OpenAI vs Anthropic: Coding Tools & Branding, SaaS Selloff Debate, Nvidia’s Blackwell Challenges

28:27Are you seeing that play out in your data at all? No, we aren't. I think there's two things that we look at when we think about what our data says about the health of the consumer. The first thing we look at is, are they engaged? Are they out engaging in commerce, completing transactions? And as you can see, last quarter, we grew GMV 36%. It's a really strong growth rate. that's a highly engaged and active consumer. And then we look at the credit performance. Are consumers paying back on time? Are the cohorts of loans charging off at the same rates that you would have expected and consistent with prior years?

29:02And all of that is very much in line. I would hesitate to extrapolate what we experience at Affirm to the broader economy only because our product is different. Our product has several features in it which actually engineer, we think, better outcomes for the consumer. For example, our products have a fixed payment schedule that amortizes very, very quickly. So consumers who enter into our transactions are much more likely to get out of them much more quickly than if you were revolving on your credit card. That makes these comparisons from what our product and our consumer is doing to the broader unsecured consumer asset class is a little bit hard to extrapolate.

29:41But for what we see with our consumer on our product, it's really healthy and you don't really see those signs of the so called K-Safe economy. I want to ask you very quickly before you go about the intersection of AI and the financial system. We talked to a lot of founders of a lot of AI startups on this show, and I'm sure there are a lot of founders out there that are working to integrate AI into the financial system. I know there are big companies like Affirm that are doing that, but as it relates to startups and emerging technology. How, which pockets of AI helping the financial system are you really focused on right now?

30:18And are you guys investing in a lot? Yeah, we're investing an incredible amount. We spent a lot of time talking about this a few quarters ago. AI is all over our business. It's first, we're a software development company. And so any of the advances that we have in AI-enabled development is certainly impacting our ability to ship code quickly. Those impacts, I think, are incredible and very dynamic. Every month, the industry is making big improvements. We also have products that leverage these models in order to target better offers for consumers or optimize our merchant financing programs. We talk about how those drive much better outcomes.

31:03The real insight there is that the quality of models just goes up everywhere. And if you can, any problem that you have, it's a math problem, it's a modeling problem. The more you can leverage these Gen.AI capabilities, the better off you're going to be at sorting risk, identifying and targeting offers for consumers and building better programs. And of course, we are in the business of building models. Everything we do at Affirm is software, and most of what we do has a model behind it. to really beneficiaries of the tailwinds of these just fundamentally better models. So it sounds like for you guys, I mean, really, it's the back end of the financial system and being able to sort through risk, being able to sort through how much credit to give people, which customers are worthy of which credit.

31:48I mean, it's really all that back end calculation, which is what AI is making a difference for you right now. Yeah, but also in what consumers are offered. And so if you're a consumer, the more that the offer presented to you fits you. The more tailored it is to your financial situation, the better off that you're going to be. You're going to have a higher probability of a successful transaction. You're going to have a higher probability of a happy path. And our models are getting better at pointing offers to consumers that are going to resonate with them. Additionally, also for the merchant side, the programs that they offer can be optimized with better modeling.

32:26And again, every part of our business has a model to it, and all of them are getting better. And very quickly, are you hiring any less this year because of AI? No, the opportunity at Affirm is just so enormous. You know, I mentioned the$1.3 trillion of revolving debt that's out there. It's so enormous that we still need more resources, but we are seeing material improvements and productivity of our engineering team because of these tools. Got it. Great. Well, Michael, I want to thank you for coming on. That is Michael Linford, Chief Operating Officer at Affirm here on TI TV. OpenAI and Anthropic have been dueling it out this week, not just because of Anthropic's Super Bowl campaign that they put out, which we were all watching, but also because both companies introduced new versions of their AI coding tools.

33:16I want to bring on Nico Gruppen, head of applied research at Harvey, to share with us some of his reflections on both of these tools. Nico, welcome back to the show. It's great to have you here. Thanks, Akash. Great to be back. So we've got Quad Opus 4.6 and GPT 5.3 Codex, which are the best named products out there, I'm convinced. Are these tools directly competitive? I mean, let's start there. Yeah, so maybe I can frame them. They are slightly different tools. So Opus 4.6 is what we consider sort of a general reasoning model. It's really intended to be placed within an agent harness. It is the same model you may use within a product like Cloud Code for code generation, but we also use it in our product for general knowledge work tasks.

34:03Codex is more specific to code generation tasks. Again, we're beginning to see a trend of adding file system management, some more general knowledge work features onto these products. They are slightly different though on the surface. So OpenAI is really, sorry, the other way around. Opus is a little more general, but Codex is a lot more specific for coding. Yes, that's right. Okay. Now, so, you know, is there a discussion around which is better for coding? You know, is this a rivalry? I mean, how do you see it? Yeah, it's interesting. So we're looking at this through two lenses. there's the coding lens where there is sort of a fierce standoff between Codex and CloudCode.

34:47I personally am a CloudCode user. I love the interface. I think what they've done with co-work on top of that is great for folks who are iterating between files and code like I am. Codex is great for long-running sort of agentic coding tasks, right? And so if you go onto the software development floor here at Harvey or in our other offices, it's not uncommon to see engineers with, you know, three or four cloud code terminals open running tasks, and then three or four codex jobs running in the background all at the same time, right? So when someone comes to you and says, which is better, what do you say to that?

35:25Yeah, I think from my perspective, you know, I think there's an interesting phenomenon here. So I think what we're observing from the field is there is this will happen once or twice a year right there's an inflection point in model capability we see a step function improvement at how the models perform on certain tasks I actually think the last inflection point that we saw with respect to model capability was in December of last year Q4 of last year and we saw Opus 4.5 and GPT-5.2 release but this pushes up the ecosystem as a whole because all of the other labs will scramble essentially to reach the same level of performance.

36:07With the Opus 4.6 release yesterday, I think you can see there is a little bit more juice to be squeezed in this particular flavor of model. And so if I'm taking on an analytical task, if it requires file system management for us, this could be a lawyer are working within Harvey for deal management, due diligence, potentially complex transactional tasks, my model of choice today would be Opus 4.6. That said, there are other constraints that might come into play, cost, latency, long form writing, GPT 5.2 still has a little bit of an edge, but that's sort of the breakdown from my end. I don't know if you'll be comfortable answering the question, but do you guys at Harvey, I Do you have brand loyalty to Anthropic versus OpenAI in terms of the products that you use?

36:58Yeah, I think, you know, our responsibility, we view our responsibility as putting the best model into the product for the use cases that our customers care about. Right. So from the perspective of the research team, we're going to evaluate every new model that comes out. and we are going to put the one that is best suited for a particular use case in the product. I know, you know, what you're hinting at is very much true. We live in dynamic, sort of turbulent, volatile environment right now. It's hotly competitive. And as you sort of alluded to at the beginning, I didn't think, you know, Super Bowl ads would sort of become the town commons for this sort of discussion.

37:40So we certainly take that into account. We view both OpenAI and Anthropic as strategic partners in the long term. Well, because it is kind of interesting. You talked about the model releases that are really inflection points. I mean, the way that I've seen this rivalry play out over the past couple months is you either have a product release, which is really making a lot of noise and is clearly a step above. And that's when people start to re-up this discussion of there is a heated rivalry. You know, if the model releases are more incremental, then you sort of hear less about it. But then you have sort of the branding side, which is like, what are the founders saying?

38:16You know, what are the campaigns that are being run? What are the Super Bowl ads, you know, focused on? And when that happens, you just said that, you know, maybe these 5.3 and 4.6, they may have been more incremental in terms of revisions or iterations. but now the brand discussion is so significant that everyone's like going back to, well, what are the products actually doing? I feel like that's why people are so focused on it right now. Yeah, that's absolutely right. On the one hand, it's extremely valuable to be perceived as the AI model that is leading for consumer applications. On the other hand, it's extremely valuable to be perceived as a leading model for enterprise applications, right?

38:53And so we see positioning from both labs on this front and you can see that in the investments that they're making for specific verticals. From our perspective, you know, we're interested in partnering with the labs to improve the models on the capabilities and the use cases that are top of mind for us. And honestly, this sort of frenzy of activity just raises the bar for our product. So let me ask you this. Harvey, how old is Harvey now, the company? Just over three years, maybe three years. Okay, so you guys are building this company from scratch. and you have to build all of the backend softwares that all of the other software companies have had to build.

39:35You presumably have a CRM. You have database management software. Playing out in the market right now is the idea or the question. Can you just code your own CRM? Can you just build your own software with AI? And you have bigger companies out there saying, well, you're never going to be able to do this at scale for 10 ,000 people, you know, just with a snap of a finger. You are building your company from scratch. How are you, are you able to build your own CRMs, your own database softwares? Can you do all this with AI right now? Yeah, so we actually did start in a unique position. And as an early application layer company, a lot of infrastructure, especially sort of infrastructure for AI, like vector databases, all of these things, evaluation, tooling, infrastructure, we self-built, right?

40:25And that's largely because a lot of it did not exist. Today, you know, I think we're already reaching a scale where we are seeing a disparity between, you know, self. From my perspective, it's really a maintenance question, not a can you do it question, right? I'm sure if you put a group of 10 engineers into a conference room together and said, build us a CRM, you could do it. But in our domain specifically, we see the difference between enterprise-grade applications and non-enterprise-grade applications pretty starkly. For us, it comes down to things like governance, security, privacy, ethical walls, all of these sorts of things.

41:07But the same is true for other sort of large-scale, what I would really consider the infrastructure of a company, both on the technical and non-technical side. So in other words, it sounds like you and a fast-growing startup, you are still buying all of these, quote-unquote, old-school enterprise software tools that everyone is scared is going to be replaced by AI. Yeah, I think we'd have to go down the laundry list and check each one off. But we are building for scale and we're building enterprise-grade applications. I think we're less interested in spending our research and engineering effort on rebuilding these tools.

41:44Great. Well, Nico, I want to thank you for coming on. That is Nico Gruppen, Head of Applied Research at Harvey here on TI TV. To close off the week, I want to bring on two of our top editors for a wide-ranging conversation on the crazy events of the past few days, not the least of which is investors' fears over large capex spend. To break it all down, I want to bring on Martin Piers and Laura Mandaro. Welcome back to the show, to the both of you. It's great to have you here. Hey, Akash. okay so martin i will start with you tech had a bit of a rough week at least from a stock market perspective uh we saw all the earnings come out capex was front and center what are your key takeaways and key questions coming out of this week in capex uh my key takeaways are people uh move in herds and they don't really think independently and they freak out they see everyone else freaking out and they freak out and i think um really these big companies which are spending a fortune on capex they can afford it uh they've got the most to actually to gain uh and i think people who are freaking out about google investing a lot or amazon investing a lot are kind of missing the bigger picture laura any any takeaways from you on on capex no i mean as martin was pointing out um i'm not going to disagree with him here in his grief and con feel free to disagree with me I'm usually wrong.

43:11Well, and I think as it's been said on the show, it's not, you know, the revenue growth is very high. I mean, that's what's a little bit confounding about the stock market action. I mean, it's not confounding in the sense that, you know, there's a lot of momentum in stock markets always. And I'm sure many traders, regardless of the fundamentals, don't want to be on the other side of, you know, a wave of selling. But, you know, for the average investor, is it really smart to time the market on something like this? I wouldn't think so. Because as we've been pointing out in our articles and briefings, the revenue growth at companies like Google have been pretty remarkable.

43:58So that's the other side of it. Okay, so let's go from CapEx then to the SaaSpocalypse, which has been playing out. We were just talking about the last segment too. So the big question, is AI going to replace enterprise software? That's what every trader is thinking about. It's what everyone is selling these stocks for. Look, I have a theory on this, but Martin, where do you think this is going to play out? God, that's a tough question. And look, this is not to sound really predictable, but this reminds me a little bit of the cord cutting debate, which, you know, obviously cord cutting has destroyed the entertainment business.

44:40And over a period of about 10 to 15 years, it has wiped out the money that the cable channels used to make as people moved over to streaming. Now, it took a long time for that to actually happen. I mean, I was writing about that in like 2009, 2010, 11, 12. It's really taken until the last three or four years for it to become a reality. And the market didn't anticipate it quite as early as it has here. So I would say, look, I don't think we yet know which of the software companies will get hurt. I think obviously some of them will, but it's a very complicated industry. I think the market is overreacting.

45:25I definitely think if you look at some of the stocks, some are trading at very low multiples, despite the fact that they're still growing fairly well. But I do think as well, I mean, Laura was arguing earlier in the week that there will be, here, Laura, I'm stealing your thunder. Sorry to do that to you. But Laura was arguing that there might be mergers because, and I think that that's correct. Actually, I should let Laura talk about that. Okay, so Laura, so which mergers are we going to see? Well, I just want to sort of go back to what Martin was saying about the long, drawn-out process of this because, you know, we and our reporters are really avidly trying to figure out what happens.

46:11I mean, if a stock is down 80 % in a year, right, there's pressure on the CEO and the board to do something. And, you know, for a big pocket of the market, this could be seen as a great buying opportunity. And for PE firms that have historically loved to buy enterprise software because of the cash flow, it's recurring revenue, you can borrow against it. This would seem to be great. And I do think we hear that sentiment. I was just talking to a bunch of late-stage investors last night. And, yes, they are thinking about bargains. But I'm getting a sense of there might be a period of paralysis because especially when you have market gyrations like the ones we've seen in the last couple weeks, people don't really quite know what's going on.

46:59And I'm not saying that lasts forever, but that could, you know, just like we saw after any market correction after 2022 when the interest rates spiked, there's this reset in valuations, whether in the public or private markets. And the solution to that doesn't happen necessarily right away. I mean, at a certain price, I'm sure some of these software companies are going to look attractive for somebody, whether it's a stronger competitor, you know, which leads to a merger of equals, to a financial sponsor like a PE firm. but right now when things are selling off so hard and there is this philosophical question out there, like, is this the end of software?

47:52Which, you know, it seems silly, but people are asking that question. I mean, that's the other side of the debate. And these are smart people who want to make a buck, right? They want to look for that return at a certain price. Can I just jump in here on this point? The same people, maybe not the same people, but don't forget there's been all these people over the last five years claiming that crypto is going to take over the world. Well, so far it hasn't really, has it? We haven't really seen very much of a use case for that. And now this week, crypto is selling off. So I think we have to be careful before we believe some of these, you know, the hypesters who claim, oh, this new technology is going to transform everything.

48:39I mean, it does sometimes, but not always. You know, I don't know the answer to that question, and I'm certainly not, like, validating it. I'm just, I mean, look, we are all in the news business, and we write about things that happen. I like to actually predict things, Laura. Well, one possibility for us is that things sort of stay stuck a little bit, right? Like there is this big catalyst to something happening. The stock market moves. And so maybe, you know, companies that look too expensive two months ago or six months ago now look very affordable. Their market cap is$2 billion. Like that makes it much more affordable than it could have been.

49:25But if the board is, you know, or the PE firm is uneasy about like kind of the ultimate direction, maybe that stalls things. I don't think they stay stall forever. I mean, I think after every one of these market corrections, and we just saw that after, you know, the low interest period, it took a while for, you know, the private capital markets, the private funded startups for those valuations to kind of get to the level that investors were willing to put more money in. But it didn't happen. I mean, we were on that story, you know, from day one. And it's actually a slow development that takes place over.

50:03I mean, it's still taking place, right? Like the Brex, we saw that takeover, which was sort of a valuation reset. And so, as Martin is saying with core cutting, this could take place over several years, right? Like the story doesn't like get resolved in the next two months. It's just that the markets move much quicker now than they do. It will calm down and then things will sort of stabilize a bit. And then, you know, then there'll be one more of these in a year or two. And this is, it's just going to take a whole long time. Well, I mean, and this, you know, all these, you know, when the AI companies go public, inevitably, whatever, you know, five, two, well, this year, we know that the big, the big companies are looking to go public, but all these application layer companies, I mean, you know maybe there will be a wave of the don't forget hang on a second this ai stuff let's just be be real we don't yet know whether any of these ai guys can actually make any money the only people who make money out of ai right now is nvidia so this we are still will open ai ever make any money who knows right right will that prevent them from going public though i mean what will happen is they will go public invest will be like oh my god this company is losing a boatload of money and So in that scenario, I think that could be like, oh, it'll be fine.

51:23It'll be fine. In five years time, you're still losing a boatload. Then he'll be fired. I mean, I'm very, very curious about it because, you know, we've reported, you know, as much as we can about the financial activities and projections of these companies. But it's just a tiny bit of what you'd have with a public company. We don't really understand some of it. And I agree when these companies eventually go public, investors are going to get a much closer look. And it may be – they may see like much more volatility or concentration – customer concentration risk, which we're already sort of hearing whispers of than investors really like.

52:06I mean, in theory, that could make some of these boring software companies look great. Look good. They'll look really good. Yeah, yeah. Yeah. So in other words, the SaaS companies just want the AI companies to go public so that they have some legs to stand on. I mean, maybe. I don't know. I mean, I don't, you know. Having a bottom line is good. OK. So that's always that's always good. Yeah. I mean, I think also with these software companies, the ones that are not showing very strong growth and earnings, but the growth is actually stalling, which you know there's plenty of the kind of flatlining software companies.

52:57And this was happening even really before AI became this big bogeyman. I don't see like why an investor would choose that. And you would just go to the cream of the crop for SaaS. Great. Well, I want to thank you both for coming on. That is Laura. Are we going to do a debate for deploying engineers? Okay, fine. Okay, fine. We'll do it. We'll do it. We'll do it quickly. Okay? But just for context, okay, for listeners and viewers, There was a debate in the office earlier this week. I was actually at home. I was calling Laura. All right. So Martin doesn't like the term forward deployed engineer. And Laura says what?

53:36It's okay? Well, look. You defend it. Come on, Laura. Defend it. I defended the use in this story. Now, for listeners who are not so in the weeds, what this refers to is a type of particular engineer. It's a consultant. Right. Right. Some engineer goes in, helps you implement AI software. That's a boring term. We have to give them this ridiculous term. Right. We didn't do it. It's Palantir, which employs people who used to work in the military. They have co-opted a lot of military terms, which is something that happens in journalism and language in general. We use terms out of sports and military.

54:18And I particularly don't like that because I think it's jargony. you know, deploys, you know, means that. But you were okay using it. I was okay using it because I think in this, you know, tech and labor market, tech labor market, I should say, not general labor market, it means something in particular. And I think there is a bit of a race. I mean, you know, I would not die on this hill, I would say, but I do think. Well, you changed your tune, so I think you proved that. I thought it was appropriate for this story because I think the term consultant means something different to most people, and it does not relay what forward deployed engineers has meant with the AI startups and the big AI startups like Anthropica and OpenAI and the small ones like Distill that are building their model around hiring these highly technical consultants or, you know, sales friendly engineers, whatever way you want to call them, and putting them in house at these companies.

55:28And, you know, and I think it means a particular thing. And I think there's some competition here because I think the companies, you know, like OpenAI are realizing if they really want to get the large corporations, this enterprise wallet, then they need to put people on the ground there. And that's an expense, right? Like that's not just like go into a website and, you know, set up your API calls. This is like you're employing people. I mean, you know, this is consultancy. And that means, you know, a different kind of cost, in my opinion. And so I think there is interest in that topic. And is that really what you're going to need?

56:07And it says something about the ability to adopt AI, I think, when you have to have these things. Well, look, Martin, you've been on the show here. You've said AI is going to take all the jobs. What are people going to do? This is what people are doing. I mean, I think this is - Coming up with stupid terms to describe - But it's a revised job description. It's a forward thinking job description. Look, at some point, I'm sure maybe you will use it in a briefing. We'll see. I am never going to use it. Got it. Okay. Well, when you do, then I look forward to bringing you back on and telling you. Okay.

56:43Thanks for coming on. That is Martin Pierce, our co-executive editor, and Laura Mandara, our managing editor here at The Information. Well, that does it for today's show. A reminder, we are on this stream Monday through Friday at 10 a.m. Pacific, 1 p.m. Eastern. I want to thank you all for tuning in. And we really do appreciate your viewership. I'm already excited for our next show on Monday. Have a great rest of your Friday and have a great weekend. Bye-bye for now.