 'cause Elon's an investor in Stripe. What happened there? Elon's an investor in Stripe. He's a co-founder of OpenAI. And Sam's, I think, an investor in Stripe. And then Elon's working with Dario on compute stuff. So maybe Cheeky Pine is like the perfect neutral ground. I think it'll probably be on national TV, actually. But we shall see. We'll follow it here. 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 of scaling, monitoring, and security. And we can, we can move on to the timeline. We can move on to other—

Quickly jumping in, we have a rate hike.

Yes, tell me about it.

Warsh hiked 25 bps.

Okay.

Uh, this was—

First time in 3 years.

Priced in. CalShe had it, I think, at like 89% this morning. So not a huge surprise. I wanted to head over to Joe Weisenthal's feed and just kind of read his reaction if he has one.

Well, you pull that up. I'll give you the highlights from The Wall Street Journal. The Nasdaq reacted positively, up 6.67%. This is The Wall Street Journal. They need a JavaScript plugin that changes it to 0.669 or something. Most officials penciled in one more increase this year. An energy shock and a surge of investment have reshaped the inflation outlook. We talked about about the Fed interest rates yesterday a lot. But the Federal Reserve raised interest rates Wednesday for the first time in 3 years, a sharp reversal that began taking back cuts as it made last year implicit and implicitly undercut the White House's insistence that inflation is not a concern. The Fed is saying it kind of is a concern. The increase approved unanimously will raise the benchmark Fed funds rate by a quarter point to between 3.75% and 4%. The Fed's meeting on Wednesday, the vast majority of officials penciled in more— one more hike this year in interest rate projections released after their meeting. So they think there's going to be more rate hikes. Chairman Kevin Warsh vowed shortly after taking office in May to end an overshoot of the Fed's 2% target, now in its 6th year, and followed through with an increase that had been widely anticipated in recent days. The rate hike scrambled an account of the White House had offered of the man tapped by the president for the job in January. Trump and his allies had cast pressure to raise rates as coming from a committee hostile to Warsh, who last year said he would have cut rates sooner than the Fed ultimately did. It also followed a lost year in the Fed's inflation fight. The central bank has made no progress towards its 2% goal since mid-2021. 25, including after cutting rates 3 times last year to guard against labor market slowdown. Instead, the Iran war has lifted energy prices, and the AI boom has driven an investment surge that has buoyed the economy and markets today. Policy action will support a timelier return to the committee's 2% goal, the rate-setting committee said in a policy statement. Analysts said that despite intense focus of late on monthly inflation data, the biggest change to the outlook has come from a run-up in energy and commodity prices. It's the fact that the war in Iran has reintensified and the energy price shock is getting bigger again, said William Dudley, the former New York Fed president. What you got for me, Jordi?

I was just reading through a bunch of different reactions on Bloomberg itself. Let me pull them back up. But they have a live blog. Some people are saying this is more hawkish than expected given that the Fed took away next year's cut. What else? Big changes in the dot plot line. The Fed September dot plot. We now have 4 officials expecting to raise rates 2 more times.

2 more times.

It previously just been 1 at that level. A whopping 12 policymakers see rates going up once more before the end of the year, and the remaining 2 see holding rates at their new 3.75 to 4% level. Reminder that in June, the last time we got these forecasts, half of the committee expected the Fed to hold or cut rates. And again, it looks like Warsh did not submit a dot. So he's going dotless here. Kind of a statement in itself. Yeah, it's so— I mean, so far, the AI trade, the buildout, Everything has been overwhelming, even in the face of headwinds like rising rates.

Yeah.

Yeah.

I mean, the mood from Silicon Valley was like, we're definitely not booming until we go back to zero interest rates. Like, this whole tech thing, it only works when the interest rates are zero. So, like, we'll just wait it out.

And yeah, the reality is there was a bunch of ideas and investing styles that only worked when rates were near zero. But yeah, it was specifically when you look at the companies that really boomed in that era, there wasn't a lot of net new, really truly innovative stuff outside of financial products, which benefited from low rates. Yeah.

Some of them benefit from high rates though, right? If it's like a savings product, the spread's higher? But if they have to borrow a lot of debt—

Potentially, but again, I'm thinking of lending companies like Pipe, right? Pipe was a company that at the time went from incorporation to billion— I forget what their peak valuation was.

And it makes a lot of sense because they're basically borrowing at 0%, and they're lending to a company at 5% or something. Whatever their spread is, it's actually justifiable to a to an earlier piece of the market. But that sort of breaks down when you have to go to a company and say, hey, you want money at 12% or something like that for an early-stage company?

Well, let's head over to Hooman, who's—

First, let me tell you about Shopify. Shopify is the commerce platform that grows with your business and lets you sell in seconds online, in-store, on mobile, on social, on marketplaces, and now with AI agents.

Hooman says, Dario, we should pump the brakes on the frontier. Sam, yeah, let's all slow down together. Meanwhile, Zuck—

The funny thing is, yeah, I mean—

Let's talk about what Zuck wrote. He said, last month I wrote about how we can build a positive and safe future for everyone. This is when Zuck said, I really want people to understand my values before we come out with our most powerful AI ever. And I at least felt like we already had a good understanding of Zuck's values, but He wrote yesterday, every lab has the responsibility and incentive to move at the pace required to train its models safely and the ability to take its own actions to ensure that happens. The reality is people don't want to use agents that are misaligned with them and don't do what they ask. So labs have a strong natural incentive to make their models more aligned. Right away, you know, starting out with this point, it's like, this is not the— this is not the— no one The safety debate has not been around the idea of, oh, they're going to create personal agents that are going to be misaligned to the users. This is not at all— this is just a point that doesn't matter.

No, no.

I mean, there is a— it's not the safety crowd, but the whole social media is brain rot addiction. That is something that— there is a separate crowd that does critique that. and says, like, I don't want the addictive flywheel of maximizing screen time to be brought to AI.

It's a totally separate debate.

Yeah, no, I agree. I agree. It's not the true AI safety debate.

Yeah.

But it is a debate.

There is a lot of debate about slowing progress on capabilities until alignment catches up. My view is that trust and alignment are quickly becoming the most important capabilities that will differentiate Again, this is not relevant to the current safety debate.

Mm-hmm.

Obviously, people want to make products that do what their customers want them to do. No one has been worried that you can't make a model that in the near term or in the medium term or even over long-running tasks can generally do what the user wants. The concern is that if you leave models and give them a task, that is more expansive, that they can start to do things like hacking Hugging Face, right? So, again, nobody is sitting here saying—

The main thing is that this is talking past the x-risk question. It is completely dismissing the discussion, which Dario is laser-focused on. And so, this feels like it's a rebuttal, but it's actually talking past it in the sense that he's like, labs face significant liability. And it's like, well, in the x-risk scenario, the liability doesn't matter. That's the whole point. Is that like no one's going to come and—

And here's, here's the best, here's the best line. Meta delayed shipping Muse for several months to focus on safety and security.

That's actually very rational and important because—

Every company delays their products for months. That's just called building a good product.

Okay.

To make it safe and secure. Like, period. This is what Meta has been doing forever. You have to do it. You have billions of users.

Yeah.

You got to make sure they're safe and secure. Every single product. Again, I felt like he was—

but you—

so there's a lot of things in here that are rational.

He's taking a victory lap on not taking a victory lap. Don't you realize that? He says we didn't call for everyone else to do this before we would. We just did it as part of our day-to-day work because it was clearly the right thing to do.

And again, every single company already does all of these things. Every single company that makes AI products already does.

But all the other companies take victory laps while they're doing it.

He's taking a victory lap for not taking a victory lap. It's not that complicated. He says everyone else says like, we're taking safety seriously, we want a pat on the back before we delay the product, and he's saying, I want the pat on the back after I delay the product.

Yeah, I just think it's— I just think it's very— No matter what, there's a lot of back patting going on. I just think it's— I think it's— I just think it's, uh, very funny to— I think it's funny how much All these points are fine. They generally are rational and they make sense. But I think it's funny that people are giving him so much credit for this note, given that he's totally missing the main point that everyone else is focused on.

X-risk?

Like, intentionally missing the point.

Because he doesn't believe it.

He has PTSD. In order to get brownie points from people that don't even understand the current debate?

No, to get brownie points from other people that have a p-doom of zero, who are like, yeah. And you can see who's supporting this. They're like, yeah, thank you. Like, just put the agents in the bag, you know, make the tokens free and just make the products. Like, I'm not worried about that at all. And for that crowd, they're like, thank goodness you didn't, like, fall in the hole of, like, stooping to this p-doom debate that I don't take seriously. That's the side.

I just wish that he would come out and say, I have a PDoM of zero. That's what he's saying. No, he's not. He's not. He's trying to position— he's saying my view— like, that's what he's saying. That's not— that's not— he's not being explicit about that. He's trying to let people say, we care a lot about safety. We slowed down our development because we care about safety. Trust and alignment are important. Yeah, right.

He should definitely come out and say PDoM zero because If it's not zero and it happens and we all go extinct, no one's going to be able to dunk on him, right?

Yeah.

So, it's pure upside. Pure upside.

Pure upside to be a P-doom zero guy. Why has no one considered this? The aura gain is so high.

Yeah. I would respect it a lot if he just came out and said what he actually thinks, which I do believe you're right.

P-doom zero.

Which he has a P-doom of zero and his P-abundance is—

I mean, that's what he said in the previous essay. He was basically like, I don't think the— he even was gesturing towards like the fear-based marketing, the doom-based marketing is just a marketing tactic. I don't think it's rational. And also, I don't think it's good for people to be in that headspace. And it's like, InfoHazard has a problem with it.

Here's the thing. One of the last lines, committing the significant majority of compute towards serving people rather than racing towards recursive self-improvement is one of the best ways to ensure we develop this technology safely. Meta has made this commitment and other labs can do this as well. Look, there's absolutely zero shot that Zuck walks into MSL and gathers the researchers and says, look, I don't want to make models that make our models better. I don't want to do it. I want you guys just focus. There's zero, zero chance. So again, this to me is just like he is being disingenuous. with his positioning of almost every single point here?

I don't know. I do think that there is a trade-off right now between making models good at things that are not on the RSI path and those that are. So, the race to become really, really good at coding is super aligned with RSI. The race to do image generation is not, and image generation does not seem to be on the RSI critical path. Although I think DeepMind put out something where they're using world models and they think they have a breakthrough there. I don't know if I saw that accurately, but maybe they're wrong. But at least the bet at, at least Anthropic has been like, we don't need to be world-class at image generation to get where we want to go because we just need to be really good at coding. Coding teaches us— teaches the model how to train new models, and then we get, you know, and then that final model, we can ask it to spin up an image generator if we want. Zuck is saying the opposite. He's saying like, yeah, we will actually go and try and build a tool just to help you book a dinner reservation, and that's a good use of compute, probably not on the RSI path, and that's like a reasonable trade-off. That feels That feels real to me. I don't know. What do you think about engaging independent evaluators? Is this like—

Yeah, and again—

He says it's already industry best practice. Is he talking about benchmark stuff or is he talking about like actually like the— you have Slack access, you have a desk, you have a badge, like you don't work here but you're allowed to just go wherever you want. I think that that's the next step and I think he's maybe talking past that a little bit.

Yeah, and he's talking past this again, engaging independent evaluators. and advisors is industry best practice. MSL already does this today in several areas because it helps produce better work. Other labs can just do this too. It's like other labs also do that too, also do that already.

Well, as of last week, Anthropic does this with Meter. They said that they were going to do that immediately. So they were doing badge and Slack access.

I know, but he's not saying that he's doing that. He's saying that he's doing the thing that every group that's making Model has been doing for the most part.

In the prior era.

For a long time.

But the new thing is badge and Slack access and desk, even though you don't work at the company. And that's what everyone's like, whoa, that's crazy, because these organizations are very, very secretive. And you're really going to let this nonprofit come in? It's like sort of a wild move. And that's why people are like, oh, wait, like, how aligned are they? And like, what is the— What's the knock-on implication? Like, can these people not leak? Can they not— can they be trusted? Are they going to go to a cocktail party and be like, oh yeah, the new model is actually really bad or whatever? Like, there's so many things that are like— like Meta deals with leaks all the time. And so the prospect of bringing in a non-employee who is there explicitly to whistleblow effectively and like is allowed to talk about anything and like is, you know, third-party evaluator, like that is a huge step. And I think that's why people are like, Whoa, this is a big deal. This is a big proposal. If this was not a big thing, if Anthropic was just like, oh yeah, we're going to do a benchmark with METER, everyone would be like, yeah, that's fine. Cool. Do that for sure. Awesome. But people are like, wow, okay, METER is going to have Slack access at Anthropic and you're asking other people to do that? We got to know who these people are. We got to make sure that this is the right team for this. This is sort of a crazy thing. This is sort of unprecedented. This doesn't happen a lot. This is new. And so, yeah, the newness is not fully embraced here. And I think that's a little bit of like, okay, you're not really engaging with what's coming down the pipe, which is like maybe a government employee in your building. That's not happening right now.

Yeah.

Maybe a nonprofit from Berkeley. Do you like that, Mark? How do you feel about that? Are you cool? It seems like you're cool with it, but I don't know if you're actually going to be cool with it because it is a little wild, right? It's a different thing. It's a new thing. It's, it's a modern— it's a modern—

pull up this image.

What image are we pulling up?

I want to see John's reaction.

Okay. While we pull this up, let me tell you about Public.com. Investing for those that take it seriously. They got stocks, options, bonds, crypto, treasuries, and more with great customer service.

Here we go, John.

Let me see. Boom. This, this meme has been applied to like 7 different people this week.

I know, but I think it's particularly relevant because you have Elon, Demis, Dario, Sam, all these people from different factions that are at or—

I've seen this applied to Theo. I've seen this applied to Cohere. I've seen this applied to DeepMind and Gemini. I've seen this be applied to MSL. People are just, whatever shot they want to take, they're applying this. It's not, it's too broad at this point.

I just find that there's people—

I think it's the inverse. It's all clowns and there's one tactical soldier.

There's groups that are at the frontier. They're on the battlefield and they're deeply concerned. And you have Zuck coming in here saying, yeah, like I built a cool personal agent. It's aligned. What's there to worry about? Nothing to worry about. It's just, you're not really, You're not really in— you're not really on the battlefield yet. And I'm not saying with Watermelon they can't get there. And Muse seems like it's—

Yeah.

Muse is getting amazing reviews. It seems like an amazing product. And that's very exciting. But I think it's just— again, I thought the whole post was silly, just like I thought the last post was silly.

Nick Carter liked the post. He said, Zuck pretty handily dismantles Dario's talking points here. People want models that are aligned with them. Subtly punches back at Anthropic's normative constitutional approach. Labs already face liability if they screw up, so incentives to release aligned models is already baked in. Meta delayed Muse for alignment reasons. Subtly questions Anthropic trying to kingmake METER. Implies METER is an Anthropic patsy. Meta doesn't need to coordinate with anyone to work on alignment, just something labs should naturally do. From the top rope according to Kevin Roose says there aren't race dynamics off the frontier. And yeah, it's a war on the culture.

Again, it all comes down to like there is absolutely zero shot that Zuck doesn't want to get to RSI yesterday and is doing—

I thought he said that they were working on RSI. I thought that was like announced as like an explicit goal maybe 3 to 6 months ago. He was like, Superintelligence, RSI.

And again, I just feel like it's not unique to Zuck. Many leaders operate like this, but he will say whatever the pick-me thing is at that time.

It would be more aggressive to be like, I'm P-Doom zero and I'm racing to RSI. Like, I'm P-Doom zero and I'm racing to RSI. All these other people, I'll see you on the other side, brother.

Meta, he said, on August 10th, 2026, Meta must build out a sufficiently large amount of compute that we can allocate enough to recursive self-improvement.

Yeah, they're doing RSI. They're doing RSI. Everyone's trying to— I mean, RSI can also mean so many things. It can mean, yeah, you looked up the— you summarized some archive papers for new strategies. Like the automated research intern is in part an RSI initiative. It's not the final RSI loop, the closed-loop RSI that people are really worried about, but it's all gradations of this. What does David Sacks think about this? I imagine that he's a fan. He says, Mark Zuckerberg, he just quotes it, he says, Meta delayed shipping Muse for several months to focus on safety. Yeah, Sacks is very much like, if you wanna slow down, go right ahead. And so he is fine with that.

All right.

Anyway, the reviews for Muse are really good.

Let me say this, because I've been a little harsh.

Yeah.

The people genuinely love Muse. I think it seems to be a great product. The race between Muse and Instinct is already Already quite exciting. Niraj says, my wife tells me Muse is good and, quote, for the girlies.

Yeah, that's extremely bullish.

Extremely bullish. But yeah, I mean, insane distribution advantage. I think it's— let's see where it's at on the charts. Number 3. We got it, Nick. We got to get this company Vinted on here. I just do not— I've brought it up like a million times. It's like, you have the most insane AI race, gambling race, short-form drama race, and Vinted is always in the top 5.

Wow.

Just like a secondhand clothes retailer competing with like a million other secondhand clothes retailers. Yeah, but they're dominating.

Look at this. Vinted, there we go.

Like Meta's like pouring every— pouring billions of eyeballs into growing Muse, and Vinted is—

Didn't— wait, didn't— Gemini and Google announced a personal agent at I/O this year. Where did that go? Because that— it's very odd that Gemini has so— like, yes, the distribution advantage from Meta Platforms is significant, but a lot of people are on Android. A lot of people already have Gmail and Google Calendar, and they have Google Maps, and they have the phone number of every business. And so when you think about booking restaurant reservations, doing all this different stuff, they— you would think that that product would have gotten more traction. But I feel like Gemini Personal Agent did not go through the same hype cycle that Muse is currently on. Gemini Spark, it died on the vine. Yeah, what happened with— wait, they call it Spark? Is that— but isn't it Muse Spark?

Yeah, Dylan in the text chat said—

Everyone's calling it—

She uses it.

Oh yeah.

It's mostly just using Gemini and toggling.

Okay.

Yeah, it hasn't had the most breakout things. There are some killer Muse use cases. I mean, everyone's talking about booking reservations.

Scott, I asked him at I/O and they couldn't describe what it did. No one knows. The world's most mysterious. It's the most mysterious agent yet.

But they got to have something soon coming. And I do feel like the Gemini distribution, like they still have a pretty significant share of the chat market. It pops up in your email and it says, oh, here, you can just email this thing. And Maybe another Gemini tab that I can open, because I can open 3 if I'm in Gmail on Chrome and pull them all open. I don't know. What happened to the Kalshi AI price tracker? Did you see this? We talked to Tarek when this launched.

Yeah.

And I was like, oh, this is cool. This will allow you to understand basically a proxy for the AI buildout. How are GPUs trading?

Banned.

Banned, apparently. Before we get into this, I'll tell you about Figma. Agents, meet the canvas. Your AI agents can now create and modify your Figma files with design system context. So this is from Semaphore. And the— do we have this open here? The US Commerce Department last month ordered Kalshi to take down one of its products tracking the price of AI compute, the crucial power from data centers that's driving the artificial intelligence boom, if you didn't know what AI compute was. The commerce officials cited national security concerns, and that's what stuck out to me. I was like, there's so many different prediction markets that I can easily trace through, like, oh, you have a flight delay one, and you could have somebody that calls in and tries to get the flight delayed. That could be very disruptive. FAA could have a problem with that. But this one, I wasn't— I wasn't worried about it at all. We talked to Tariq about it and we were like, yeah, this one seems sort of informative and interesting. So the product pulls together data from several markets that allow users to bet on the cost to rent NVIDIA chips to create an overall picture of where AI compute costs are heading. Kalshi quietly complied, though many of the underlying markets remain open for trading. Separately, Commerce has pushed the Commodities Futures Trading Commission, which oversees prediction and future markets, to effectively freeze approval of new compute contracts for 60 days. And so even though some of the contracts are still open and will close, maybe in 60 days there won't be any compute prediction market futures, which is very interesting. But Kalshi declined to comment, while a Commerce spokesperson said the department, quote, has never once asked Kalshi to take down this market or any other markets. Interesting. So the Commerce spokesperson said This story is false to Semaphore. So lots of people going back and forth. It's unclear why Commerce is worried about the nascent market, which aims to do for AI computing what oil futures do for crude, let buyers and sellers of compute lock in prices and give traders a way to bet on where those prices go. Now, the markets were always a little thin. You know, these new markets, they tend to be sort of thin and they tend to be more speculative, just people that are trading them. On vibes or news, not necessarily used by the actual companies like, you know, oil futures. Those will be actually employed in Treasury strategies by like airlines.

And they—

and like big, big firms will trade those in size. But that was what— that's what Kalshi and Tarek were pitching as like where this all goes. I was a little skeptical that it would wind up on Kalshi. I thought that they could have a decent chunk of the market, but I thought a lot of it might just go to insurance companies and reinsurance companies and sort of like the traditional financial rails that do these types of deals. But nevertheless, it's going away and let's figure out why. One potential reason floated to Semaphore by market participants is that compute futures could be manipulated to show a sharp drop in the cost of older chips, which might destabilize AI stocks and debt markets. Some of these markets are thinly traded. Which could lead to volatility even without bad actors. So you're trying to wipe out situational awareness. You're short the Kalshi prediction markets on AI compute futures. Everyone thinks, oh, AI is bust. The market trades down for a couple of days. You clean up and then you buy back in or something like that. I guess that's what the rumor is here that some folks are reporting on. The cost of compute has become one of the most important numbers in the US economy. One side of the debate fears that older chips, which serve as collateral for billions of dollars of borrowing by NeoClouds— NeoCloud, fun fact, coinage by SemiAnalysis. I didn't know that's where it came from, but they actually— they didn't just rank them all with ClusterMax. They actually coined the term NeoCloud. It had a rude name beforehand, which I didn't realize. You can go look it up. I'm not going to say it. But on the other hand, on the other are concerns from big companies adopting AI that shortages of power and infrastructure will send prices of tokens soaring. That uncertainty has given rise to futures market that was starting to take off this summer. The CFTC's 60-day pause could delay plans by exchange operators like CME and NYSE parent Intercontinental Exchange, along with upstarts like Architectural Financial Technologies, to list two-sided betting parlors. Interesting. Well, let me tell you about Cisco, and Jordi will pull up the next story. Critical infrastructure for the AI era unlocks seamless real-time experiences and new value with Cisco, where you want to go next?

Zach Yadigari is back.

Where do we know him from?

He says Cal AI.

Cal AI.

Been on the show. He says everyone's been asking what's next after selling Cal AI. Introducing Persona. Others couldn't figure out the interface. We did. Let's pull up the video. We got to ask Zach how his deal worked to sell Cal AI because he seemingly had to stay for like a week. He ended up However he worked it out, he basically just got to to bail.

You can't keep this guy out of the arena. You can't get him on the sidelines. It's impossible.

Yeah, and Cali AI is the number four fitness app still.

Cooking.

Yeah, so good. Yeah, let's play this video.

Every major AI company is racing to build the best assistant, but they're all building within the same two by six inch rectangle we wear in our pockets every day. We believe AI is supposed to free us from. our screens, not make us use them more. So we built a new interface. Introducing Persona and the Persona Band. To talk to your Persona, simply flick your wrist or hold down on the face, then ask anything. Hey, can you ask Mac how many pre-orders we got today?

Sir, I just emailed Mac.

I'll let you know his reply.

Your Persona will proactively learn your habits and to make your life easier. For example, when you get off a flight, it should know to get you an Uber from the airport to your hotel. Reactively, it will be able to help with all kinds of things like canceling subscriptions, negotiating bills, or even DoorDashing your usual Chipotle bowl, knowing exactly what you like.

Elite.

Persona is also the only AI assistant—

Wait, it's the predictive burrito! It's what the DoorDash guy was talking about. They ordered the burrito before you even know. He's actually building it.

Privacy for your and those around you's data. This isn't a continuous listening band. The microphone will only activate under your control. We are launching with 3 different materials and 4 different colors. Pre-order your band and start texting your persona today.

I've seen enough. Time for Zuck to come in with the $500 offer.

Yeah. Pretty good pitch. A little bit harder to get people to install it, right? Because they got to buy it, wait for it to ship, a little bit of a barrier to entry. But also, I just love these AI devices.

Look at Pocket.

Yeah, Pocket really worked. Yeah, I don't know. There's—

this is—

he's gonna really blow this thing up. I have a feeling this is gonna be big. That's cool. What did Prompter say? Never bet against a guy who managed to sell a calorie-tracking GPS. GPT wrapper, which in no way could have possibly been accurate at calculating macros for $100 million. No, he is a master of—

Growth.

Growth and scale and just like—

So if you look at the charts right now, CalAI is number 4. The company that bought CalAI is number 9.

Wow.

Yeah, I mean, that's a good example. It's a good growth team. Doesn't he also have another company, EcoGPT or something? Is that not him?

No, no, that's his co-founder.

Got it. Okay. Yeah, that's a separate company.

Yeah.

So I'm still— yeah, we— Tyler, preorder one of these.

Oh, yeah.

Definitely do that. Put it on the ramp. So yeah, I'm excited to check it out. I'm still not— I'm still overall not convinced that I need a new device. Like there has to be something—

Yeah.

Like a lot of the demo was screenshots of the phone. And so, and I do spend time away from my phone, but I'm typically in the water.

Is it waterproof? Can you wear that while you're surfing? You said you had a brilliant idea while you were surfing, needed to hold it in your head the old way before you got back to your phone.

Like an idiot.

Like an idiot, using your brain. No, it really does feel like a wave of new devices. Is coming online in the war for the home device. I mean, you can see the battle between Apple and Meta reemerging with Muse. How deep will they be able to go before Apple pushes back? Every other firm is working on hardware. The Meta glasses with Muse should act very similarly, right? You have an You can talk to it. You can do some of the similar actions. The trick is people don't wear the glasses all the time, whereas a wristband like Whoop, people wear that all the time. And he seems very in tune with, I don't know, just the obvious pushback, like calling out that it's not listening all the time. There are just people that would just forget about that. And it's very obvious that that actually—

Yeah, I'm just using the burrito example. I'm struggling to see when I would need a burrito but wouldn't be near my phone.

It's more just like if you don't— if it's faster. Like, people just do anything that's faster. That's how we got to phones. Like, you used to— there was a time when it was like—

I know, but if you're gonna shop online, you go on desktop. I'm not saying like the overall form factor is bad, but if you had the ability to just hit a big button and talk to the Persona app and say, order me the usual from Chipotle, it's not like the band is necessarily faster if your phone is sitting right there.

Yeah. I wonder how fast they're gonna update Siri because it feels like—

No, it is so funny to me that Apple's finally coming out with their knowledge retrieval LLM chat app. And right, literally right before they could really get it out to the public, there's like a whole new paradigm of agents that are You can just tell it's going to take Apple like 2.5 years to get a beta out.

Yep.

And so again, like right as it felt like they were, oh, we're going to catch up, you know.

Will it?

Will it? Will it? Is there a way that the next version of Gemini is so good, if you've talked to Gemini in Gmail, it links it? And so when you talk to Gemini through Siri, it knows that it's you and it can go in your email and it can email people and text that restaurant to get the reservation? Like, is that a— I'm always interested in, like, how much innovation will Apple get for free because they're partnered with Google? Like, if Google advances the model, does that advance automatically come or is Apple like, we're buying Gemini 3.1 Flash from you. Thank you. We'll let you know if we want the new thing. Because if they just want Gemini and all the progress, like, There's going to be progress there. There's going to be new capabilities. The models are gonna get better. The functionality is gonna get better. I would bet that even though people were sort of—

Yeah, Google has just really, really, really struggled to make good AI products.

Yeah, I don't know. I would be surprised if they can't figure out a way to order you a burrito in the next 6 months, right? Like, that's not some like RSI critical thing. It's like—

I don't know. You would think that— Having spent billions of dollars buying Windsurf, that they could have a competent codegen product.

Yeah, I haven't seen— yeah, it's tricky.

I don't know.

Like, there's small companies that have made good coding harnesses.

Yeah. Anyway, let me tell you about CrowdStrike. Your business is AI. Their business is securing it. CrowdStrike secures AI and stops breaches. We have Jeremy Allaire from Circle. You want to talk about Sam first?

Let's bring— let's bring—

Let's bring Jeremy Allaire from Circle, co-founder, chairman, and CEO, back on the show. Welcome to the show, Jeremy.

How you doing?

Great to be back.

Crazy week, tons of news. Let's start with Circle. Let's start with what's new in your world.

Yeah, well, so we turned on a new global operating system today called Arc. It is an economic operating system that we've been building for 2 and a half years. Hundreds of companies launch their apps and services on it. It's being run by not just Circle, but like many of the biggest financial infrastructure companies in the world, the people that clear the securities of the world, run the payment networks of the world, manage the assets of the world. And so it's got everything from social trading and DeFi to major payments firms and capital markets. And, you know, we built this for AI first. Everything is accessible, all the builder tools, everything. But also we've We've built services so that AI agents can use this themselves as an economic layer for transactions, for contracts, for coordinating. And it's a big milestone. It's the biggest, really the biggest platform launch in our history. And we think it's a big upgrade for the internet too.

Amazing.

A lot of people 6 months ago would be like, awesome, it works with AI agents. But today they'd be like, definitely don't make it work with AI agents. That's not what we want. How do you think about the risks and the rewards that come from enabling a platform to be AI agent accessible. I mean, sometimes they'll access things that aren't even accessible.

I mean, this is the key, right? Yeah. The biggest challenge with this explosion in agents is basically trust.

Yeah.

And verifiability.

Yeah.

And so, you know, it turns out that cryptographic computing and these, these, these networks, these network operating systems of cryptographic computing are built for that. And so we've, we've kind of created the most trusted dollar on the internet, USDC, and these operating system layers actually have features that AI can build on top of. So you can basically have proofs of the, of the entities and identities behind agents. You were actually demoing today something called Arc Agent Sector, which actually allows you to have an AI agent prove the work it's done, the data it's used, how it's done its work, and then provide that in cryptographically provable way to the internet. And so you can have more trust and verifiability around what actually AI is doing. And right now, the black box issue is the scariest thing for people out there, whether it's the labs or it's actually just out with the agents that we're building. And so it's this convergence. I, you know, I talk about the agentic economy. It's this convergence of these operating systems for intelligence and these operating systems for economic activity rooted in, in cryptographic computing, which is what these blockchain networks really are. And so We're building a layer that is able to begin to provide that trust and verifiability to the agentic world. And so that's how we get comfortable with it. We need this. We can't move forward in the agentic economy without technology like this.

Are you equipped to explain to me the history of Walmart's misadventures in payment infrastructure? I was reading on Srethakari that they were trying to push people to debit card and ACH rails for years. And then they finally sort of succumbed and went with a different one. And I guess I'm just interested in, like, is this a transformative moment for any particular sector of the economy where you're like, oh, someone's going to be saving half a percent now, or some big pool? Yeah.

I mean, for sure. I mean, look, when we started Circle, right, we had this basic idea that there could be a protocol for dollars on the internet. That the marginal cost of storing and moving value, just like the marginal cost of storing, moving data and information and communications, went to zero. That happens with, with the movement of money. And now we have that, like stablecoin digital dollars are now, as of January of this coming year, like legal digital dollars in the financial system. You can move them programmatically, instantly, anywhere in the world in a fraction of a second for a fraction of a cent. And so payments and settlement becomes commoditized. And so as you get the proliferation of wallets that can talk to these computer networks, which is now everywhere, like from Cash App to Revolut to all these wallets around the world can talk to these networks. Well, now you can directly settle. So if I'm a business, I can just say, you know, pay me, pay me over this rail and I get it as basically cash instantly. And, and it settles, you know, like a cash transaction. And so over time, basically that compresses out this rent extraction that's existed for a very, very long time and all kinds of other things like time delays and the like. But, and so, yeah, I mean, it's, it's purpose-built for that world. And a lot of stuff is converging all at once. Operating systems like Arc, the legal frameworks for digital dollars, the people getting comfortable that they can use this in their, in their businesses, know how to operate it, etc. And so it creates a huge opening for, you know, the Amazons and the Walmarts. And really, you talk to any major retail-facing CEO and they'll tell you how much of their margin goes to processing fees. And so I do, I do think it's significant, but the whole utility model of payments is going to change very, very significantly. That's a big part of what we see happening here.

You mentioned what's the most boring payments happening on, on ARK. Like the agent stuff's awesome, but I imagine you have a bunch of different partners.

I'm assuming some of the boring stuff is, is like You know, there are businesses that have to move money around the world and they're finding that they can do it better. And they— if there's someone who's importing from Asia in Latin America and they need to do that and they want to like have that done quickly and they don't want to have a lot of intermediaries and they want to just like directly do that. And so volumes of kind of how people need to move money around is, you know, it's boring to all of us, but to those people in those countries and to those suppliers and those other people, it's actually really significant. I mean, other boring stuff is like, hey, I have a, I have a tokenized version of a stock and I'm someone in an emerging market that wants to own a stock. I can now do that and I can, I can purchase it and trade it instantly. I can take out options on it instantly. And I don't have to be in a US brokerage account. I can just be on the internet with a wallet that connects to these networks. And so, you know, from, you know, from the average individual who just wants to participate in capital markets in simpler ways to these businesses that have to move stuff around. That's really important to the, you know, the kind of efficiency and cash efficiency of their business. Yes, agents are exciting. That's frontier, so to speak.

Yeah.

And then even, even simple things like people love to watch memes on TikTok. They also like to trade memes. And so if you want to trade memes, go for it.

Sprinkle some of those in.

I want to talk Legal. You mentioned that these are now legal US dollars. But yesterday, CoinDesk reported Crypto Clarity Act flames out in failed US Senate vote. The years-long effort to set US regulations for crypto markets couldn't muster enough support to make the leap over the Senate's final 60-vote hurdle. I would love your reactions, context, history, anything you can. Do to get the audience up to speed here?

The key thing is that a year ago, there's something called the Genius Act, which did become federal law with massive bipartisan support in the Senate and House. And the Genius Act makes digital dollars legal in the US financial system. The Federal Reserve, the US Treasury Department, all the regulators have made the rules. And as of January 2027, stablecoins like USDC become legal digital dollars in the US financial system and in the global financial system. So a lot of our work is done, like running this new money, being able to support it with all these markets. That's here. We've got that, which is great. Clarity is about like how trading venues get set up, who can trade in a trading venue, the digital asset trading and markets side of things. There's other pieces to it, clarifications around like how DeFi is, is sort of treated and stuff like that. But that's what— that's like market structure, which has to do with like the capital markets. And so I think, you know, there's still a lot of desire to have like good regulations there. But at the same time, I mean, essentially the agencies that regulate these things like the CFTC, the SEC, even the bank regulators too, they're all just going to move forward and regulate it themselves. And then eventually we'll get additional laws passed through Congress.

Got it. Jordi, anything else?

Not for now. Congrats on the launch.

Congrats on the launch. Thank you so much for coming on the show. We'll talk to you soon.

Great to see you.

Have a good one. Let's talk about Sam Parr. He was offered $600,000 to sell his company's data to Micro One. He's not doing it. 100% not doing it. He turned it down. He says, it does interest me like crazy. What a crazy business model, isn't it?

He's tempted.

Yeah.

He's tempted. So apparently— let's finish this post. He says, if you don't know, here's All I know about them. They go to private companies, they ask for your Slack, Notion, email info, offer you 6, 7 figures. What are we doing here? Somehow anonymize that data and they sell it to OpenAI, et cetera. And the interesting thing is that I've been getting these Instagram ads and they're trying to get me to sell the OpenAI data to Micro One.

To sell that.

To sell to OpenAI. And I think I could maybe, one hand washes the other on this one, you know? No, I would never, to be clear. Seems like MicroOne has grown to 9 figures in revenue very fast. The ad on Instagram that at least I'm getting targeted with is $1 to $2 million offer, but maybe that's how they get you in and they work you, look, you are a podcast, you only need this much, I don't know. But he says, what else am I missing here? I get the value of this, but doesn't seem right to me. Interesting. For some companies, I would imagine this being super valuable. The range seems really, really wide. I don't know. But what do you think about data brokerage in the modern era? Are you selling your data?

I just think it's, you know, that in our team chat, like, you got a few people working back and forth on a meme.

Yeah.

A bunch of crying emojis, laughing emojis.

We know that the models are of humor, how will they get funnier if not training on our internal message data? They need to see, oh, that one got 5 crying emoji reactions, reinforce that. That's what happens. This is how we create comedy superintelligence through acquiring the Theo von data trove.

Yeah.

I mean, I think it's quite interesting to look at how projects and tasks get completed in organizations, how products get launched, how deals get done, all these different things. It's just like—

Yeah, wait. I have an interesting question for Sam. We'll have to follow up. Which company? Because he is, of course, affiliated with MedTech.

I would assume Hampton.

Hampton would be the one that— yeah, that actually makes sense. Because, again, you wind up in the situation that it's like, oh, we'd like to buy your data from My First Million, and everything HubSpot HubSpot has as well. Just give us access to the HubSpot Slack, right? I don't think that's going to happen. They probably don't use Slack there, right? CRM competitor? Probably use something else.

Anyway. We can ask Tomasz what he thinks about selling your company's data.

Yes.

Let's pull up this video of PT talking about why the name of your startup is predictive of success or failure.

Oh, did you notice what Jeremy did there? Jeremy, a layer. He was like, we built a layer for transactions. Yeah, nominative determinism in action. I caught it. Did you? Let's play the video.

This is sort of like a slight aesthetic thing I believe in very strongly is that the names of companies are often very predictive of future failure or success. So I'd say PayPal was a was a very friendly name. It was the friend that helps you pay. Napster was a bad name. It was the music sharing site. You nap some music, you nap a kid. That sounds like sort of a bad thing to be doing. And it's no wonder the government then comes in and shuts the company down within a few years. So you want to be very careful how you name companies. In the sharing economy context— Tell me more. I like Airbnb way more than Uber. Airbnb sounds like this very innocent, virtual bed and breakfast, this very light, non-threatening sort of company. Uber, that sort of sounds like a bad name from Germany sometime in the 1930s. You know, what are you exactly above? Maybe the law? And this is probably something that, again, from a government regulatory perspective—

Certainly Facebook-wise, yeah.

I think Airbnb is a vastly better name than Uber. And on the social networking side, I would say that I actually think Facebook was a very good name. I think MySpace was sort of a more problematic name.

Um-hum.

Facebook was— you can say that all these social networks involve both reading and writing. Unlike real life, you learn to— you have to write before you read. You first have to write some things about yourself, then you read more about other people. Over time, reading dominates writing. Facebook was about learning about people around you, about the real identities at Harvard. MySpace started among wannabe actors in Los Angeles, and it was about them coming up with fictional narratives around themselves, and then sort of a lot of other people in LA who are generally like that.

Shots fired, always.

And because reading dominates writing, Facebook would ultimately dominate MySpace. So I think you could sort of— there's certain version where the whole arc of the company was— the whole product arc was implicit in the names.

Interesting. How important do you think naming is?

Extremely important.

TBPN.

I'm overly obsessed with dot-coms at the inception stage of companies too, because there's— one, they go hand in hand, but names are important. I feel like names and dot-coms— And .coms are maybe— the name itself is like 70%, and actually having your .com and owning that is like maybe 30% over time. But when I see founders that are super ambitious and they plan to build a really big company that millions of people are going to interact with, and then they pick a name where they will never, ever, ever get the .com unless they can— Oh, sure, sure, sure. Unless— we actually just saw Runway, the AI company, bought runway.com from Siki's.

Oh, no way.

Yeah. Let me see that.

Siki Chen's company. And again, I expect that Runway had to pay like an insane price because, you know, Siki had a well-funded company himself.

Leverage.

Anyway, so important. Bending Spoons.

Great name. We do have our next guest though.

Ben experienced this later. We can get into this next.

Let me tell you about Console while we bring in our next guest. Console builds AI agents that automate 70% of IT, HR, and finance support, giving employees instant resolution for access requests and password resets. We have Tomasz from Theory Ventures back on the show. It has been way too long. Thank you so much for taking the time. I've read a bunch of stuff that you've written, seen a bunch of stuff, and I'm so excited to get into this talk. How are you doing?

I'm doing great. Thanks for having me back on the show.

To have you.

Are you thinking about AI regulation as much as everybody else?

It seems like it's the topic of the week, that's for sure.

Okay.

You know, we saw Zuckerberg tweet yesterday. I thought with actually a very sanguine view, probably one of the most level-headed views.

Okay.

Debate. Jordi thought it was talking past the x-risk discussion, that he wasn't engaging with the real discussion. Dario Amadei is putting out—

No one— let me pull up the post again. My point of view is, like, totally level-headed, totally rational, saying, like, if Zuck had a leading model and he was talking like that, it would be everyone— it would actually truly be incredible. The reality is that his— the sort of bright spot with MSL right now is Muse, which is a great product. People like it. And literally no one is worried about, like, personal agent app safety. And so when he says people don't want to use agents that are misaligned with them and that don't do what they ask, so labs have a strong natural incentive to make their models more aligned. My point of view is like, no one— of course people don't want to use products that don't do what they want them to do or do things that they don't want them to do, right? But that is not at all like the debate right now. And so I feel like he came and made a bunch of points that sound good if you don't fully understand the debate. Later on, he says, committing the significant majority of compute towards serving people rather than racing towards RSI is one of the best ways to ensure we develop this technology safely. Literally, just over a month ago, he was saying, we need to have a lot of compute so that we can keep up and get to RSI. And so, I feel like he's consistently been focused on saying the thing that will make him sort of popular in the moment that sort of misses the bigger picture. But overall, our takeaway from all of his actions is that he actually has a p-doom of zero, and he just thinks the technology is cool and wants to build, like, successful products.

That's great.

Like, that is what— that's what I consistently, consistently take away. And I, I, like, respect that point of view. I'm glad we have someone in the arena that's, like, playing at massive scale that has a P-doom of zero. Hopefully, it doesn't increase x-risk.

But yeah, what do you think?

Yeah, I mean, I think he echoed a couple of the thoughts that Lina Khan put out, which I thought were excellent, which is there's an existing regulatory framework for companies having responsibility for the products that they develop. I thought that was extremely— it was right on point. And so I think he's arguing for control systems in a way. Like, I think the debate of AI safety has really focused on alignment, not really touched very much on control and how do we build the guardrails around these models for them to be effective. I think that's where this all needs to go.

The liability question I think makes so much sense. Are you grappling with any of the questions about like, does liability apply if no one, if no human says to go do an action, And there's no economic harm related to the action that's taken, but it violates something in the cybersecurity context? Like, I just tell my agent, go solve a math problem. It hacks into your system. Shouldn't do that, but I didn't ask it to. And it didn't take down your e-commerce site, so you didn't lose any revenue.

Yeah, we haven't seen it across company boundaries so much, but we have seen it within a company. So let's say I have an agent and it does something bad, right? So the initial— challenge, I think, is primarily just spending a lot of money. And we see that a lot of executives are worried about blowing past their AI budgets as a result of unprecedented agentic spend. And this is probably the first instantiation or first case within an enterprise where an agent does something it's really not supposed to.

Yeah.

And I, you know, you receive a warning. I think most of the times you might be fired for that. So, in that sense, you're really responsible for the agent.

Sure.

I do think if it starts span across companies. There's law. And I think you will ultimately be responsible for the agent that you start.

Yeah. Is it fair, the conception of just collapsing the debate down to the P-doom equals 0 versus P-doom equals greater than 0 crowd?

No, I think— I mean, we've had a lot of conversations about this. I think whenever you talk about doom, okay, let's define doom, let's figure out what all the conditional probabilities are.

Sure.

The reality is like doom exists in a car, doom exists in an airplane.

Sure.

And so, we accept some probability of failure in those. But I think the conversation, I think the Dario letter did a really wonderful job of taking the conversation of pro-AI, anti-AI, and then stepping it forward into, okay, what is it that we're talking about? And the way I read his letter is, There's action that needs to happen at the individual company level. There's action that needs to happen cross-company, and then there's action that needs to happen at the international level. And that increase in resolution allows us to have the next step of the conversation.

Yeah, and it feels like we're stepping through those pretty efficiently. It feels like the first step of company regulation and independent evaluators, all the companies have sort of taken various levels of steps to either agree or promise or make current commitments around that. The national conversation is happening right now. There are some bills, there are some politicians that are weighing in. It feels like that is progressing. It's going to be slower, but it is happening. Is there any hope for international collaboration? Demis and Sebastian Malaby were talking about, hey, I think that the Chinese are open to something. I've heard other people that are like, this is crazy. There's no chance that there's any sort of international cooperation here.

It's really tough. I mean, you have a lot of geopolitics here at play. Clearly, you know, for the US, this is probably the defining issue of the midterm election. You have data centers contribute 2/3 to US GDP growth. We just saw the Fed raise rates just a couple of hours ago. And so, you know, I think at least within the world of the United States, One of the most important topics of conversation is the economic one. And first, we need to reconcile what happens internally, and then maybe perhaps we'll pursue international.

Yeah. Where are you investing? Where are you thinking of opportunity? It feels like there's sort of— a lot of people have made bets on the frontier labs. A lot of venture capitalists are sort of happy with where their positions have have, you know, or like the ship has sailed on those. The IPOs are almost imminent. But then there's so much work to be done in diffusion. There's a ton of application layer, applied AI, working to actually create legal software engineering work. There's so many opportunities there. And then there's also folks who are going deeper in the stack on semiconductors, neo-clouds, inference engines. There's a whole bunch of other plays that go way deeper down the stack. Have either of those appealed to you recently? Do you think they're equal opportunities? How are you shaping those?

Yeah, I mean, the single biggest market in software today is inference. It used to be the database market. Now it's inference. Just the way the database market's segmented and you have fast databases and slow databases, image databases, video databases, you have the same thing for inference. And so we've invested in one company called Sail that, you know, you can save a tremendous amount of your inference if you're willing to wait 5, 5 to 10 minutes on that inference. Today, most of the AI that we use is instant, and we all pay a premium for that. But a lot of business use cases are slower, like Databricks.

Yeah.

And then we've been researching categories around like voice inference. When you chat with a voice AI, it's actually very different optimization. What you care about is how quickly the AI responds to you. Robotics. And so we've been splitting that up, and we're also investors in a company called Ollama, which is about 10 million people using it initially for local. So you can actually run a lot of AI on your computer. Stanford released a study about 3 or 4 weeks ago showing 90% of white-collar AI use cases can be solved on your MacBook. And then also, if the AI needs more, go to the cloud. Yeah.

What do you think is key in voice AI? Because that can be done on device in many ways, but it can also be done on the edge. We've talked to Matthew Prince, Cloudflare, about putting GPUs on the edge. Then there's also just inference optimizations that even if you're pinging to a data center that's 100 milliseconds away, if the inference is fast enough, it doesn't really matter. Have you consulted that or thought about that trade-off?

Yeah, it makes a lot of sense. So there's 4 different layers of the AI stack. One is you have a phone, goes through Twilio, then you need to convert that phone call into digital.

Yeah.

Then you need an AI model to process it. There are 3 different steps there. Where the inference happens matters a lot. The edge definitely makes a lot of sense. And then there's also time zones, like customer support. You'll have different time zones coming up. You want to make sure that the AI systems are ready for when those loads come. That's called KV cache warming. But so I think ultimately it is a really important category where late You can see it matters a lot. The optimization of those systems is pretty materially different than the way that we use a lot of large language models. And it will likely start initially in the centralized data center, but very likely pushed to the edge.

What— how are you seeing Chinese open source fit into the current AI build-out, AI story? Are enterprises actually deploying them at scale? Is it more like there's an intermediary where an American company will fine-tune or serve that model? Like, how many layers of abstraction are most of the companies away from the Chinese open-source models, which have been doing incredibly well in benchmarks and seem to be very capable but haven't been eating a lot of AI spend that I've been seeing in panel data, at least?

Yeah, the Chinese models, I think, are— some very large companies are using them in production. You'll have different perspectives, particularly from the chief information security officer.

Sure.

Some people will not use Chinese models at all. Most of those Chinese models are typically run on US inference, NeoClouds, and they're incredibly capable. We use them internally. We have backed companies that are using them. We think open source is an absolutely essential component of the ecosystem going forward. And I'm excited to see, you know, the Meta models, the Gemma models, Clearly, NVIDIA is pushing quite a bit with the NeMoTron and the acquisition of Hugging Face plus the Poolside semi-acquisition, let's call it.

Yeah.

So, it's essential. I think one key question that hasn't yet been answered is if an American company fine-tunes a Chinese model and runs it on US infrastructure, does the market perceive that both economically and politically as a US or an American model or as a Chinese model?

Depends how American the CEO of that company is, I guess. If they announce it on Joe Rogan, I think it'll be well received.

Walk me through how you think that platform VCs are justifying doing some of these vertical AI companies at like 50 times, 50 to 100 times revenue today. It feels like a year ago, a lot of the conversation was around like, These aren't software budgets anymore. They're capturing some amount of labor spend. Unclear how true that is today, even if you're not seeing a correction in the labor market. But again, maybe firms are just hopefully doing quite a bit more, and so that you are actually getting into labor spend by just replacing an incremental hire or something like that. But comparing some of these vertical AI companies to the prices that Bending Experience has been paying, it just feels like an incredible disconnect unless these vertical AI companies, like, you just look at them and I'm wondering how they get into the billions of dollars of run rate, which I think right now the industry is pricing them That at least some of them need to hit that for the category to work out.

Yeah, you're right. I mean, we've benchmarked the, you know, AI harnesses are now the fastest growing ones are trading between 100 to 150 times current ARR, which is an enormous multiple, particularly at that level of scale. Sometimes it's subsidized with lower gross margins. And the idea there is primarily just let's get the product out as far as possible and then ultimately ultimately will improve the economics. In their favor, I will say we made a prediction at the end of '25 that 2026 would be the first year where employers would pay agents at market level for a person or more.

Hmm.

And that happened. The first time we noticed that was actually in March or April of this year. So, we have a portfolio company that charges at par, and it will likely charge at a premium To a person. And so, if that's the case, and it's not necessarily human replacement, it is more augmentation, is the ability to do more, process more. But if that's the case, and you could argue economically, there's no management, there's no healthcare, and so you should actually pay a premium to a human user.

Well, and it's always been that, like hiring a freelancer, like an expert freelancer is always more expensive on a per-day basis than a long-term hire, right? Because you just want to be able to get that expertise immediately, but then not pay for it when you—

It's super hard to get 10 freelancers to show up for 1 week and then go away for 2 weeks, and then 2 of them come back for 1 week, and then 2 days over here. And obviously, a lot of services, depending on how—

Do you think we'll get an American Bending Spoons? I've had this idea recently of how many How many companies need to be acquired at like 2 to 5 times revenue for American VCs to say, well, why don't we put a few billion dollars into a firm that does at least so we can monetize the way up and the way down? Do you think that'll happen? Because it feels like I look at some of these acquisitions and again, these are not companies that people are generally excited about, but I think they're going to be pretty durable for the most part. And I think part of the reason they're getting such good pricing right now is no one else has basically the stones to go out and say, yeah, I'm going to buy Airtable, right? Like, yeah, I want to be that. I want that to be my problem, you know? But a company set up entirely to do that, and that's the core bet, I think could make sense.

Oh, absolutely. I think the multiples for some of these companies are really quite small. The key metric there is just net dollar retention and gross dollar retention. So, some of the legacy, we'll call them legacy software companies for a second, their revenue retention rates are still phenomenal. And so, there's a multiple arbitrage opportunity. There are a handful of companies that have been, or investment firms that have been doing this for a long time. And I'm surprised we haven't seen them be more active. And then you also have holding companies that have bought a lot of these businesses, like Danaher would be a publicly traded example of a holding company.

Yeah.

Constellation Software would be an example of another one. And so, I would expect them to be pretty active just given— I haven't looked at their activity in a while, but I would expect them to be pretty active just given how attractive the multiples are.

Is there demand or at least gesturing towards demand from LPs for VC firms with companies from that vintage, companies in that position to get cleaned up? You know, that's something that bending spoons offer is like there's a logjam, maybe you still have a board seat and it's like, hey, we actually want you to maybe realize not a great return, but then you'll be focused on the next era. You can be 100% in on this wave, which we still want to back you on. Is that something that actually will come from LPs or is that just a separate calculus?

No, I think that's right. I think GPs and LPs LPs definitely want to, I mean, in certain many cases, move on from those positions just because the industry has changed and liquidity is absolutely important, although '26 is a monster year for liquidity.

Yeah.

But just cleaning up those fund vintages and migrating those companies to the next person who might be interested in managing them the way that you talked about is a big part of the industry. It's in every LP conversation that we have, managing liquidity is is among the top 2 or 3 topics.

Yeah.

What— how do you expect the economics of the personal sort of new category of personal agent companies to evolve? I'm thinking Instinct and Muse. A lot of people are using these agents to just say like, hey, go buy this product on this website, right? It's— there's no like discovery happening. There's no— it's truly like the end customer just telling a piece of software, go buy this thing. And so I think Instinct has said they don't want to charge for it. Muse clearly is going to just try to make it free effectively forever. But I think both of them— it's hard to imagine ads in an Instinct workflow right now, given that— just thinking about the surface area. I'm sure Meta will figure out a way to put ads in it. That's going to be a problem. But when you think about affiliate and taking cuts when the intent is just coming from the user, like if I'm a retailer, and an agent comes to the website and buys something, but I know that the individual just directed the agent to do it, I'm not exactly sitting there being like, thank you, now here's your cut, because you didn't actually drive the demand. Like, other marketing or other things were happening in the world that drove that person to decide, hey, go buy this thing. So I'm wondering, like, how the sort of economics of this new category will evolve.

I think it breaks down into 2 phases. The first phase is really about data acquisition. But the most valuable data, I think the estimate was about $10 billion in data, is being spent this year to train some of these models. And so, as we think about like what Cursor has done or what Meta is doing with Muse, they very likely want to develop very specific models that are incredibly efficient to serve. And to do that, they need the data acquisition. Makes sense to subsidize it for a while until they get enough trajectories to be able to fine-tune and serve a very efficient model. So, I think that's probably Priority 2. Priority 1 is distribution. Priority 2 is scaling the cost. And then, and then third is monetization. And just given like, you know, Google's distribution and consumer, Meta's or Facebook at the time's distribution and consumer, Snapchat, same thing, Pinterest even, the ultimate monetization can be very powerful when there's a new dataset to monetize. Google's with search, clearly Meta's with social. Snapchat, Pinterest, new targeting mechanisms. I used to— was a product manager on the ads team at Google. And so every— this is the way I think about that ecosystem. Anytime you have a new targeting criterion, you can have a multi-hundred billion dollar company.

Yeah.

And these trajectories are unbelievably valuable. The average user on Google in the US generates about $120 in ARPU. It's very easy to see a doubling or tripling of that with these systems. And so I think a lot of us in the venture markets are willing to take that on faith.

Yep. Yeah, no, that makes a ton of sense. Um, well, thank you so much for coming on the show.

Yeah, I wish we had more time.

Yeah, there's so much more we could talk about. I have a million more questions, but our next guest is in the waiting room. So have a great day.

Let's do it. Wait, wait, last final, final question.

Final question.

Have you ever seen Spencer surf?

You're, uh, he's incredible. Yes, he has He is like—

I would go out on a limb and I would say he's probably the best surfer in tech, period. I don't know of anyone— I don't know anyone that would actually be able to go head-to-head with him and come out alive. He's an absolute animal. Seeing him surf made me want to do a TBPN surf invitational at Kelly's ranch.

The surf ranch.

Yeah, yeah, yeah.

He's incredible. He has this video inside of a barrel. He's just an exceptional surfer.

So he's definitely the best one I know surfing the capital markets with you, though. So that's great.

Great to see you guys.

Good to see you.

Cheers. Great one. We'll see you soon. Let me tell you about the New York Stock Exchange. Want to change the world? Raise capital at the New York Stock Exchange. Just do it. Our next guest is already in the waiting room. So let's bring in William Layden from Rune. Co-founder and CEO, building a modular data center that converts unused solar power directly into AI compute. How much unused solar power is there? I feel like we'd be using it all.

Hey guys, that's the trick, isn't it? Yeah, over 50 terawatt-hours every year in the United States alone.

And is that residential? Is that just solar farms out in the middle of the desert? Like, why did we build it if we're we're not going to use it?

Yeah, I think— so we're exclusively focused on utility-scale solar farms, which is the large assets. Think 50, 100, even 400 megawatts of installed capacity. And I think that abundance or excess is actually a feature of renewable energy and not a bug. So you typically build a power plant to meet the highest hour of demand. But solar, it's relatively cheap to build. The sun's zero-cost fuel, so we tend to overbuild, and as a result, there's a lot of excess power.

Interesting.

So are tokens going to get cheaper in the summer in the long term?

Well, if we get big enough, then sure, maybe we'll have seasonality impact on tokens.

Okay.

Talk to me about the importance of modularity, because You mentioned a few different scales of industrial solar farms. And it feels like those would require— those would support very different configurations of, I imagine, inference GPUs, right? So you have to be able to show up. And you can't show up with a gigawatt worth of compute for a solar farm that can only, at max, put out 100 megawatts, right?

Yeah.

Yeah, that's right. So we are deploying what's called, you know, our product is called the Relic. It's essentially a micro shell that has a server inside of it. And we deploy many of these data centers, couple them to create a large cluster size, basically the largest, as much of, as large a cluster size as the solar farm can tolerate, we will deploy. So we're working with like a 400-megawatt solar facility. We think we can probably put 100, 200 megawatts of solar there or of data center capacity there.

What's the go-to-market like? Who do you actually have to negotiate with to, A, get space and supply and actually deploy these? And then is it as easy as just hooking the system up to OpenRouter and serving up tokens? Is it, is it if you build it, they will come at this point?

Yeah, I would say so. So, you know, we manufacture, as you said, modular AI data centers and we use power electronics to plug them into these facilities. We can deploy, you know, a unit in about 60 minutes. So we have a guy with a forklift come drop it down. There's no concrete, there's no construction, there's no modifications. No, we drop it right on the ground. There's no concrete, no modifications. We take 2 wires. And plug it in. And it is truly the fastest, least expensive, least intrusive data center out there. And then, in terms of our customers, we're actually selling additional capacity.

No noise, no pollution.

Yeah.

Like, it feels—

This seems popular. Somebody was asking me yesterday, last night, like, what does it take to make data centers popular? And I was like, solar for sure. And it seems like you've been obviously ahead of the curve on this. What if you go more modular? What if every solar panel came with a GPU attached to it by default? It could always be in inference mode and flip over. Is that the future? Or do you think that this like, you know, mid-scale modularity, like there's some economy of scale that comes from like marshaling a lot of energy together, or maybe even not a lot, but like some threshold amount?

Truly, I think every solar power plant is a latent data center, and we can convert it with our technology. The same is true for wind. So I can see a future exactly like you said, where every solar plant, every solar module, every wind turbine comes with a Roon data center.

Interesting.

And that's how we power the future of compute.

So I imagine that although you're focused on solar, you're already thinking about wind, as you mentioned, hydro, and other places where there might be some latent energy, but it requires the modular solution that you're building.

Yeah, the modular solution, the power electronics, these are absolutely kind of our secret sauce to make this work. You know, solar is the most abundant energy source we have, and it's the one that we are the worst at using. And Roon is designed to make it, you know, make it useful. Let's make the sun power the future of compute.

What were you doing before this?

I started my career working for President Obama in the White House. I joined a hydropower company and then was most recently at SB Energy.

Nice.

Why—

feels like the most perfect entry point into this company.

Overnight success. Why is it the— what was the chart that we were pulling up from the IEA? Was that right? The International Energy Association or something like that? Agency? The IEA, every year they predict how much solar we're going to build, and every year they get it wrong. What's going on? What's going on with the IEA? Why can't they forecast accurately?

Well, it's Paris-based. I don't know.

Oh, have they never heard of exponentials or something? What's going on?

I think that's the power. You know, you guys were kind of focused on modularity, and that is the power of modularity. We're a product company. Solar is a product. It's not a construction project. Sure, it can scale much faster.

Okay, but why haven't they internalized that yet? I feel like at this point it's like, just fit a different curve. If like the linear line is not working again, let's go with an exponential this time.

Yeah, we'll have to make a pit stop in Paris and talk to them.

But separately, like, do you have any worries about us hitting a ceiling on deployed solar? Because I see that curve and I'm like, this is amazing. This is the ultimate technology white pill. I've never met anyone that doesn't really like solar. Nuclear, I still get people are, oh, what if it blows up? You got to work through that. Obviously, natural gas and oil, there's a whole bunch of complaints there. But solar is really, really popular. But at the same time, I know that there's a complex supply chain, there's geopolitics.

Is there any risk that I think geopolitics is always a challenge, particularly in today's environment. But truly, I'm a solar maximalist. It is the fastest and easiest way to create energy. And even now you're seeing companies that use robots to install solar power plants. And when you go to these sites, you know, you go to Texas, you go to Nevada, you go to Arizona, and you see these massive solar farms. There's so much unused land, and we can really use that land to create abundant energy via solar. We're nowhere near saturating the solar market in terms of how much we've deployed.

Last question, and if you don't have a strong answer, you can just email us. But I've been on the hunt for, like, the Elon Musk of solar. You know how there's, like, the Palmer Luckey of defense tech who makes, you know, big waves and, like, sort of popularized—

This guy could be looking at the Elon Musk of solar in the eyes and not even realize it.

Maybe, but I'm talking about someone who's specifically focused on building a company that will deploy the most solar panels, build the panels, really be like the mega winner, the public company, the voice of the industry. I don't know if that person is on the tip of your tongue or someone you need to think about, but I'm really— I am hungry for that person. So if you know them, let us know. I'd love to have them on the show.

You know what? I've always been an admirer of Sheldon Kimber.

Okay.

He built Intersect Power.

Okay.

And I think he's been a visionary in the solar space. Okay.

Amazing. Thank you for the recommendation. We're definitely going to check it out.

Who did the round?

Oh yeah, let's hit the gong.

Yeah, yeah. Spark.

Boom!

Just needed one word.

Very cool.

Yeah, Spark.

Spark Capital led the round.

$40 million.

Congratulations.

Thank you.

Thank you so much.

I'm glad you're doing well.

Great to talk to you soon. Have a good one. 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. Our next guest is Justin Barrows from Rainco. He's the founder and CEO, and he's here to talk about Navier-Stokes. What does it matter?

Doesn't it?

Does it not matter? We were having this debate.

Yeah.

Seems like a cool demonstration for the power of mathematical models and artificial intelligence. Great benchmark. Clearly, everyone was trying to do it, so it's cool to be first. Controversial among mathematicians, but my question was like, this is fluid dynamics. We all want planes that use less gasoline or diesel to get around, jet fuel to get around. We all want more efficient wind energy, all the different things that you can get from understanding turbulence. Does this have any real-world application?

Yeah, man. Well, first of all, thanks, folks, for bringing me on. It's certainly a pleasure to be here. So regarding the practical aspect of this, so I should— short answer is no. Long answer is let me tell you about it.

Okay.

So, you know, Oftentimes the way I view this is there's kind of 3 hierarchies of theory. So there's the mathematicians at the top, the pure mathematicians, and what they say, that becomes the applied stuff for the theoretical physicists, which then becomes the applied stuff for the engineers.

Yeah.

So I should describe what is this open question. It's really about a particular mathematical aspect of the Navier-Stokes equation. So what's been a long-running problem is, you know, can we For instance, can you come up with any scenario in which you start with a physically meaningful initial state of the flow, and then can it time evolve to do something that's not physical? So in particular, what the worry has been is, is there any point in the flow that can attain an infinite velocity? Obviously, that doesn't happen in real life. However, we don't have any firm mathematical proof whether that can or cannot happen. What OpenAI has done is they've essentially found a counterexample. So they constructed a specific example where you do, in fact, get blowup. You get a certain point within the flow that attains an infinite velocity. Mm-hmm. And so if this proof holds up, you know, first of all, it's important to mention that they've submitted a proof, but, you know, it may take years for the actual experts to review this thing and figure it out. But essentially what it does is this is putting a guardrail on the Navier-Stokes equation as far as its applicability. So you can imagine that if you have a— well, in any real flow scenario, obviously you don't see this, right?

Mm-hmm.

But it is hard guardrails in the sense, like, you know, if you're running a simulation and if it's not converging, You know, this is another thing to add to the checklist. Like, you know, maybe there might be a singular— excuse me, a singularity developing in flow.

Sure. So, there are at the same time on the opposite side, the engineering side of the world, there are lots of companies that are trying to develop supersonic airplanes, hypersonic airplanes, drones, all sorts of different things that need to do CFD and model fluid and turbulence. There's real-world work to do. What is that community clamoring for? What do they want to be advanced? What are their biggest problems?

Yeah, so I'm reminded of the quote, mathematicians can tell you if a solution exists, they don't tell you what it is. And that's kind of relevant here. So, you know, the proof that Navier-Stokes did, you know, it's a point in terms of a specific mathematical property, but the real-world problem of give me a flow scenario and tell me what exactly is going on, what are those dynamics? That is still a wide-open problem. And I should mention that this is one of 2 big problems related to Navier-Stokes. So there's the million-dollar math Clay Institute one.

Yeah.

And then the other one is called the turbulence problem. This is more of a physics one. So, you know, so what's known, you know, the difficult thing about fluids is that typically they destabilize and they make these complicated swirling flows. So we've probably all seen smoke coming off of a cigarette and then it starts tumbling around. That's an example of this phenomenon of turbulence. And having a complete mathematical framework to describe that, that is also an open question. And that's been around for about 200 years. That's actually the one that we're working on as a company. So I should mention my background. I'm also a theoretical physicist, and we're working on coming up with the general computational and mathematical tools to solve real-world problems.

And are you trying to build deterministic tools, or are you trying to build AI tools that will just approximate turbulence at such a reliable level that the engineering will be advanced?

No, yeah, so this is all pencil and paper, man, you know, we're mathematicians, you know, we're working out the general mathematical framework. Yeah, I mean, this is, you know, I should mention that the current state of the art, you know, because we don't have a general method to solve this equation, if you actually look at the kind of simulations we run, you know, it doesn't matter what the example is, maybe this is aerodynamics or hydrodynamics, it could be weather and climate modeling, heat transfer, chemical mixing. These are all examples where fluid flows are really core to the industry. The state of the art is, you know, these simulations are really not predictive. So if you peer under the hood, what you see is we actually don't know what the equations of motion should be that we are solving. So, you know, for the technically minded audience, I should mention that, you know, these are called the so-called closure equations. But basically, we have— when we run simulations, you have a bunch of free parameters. So you need to start by doing experiments, and you take measurements from the experiments to inform what these free fit parameters in your simulator are. And as you can imagine, having physical prototyping as part of the iteration design loop, that's the most expensive way to do it.

So step back a little bit. We jumped straight into this, but give me a little bit on your background and then I want to walk through to today and the actual structure of the organization and the plan for what you're Yeah, I guess, you know, quick background on me.

So I started out as a mechanical engineer. I have bachelor's through PhD in MechE, primarily focused in precision mechanical design. So originally I was one of the guys who was designing, building stuff in the machine shop, and evidently I went through some sort of quarter-life crisis and ended up picking a second— picking up a second PhD in theoretical physics as well. And really, the company that we're— that I'm building now is the outproduct of that work. You know, as I mentioned, you know, this problem about solving turbulence, this is a longstanding open problem. And I got interested in this a few years ago. And after plugging away at it for a while, I actually penned that closed-form framework. So, essentially, what we're doing as a company is we're hiring a bunch of PhD and postdoc-level theoretical physicists and applied mathematicians. And we're taking this new theoretical framework for Navier-Stokes, and we're, you know, basically going after everything that relates to And how do you plan to productize this?

I mean, it is a business. It's not a nonprofit research organization. I imagine that there's a different path that you could have taken, but you chose to build a business. How do you see that developing?

Yeah. So, I guess a couple of things to say. I mean, one is that my engineering background has given me a very good sense that We really, really struggle when it comes to engineering fluid systems.

Mm-hmm.

So, you know, I'm well aware, like, you know, the market need, the market need for a solution to the turbulence problem has been here for like 400 years. You know, that hasn't gone anywhere. The way we're thinking about it is essentially breaking this down to 3 steps. So, as I mentioned, the first thing we're focused on is better computational software. You know, I mean, step 1 is you want to be able to simulate something in a computer, know that you're getting an answer that you can trust, and do the design and optimization in the computer first. And then you build the one prototype at the end to confirm that things are right. So I should mention, like, this is what's done in every industry except— or every engineering discipline except for fluids. It's really the one holdout where you're really wedded to experiments. So we're essentially— the company's going to develop in 3 stages. So we're starting first with better software. We're a couple years away from releasing our first commercial product.

Mm-hmm.

But afterwards, there's a couple things we're going to do. So one, of course, is modeling. So there's a number of examples of industries where there's important hair-on-fire turbulent flow problems, and the most important ones we want to go after ourselves. And then long-term, we want to get into hardware. So, think hardware, software, control algorithms to control and mitigate turbulence in number of canonical flows. So, this will take us, like, way into the future. I mean, you know, it's kind of like if you solve the turbulence problem, you've just opened the doors to the playground, but there's so much to do and explore and so many ways to add value.

So is the team right now all focused on that phase 1, all sort of theoretical physicists and scientists effectively, and there's no sales guys running around just yet?

Well, yeah. So as far as the technical team, yes, it's—

Oh, so you got some sales guys? I'm hearing—

We do.

That was actually our most recent hire.

You got a sales guy, yeah.

We're a small startup out of the Brooklyn Navy Yard. So we're a team of 8. Our most recent hire was in fact our chief commercial officer, and he's absolutely fantastic. So, you know, I would say like the primary thing is the technical people, the engineers and the physicists and mathematicians, but then there's a glue that's holding the company together. So my COO, myself, we have our chief commercial officer and our administration person.

Very cool. I'm interested in one of the critiques that I heard on the Navier Stokes' solution was that it sort of like ripped the problem apart into such an extreme position that it was like, it was sort of impractical, or it was not like the most elegant or like it wasn't the way you would expect a mathematician to solve it. It was sort of brute force. And I'm wondering just, I'm sure you're using different AI models. Is it accelerating your work? What part is accelerated the most? Are you running into any guardrails or any hurdles as you try and collaborate with artificial intelligence? Like, what's your experience been?

Yeah, that's a great question. So first of all, I don't know who you heard that from, but that's exactly right. So when I read the proof from OpenAI, I had the exact same feeling. You know, I was kind of naive and I kind of assumed that the PDF write-up, you know, had some sort of human involvement. And after the first 2 pages, I was like, wait, this doesn't sound like something a human did. And it's not. The whole thing is AI-generated.

Yeah.

Yeah.

So, I mean, from, you know, it's not elegant, but if it's right, it's right.

Exactly.

You know, one of the big things— and I'm sorry, I'm taking a long, circuitous route to answer your question. But, you know, to me, I see an analogy between what AI is doing for math and what, for instance, we do by running experiments in more applied fields. So in some sense, you know, up until this result, the idea of proving something and understanding something have been synonymous. And there's an underlying assumption that it's really human beings who are jumping in and putting the proof together and developing the understanding along the way. But what we're seeing here with, with, with AI is that that's no longer the case. You can actually produce a proof that may in fact be correct without understanding anything about it.

Right.

So the way I see this is this, this is sort of like running an experiment where you can observe the right result. But, you know, the whole process remains in front of you on how do you actually understand it. And, you know, and this is You know, part and parcel in engineering and physics, right? You run experiments and you get results. You know that they're right, but you don't understand it for a while.

Well, good luck and thanks so much for coming on and giving more context about it. I love that you're thinking in multiple acts. You're thinking in applications here. It's going to be exciting to follow your journey. So have a great day and thank you so much for coming on the show. We'll talk to you soon.

Thanks.

Have a good one. 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. Jordi, why did you put on that Tyler mask? It's so— it's so photorealistic.

Does it look realistic?

It looks extremely realistic. Now, through the power of switching technology and cutting camera angles, Tyler has subbed in for Jordi because Jordi has to take— had to get out of here early. We have a bunch of We have three more great guests, so stay with us. Thank you so much for tuning into TBN today. Up next, we have Eli Walks from Footprint, building an AI operating system for financial crime compliance. This is going to be important for you to pay attention to because I know you're thinking about doing something. Tyler over here, he's one step away from committing some heinous financial crimes. So how can I keep an eye on him? Do I need to pay you?

It's a great question. Look, we are at a time where bad actors are using AI far quicker than any bank or financial institution, and it's become the 4th largest economy in the world. What is $4.5 trillion moved illicitly each year?

Wait, wait, wait. So, so like financial crime, it's like US, China. I don't know who the 3rd is. Is it Japan or Germany? Germany. Okay. Germany. And then financial criminals.

Wow.

Like the Sinaloa Cartel and their friends.

And that probably also includes like crypto heists, North Korean hackers, that type of stuff. But then also just like little phone scams trying to get grandma to send you $100 of gift cards, that type of thing.

Exactly. So you have them on one hand.

Yep.

Using AI since GPT-3. And then the other, you have 1.5 million people who are supposed to by hand investigate any suspicious wire that comes across their desk.

Wow.

Days or weeks. It's a system that's been very unprepared.

Yeah.

And the gap's only growing.

Okay. So how broad are you focused? It seems like AI operating system. I imagine you're plugging in deep into financial rails and trying to at least set your system, your operating system, your company up to both be able to handle the, the big sophisticated crypto heist and also the, the fraudulent bank wire for a couple hundred bucks?

Yeah. So we work with some of the largest fintechs, crypto companies, financial institutions. The idea is at both the second line of defense, which are core compliance use cases, AML, transaction sanctions, sexy stuff. And then the other one would be called first line defense. So payments, ATO, any other type of fraud scheme. The idea is that we are giving these customers essentially infinite time and infinite memory. So it used to be cartels win in numbers, similar to how you ship drugs across the border. You put cocaine on enough trucks, some of them are going to get through.

Sure.

You create enough shell companies, enough of them are going to receive money.

Sure.

And people didn't have the ability to investigate each of them. So the first thing we do is we give them the compute to investigate every case that comes across their desk. And the second is very unique to Footprint, we give them the ability to remember every case that they've seen. So that way they're able to essentially compare vectors and prevent financial crime before it even happens.

Okay. You said you give them compute. If we were talking a couple of years ago, I would be thinking this is an annual contract. This is like a SaaS product almost. But it feels like the business model for this business might be a little bit different in the modern era. How are customers actually compensating you and, and working with you in an elastic way that allows them to scale up and down?

We charge by usage.

Okay.

It's essentially credits based off of the complexity of the case. So you may have the same genre of case, but one, we're only looking at a couple of pieces of information. Let's say we're investigating a fraud scheme. Maybe one's very cut and dry. Versus there may be one where we actually want to have recall and go through embeddings of thousands of similar cases. We had a case where there was a flagged Russian individual and our agent went back into Ukrainian newspapers, read the initial Cyrillic because Cyrillic can be translated improperly, then tracked down this person's wife to a home in New Jersey and found a real estate listing in their name. So you can also get into much more complex things that you never expect a human to done before.

Yeah.

So we charge the treading caps like that. We also have our own model routing internally.

Okay. So on model routing, it feels like some of the work that you do, it's going to wind up looking like fraud because you're— a lot of the prompts are like sort of sketchy adjacent, even though they're the opposite and you're actually fighting the financial crime. Your prompts are going to look financial crime related. You might get flagged, you might get, you know, refusals. Have you solved that by fine-tuning models using open source or partnering with labs and doing deals to get whitelisted? Like, what is your strategy for maintaining access to effective intelligence to fight these crimes?

All of the above. We have really good relationships with the frontier labs. You know, I think that we all view this as one of the most important use cases of AI. I'm very biased, but I think that we either— if AI grows the economy by 14% by 2030, that means we either add $600 billion into the pockets of the worst people on the planet, or we bring that down to zero, which I think is possible. So you're right, our prompts do look interesting.

Yeah.

Because we have to think like financial criminals to do it. And then we do some of our own model hosting, but we also have routing in that some banks may only want to use 2 models.

Sure.

We also, we may take— we recently had a bank put in a 150-page policy document that's taking thousands of rules that have to be followed. And each of that essentially is getting its own subagent in its own environment. That's why I think compliance AI is 2 years behind legal AI, because there's so much more regulation and such a higher barrier of trust to overcome.

Do you—

I guess I'm thinking if, like, do you have either benchmarks as you're fine-tuning models, testing models? And do you notice anything particular about models where you're like, oh, it spiked on this particular vector of fraud detection, do you notice the spiky intelligence behaving differently? Because the X algorithm will be obsessed with a particular model typically by how viral the creation can be. So if it's a Blender 3D model or a video game, everyone crowns that the best model of the day. But no one's really like, wow, this model is incredible at detecting financial fraud because it's probably just collapsed into some sort of score. But what are you doing on actual model benchmarking evaluation? And then how do you communicate that to your customers?

Yeah, I think evals are pertinent to any vertical AI company, but especially us, in that we need to make sure that when a new model comes out, it's not spinning its wheels and it's actually getting to the right answer that we expect.

Sure.

So we have a huge library of synthetic cases that are based off real cases that we're leveraging to understand. I think one of the powerful things about Footprint versus, you know, like an AI wrapper is that we are built with access to hundreds of databases around the world to fact-check the open-source research of an agent. So we have cases which may involve an agent going and verifying a business in China as part of an investigation. It used to be in this space you had ML-based detection tools that would flag an alert and humans would investigate it. it was like a very cowardly industry of vendors in that we're saying all of the tough stuff people actually have to figure out. And what Footprint's done is we've essentially taken AI and we let AI orchestrate those initial detection tools and databases with AI on top of it to essentially make this a dynamic process of detection and investigation.

What about sort of like benchmarks or like KPIs for the overall fraud industry? How accurate are you on that number of The fourth largest economy. How and how rigorously is that tracked? Like in a couple of years, will we be able to run this interview back and say like, okay, yeah, it's like the ninth biggest economy now. We're making progress because that's what success looks like here, right?

100%. That's the goal. Fraudsters are really smart. If the cartel has $5 million of cash in the US, 70% of the time that goes to a Chinese money broker because you can only get out $50K. Acquire, and then the cartel sends 10 million dishwashers to the cartel in Mexico. They charge them $5 million, they sell the dishwashers. Really good operation. I would love to invest. I would love to buy away. I think that it is possible for us to go down. I think when you look at benchmarks, it may sound ironic, but you're often hearing BSA officers say 95% of our AML hits are false positives, 75% transactions are false positives. We actually want that to be 99.999%. Because we're only looking at what's flagged as the highest risk transactions in the future. We want to look at every case because you don't want an agent to have memory of every transaction that's ever flown through every account that's ever been open. And then you can compare those. That's what we've built with Trust Fabric, with memory footprint and that recall. Yeah, ideally we go from 4 to 40th. I say our mission is to topple cartels and corrupt regimes. So, look, I don't think we're going to start with, with Russia, but Maybe Turkmenistan or someone out there.

Do you expect any cartels to spin up clusters somewhere, start doing pre-training, completely build their own fraud models, fraud OSI? Do you expect this to happen soon?

I think they would do well on Sand Hill Road. I'm not sure. What I will say is they're sophisticated. So you have the cartels, the most sophisticated is the Golden Triangle in Southeast Asia. Where you essentially have these multi-thousand-person factories. It involves human trafficking. At one point, China— if you look at the rise of pig butchering in the US, it's essentially because Xi Jinping sent in a junta, or sent in essentially, like, funded a junta to go and destroy one of these centers in the Golden Triangle. Because all pig butchering going back, like, early in 2020, 2021 was targeting Chinese nationals because of the language.

Oh, interesting.

So he struck a deal.

Wait, it was targeting Chinese nationals Who live in America?

No, who live in China.

Oh, who live in China. Okay, so he was like, you're attacking me, I'm gonna go after you. Interesting.

Yeah, and he said, look, if you wanna keep these centers open, you gotta go after somebody else. And then you see a massive spike in pig butchering in the US, you know, 2023, 2024. But this is all geopolitics coming together.

Yeah, yeah, fascinating. Anything else? No, I think that's great, yeah. Well, thank you so much for coming on the show. Have a great day. And you raised some money, $25 million, right? Series B?

Yes, we did.

I gotta hit the road.

Series B.

Thank you so much for coming on the show. We will talk to you soon. Have a good day. Goodbye.

See ya.

Uh, quick hit on the news, breaking news. Chipotle has partnered with Palantir. The burrito chain has enlisted Palantir for a platform that monitors pest incidents, employee illnesses, and other factors The partnership comes after a nightmare summer for food safety across the US. Good news. Good news for both companies. We want safe food. We want burritos that don't make us sick, especially if they're coming before I even think to order them with burrito superintelligence from one of the companies. It feels like it's a race now. The DoorDash guy was talking about building it, and then that other entrepreneur just did it. Anyway, we have our next guest in the waiting room.

Hi.

I'd like to bring in Sean McCarthy from BackOps. How you doing, Sean?

I'm great.

How are you?

I'm great. Welcome to the show. First time on the show. Please introduce yourself and the company.

Yeah, I'm Sean McCarthy. I'm the co-founder and CEO of BackOps. BackOps is an AI-native resolution layer for supply chain. So we target companies that make or move physical goods. You can think as a consumer, something goes wrong, You order something, it's broken, what happens behind the scene? That's going to be us. So happy to chat a little bit deeper into what we're doing, but looking forward to it.

Yeah. What's the median back office supply chain tech stack look like these days? Has everyone moved off of pen and paper? Have people moved off of spreadsheets? Are they at least in legacy ERP systems with green screens?

You would hope, you know, that's where we started. I was at Amazon prior to and spent a lot of time in customer warehouses. And, you, we started around the consumer side, you know, more like you or I order up a lamp and it's broken again and something happens. And now we really service the enterprise. But it's the same thing throughout from the mom-and-pop shop to the, to the large enterprise. So some do have, you know, SAP and—

Yeah.

Oracle and some of these other, you know, common systems of record. But the Excel, I don't think we've met a company that doesn't use Excel spreadsheets. At least pre-BackOps.

Well, I mean, it's good that all the models are getting really good at speaking Excel. Is it easier to build integrations? Did you have to sort of like pick a landing zone where you're going to be really great at SAP first, or are we actually in an era where you can sort of go customer by customer and say, look, I don't care if you're using Excel or Google Docs, like we will build the integrations that we need in a weekend and we're good to go?

It started more— so we started in 2024. When we started, it was more the former. So you really had to kind of pick your lane. We started in the warehousing space. And so we picked kind of our top 10 integrations. I think a lot of it was just that the browser integrations just weren't quite there, right? And a lot of times folks were also hesitant to give API access or just didn't have it. And so you kind of were pigeonholed. And in that sense, now, you know, we're at a point where it really doesn't matter. We have customers that have mainframes all the way through, you know, all the common systems records. So that is a good thing because I think prior to that was a big hindrance in just like getting something live.

Are you seeing acceptance or pushback around computer use agents being deployed by a third party? Like if I have a If I have a computer in my warehouse with some, you know, bespoke piece of software and you say like, hey, let me install something that's just going to move the cursor and type stuff in and use it remotely, that could be awesome. It could also be sort of weird. How are customers like ready to have that conversation?

Yes and no.

Okay.

There's kind of 2 frames to that, right? Like you have the internal side of, and normally we're asking them to provision us like a contractor login. So they're not necessarily seeing it in front of their face. The other side of that is when you look at terms and conditions across like third parties with bot detection and things like that, that's getting much better. And so we kind of have a twofold hill to climb there. But I think in general, when you look at the first problems we're solving for customers, it is a lot of the stuff that the team doesn't want to do. So they are happy to like handed over on a silver platter to us. But luckily, they're not necessarily watching a screen with the mouse moving and clicking into their portal.

Yeah. What's the killer sales pitch to someone? Is it cost savings on the back office supply chain? Are you actually able to drive incremental sales or revenue, or are you selling it as like you'll be able to scale your business more with less headcount, or is it like, we'll actually just save you a couple extra lamps that didn't make it back onto the shelf, so you're going to be saving cost of goods? Like, how are you pitching the value?

It could be—

it's a combination of all of the factors. I think the biggest thing that we see is an OpEx reduction. So where you might have had multiple folks that needed to be hired to fulfill like a large claims team as the business grows, you don't have to hire those folks on just the approvals, we're talking about claims alone. We see at least a 13% higher approval rating in the system filing versus a human. Right. So that is actually money back in their pocket as well that they would have missed. And then I think that obviously the customer experience—

really quickly, approval rating, is that like someone sent back the lamp, you can actually restock it because in this case it's not broken, it was just the wrong lamp. So restock it as opposed to send it to the dump?

It's more of, let's say you order the lamp, it's delivered, right? You call and you say, hey, send me a new one. Yeah, they're going to send you that new one, but then they're going to file a claim with FedEx or UPS, whoever broke it. Right. And a lot of times a human's filing that. If it gets denied the first time, they'll just say it was denied. Well, we'll try 3 times. And so a lot of times on the second or third, we're getting it through.

Yep.

And/or you can think of what if we have to reach out to you and ask for a picture of the damaged lamp? Or, you know, on, you know, one of our largest customers is one of the biggest grocers in the United States. A lot of times the complexity of the things like, let's say it's a cold temperature breach that we have to go find from just the fact-finding and the documentation is pretty deep. And if they forget some of that, technically it might not be approved. And so it's more on the approval side.

Yeah, it's almost like insurance adjuster work. Inspecting, like, what went wrong and creating, like, a chain of responsibility. Do you have anything? Because I have another one.

Yeah, I mean, you raised some money, right?

Oh, yeah. How much did you raise?

You want to hit the gong?

Hit the gong.

We announced our $42 million Series B today.

What's next with the $42 million? Like, what is the What is the gating factor on growth? Where do you plan on deploying the most capital? Is it hiring people, sales force, just, just top of funnel awareness, brand advertising? What are you thinking?

You know, I think for us, we want to go wide and deep at the same time. We have customers that are adding flows daily, and a lot of that is just that these things compound, right? Like in the simple scenario we talked about, it was customer service and ops and finance in one lamp. And at enterprise, it's much, much more complex and higher scale. So we want to go deep, but we also want to go wide. You know, we're currently servicing across retail, manufacturing, getting into pharma because we can reuse a lot of this infrastructure, like cold chain across grocery also applies to pharma. And so we really want to expand the scope, and this will help us hire some folks that are domain experts. And that's the second thing, like when we're applying, this to customers. We're going with like, hey, we know you have this problem. This is what we built to fix it. And this is why you should pick us.

That makes a lot of sense. Well, congratulations and thank you so much for hopping on the show and breaking it down. We'll talk to you soon.

Thanks for having me.

Thanks, guys.

Have a good rest of your week. Goodbye. Up next, we have Tom Mueller from Impulse Space, the founder and CEO. Keeps putting up huge numbers. We'll bring him in from the waiting room in just a moment. Tom, welcome to the show.

How you doing? Hey, I'm good.

Good to see you again. Always a good day.

You too.

When there's good news in Impulse Space. I want to hear the news. I want to hear the progress. And then I want to zoom out and just talk about a little bit about your journey, because I think it relates to the current moment in some interesting ways. But let's start with the news today. What happened?

Well, we did our Series D round earlier this year and raised $500 million, and we've added another $308 million to that round. For a total of $808 million.

That is incredible. Congratulations.

Thank you.

Why the extension? Is this about production? Is this about hiring? Is there a talent war going on? Is this related to the SpaceX IPO? What's the rationale? What's driving the market?

Oh, probably all of the above, John. Certainly You know, we're seeing a lot of demand. Our customers really love what we're doing, orbital, you know, orbital transfer, orbital maneuvering.

Yeah.

Our investors love what we're doing. You know, there seems to be a lot of excitement in space these days for many reasons. And we seem to have been the right, you know, the right type of company in the right place at the right time. So I'm really happy with, you know, what we're achieving. So we need the money right now to, you know, to build out, to continue hiring at all levels of the company, and to continue to build out our facilities here to get ready for a higher production rate.

Yeah. Can you refresh everyone on the core product, the capabilities? And then I'm interested in hearing about how this fits into the modern trend. The data centers are going to space. It was controversial for 2025, people were debating it. And now I think everyone's come around and said, okay, maybe it's not this year, but it's gonna happen at some point. There's going to be a lot of attempts and there's probably gonna be some compute in space sooner than later. So how does Impulse fit into that?

Okay. Well, first of all, we do— we take over where launch leaves off. So basically we get to space on an existing launch vehicle. Typically it has been Falcon 9, but we're gonna be on new vehicles in the future. And then we have 2 products. We have MIRA, which does precision maneuvering, and that's the one that just did a flyby here that we just announced, came within 200 meters. MIRA-1— or I'm sorry, MIRA-2 and MIRA-3 came pretty close to each other.

Right.

And then we have HELIOS, which is basically— we call it a rocket on a rocket. It's an upper stage that we add to a launch vehicle. that goes in the fairing and that can take basically a lot of cargo to very high-energy places like geosynchronous orbit or out beyond Earth gravity, out to the Moon, out to Mars. So we have— so either precisely move around or just move big distances fast.

So, I mean, I'm like, it seems like with orbital compute, not necessary that it's in geostationary that I'm aware of, but super valuable to keep it in orbit longer. And because you paid to put the chips up there, you probably don't want them burning up. Is that how this might play out, or have you thought about how these 2 technology trends fit together? And only if you're willing to disclose anything, but—

You know, it's speculation.

Yeah.

I don't know if there's a big fit right now.

Okay.

You know, most of the people that are doing orbital data centers just need the rocket to get to where they're going, taking a whole bunch at once. If that changes, we're glad to help. Certainly SpaceX, you know, pretty much keeps everything in-house and, you know, does everything themselves.

Sure.

Others that I've seen plans, I think, are gonna just go up to LEO on existing rockets, you know, on a rideshare, many at a time. Yeah.

So what else What else is exciting to you on the near term in space? I mean, we all watched the moon mission. It feels like we're closer than ever to economic activity on the moon. There's also so much happening in aerospace defense, imagery, communications, but where are you seeing the most near-term opportunity for Impulse?

Near term, like for MIRA, we just signed on Victus, Solo, follow-on. So 2 more spacecraft for Space Force.

Mm-hmm.

We have— we just got onboarded to NSSL for HELIOS, which is the National Space Launch Lane, which allows us to fly the US government, which is, you know, the biggest customer in the world for launch. And we'll be the only upper stage that's on that program now. So that's, you know, that's a huge opportunity for us. But going further than that, you know, we're building Moonbase Alpha, and hopefully Ampulse can somehow be involved in that awesome project.

Yeah, that's very cool. As you think back to your time at SpaceX, I'm interested in the parallels between the existential risk question that SpaceX was in many ways founded to resolve, the making life multi— making humanity multi-planetary. That was a very animating force. And today with the AI lab leaders, they are similarly animated by existential risk. And I'm wondering about, like, what was the mood at the time? What What was— how much did that— was that a motivating factor? Just what was the color of the mission to make life multiplanetary? How did that play out back when you were at SpaceX?

You know, it was the core vision, and I think that's why a lot of people were excited to be there. But we were mostly just head down, just trying to make the rocket work and become— be reliable, especially us guys in propulsion, which is, you know, always the That thing that seems to break.

Yeah, just focus on the engineering. Uh, have you been able to maintain that culture? Is that the main thing that you want to take forward to Impulse? Like, how, how has the culture of Impulse changed or remained the same? What have you pulled? Uh, what can you share about how the team operates?

Yeah, it's very similar. I mean, I think, I think, you know, myself and my team had a huge effect on how the culture at SpaceX was formed. And I brought that culture here for the most part. You know, very merit-based, very ownership of the company. Everybody gets, you know, equity ownership in the company. So we're all pulling towards a common goal.

Yeah.

And also, we want it to be interesting and fun. So, you know, like we work hard and And see the results of it. And it's, you know, we're stoked. It's pretty cool.

How has the team embraced or even just dealt with the diffusion of AI tooling? Like, we're seeing these incredible mathematical results. A lot of them don't immediately apply to, you know, specific engineering problems. AI has hallucinated a lot less, but the tolerance for hallucinations is basically zero when you're launching rockets, but how have you integrated or paced the adoption of AI tooling at Impulse?

Yeah, I have a little anecdote on that. I think it was, you know, about a year ago, I was trying to do an engineering problem, actually a propulsion problem, using chat. And I think I mentioned that I kept telling it what it was doing wrong and it kept doing it over again. And I said, if this was my intern, I would fire it. Now it's a lot better. Now I'm finding just in a year how much it's improved.

Yeah.

We're a little bit choked on what we can use here just because we do classified programs and we have ITAR issues. So we can't just like use the latest model generally, but we try to use as much as we can. I'm, as you know, as a founder and CEO, a little worried about us getting left behind because we have these restraints on us where many other tech companies don't. So, it's, you know, we got to figure out how to wade through that.

Yeah, yeah. It does seem like AWS has done some FedRAMP stuff and is really moving on the ITAR stuff. So, good luck there. I'm interested in your— do you think as CEO, you're more uniquely equipped to evaluate these tools because maybe you're not actually deploying a propulsion problem that you're sort of sketching out with an AI system, but because you'll have a team actually go and finalize everything, but you can sort of take an idea halfway and communicate in a fuller way with your team. Is that how you're using AI these days?

Yeah, I think mostly I use AI a lot just to ask questions, you know, just answer— just I'm working on a design and I want to know, like, Like one of the ones recently was like, what would be a good baseline seat pressure for a Vespel seat, like a valve seat? It's just like, it's out there and I can go find it, but all I do is type it into Groq and seconds later there it is between 3,000 to 5,000 PSI. Like, there, got my answer. So that's how I use it a lot. But then like our guys doing coding.

Sure.

We'll use it, you know, I think that's where probably the strongest use within the company might be.

Yeah.

Is there still— I remember hearing all these stories about, like, NASA, everything on, like, the space shuttle having these, like, switches and everything built with, like, double or triple fallbacks and risk tolerances. Is that culture still there, or When you don't have a human on board, that level of engineering is less relevant.

Yeah. And this is something that we went through when we started flying humans on Falcon, you know, back at SpaceX is like, we had to increase our level of, you know, backup of failsafes. So it adds complexity and adds cost and adds mass. So if you're trying to If you're trying to develop a low-cost vehicle that you're going to make a lot of, you might not put that many. You'll figure out the things that are likely to break and to have some backups. Like we have dual computers, or even on Helios we have triple. So, yeah, 2 agree, if one doesn't, it's out.

Okay.

For MIRA, we have 2, and if one goes out, we can— even if both go out, we can reboot. You have time. On a lot of people, you don't have time.

Yeah.

When, you know, when you're in orbit, you're in a stable orbit, you can, you know—

Oh, that's interesting.

Reboot everything and turn it back on. But, uh, so there's critical places like that that you'll have redundancy. But like, in many cases, you'll have one engine like we do on Helios. You might have dual igniter to light it, but, um, you just put all your effort to make sure we get that thing lit and running because it's the only engine you got. If that engine fails, the mission's over.

Yeah, that's really cool. Uh, Tyler, do you have any questions? I have one more, but no, you got it. I'm interested in in the talent wars, the hiring market. There's two effects. One is like AI so hot everyone wants to go work in AI. The other effect is that everyone's worried about AI taking all the jobs. They want to go and work at real things. Have you noticed anything out of the younger generation? The shape of the type of person that you're trying to hire? What does it look like? to make a career at Impulse these days?

Yeah, so hiring, we're hiring like crazy right now. I think we're doing pretty good in this environment. There's a lot of startups around here that have really drained the talent pool. It's a little harder to find more seasoned people, experienced, like at the senior level where it's more difficult. We're getting a lot of smart kids right out of school People from other industries that are moving to our exciting industry. The hardest place, of course, is in the software and coding, that there's so much competition for those guys that that's always been, from when I started the company, it's always been you pay a premium to those guys and it's just hard to get 'em 'cause AI is like the big moneymaker.

Yeah, yeah, yeah, yeah, it's a crazy time. Are you relocating people from the Bay Area Bay Area or pulling from universities that are outside? Everything? All of the above? Wherever you'll find them?

Everything. All over. As you know, we have an office in Colorado. There's a lot of talent out there. Our guidance, navigation, and control is out there. So, we got quite a bit of software people. We actually put a machine shop out there because we got some really good machinists out there that are making precision parts for us. So, Yeah, we're expanding to other other areas where the talent is.

Yeah, that's really exciting. Well, thank you so much for coming on the show. Congratulations on the extension.

Thank you. Great talking, guys.

See it.

I can't wait for more progress. Have a great rest of your week. We'll talk to you soon. You too. Bye. And with that, that's our show. Is there anything else that we didn't talk to? Any breaking news? We know the Fed hiked. We know that. Palantir partnered with Chipotle, the two most important stories of the day apparently. I think we've gotten through everything. If we didn't, we'll get to it tomorrow at 11 a.m. Pacific. Get that flashbang ready. Leave us five stars on Apple Podcasts and Spotify. Sign up for our newsletter at tbpn.com, and we will see you tomorrow. Throwing flashbang.

Boom!