James van Geelen on His Viral AI Doom Scenario

Something very unusual happened in the market in the last week of February. It sold off, in part, thanks to an article on Substack. James van Geelen is the founder of Citrini Research, which published a piece a week ago titled, “The 2028 Global Intelligence Crisis.” It was not written as a forecast of an imminent disaster, but rather as a scenario analysis in which AI capabilities lead to widespread white collar job losses, triggering a deep downturn, and a financial crisis. Nonetheless, the piece went extraordinary viral, gathering all kinds of responses from economists and research shops and even Citadel Securities. On this episode, we speak with James, the piece's co-author, about what Citrini Research actually is, why he wrote the piece, and why this is a scenario worth paying attention to, even if it's not the most likely outcome. Read more:Bank Shares Walloped by More AI and ‘Cockroach’ Credit WoesPentagon Casts Cloud of Doubt Over Anthropic’s AI Business Only Bloomberg - Business News, Stock Markets, Finance, Breaking & World News subscribers can get the Odd Lots newsletter in their inbox each week, plus unlimited access to the site and app. Subscribe at bloomberg.com/subscriptions/oddlots Subscribe to the Odd Lots NewsletterJoin the conversation: discord.gg/oddlotsSee omnystudio.com/listener for privacy information.

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Tracy Alloway?

  There are some market stories where you want every detail. Joe and I have made quite a few podcasts on that basis. But sometimes you've only got 10 minutes and just want to know what's moving markets fast. That's the Barclays Brief podcast. Every week, experts from Barclays Markets and Research get you up to speed on what's happening and what to watch next. Concise, focused, and brief. Search Barclays Brief wherever you get your podcasts.

• 0:26Speaker 01

  Your team is on a deadline, but moving all your enterprise data will take time your business simply doesn't have. IBM technology lets AI work with your data where it already lives with an open hybrid cloud approach designed to build on your existing investments. It's another way IBM helps you put your data to work. Let's create smarter business. IBM. Visit ibm.com to learn more.

• 0:57Speaker 02

  Bloomberg Audio Studios. Podcasts, radio, news.

• 1:13Tracy Alloway?

  Hello and welcome to another episode of the Odd Lots Podcast. I'm Tracy Alockway.

• 1:17Joe Weisenthal?

  And I'm Joe Weisenthal.

• 1:19Tracy Alloway?

  Joe, we're in the media business.

• 1:20Joe Weisenthal?

  That's right. That's right.

• 1:21Tracy Alloway?

  Have you ever had an article go viral, unexpectedly viral?

• 1:26Joe Weisenthal?

  Yeah, I can't, I'm like trying to like remember like specifics, but yes. And it's one of those things typically where you're like really excited, like, oh, a lot of people are, you know, there's getting a lot of traction. Cool, people are talking about this. And then it goes like multiple orders of magnitude bigger. And you're like, oh, this is like super weird and no context for what this is. And you're like, sort of want to hide in your home and like close the laptop because then you sort of like make it all go away and stuff like that.

• 1:53Tracy Alloway?

  Yeah, it's kind of like, once you release it into the world, you don't actually have a lot of control over how people use it. And I think back to, I wrote a piece about some investors trying to revive claims on Chinese imperial bonds, like antique Chinese imperial debt from the early

• 2:10Tracy Alloway?

  1900s and somehow this went absolutely viral in Hong Kong at the time of the pro democracy protest. So I would walk down the street and I would see these homemade banners that people had created saying that China owes the US like 20 billion in payments on old debt and it was just surreal. Absolutely surreal and like completely unexpected because you you wouldn't think that some like intricate debt story was suddenly going to become a pro democracy protest slogan. But the world works in mysterious ways. And speaking of the world working in mysterious ways, there is something that went viral this week.

• 2:49Tracy Alloway?

  We are recording on February 27th and if you haven't heard of this particular thing, you have probably been living under the proverbial rock.

• 2:59Joe Weisenthal?

  Right, so past Odd Lots guest, James Van Galen, a co-author to a piece on his Substack, Satria Research. Talk about a potential AI doom scenario, which a lot of people talk about. And there's been a lot of talk about mass white collar displacement as a possible thing that could happen as AI gets adopted, etc. But you know, we know that the market's been very skittish about this specifically. We've been seeing the software stock sell off all year, which we've talked about plenty on the podcast and the some of the private insurers and all of this. And something about this moment and this particular piece, I think it came out on Sunday, last Sunday, landed with a sort of like unbelievable thud and so it evidently started moving markets on Monday.

• 3:42Joe Weisenthal?

  And then throughout the week, and this is the part that really flabbergasted me, was you see like all these banks and every economists, etc., like weighing in and many of them very critical and like Citadel Securities, which I didn't even know they like publish stuff. Because that's just a market maker, like they put out all this stuff. So of responding to it and trying to take it out, it was a as a market story and a media story, a wild week.

• 4:08Tracy Alloway?

  It has become the discourse du jour. There's actually a prediction market on it, which you were telling me about a few minutes ago. Like, this thing has just become much bigger than the initial Substack, which to me, again, says much more about the nervousness of the market and how little anyone actually knows about how AI is going to unfold at the moment that people are so keen to just latch on to any scenario that comes out.

• 4:32Joe Weisenthal?

  I get these notes from like sell-side or research shops, and they're like, clients have been asking us about the Citrini scenario. And it's just like, wow, this is wild. Like, this really like.

• 4:41Tracy Alloway?

  That's right. People calling up Capital Economics, being like, I manage a portfolio of 100 billion and I am concerned about a Substack. Okay, well, we should talk to the author of this Substack. And as you said, we've had him on a number of times before, often talking about AI. It is, of course, James Van Keelen, the founder of Citrini Research. So, James, thanks so much for coming back on the podcast.

• 5:01Speaker 04

  Thanks for having me.

• 5:02Tracy Alloway?

  Why don't we start with what Citrini Research actually is and what it is that you actually do in some of your other enterprises? Because I think this has become also a source of confusion, or at least interest, for people who are reading this.

• 5:18Speaker 04

  Citrini Research is a pure investment research firm. We focus primarily on thematic equity and macro research. The progression of it was, I started it as a newsletter, just speaking about stocks and bonds and whatever else. And as we had a kind of string of good calls, which you were kind enough to have us on with the GLP-1 early July 2023, I think it was.

• 5:43Tracy Alloway?

  Yeah, that was a great call.

• 5:43Speaker 04

  Yeah. And the first piece we ever published was a piece that was very bullish on the AI infrastructure complex. So that's been an area, the AI robotics has been a big area for us in terms of thematic equity. We've kind of covered this winding road of bottlenecks in terms of optics, memory, power, whatever else you can possibly allude to. We've probably covered from a what stories are people telling about the movements that are going on in stocks? I remember the last time that I was on Odd Lots, it was about this massive Stargate data center build-out. And Joe was very surprised to see that Caterpillar was, and I think very happy that the old economy was getting a bit of a boost.

• 6:22Tracy Alloway?

  He's an old economy stan. That's right.

• 6:24Speaker 04

  And really, that's what we've been doing for the past three years. I've built out the team. And this piece very much was just a response to what the market has done year to date, which is bonds have rallied, software companies have gotten sold off, a lot of fintech companies have gotten sold off, private equity has sold off. And we're always kind of looking for the cohesive narrative that can connect disparate market moves. And the piece's co-author, Ala, posed to me a question, which was, we've been focused on the bullishness surrounding AI infrastructure for a while, and it's translated into this capability curve that is moving a lot faster than anyone could expect.

• 7:04Speaker 04

  If you imagine this exponential analog logarithmic chart, it's just a diagonal line. It goes up and to the right. People have been trying to put sigmoids or kind of level that that curve off for a long time, and it hasn't. So we basically drew that line out and said, what could be the implications of this happening? It's a scenario which we would ascribe maybe 10, 15% towards. And it comes from a place of, everybody talks about equity markets being forward looking, but really, a lot more of what you see is people justifying historical moves with new narratives that they come up with afterwards. Very little of it is driven by, let me think of potential future outcomes.

• 7:47Speaker 04

  As an investor, which was the audience that this was meant to go out to, I feel a lot more comfortable when I can envision the bull case, the bear case, the base case. And the most uncomfortable that you can be as an investor is when you can't see the bear case at all. So every time that we get into a market that's similar to this, people start asking, what if this time is different? And I guess the thing that this piece did differently was it asked, what if this time is different, but not so much in a SK, Hynix, and Micron are going from price to book to price to earnings, but in a way where, what if this time is different, where the period of transition has to respond to a very, very fast accelerating capability curve?

• 8:31Speaker 04

  And you start from a place where there's a strong kind of historical precedent for the past century or two centuries. Every time you've had a technological revolution, it's been great. It's been awesome. And you see that when you go from 95% of the population working in agriculture to 5% of the population and you create all these amazing jobs. But it happens over a period of 50 years. And now we have this capability curve where you go from two minutes, agents are capable of two minutes of autonomy on intellectually complex tasks. And now, depending on who you ask, it's eight to 16 hours. And that's happened in two years. That is an exponential curve. What happens when we get to multi-day? You know, what happens when we get to multi-week?

• 9:13Speaker 04

  And really, the core of this is if this capability curve continues being as fast and exponential as it is, what does the world look like? There are a lot of very good reasons why that capability curve could level off, but that is the core of the argument.

• 9:26Joe Weisenthal?

  I do think that's just like an important sort of level set for people here, which is that the progress that we've seen since ChatGPT came out, whenever that was, late 2022, has exceeded all of the expectations of everyone who's working on it at the time, including the people who are in the space and the most bullish and like the true believers. And there are various like measures and stuff, but you know, you mentioned the length of time, you know, that it could replicate a human focused on stuff, like all the people, like they made like these bets, right? And there were even prediction markets on their capabilities. And so like, as you say, like, it seems very plausible that the gains will level out in

• 10:08Joe Weisenthal?

  some way, or that perhaps simple computer tasks don't actually replace a lot of white-collar work because there's more to white-collar work than what could be done on a computer, including personality and all kinds of stuff. All of that seems very plausible, and I probably even buy some of that. But this point that you make, it's like, yeah, sure, but it is still improving very fast.

• 10:29Speaker 04

  And it's, it's something where the overall trend of the cost of inference per cognitive task has gone down so significantly, maybe depending on the forecast, 10 to 30 times over the past year. And a task that was uneconomical in the first quarter of 26 might cross that threshold in the third quarter. And the other interesting thing is this capability gap where AI is capable of a lot of things.

• 10:59Speaker 04

  And a lot of people don't know that it's capable of that, right? So is it about the capability improving? Or is it about people becoming more familiar with that? And as AI infrastructure, it's been a great trade and it continues to stay tight. And I think the best rebuttal to this piece has been, well, I think Gavin Baker made this point, which is the world is short on Watson waivers. And that's true, absolutely true. But technological revolutions are volatile, right? Improvements come from places that you don't really expect them to. And I think you can't fully underwrite the idea that there aren't algorithmic improvements or there aren't improvements to the computer infrastructure. So we should look at, okay, if this capability curve continues improving, what are the downstream impacts there?

• 11:43Speaker 04

  And has the financial system ever been stress tested for a scenario like this? Because even if it takes five years, even if it takes seven years, eventually we will get there. And that's not a bearish take, it's a very bullish take. I think that there will be great opportunities that arise because of AI. But that's not to say that there won't be a period of transition. And the faster that it comes, the more aggressive that transition is. And I think the point of the piece really was to get comfortable with what monitoring that looks like. And I'll just make the point that the piece also starts out with an S and P that goes to 8,000 because AI infrastructure is a very bullish trade that makes up a lot of the index.

• 12:25Speaker 04

  And that's a, that's a very strong and very momentum having trade right now. And it ends with the reminder that it's still February 2026. But in the middle of it, it says, how do we kind of get comfortable with the non immediacy of the replacement? If a company decides whether they're doing it because AI has gotten better or because the market likes it when they cut jobs, what is that?

• 12:51Tracy Alloway?

  Which we're seeing already.

• 12:52Speaker 04

  We saw it with Block last night. And you can argue whether that's because of AI or whether that's because of overhiring during COVID. But Keynes said that by the end of the century, we'd have a 15 hour work week. And he was wrong. And there's a lot of, you have to kind of look at why he was wrong. There are a few explanations. David Graeber says that we just kind of created all these bull jobs. This is the title of the book, I'm not cursing.

• 13:17Tracy Alloway?

  People have said worse on this podcast.

• 13:19Speaker 04

  The other explanation is that, you know, human wants and desires you can't really model for, and we will create whatever we need to fill that. At the same time, that required mechanisms by which humans kind of are involved in the process of making those machines better. It's kind of not necessarily in every scenario concurrent with the idea of a piece of software that has the ability for recursive improvement. This isn't to say that tomorrow, every single company in large enterprise goes out and replaces half their workforce, but you do have to take a holistic picture, which is everybody in venture capital has been talking about who's going to be the first one-person unicorn because of agentic AI.

• 14:03Speaker 04

  I don't know if we're there yet. I haven't really kept on top of that, but that does seem like something plausible to me. And I think one of the better lines of the Citadel Securities counterargument, yeah, was recursive capability doesn't imply recursive adoption. That's extremely true. The S-curve framework, though, is kind of describing the wrong variable. And it's a variable that's really important when you don't just have incumbents adopting, but you have startups threatening. And that variable is not necessarily breadth of adoption, it's intensity of adoption and capability of adoption. So you might have a flattening out S-curve, and the seats that you've already enabled with these AI tools are just constantly getting better.

• 14:47Speaker 04

  And so that is, you know, the other thing is, the S-curve is very kind of related to consumer adoption of new technologies. And what I would ask is, was there an S-curve for the adoption of spell check? Everybody already had a PC. Everybody already had, you know, word processing software. It was kind of added as a feature. There are a lot of people in the world today that have no clue how to use ChatGPT that are using AI every single day. It's probably what is going to recommend you this podcast. It's probably what is making these decisions of what items you see when you go on Amazon. So if these agentic capabilities are introduced as features to a technology that everyone has already adopted, you have to adjust your model for that.

• 15:48Joe Weisenthal?

  There are some market stories where you want every detail. Tracy and I have made quite a few podcasts on that basis, but sometimes you've only got 10 minutes and need to know what's moving markets and why. That's the Barclays Brief Podcast. Every week, experts from Barclays Markets and research get you up to speed on what matters and what to watch next. About the time it takes to grab a coffee. So, search Barclays Brief wherever you get your podcasts.

• 16:14Speaker 01

  Running a business means keeping pace with the latest tech. You put a lot into building your enterprise software stack, and scrapping it to integrate AI is out of the question. That's why IBM Technology works with all the apps, data, and models you need. It's another way IBM helps you build on the technology you rely on. Let's create smarter business. IBM. Visit ibm.com to learn more.

• 16:43Tracy Alloway?

  I have so many things to say about this, but first of all, there's something very dystopian about living in a world where like, the upside is, well, we have a lot of bull jobs in existence already. And so, maybe some of those bull jobs will continue to exist even with AI. But the other thing is, like, the self-reinforcing nature of AI seems really important to me in the sense that, as you pointed out, James, like, it's not necessarily that people have to go out and find these new capabilities themselves. It's that the technology itself that they're maybe already using just to substitute search or something like that can do it on their behalf. And so, you just get this feedback cycle where like, one AI thing creates new AI things,

• 17:27Tracy Alloway?

  and it just builds and builds on itself.

• 17:30Speaker 04

  I really would be remiss if I didn't say this again, which is a lesson that I've learned over the past five days, that you can put something in all caps, you can bold it, and people will still not read it. But maybe this is different because I'm speaking. My base case is probably a lot closer to a lot of the people rebutting this article than the article itself. The point of this- Really was to explore what the bear case is if we continue to have a very bullish world in AI infrastructure. I think that any investor that reads it and thinks and, you know, disagrees with half of the things that we say, maybe agrees with half of it and forms a more nuanced understanding of what to watch out for. That's kind of our job.

• 18:13Joe Weisenthal?

  So this is important and people who haven't read the piece should know that like right up front, you do say this, you say this piece is not a forecast. This is a possible scenario and how it could go. And we're going to get into some of the details, but you know, one counter argument to sort of the idea of macroeconomic doom or financial crisis or whatever is, okay, if you have AI and it's driving incredible productivity gains, if it's very disinflationary and so forth, if some people are becoming fabulously wealthy in part of this big redistribution that would happen, well, then the government has a lot more fiscal capacity to stabilize this, right? Then the government can spend a lot of money, rates have come down, they can counteract

• 18:57Joe Weisenthal?

  the disinflation, not totally unlike perhaps COVID would be like a great example. But it strikes me as like, well, if we're ever going to have a government that's thinking about these things proactively, that strikes me as a good reason to write them out. And it's notable, like that many of the executives at the top AI labs, they talk about exactly this. In fact, it seems like they're pleading almost with the government to take this more seriously because if we're going to have this big disruption and redistribution, we're going to have to start thinking about what are the fiscal mechanisms to counter it out?

• 19:31Speaker 04

  A hundred percent. I think that it's something where it's perfectly fine and good to say that the government will be able to deal with it. But it's probably better to formulate a framework in which the government is more able to do that. And in order to do that, you kind of have to have an idea of what to keep track of. And I can say that in the discourse that I've seen, I don't think that there's a very strong kind of data collection. On this specifically, one of the big rebuttals has been that software job postings have gone up 11% year over year. Those job postings include AI and machine learning engineers.

• 20:11Speaker 04

  So you're really seeing a composition shift where these new, you know, AI engineers are coming in. And they're creating software that will improve itself. And when it comes to the government response, Jolts doesn't really speak about composition. In my opinion, there's not a great amount of data on white collar specifically. And yeah, it was, it was, it was almost worrying in itself to see this reaction. Where we write this article that's kind of saying what I think most people are thinking. We're putting trillions of dollars at the white collar productivity machine and oh, that might, you know, have some level of disruption. And I get it. The thing that I'm very thankful to a lot of the rebuttals for is that they remind people that it's 2026,

• 20:56Speaker 04

  which we tried to do three times in the piece, but apparently we're not successful. Thank you to everyone that reminded, that, that made sure that this isn't like a spin out crazy, whatever. But the worrying side is, well, everyone seems very, very comfortable that this is all going to be okay. And I think that that reasonably, I'm also a student of financial history. That reasonably comes from when you look back at the past and you say, well, we had this industrial revolution and it was amazing. And we've had mechanization and it was amazing. And we've had the internet and it was amazing. And it created all these jobs that we couldn't have possibly foreseen beforehand. And you're looking at that from a hundred or, or more years in the future. We have the term Luddite because of the fact that the transition was so abrupt.

• 21:41Speaker 04

  And marked that people were moved to physical violence, right? We don't want that to happen. The transitions do occur and the faster that this happens. If this were going to happen over the next 20 or 30 years, fine, what, you know, that, that's going to be great. Everything's going to be awesome. I think that the the real timeframe is closer to five to 15. And obviously, this piece extrapolates where it's three years. We should be prepared for anything because the the government isn't going to accurately forecast technological advancement, but they can accurately forecast what they should watch and what the best policy response would be.

• 22:16Tracy Alloway?

  Yeah, this is the thing, the Luddites were like ultimately on the wrong side of history in terms of thinking that resistance to new technology would actually matter. But that doesn't mean that there wasn't major resistance and disruption on the way there.

• 22:28Speaker 04

  And that it wasn't absolutely awful.

• 22:30Tracy Alloway?

  Yeah.

• 22:30Speaker 04

  Yeah.

• 22:30Tracy Alloway?

  No, exactly.

• 22:31Speaker 04

  Right from their perspective, from their lives.

• 22:33Tracy Alloway?

  Exactly. You know, you mentioned software job openings still rising. And one of the reasons that's able to happen is because we still have a financial system that up until relatively recently has been very comfortable with extending credit to software companies. And there's obviously a reflexivity between the financial system, the market and the real economy. And you dig into that in your piece as well. And this is the part of it that I actually found the most interesting, where you describe how AI could actually and the disruptive effects of AI could actually end up becoming problematic, especially for private capital. And this, again, is something that is very much in the public slash market psyche this

• 23:16Tracy Alloway?

  week because we've had a number of private credit blowups starting to become public. Talk a little bit more about how you see that kind of private credit AI disruption, now insurance as well, nexus unfolding.

• 23:30Speaker 04

  Just to reiterate, I don't see it unfolding. But I think this wasn't like a singling out of private credit. It was very much a response to the price action of the market. But it is something worth considering that it's relatively new in the grand scheme of things. And there's a system that's built upon the assumption that things stay relatively stable. And if things aren't relatively stable, then What could possibly happen? We're not really private credit analysts, right? We're thematic equity and macro research. This was something where we presented kind of, if you were to have a wave of defaults in one of these disrupted industries, what would happen? And then the other thing is maybe the job losses are fine

• 24:13Speaker 04

  and we go back to an economy like the 1950s where the participation rate is much lower, but productivity is much higher. That's great, too. In the transition, the people that are at the highest risk of being replaced by AI have like 780 FICO scores. And they're not classically what gets modeled as a risk in terms of a default. So these are all things where it's not saying that this is going to happen. It's saying, has a private credit lending? And, you know, to their credit, I will say Apollo, much earlier to the software thing than even I was or the market was, right? Apollo reduced their software lending pretty early on. I think it was in early 2025. Though for the rest of it, you know, like,

• 24:57Speaker 04

  has there been enough changes to the assumptions about the income and about, you know, does ARR stay recurring? That's just something to consider, I think.

• 25:06Tracy Alloway?

  What's your base case on private credit, then? Is it the sort of Jamie Dimon cockroach scenario?

• 25:12Speaker 04

  So I think that private credit isn't banking, right? The like run on the bank dynamic doesn't necessarily play out. They are in possession of permanent capital to a certain degree. And that's through, in a lot of areas, the acquisition of these life insurers. So I think you could definitely see the contagion being very minimized if there were to be. I don't think there have been any like very high profile blowups yet. Everything's pretty much fine as right now, as I understand it. The progression of it, though, I don't think that you're at a very high risk. My base case would be just like that. And the only kind of added risk is

• 25:51Speaker 04

  If you were to have some sort of change to how private credit is treated from a regulatory perspective on the balance sheet of these life insurers.

• 26:00Joe Weisenthal?

  So there's sort of two major components to the piece that you wrote. And one is obviously the macro scenario. And the way it's framed is like, OK, the year is 2028. Unemployment is above 10%. The stock market has fallen 40%. So there's the macro story. But then there's also the sort of secular micro story. And I think this is really interesting. And this is the part that I've been like trying to work out and trying to understand better. This idea that like there are all these businesses that have essentially been built up around building a moat based on network effects, you know, payments platforms and so forth and whatever. And so this idea that AI and agentic commerce will fundamentally change the way a lot of these businesses operate. And these moats will disappear.

• 26:45Joe Weisenthal?

  And talk to us about that because I have a harder time wrapping my head around what is it about AI per se that's like, here you have these legacy networks, delivery drivers, payment companies with whatever they have on the desk. You swipe your card and stuff like that. What are those called?

• 27:05Speaker 04

  Point of sale?

• 27:06Joe Weisenthal?

  What? Yeah, the little point of sale machines. But talk to us about like, just from a pure tech standpoint, what is it about agentic AI that can sort of evaporate this mode?

• 27:16Speaker 04

  So I will say, if I had to go back in time and write the piece differently, I would not have singled that. I would have just kept it on a sector basis, right? And I think that if I knew that it was going to get 30 million views, I would not have mentioned single stocks at all. So I won't do that here. But what I will say is, and this future could be wrong. But if you envision a future where, I remember talking to you guys about this in 2024, when I was using it as a bull case for Apple, which didn't end up coming. You know, they, Apple was kind of, let the chips fall where they may, and then we'll come in afterwards, which they've done a lot in the past 10 years. But the idea is you have this agentic assistant and it's in your phone and it knows everything about you.

• 28:01Speaker 04

  And then you kind of extrapolate that to a lot of people spend a decent amount of time shopping. What they don't spend a lot of time doing is price matching. If you're going to buy a box of protein bars, you don't really check five different vendors because it's tedious. AI agents do not experience tedium, right? So the kind of way that There are a lot of layered intermediation and rent kind of extraction layer in the economy. And then there are a lot of places where having a like an oligopoly essentially has allowed margins to really be artificially increased. So just to address, I don't think that code is the moat on a delivery network for like,

• 28:43Speaker 04

  like, like, that's, you have the drivers, you have the customers, I get that. What I could see happening is something that's already kind of happening where these startups are enabled to create something that's similar. And, well, you don't have the network effect, okay, but if you have an AI agent that has the explicit instructions to go out and find the cheapest option, then it doesn't really care about using this thing that has a network effect, it cares about using the thing that's the cheapest. So if you have an order aggregator, that's an agentic kind of aggregator on the driver side and the customer side, then the customer says to the agent, hey, I want this burrito from Chipotle.

• 29:24Speaker 04

  And then there's a bunch of different platforms that the listing is on because the restaurant has used one of these agentic aggregators to go on every single one and put their thing. And the driver also has the one that will get them paid the most. So the idea of, you know, taking half of the delivery fee as the company kind of goes away because your margin is my opportunity. And if someone that's five people, that's kind of coding up this maybe shoddy replacement is very happy to, you know, obviously there are other moats here, but that's just one example of how you might see a world in which agentic commerce and it's very similar to like the paperclip problem. If you tell a machine to do something, it's just trying to get you the best price.

• 30:09Speaker 04

  And maybe that includes finding a way around interchange.

• 30:11Joe Weisenthal?

  Just to push back on this or just to pressure, I mean, like, Comparison shopping websites have existed for a long time, almost since the beginning of the internet, right? And you know, in theory, you can Google, I don't know, it's just like Google shop had a thing for a while. I don't think people ever that ever took off. But you know, it'd show you like, here's the price of a computer monitor on Amazon and Walmart.com and newegg.com and a few of these sites that like, don't exist anymore, et cetera. Like in theory, like, isn't that describing the same thing? That like, from the customer's perspective, it's like, okay, I just, they're all the same. I'm going to click the cheapest.

• 30:48Speaker 04

  Totally, I get that. And that's an entirely possible case. What I will say is there's a big difference between actively going and taking the effort and taking the time to go to one of these comparison shopping sites to get the best price versus just telling your phone, get me a burrito, get me the best price. Right, those are, they're two kind of fundamentally different things. This will play out over the next five or 10 years and we'll see. And also, I'm sure that we're not going to just delete friction overnight. Right, so that's why it was so shocking to see this kind of like immediate reaction. It's like, no, this stuff hasn't happened yet and we don't know exactly how it's going to happen. It's just a future scenario where things happen a certain way, so.

• 31:48Speaker 01

  Your team is on a deadline, but moving all your enterprise data will take time your business simply doesn't have. IBM technology lets AI work with your data where it already lives with an open hybrid cloud approach designed to build on your existing investments. It's another way IBM helps you put your data to work. Let's create smarter business. IBM. Visit ibm.com to learn more.

• 32:16Tracy Alloway?

  Can you talk to us for a second just where you see AI valuations at the moment? Because I think this is also part of the reason that people are very nervous at the moment, which is like, okay, on the one hand, we think AI is going to eat the world. But on the other hand, it's not entirely clear that a lot of AI is going to make money in doing so. And if you look at, you know, some of the big hyperscalers at the moment, they're still losing money on certain power users. So how do we think that AI is actually going to make money as it sort of eats the world?

• 32:50Speaker 04

  I think that that's the other thing that's important here is that these companies need to go out and search for ROI. And there are a lot of threats. You saw Anthropic respond to the Chinese distillation of models. And, you know, if you go and you use Minimax, it's relatively comparable, but it's also 90% cheaper. So this is like, there is a race happening right now. And the economics are, they span the gamut, right? Good and bad on both sides. The thing that drives this kind of capability improvement is you do need customers to pay for these things that you have spent so much money on.

• 33:32Speaker 04

  And that means making it capable in a way that's useful to your customers or integrating it in a way that's useful to your customers. So I personally think that that will happen. How quickly it happens is anybody's guess. But I think valuations right now are reflective of this expectation that we are going to continue adding compute capacity to be able to handle this. And I think that if you spend eight hours just thinking about it, you can see a lot of places where AI is pretty valuable. But a lot of those places are places where you might otherwise pay a human right now. So, yeah, it's just, you just have to balance it. And there's a lot of ways that it can go well. And then there's a couple of ways that it doesn't.

• 34:12Joe Weisenthal?

  Let's talk about enterprise software for a second because, okay. The public facing these moats, these network effects, et cetera. Maybe AI agents allow us to get the best price wherever. Is it different economics if we're saying the enterprise? We know about the enterprise, the SaaS sell off, et cetera. What is the scenario? How would you articulate the fear in the market right now? That all of these incumbent software companies could theoretically get ripped out because something, something AI will make it so that customers don't need them.

• 34:40Speaker 04

  So you can separate software. You have kind of like this long tail of SaaS that includes these, you know, workflow automation tools. And then you have like the systems of record. I think that it's very likely that the, at least the systems of record have like a short squeeze in the sense that right now they kind of just have upside in that they are, they are most situated to be able to improve their margins because of AI.

• 35:04Joe Weisenthal?

  Right, because coding is a cost for them, right? It's like they can theoretically maintain these things much cheaper than they are.

• 35:09Speaker 04

  Yeah, 100%. And what we said in the piece, which will be, you know, interesting to see in real life, and I don't necessarily, it's a good point that enterprises don't really react as quickly as this. So the timeline is probably aggressive. But the way that these kind of contracts are negotiated last year, when you had the first half, the kind of budget resetting, these CIOs and procurement teams, they, Agentic AI was still kind of a buzzword, right? It wasn't until the end of November that it became insane. You know, I saw you have vibe coded a couple of things yourself, so there was.

• 35:42Joe Weisenthal?

  I have something cool coming out next week.

• 35:44Speaker 04

  Nice. There was a great kind of jump in capability. What is it, by the way? Are you, can you speak about it or?

• 35:50Tracy Alloway?

  No, he can't. It requires some finesse, I think. Well, this is the thing. This is the thing, like, I used to blame Joe for the SaaS sell off, right? Because he was the one vibe coding and publicizing vibe coding. But now we can all blame Citrix.

• 36:05Speaker 04

  Yeah, I'm off the hook. You're welcome.

• 36:09Speaker 04

  But the the strategy that's been adopted by OpenAI is very similar to Talent here, where they say, we have these forward deployed engineers and we're just gonna install them at your place. And so maybe, you know, I don't necessarily think the enterprises are gonna jump to vibe code their own system record. But what I do think is that when you have these sales teams that call up their customers and say, hey, remember last year, we said this was what inflation was, and then we added a couple percent on top of that. So you're getting a 5% price increase. All good. Okay, you're not going anywhere because you don't have anywhere else to go. Done. Now, the person on the other side of the phone can say, you know, OpenAI called me the other day, even if they're bluffing, right?

• 36:49Speaker 04

  So so you do see like some potential downside to pricing power. And that's that's in the places where it's very unlikely that these vibe coded alternatives actually pose a threat. And then you see, it's been interesting how Anthropic has handled it where they've recognized this capability gap where they say, oh, these people, the people don't really understand what these tools can do. So they've started releasing like suites of AI tools. I don't know if you saw the wealth management one, right? It's they released the wealth management one, I think, a couple days ago. It's like you could have done this yourself with Claude custom skills.

• 37:26Joe Weisenthal?

  This is a really good point. And I hadn't really thought of it in that terms because these things that like Claude announced or Anthropic releases something, they're not that incredible in some sense. But they're essentially just very simple reminders. You hadn't thought to use this for, you know, modeling various retirement scenarios. Actually, it's very simple. You could do that. You hadn't thought to use this. So because they're simple. They're like markdown files. They're not like particularly exotic pieces of software. But they are reminders that this thing you didn't think of. Yeah, just do it.

• 37:54Tracy Alloway?

  It's like a thing that you can use to hammer your supplier over the head with, right?

• 37:58Speaker 04

  Yeah, I don't know exactly what the timeline that that happens on, but. There are going to be adjustments to pricing power because of it. And yeah, it seems that this is kind of the reason why in the beginning I thought that framing the piece this way was valuable to our client base and reader base was because as an investor, you don't really care if you're presented with 10 scenarios and nine of them are wrong if one of them makes you money. Right? So I obviously knew that some people who had already bought the dip in software would disagree with the software part, but maybe they would agree with the, you know, with the disintermediation part. But then it kind of escaped containment. And in retrospect, if I was going to write a piece for broad distribution, it would probably be pretty optimistic because I'm a pretty optimistic guy.

• 38:43Speaker 04

  Like, so yeah, that's been an interesting experience.

• 38:47Tracy Alloway?

  What was the most surprising thing from this week for you?

• 38:51Speaker 04

  Well, I had someone that, that really strongly disagreed with me. And then when I asked why, sent me a Claude readout.

• 38:59Joe Weisenthal?

  The, the Kelsey is cool that there's a

• 39:02Tracy Alloway?

  You can use this as a hedge for like your own company.

• 39:06Joe Weisenthal?

  There's now an instrument in which, let's see, Kelsey, I'm going to look it up. Kelsey, Citrini scenario. Like if you start typing in Kelsey and then start the word C, it autofills Citrini scenario. Will this, I love that. Will the Citrini scenario happen? It said 11.6%.

• 39:22Speaker 04

  Is that basically the rate that you would get if you put it in the money market?

• 39:24Joe Weisenthal?

  It probably is. So this is, can I read the specifications of the contract?

• 39:28Tracy Alloway?

  Fine print matters.

• 39:29Joe Weisenthal?

  The rule summary. So if at least three of, colon, unemployment rate exceeds 10% for the BLS, S and P 500 declines more than 30% from its closing level of issuance. That's weird terminology. Zillow home index declines more than 10% in any of New York City, LA, San Francisco, Chicago, Houston, Phoenix. Labor share of GDI falls below 50% and CPU falls below 0%. If any of those three things happen, then the Citrini scenario.

• 39:55Tracy Alloway?

  That's crazy because like most of that is just a financial crash, right? Like it's not even necessarily tied to AI.

• 40:02Joe Weisenthal?

  It's cool. Like, do you like that? That's like, this is now going to be known as the Citrini scenario forever. Like when, like when we get the next crisis, whenever people is like, oh, this is like an omen.

• 40:12Speaker 04

  I feel like anybody consider, like, I feel like you could make a lot more money on TLT calls if three of these things hit, but.

• 40:20Joe Weisenthal?

  There's $125,000 been traded in this market. So it's still pretty minor.

• 40:25Speaker 04

  Deep liquidity.

• 40:26Joe Weisenthal?

  You can't, right, you can't probably hedge, you can't hedge your whole life or, you know, your whole portfolio.

• 40:31Tracy Alloway?

  The whole business.

• 40:32Joe Weisenthal?

  But, you know.

• 40:33Speaker 04

  If I was gonna pick a thing I'd be known for, it probably would have been not this.

• 40:37Joe Weisenthal?

  Yeah.

• 40:37Speaker 04

  But, you know, you don't get to pick, so. I, I still stand by what we've written and I think that it's, uh, as a scenario, useful to consider.

• 40:45Tracy Alloway?

  All right, James, thank you for coming on during a very busy and I'm sure surreal week for you.

• 40:51Speaker 04

  Thank you for having me.

• 41:05Tracy Alloway?

  All right, Joe, I, I'm very glad we got James on to discuss that.

• 41:09Joe Weisenthal?

  Yeah, me too.

• 41:09Tracy Alloway?

  Because obviously, this is the talking point of the week, at least. It is just fascinating from a media perspective, how you can have these viral pieces that kind of get out into the world and develop a life of their own, but obviously, the major point of interest in all of this is, these are the things that the market seems to be actively considering at the moment, right?

• 41:29Joe Weisenthal?

  Paul Krugman wrote a good piece, he disagreed with a lot of it, but he pointed out, you know, when, uh, the radio broadcast of War of the Worlds happened and a bunch of people panicked because they thought there was some big invasion, it occurred in the environment of a very, it was like, you know, during the depression.

• 41:43Tracy Alloway?

  Yeah, of like existential dread.

• 41:45Joe Weisenthal?

  And look, like, this is the worry that has been, people have been talking about all year, long before this piece, and so like, the whole reason people are like talking about, oh, are all these software companies that have thrived forever, the reason why they're, many of them are at all time lows is because of like, wow, people are very impressed with the capabilities, and you have a lot of people talking about the potential for mass white collar layoffs, and so therefore, you know, I read it as a sort of, let's put this all together, and to the point, it's like, you want to be thinking about scenarios, particularly from the public sector response, like, let's actually talk about what this could look like, it strikes me as a useful exercise.

• 42:19Tracy Alloway?

  Right, and the reaction itself is informative.

• 42:22Joe Weisenthal?

  Yeah, totally.

• 42:23Tracy Alloway?

  Right, so, again, we should not be in an environment where you can have a think piece, a single scenario, that actually causes a broad sell-off.

• 42:33Joe Weisenthal?

  Yeah, I know, I know.

• 42:33Tracy Alloway?

  That lots of people start like, pinning on this particular. Particular piece. And likewise, we shouldn't really be in a scenario where Citadel Securities publishes a rebuttal and then everything starts rallying. All it underscores is that no one really knows anything at the moment.

• 42:44Joe Weisenthal?

  And this is on tenterhooks, right?

• 42:46Tracy Alloway?

  Yeah.

• 42:46Joe Weisenthal?

  Like there's like, people are extremely stressed and no one, it's, you know, it's like, it's genuinely, it's uncharted territory. It's uncharted to have a technology that is improving as fast as it is. It's uncharted to have it, you know, it's not like one lot, one specific industry is, it's like a broad range. No one knows where it's going to be. So it's like, people are like deeply anxious about it. And it articulated a lot of views and it landed at a moment where this was just top of mind for everyone.

• 43:15Tracy Alloway?

  The one last thing I'll say about this is I'm really glad you asked about policy because this also seems to be the wild card in this entire discussion, which is like, the outcome of all of this could end up being very different. Depending on what policymakers actually decide to do about it. And so far, we haven't really seen any, like, not even early signs of how people are thinking about this.

• 43:37Joe Weisenthal?

  There's virtually no discussion in DC about anything substantive related to, like, the actual impacts of AI. There's almost none. And there's, it's this very weird chasm that's opened up between how much of a big deal so many people are thinking about this? And how politicians are like, they'll talk about anything but this. It's very strange.

• 43:59Tracy Alloway?

  It's wild, really.

• 44:00Joe Weisenthal?

  It's starting to get pretty surreal on that.

• 44:01Tracy Alloway?

  Yeah. All right. Well, shall we leave it there?

• 44:03Joe Weisenthal?

  Let's leave it there.

• 44:04Tracy Alloway?

  Okay. This has been another episode of the Odd Lots Podcast. I'm Tracy Alloway. You can follow me at Tracy Alloway.

• 44:09Joe Weisenthal?

  And I'm Jill Weisenthal. You can follow me at the Stalwart. Follow our guest, James Van Gielan. He's at Citrini7. Follow our producers, Carmen Rodriguez at Carmen Armin, Dashil Bennett at Dashbot, and Cale Brooks at CaleBrooks. And for more Odd Lots content, go to Bloomberg.com/oddlots for the daily newsletter and all of our episodes. And you can chat about all these topics 24/7 in our Discord, discord.gg/oddlots.

• 44:32Tracy Alloway?

  And if you enjoy Odd Lots, if you like it when we talk about the AI doom scenario of 2028, then please leave us a positive review on your favorite podcast platform. And remember, if you are a Bloomberg subscriber, you can listen to all of our episodes absolutely ad-free. All you need to do is find the Bloomberg channel on Apple Podcasts and follow the instructions there. Thanks for listening.

• 45:24Speaker 05

  Congratulations, you've built a dashboard. But now comes the follow-up question, and the follow-up to the follow-up, and the dashboard for that. Meet Hex. Give everyone the power to ask their own questions, explore data, and build with AI, all in natural language. Because every answer should spark another question, limited only by your curiosity, not your data team's bandwidth. Go deeper than dashboards. Join over 2,000 leading data organizations. Visit hex.ai.

• 45:55Speaker 06

  If you're a business owner, you need a card that can keep up. With the Delta SkyMiles Reserve Business American Express Card, everyday purchases transform into miles you can use towards future travel. So business as usual can lead to a well-deserved getaway. And for a limited time, card members can earn 200,000 bonus miles, the highest miles offered this year. Whether you're traveling for business or planning your next adventure, now's a great time to make every purchase go further. Apply today at www.deltacom/podcast. Offer ends November 4. Minimum spending requirements and terms apply.

• 46:31Speaker 07

  A business gift should do more than check a box. It should reflect your brand and show someone they're appreciated, recognized, and truly seen. 4imprint offers thousands of high-quality customizable products, like premium apparel, drinkware, tech, and more, making it easy to create a gift that feels meaningful and on-brand. And with 4imprint's expert support and their 360-degree guarantee, you can be 4imprint certain your order will arrive exactly as intended. Explore gifting with confidence at 4imprint.com. 4imprint. For certain.