Carly Rector: the slightly optimistic take that I will have on this ⁓ is that this is Bloomberg interactive chart that many people that I have known have referenced independently. ⁓ And many people are aware of this at this point, ⁓ that this is a thing and that it's a great thing and people are talking about it.
Oz: Mm-hmm. Mm-hmm.
Carly Rector: Which, again, the first step to solving a problem is acknowledging you have one, etc. people are at least noting that this is not great, including Bloomberg, notably not a socialist bastion usually. ⁓ is biggest one here, obviously. It's the biggest ⁓ circle this chart right now, and also the one with the most links here. And their stock has been booming lately, even. Like, I think they have 4.8 trillion valuation on here, and I think they're coming up on six ⁓ with stock rallying recently, $6 trillion, Because... ⁓
Oz: Mm-hmm.
Carly Rector: Everyone all the investors are being like well We don't know who's gonna win the the AI app race except that we know it's gonna be Nvidia because look at this chart Doesn't matter who wins Nvidia is gonna come out on top we think so So everyone's trying to get into Nvidia and they are notably a public company so everyone can just buy their stock
Oz: Yeah. Yeah. All right. Let's see how that works.
Carly Rector: Yeah, I mean, they're also the what they like to call picks and shovels company of AI, which is like if it is a gold rush analogy, if you've never heard that saying before. It's a startup thing, but it's basically like in the gold rush, you don't want to be the person who's digging for gold. You want to be the person selling picks and shovels. ⁓ then it doesn't matter if anyone actually finds gold. You're doing fine. You got your
Oz: I have not. ⁓
Carly Rector: You got your money, more or less. So, is here, picks and shovels of the AI boom, which is people have to buy their chips. Doesn't matter if they do anything successful with them.
Oz: Boy, what a fun way to live. ⁓ Historically, ⁓ Gold Rush noted as real cool and fun. ⁓
Carly Rector: Mm-hmm. Mm-hmm. Real great for the economy.
Oz: Good for the economy, good for the environment, good for the people involved, good for the people not involved.
Carly Rector: Yeah. So yeah, that's that. And there's some other picks and shovels companies on here, but basically a lot of the hardware chip companies are getting that treatment. ⁓ Or any of the more foundational tools to it, just like the companies doing data labeling are also sort of in there, like that kind of stuff.
Oz: Yeah, I makes sense. Yeah.
Carly Rector: the things where it's just like, is foundational to the whole thing and it doesn't depend on anyone particular successful bet.
Oz: Right, right, right, because everyone's gonna need these things.
Carly Rector: Yeah. And I guess slightly positively, you could use chips for lots of other things too. So some of the what happens when the bubble pops takes that I've seen have been like, but it's not necessarily bad if we have a bunch of chips in data centers, because there's a lot of things that you can compute, ⁓ which it better than some other bubbles. ⁓
Oz: Yeah, yeah. and afford graphics cards again, that'd be great.
Carly Rector: Yeah. Everyone build their cloud gaming companies. They've never been able to make work until now.
Oz: ⁓ yeah, this time I swear, I swear the cloud gaming company, just give me one more shot at it. Making a viable business model.
Carly Rector: What's wrong? So yeah, so our final topic is your topic. Do want to give a little summary?
Oz: Yeah, kind of ⁓ pursuant to what we've been about with the technological problems versus systemic problems there has been a big discrepancy in how various, either laws or, know, of service have been enforced on large versus small creators. And kind of the two, of that I've been seeing the most of is one, ⁓ the way censorship been coming down on smaller creators, mostly driven by these payment processors, know, Visa and Stripe and such, disallowing, you know, adult on storefronts, but not necessarily from large production companies. ⁓ then sort of, ⁓ isn't it funny how copyright law doesn't matter anymore? ⁓ Like, you know, remember ⁓ they were putting grandmas in prison for ⁓ Napster and now we just like, yeah, just kidding. ⁓ But the payment processor obscenity stuff, ⁓ man, now that Kickstarter is also like really cracking down ⁓ ⁓
Carly Rector: Mm-hmm. Mm-hmm.
Oz: you know, adult material. ⁓ kind of depressing. Like I saw a tweet from someone on the Baldur's Gate creative team being like, hey, shouldn't, doesn't our game not meet Steam community guidelines? Kind of interesting that you're tossing off a whole lot of, you know, small and, you know, marginalized queer creators, but not, not Baldur's Gate. Baldur's Gate's not a problem, huh? And, you know, I think, I think,
Carly Rector: Yeah.
Oz: some platforms have tried to be more judicious about it than others. know, HIO has been like quite imperfect, ⁓ but at least are like showing some sympathy and sensitivity for creators. But I think it's really scary, especially when, you know, a lot of legislature is also rolling back protections for minorities and queer folks that like,
Carly Rector: Mm-hmm.
Oz: You see things with any queer themes being called quote unquote adult. If there is just. Any whiff of anything remotely sexual about it, ⁓ things are getting tossed off for stuff way more tame than ⁓ the victorious secret, display that I would walk past every day going to the mall job.
Carly Rector: Yeah, think the Baldur's Gate comment thing I think is really interesting and pointed, which is just like, yeah, you will toss people off your platform if they don't matter to you. And the big ones are fine because you can't afford to lose them. ⁓ Sort of on that topic, I think the most egregious sort of platform moderation case of this. recently was X not getting kicked off the app store when it was full of deepfake porn. Yes. Yeah, like, because it's like, it has more and worse stuff on it than like so many things, so many things. And like Apple is historically like really strict about that stuff. ⁓ And no, no, no comment there because they're not going to pick a fight with Elon Musk. The only reason, so.
Oz: Yeah! Right, what do you mean? Yeah, yeah. Yeah, it's just kind of going mask off. mean, like, it's always been this way, and especially for certain populations, it has always, always been this way. But just how many things go rules for thee, but not for me. That lives were ruined, people were killed over piracy. ⁓
Carly Rector: Mm-hmm.
Oz: And now Disney wanted to invest in Sora, you know, Soraia. I mean, they didn't, like it fell through, but like just the fact that they were even talking about that is insane. like the number one rule in all, you know, that you hear people talk about all the time is don't mess with the mouse. just how draconianly protective Disney has always been of their intellectual property. And still just don't know.
Carly Rector: Yeah. Yes. Yeah. Yeah, like that, when that was announced, that was just so wild to me because, like, Sam Altman thinks that he can get away with everything, but he appears to kind of be able to, for some reason that I don't understand. Like, less so now. Like, he's hit a few losses now and therefore is being very confused and struggling about it. But like, got a few L's for Sam Altman on the board, but like, he was on a real run of W's there and the-
Oz: Mm-hmm. We've got some L's on the board.
Carly Rector: convincing Disney that they wanted to him money for giving him their intellectual property. The ⁓ of the deal there were never completely revealed. And I did see some reporting that it ⁓ going to involve some licensing fee being paid from OpenAI to Disney in addition to Disney just investing in OpenAI. But that was just wild. ⁓ we'll only slightly get into this podcast, but like the Elon Musk open AI trial has been a lot of digging up receipts, like it's internet drama only it's for a a trillion dollar almost company and, the richest guy in the world. Yeah.
Oz: Yeah. Yeah. And just, it just, it just doesn't, it just doesn't matter anymore.
Carly Rector: Yeah, or like it does matter for some people and it's doesn't matter if you're an AI company and you can just be like, but this is the future.
Oz: I feel like more things in human history function like forum drama than we care to admit to ourselves. But this is future.
Carly Rector: And yeah, on the topic of does copyright even matter anymore, I always think about Mark Andreessen's comments to the Copyright Office, like a couple years ago, about like the whole question of can you use copyrighted data for AI training, basically. And his comments were more or less like, it has to be allowed because the business model doesn't work otherwise. And it will kill American innovation if you don't make this be the case. Yeah, I think that's definitely true and unfortunate. And I think more people should have gone through forum drama in their formative years to have a better idea of how to handle it.
Oz: Hahaha! Get on Reddit, start some flame wars, pay for it. Anyways, it's like...
Carly Rector: Sincerely, like...
Oz: We should be having high schoolers signing up for something awful. My business model doesn't work if everyone I make eye contact doesn't hand me a crisp Benji.
Carly Rector: intro freshman year course, have a flame war and come out of it and then try and make friends with everyone in your class afterwards, I guess. ⁓ Yes, exactly. ⁓ But because he's Marc Andreessen and he's the head of the biggest venture capital firm in the country, he could just be like, but the businesses won't work if this doesn't happen. And people listen to him for some reason, it is like, okay, but like, there are many businesses run by creators, or even just like small studios are journalists or like
Oz: Yes, I like that. mod a large unruly community or a small insane community.
Carly Rector: Yeah, just like organize any group just at all I guess because like again the thing is most of these guys have done ⁓ nothing nothing even remotely like that they don't even like plan ⁓ outings like ⁓ their their wives or secretaries that ⁓ so they just ⁓ many many businesses that don't work if you can just take their intellectual property for nothing that's also a business that doesn't work so but we see which one matters and it's the one that can donate hundreds of of dollars to a political campaign so
Oz: They don't plan outings. What?
Carly Rector: Mm-hmm.
Oz: they united those citizens, didn't they?
Carly Rector: Yep, yeah, it's bad. So yes, I think it should be mandatory you have to organize a social group before you're allowed to make any consequential decisions for the world economy. ⁓ But I think that's hard to enforce. Yeah. Yeah.
Oz: ⁓ Carly, I think it might be broken and I think it might be broken fast. You ran ahead and how are you enforcing this? Oh, oh, thinking, yeah, this is a new department and went under the attorney general to like enforce LARP participation among venture capitalists. Mm-hmm.
Carly Rector: ⁓ theoretically there are more things to say about that, Yeah, definitely send them all out on LARP weekend first. I mean. The whole issue of payment processors dictating content standards, I think we have talked about the podcast before, but it's just like such a wild artifact of how we have ended up in this situation.
Oz: Right, right. What do you mean, Visa gets to decide whether or not I get to look at boobies? that's not Visa's business.
Carly Rector: Yes. Yeah, for people who maybe as familiar, this is a common pattern among platforms where they will be like, we as a platform need to impose more content standards, especially on like pornographic or ⁓ Hello and welcome to the Fast and Broken podcast where we take a critical look at the fast moving world of tech and online culture. I'm Carly Rector, a software engineer and startup founder who has been using Claude code since its release and cursor for even longer than that. And I've said that before, we're going to get into it again, sexual for it's some by some usually unclear definition content because our payment processors allow it to be for Adult content and payment processors meaning Stripe in the case of Kickstarter But Stripe is doing it because they are on top of like credit companies and things like that specifically ⁓
Oz: Right.
Carly Rector: ⁓
Oz: ⁓ Well, I'm Oz Brown. I am a ⁓ support enthusiast who has somehow, despite my efforts, also become a software engineer. And I have been your software engineer.
Carly Rector: you have, very exciting. ⁓ Oz is a friend and mentee question mark and past employee when I was running my startup that I founded, per my intro. So, yeah, which for anyone who doesn't have the context shut down like two years ago now a little over two years ago now. So Oz is back at Amazon ish.
Oz: Yes.
Carly Rector: And so because the credit companies don't want their money being used for porn, they just get to dictate the content standards for any website in the world. ⁓
Oz: Yeah, I am.
Carly Rector: I am doing other stuff, but we've got some background. In to like, being friends from the internet, ⁓
Oz: So I was reading a little more on this, months ago, believe it was ⁓ Denise Dreamwith was talking about this, that basically, you have payment providers that handle high risk and low risk transactions. And so the ones we're mostly used to, Stripe and everyone that Stripes sitting on top of, That's, that's fair. That's fair. And I've also been using Claude Code for longer than some others, ⁓ due to, your encouragement.
Carly Rector: Yeah, which we can talk about a little bit. So the topics we're going to go over today, we're to talk a little bit about AI coding tools, as alluded to by that intro. how Code is doing, our software engineers rotting their brains, some other developments in the world of ⁓ everyone is Claude Code now, or maybe they're stopping for various reasons. ⁓ Then we're to talk a little bit about Cool.
Oz: they're expecting to process a certain volume of fraudulent activity. whether this is based data or this is just kind of how the industry is, like quote unquote adult entertainment has always been considered like high risk. ⁓
Carly Rector: the ongoing saga of circular finance deals in AI. in particular, I do want to talk a little bit about the anthropic SpaceX deal, because I think that one is interesting for its own ways in addition to the fun circular finance nature of it.
Oz: Mmm.
Carly Rector: Yeah, this is starting to sound familiar. I probably also read this post at some point. And I do think there's some data around that. I remember hearing anecdotal stuff from even just back in the day when you had to subscribe to cable channels and things like that. Dudes coming in and being like, ⁓ no, I didn't subscribe to porn.
Oz: Yeah, you probably did. ⁓ I'm very excited to your take on that one. ⁓ Yeah, yeah.
Carly Rector: Yes. And then we're talk a little bit about ⁓ Kickstarter's of clarified, clarified content policies and content and copyright policies online in general and the uneven enforcement of them in some cases and how that sort of affects the online creator and content ecosystem, Channel or whatever who charged that to my credit card. No, I didn't I didn't do that No, so like I think there's there's definitely more of that in that area ⁓
Oz: Who charged that to my hotel room? Yeah, yeah. But just the problem is like what's getting defined as that is getting tighter and tighter. And, you know, I mean, I'm a bit more liberal in, you know, what I...
Carly Rector: Let's jump into AI coding tools. I think the first thing I want to do is actually do a little bit of definitions here because I feel like people who don't use them don't really know even like exactly what AI coding tools are, or I will particularly like narrow this to agentic coding tools, which agentic is a word that can mean
Oz: you know, I support sex workers, I support sex work. I'll just say that on the podcast with my professional name attached. ⁓ That, you know, I think that stuff should be recognized and protected and it's, you know, not necessarily a payment provider's business. you know, what within the law a person wants to do with their money. And also there's kind of the double problem of like what...
Carly Rector: I'm just... Mm-hmm. whatever people want it to be, perhaps as much or even more so than AI does. But the way Anthropic usually talks about something being agentic is it's an LLM using tools in a loop.
Oz: gets defined as acceptable and unacceptable eroticism, which is ⁓ obviously way outside the scope of this podcast.
Carly Rector: So what that means is it's not that you ask it to generate something and it just gives you some text and it's done. It's it generates something and part of that something is deciding to take some action and then it can get the results of that action back and take more actions such that it can do multiple things and just sort of proceed and do a whole bunch of work on its own without you having to prompt it every single time. ⁓ So in the case of agentic coding tools that might mean you ask it to do something and it not only like generates code but it also actually edits the Yes. But yeah, we've definitely talked about supporting sex workers on this podcast before. And in fact, it reminds me of another related topic that I think we talked about a little bit before, which is ⁓ open AI doing erotic chat bots. of speaking of double standards, which they have they have put on pause for now. ⁓ So that was the whole thing because
Oz: ooo Mm-hmm.
Carly Rector: First of all, Sam Altman said like two years ago in an interview that they wouldn't do erotic chatbots because that seemed like a hack to juice user engagement or whatever that they wouldn't need to do except as a last resort or something like that. ⁓ And then they announced that they were gonna do that in like December or something like that. So everyone's like, you're at your last resort already, huh buddy? ⁓ But also one of those things where it's just like, okay, like so. file and puts the code in there, and then runs some ID linter checks on it to make sure that it's syntactically correct. And then it runs some tests. And then it finds some errors. And it goes back and fixes the code. And then finally, it comes back to you and says, OK, I'm done. ⁓
Oz: I accept it's a lot, all right. My favorite part is when it clarifies requirements. So I'm like, yeah, you're right, Claude Code. That is actually what I meant, not the thing I asked you.
Carly Rector: Yes. Sam Altman, OpenAI gets to make erotic content because again, they're not real sex workers. So you get to build the content scraping company and then use everyone's work and generate that and you get paid for it, but the people who actually do it, no, they're terrible, dirty people who cannot use Visa cards or whatever. That's cool. So like... Yes,
Oz: Yep. Yep. Yeah, yeah, amazing. Right, right. That is a great point of like, well, what what are you trading your erotic bots on?
Carly Rector: one way to think about that is like the LLM ⁓ a bunch of knowledge about how to do things like how like the actual syntax of code and like knowing a bunch of design patterns and libraries. but what it doesn't ⁓ know how to do ⁓ is exactly how to take actions or like Mm-hmm. I mean, for sure, a lot of the existing chatbots are trained on Archive of Our Own. That is explicit. They have said as much, even in some of their official publications. consistent workflows. So if you imagine, even if you had some really smart person working on something, you generally don't just, like, give them a task. like, go, without giving them any more information than that. Or, some to work within. Like, ⁓ an SOP or, like, guidance about, like, we usually do it this way, like, following these steps. Like, you write the code, then you run the tests. Or... ⁓ we like these kinds of standards, et cetera, or like, if you don't understand, you should clarify. And so basically what Claude Code, the product tool is, building all of that kind of like common sense workflow stuff into a wrapper around the LLM that makes it just work better. ⁓ So for example, they added fairly late into Claude Code's ⁓ actually, think maybe five or six months ago. ⁓ the Ask User Question tool, which specifically lets it like... give you a structured set of questions where it's like pick A, B, or C. I don't know if you've even seen it do this at all. Because it only, yeah, because it didn't used to do that. Like it would ask you questions sometimes, but it would just like give you a list of questions, free text, you would have to type them back. Whereas now it will just like give you structured questions with multiple choice responses, and you can click them and go next question, etc. Making it really easy for it to be like, okay, you gave me some instructions, but it was unclear in like these different ways. And
Oz: ⁓ yeah, I hope so. you're barista, a lonely single barista, or ⁓ you're botanist, or you ⁓ run flower shop new in town and you've broken your leg ⁓ and there's one bed. ⁓
Carly Rector: Mm-hmm. Yes. Yup. Yes.
Oz: Martha Wells visionary, she knew that the robots were gonna be watching pulp garbage to learn to be people.
Carly Rector: these are the questions that I want to clarify. And like, here's what I think the recommended thing is basically like allowing you to clarify in a more structured way. ⁓ Yeah, the one thing I tried to put in here, but it was paywalled for me, was a Forbes article that was released on Janitor.ai, which is a AI erotic roleplay platform. And it is run by three dudes. this article was like the three dudes who are in charge of like the erotic roleplay website that so many women use, et cetera. So there's, to be on topic here, there's a lot of aspects here of
Oz: You know, as you're saying that, what it kind of reminds me of is, made a joke at the beginning about customer support enthusiasts, but I did get ⁓ my start doing customer support work and in particular ⁓ support over forums, over the Xbox forums. And since a lot of times we'd only have like one customer a day, ⁓ Mm. Okay.
Carly Rector: appropriating some things and then monetizing them because if you're a cool guy running a company with AI in the name then it's fine and you are a woman who is actually just doing the thing then it's evil yeah
Oz: per issue, you'd really want to give them a good answer so that it's not like ask them a one line question and then come back another line. And so we talked a lot about how do you format something to the customer where they will give you back the maximum useful information? Because of course, anyone who's ever written an email or a message with a bunch of different questions in it, you do not get all of them answered. ⁓ we a lot of time discussing how do you maximize the odds of getting the important.
Carly Rector: Okay.
Oz: Right. that. ⁓ boy.
Carly Rector: Mm-hmm. You do not.
Oz: information back to move forward on an issue. it's kind of interesting that ⁓ that's also kind of being built into these ⁓ agentic systems.
Carly Rector: Yeah. But yeah, I think there's a lot of really concerning things that follow that pattern of like, the deciding factor is not the content, but the entity or person that is presenting it, and therefore it is really a judgment on that person and not on the content.
Oz: Yeah, yeah.
Carly Rector: to sort of segue to our first bullet point here, which is like ⁓ the Claude code leak the Claude code harness itself is a separate thing from Claude the LLM ⁓ And there's a lot that is in the harness actually because it turns out that all that kind of stuff is what makes it work really good. And there is something like 500,000 lines of code and just like the harness for Claude code, which is probably I think so. Let me let me make sure that's right.
Oz: 500k in the harness?
Carly Rector: 500,000 lines of code. Which I think is bloated, probably, because they are having Claude write all of their code for Claude code. But there's just a lot of stuff in there like prompts and tool use guidelines and just a bunch of things in there that are
Oz: Yeah! ⁓ Mm-hmm. Mm-hmm.
Carly Rector: best practices for how to work with code, basically. ⁓
Oz: I mean, I think one frustrating thing about being an engineer is that people don't want to believe that tech requires engineering. You know what I mean? Like they really do want to think that we're wizards and like, it's almost like, you know, my wife's an artist. And so people are like, ⁓ they, know, they think that like drawing is magical. And like, it's like, no, it's a lot of thought and skill. like, so people, you know, think that like tech.
Carly Rector: Yes.
Oz: is magic and like it's not like no it's all just thinking.
Carly Rector: Yeah, it's like for a long time, I have felt that it's difficult to explain to people what a software engineer is and what they do. Like kind of both in the non-tech world and just like to non-engineer tech people even. ⁓ Because like non-tech people just don't understand at all.
Oz: Mm-hmm.
Carly Rector: like people just don't have the concept of like what it is involved. I think they do a little bit more now in terms of like you write code and stuff, but even still, think will imagine like HTML, and not like the amount of logical putting together of systems that is the actual engineering part of it. And how little that has to do with code in a lot of ways. And. Like I took an algorithms class in school where we didn't write code. Most of it we didn't even do on a computer. We did it on paper because it was just like doing algorithmic proofs, Cause it wasn't a matter of just like how, how did the computer runs it. It was more a conceptual. this one is especially depressing to me because ⁓ there's so much like protect the children bullshit around it and it's just people who have done much more research and are much more eloquent than me on the topic have talked about why protect the children is
Oz: I know. it is never about projecting children. It is.
Carly Rector: ⁓ But that like there are definitely a certain set of conservatives who, for example, don't just want to stop children from seeing porn. They want to stop everyone from seeing porn and that they will do that just by making it really inconvenient for everyone under the guise of protecting the children, for example. like that's. thing. And as people are figuring out that just because AI can write code doesn't mean that it can engineer, people are having to sort of discover that more. ⁓ And so I think that a lot of Claude Code and a lot of similar people have to build agentic harnesses for things is them figuring out that just having the raw capability doesn't do it for you. ⁓ You need to actually do some systems engineering there. ⁓
Oz: Yeah. Yes.
Carly Rector: a real motivation that some of them have said out loud, literally, but yet there are still a lot of Democrats who are like, well, I'm not going to be the person who says don't protect the children, so I'm on board this too. And it would be really cool if anyone had any amount of spine, but, you know.
Oz: one thing I kind ⁓ like about vibe coding as a concept, like I'm not as against vibe coding as lot of people, at the same time that ⁓ people are finding out that a computer can't do engineering, I think there are other people ⁓ are finding out that they can. Yeah, for real. Well, it's fun to have dreams.
Carly Rector: Mm-hmm. Yes, sure. And I'm actually quite pro vibe coding ⁓ to many people. Like, not only because like for my entire life, basically, but definitely my entire adult life, I have felt that more people could be software engineers than are. Yeah.
Oz: Hey And I think in conclusion, ⁓ all technology was a mistake. Rocks weren't meant to think. And we have done the devil's work.
Carly Rector: That's it. Yeah. ⁓
Oz: ⁓ yeah. Bring back carrier pigeons.
Carly Rector: Like especially women, obviously, but just like in general, because it's framed as like you're sitting at the computer and you're just like typing basically. You're like the hacker or whatever, and you're putting in too many semi-colons. It's like, no, that's not, that's not the fun part though. Like the fun part is like you're building a thing that does something and you're figuring out how to solve problems. And like that happens to incidentally involve code because that's how you like tell the computer how to do it. But that's not necessarily the most important part of it. ⁓ I don't think all technology was a mistake. Some of the internet is pretty cool, but...
Oz: Yeah.
Carly Rector: and Vibe coding allows people who don't do the code to do some of that other part of like having the fun part of actually making things. ⁓
Oz: Yes, yeah, and I think, know, vibe coding isn't a bad way to learn coding if you're going back and studying it also, you know, like...
Carly Rector: Mm-hmm. Yeah, and like I think there's definitely some pitfalls of it and a lot of them are like people getting too over excited about it and being like, yeah, I could just launch an app, I don't need software engineers anymore or like let's just vibe code everything now and you don't have to take a step back and be like, is this a problem that we actually needed to solve it all like let alone with a vibe coded app? So like, but like which I have some empathy for because it's like people are getting suddenly unfathomable power. Like they just get really excited about it. It's like that's fair, but still. And like there's some risks of obviously security things. You don't want to use it for things where it's important. But I also am not as a big stickler on that for a lot of things. Cause if you're just building an app for yourself that runs locally, it does, it genuinely does not matter. Vibe code your heart out, like, et cetera. ⁓ But, and then yeah, like.
Oz: Yeah, yeah.
Carly Rector: really concerning because it's like it's not like tech was great a decade ago, or even like 15 years ago, I do think it has just gotten way worse ⁓ it'll Yeah, and just like people when people run into issues with it, they do have to figure out like, ⁓ I actually like need to learn some basic engineering concepts here more or less to figure out how to go past like the limits that they hit there, which I think is really interesting as well. So.
Oz: Yeah, it's gone mask off. Yeah, yeah, no, I agree. just like demystification of it all, you ⁓ You can be a software engineer and Claude can help you, or ⁓ maybe you should just learn software engineering. can do that too.
Carly Rector: in a lot of ways, like again, you can just be like, ⁓ is XAI posting a bunch of deep fake porn on Twitter? Is that against my terms of service? Yeah. Am I going to do anything about it? No. Is anyone going do anything about it? No. Because there's nothing anyone can do about it. Like, okay, ⁓ like just where we live now. ⁓ And depressing. It's bad. ⁓ Yeah.
Oz: No. Yeah. Yeah.
Carly Rector: Yeah. But yes, I think the biggest thing there is like. Like, mean, part of it is, again, that like I'm a very senior software engineer and I have gotten to the point where like I work at very large scale systems. I'd like, if you're say a principal engineer at Amazon. A lot of things that they would say about principal engineers at Amazon is that often some of them just did not write code for months or occasionally years at a time. Cause like that's not the most efficient use of your time at that level. ⁓ other people can write code. Other people cannot think at what the level of a hundred person organization is basically. ⁓
Oz: Other people can write code. Not everyone can go into a half hour meeting with that person who wrote code and was like, yeah, here's how this is going to break something you've never heard of.
Carly Rector: Yes, or like plan the six month roadmap for like the entire organization, technically based on like an understanding of what the goals of the business are or whatever. And like that kind of stuff. So that is sort of like the angle that I come at it from, which is like, I have not felt like code was the point in quite a while more or less. ⁓ So like my, my experience with Claude code has been like,
Oz: I do have a hard time when people who have assured me that I'm a rube ⁓ notice what I noticed 10 years ⁓ and are like, damn, that's crazy. I'm like, ⁓ is it, are we going to like reconsider everything else you've been wrong about and I've been right about? No, we're not going to do that.
Carly Rector: this is amazing and I'm not sure I could go back to not using it ever because like the gap between what like I could come up with something that in my head that I know exactly what it is and that I could make it it happen has just narrowed so drastically because that was so big in the first place. And I think like that that gap depends on your feelings about it because it's like for me I could be like, ⁓ I figured out exactly how the system should work say at Amazon and then Yeah. Between that and actually getting the system launched involved spending another couple of weeks writing up a design document, logically breaking it out into pieces that other people than me could do and flesh out further ⁓ in further depth that even more people could do because it's just like there was just too much work there that I was not gonna do all of it. It was a team's worth of work and therefore I had to like... define it and break it out into the like such that a whole team of engineers could do individual pieces of it at the level that they were able to do and work on that over the course of like five sprints or whatever. And then it got launched and that's just a lot of work. And now I can just be like, Claude, I vaguely described this spec to you. And then it's like, great, I created a plan and you want me to do it? And I'm like, yes, and then it does it. ⁓ So like.
Oz: Mm-hmm.
Carly Rector: the level of speed, like people talk about, is this like a 10 % improvement? Is this a 10x improvement? Is this a 100x improvement? Like genuinely for some of the stuff that I'm doing, I do think it's a 100x improvement because otherwise it would have taken months and I can do it in a day. Which is crazy. And that's why every time everyone's like, well, I don't know if coding tools are actually that, that much improvement. like, no, you don't understand. ⁓
Oz: Yeah, that's crazy.
Carly Rector: ⁓
Oz: just wish there was some humility in tech, you No. ⁓
Carly Rector: No. I mean, yes, I also wish that, but I guess I have stopped having any hope for that to be the case.
Oz: And think that's, you know, I think that's where a lot of, but I don't even know what the word I'm looking for is, but like, kind of like the tension between there's some stuff that these tools are really, really good at. And then there's everything else that is being forced to adapt to it that like, it's like, well, no, it's, you know, it's bad there. You know, it's not good at doing that thing or it's extremely energy intensive or it only works because it's, you know, Denethor, yes I do wish that.
Carly Rector: Yes.
Oz: trained on a bunch of stolen artwork. ⁓ know, again, like, wife is an artist, artist. ⁓ if ⁓ she was a digital artist, ⁓ we might have some more contentious feelings about, ⁓ you know, AI use the home. But luckily, ⁓ computers can't stab you yet.
Carly Rector: Mm-hmm. yet. I mean, they can probably, but not very well.
Oz: Actually, I guess that's sort of radiation therapy as a computer stabbing you.
Carly Rector: Yeah, was going to say there's definitely medical robots for that kind of, not tattooing obviously, but.
Oz: Not tattooing. Yet.
Carly Rector: Yeah, only because there's not that much money in tattooing probably, but ⁓ yeah, but yeah, like there's, the theme of this podcast is that there's a lot of tension here and that I think there are some good ways to use AI and bad ways to use AI and we should talk about both of those things instead of being either on the hype train or the throw it all in the trash forever train ⁓ because.
Oz: I am on the technically impressed and practically concerned train.
Carly Rector: Yeah, like, the other part of it is like, the, could do something in a day that I could do in months is like cool, first of all, and fun because I can get things done, but also as someone who likes to think that I'm working on things that are valuable, like that feels really valuable because that means you can do so many more valuable things with that. So, and like,
Oz: Mmm. Yeah, yeah. I mean, Carly, if the people, there were more people like you doing things and less people unlike you doing things.
Carly Rector: Yeah, and well, so the thing is, I think that's the problem with a lot of the, at least like the scientist side of the AI hype thing is like scientists and engineers are all people who work on stuff where they're like, this is all really valuable, but it's really hard. And if I could just do more of it, it would be better. And so, and often they are very narrowly focused on that and don't imagine that there are other fields in which that is not the case. ⁓
Oz: Mm-hmm. Right, right. And you know, and and and I industry loves to downplay its its harmful effects, especially within itself. you know how hard it is to get someone to see the flaws in the things that puts food on their table, you know.
Carly Rector: Mm-hmm. Yeah.
Oz: You know, my, my, the irony of my dad, who's extremely concerned with global warming, but like, you know, I grew up when he was working for oil companies and like thought everything I said about environmentalism was like, you know, weird hippie stuff. And now he has ⁓ a ton of climate anxiety and he's right.
Carly Rector: Yeah. Yep.
Oz: And so it can be hard when we're engineers and we're in tech and you're getting inundated with these constant messages of all these ways that it's improving things and innovating and yada, yada. And it's like, well, yes, it is doing that, but it's enabling scammers ⁓ and it's allowing unscrupulous companies to
Carly Rector: Uh-huh. The amount of ⁓ discovering that tech people are doing of stuff where it's just like, this is literally the foundations of the humanities. you had not been dismissing the entire value of a humanities education your entire life, you would have known this. you're not the first person just discovering this now.
Oz: lie about where their product's coming from ⁓ ⁓ causing a ton of environmental damage and it's really annoying on Google searches. ⁓
Carly Rector: Yep.
Oz: And so it's like, yeah, know, Claude code does make a lot of things a lot easier. And also I understand ⁓ why I have a lot of friends who are really angry.
Carly Rector: There's a lot of that stuff, but at least they're discovering it now, and we can have these conversations a little bit more.
Oz: Hubris!
Carly Rector: Yes, and like I also understand both of those things, which it makes it hard, but
Oz: I wish, my wish is that more times when someone discovers or newly learns to value like a large branch of knowledge or skill that they previously hadn't, that like their first instinct wasn't like, I am so much smarter now. Like their first instinct was like, ⁓ shit, ⁓ was something before. Maybe I need to ⁓ reevaluate the.
Carly Rector: I hope the things are separable to some extent, I guess, is the idea. don't think AI is to die forever, partially because it is actually very useful for things and is proving so in real ways. yeah, so given that... ⁓ Yeah.
Oz: Well, the technology's not gonna go away. the value of my contribution here, no, it's always now I'm extra right.
Carly Rector: I would like us to be having a conversation about like, okay, what if we didn't do some of the bad things quite as much ⁓ and not the like, it should just all die forever because it's not gonna die forever. So like that's the theme of the podcast, but more or less. ⁓ Yeah. Yes. On that topic, I don't think I've talked about this one on the podcast. I've ranted about it on LinkedIn a little bit. The whole thing about AI people talking about how taste is the differentiator now.
Oz: Well, I think, think,
Carly Rector: Like, anyone can build a thing, but having good taste about what the right thing is to build and the aesthetics of it and the interaction of it is important. And I read one whole long blog post about, like, okay, how do you develop taste? What does taste mean talking about looking at really good music producers ⁓ and art studies things like that. I'm like, yeah, because this is a whole thing. ⁓ Guys, like...
Oz: systemic problems aren't solved by individual actions. And I think that's true on a personal level. And it's also true on a technological level, right? Like while we have a system that is rewarding a race to the bottom and, capital as the be all end all success metric, whatever the technology is, it's going to be used in a deleterious way because ⁓ the people who ⁓ control ⁓ Okay.
Carly Rector: Mm-hmm.
Oz: Oh word? human creativity matters and a breadth of knowledge and experience? That's crazy.
Carly Rector: Yes.
Oz: have absolutely no reason to be responsible with it.
Carly Rector: Yeah, and that goes to something else that I can maybe like bring us a little bit more back on the list of things that we were supposed to talk about on this list. But like, which is like a lot of the problems that AI is causing are like not new problems. They're just drastically accelerated problems. And like, if you got rid of AI, they would still be there, but less so. like, maybe that's still that's
Oz: ⁓ yeah, yeah, yeah, yeah. Correct.
Carly Rector: But, that's also an AI-positive thing, is that people do have to start caring about that stuff A little bit more. still fine and better and it's like brings it to a less so level that's acceptable because it was before. But like maybe we should just figure out how to solve the underlying problem to some extent. And like the more the most like simple version of this one is with one of the things that came up in this 404 media article about software engineers rotting their brains, ⁓ which is the problem of like AI generates more code than humans can review, which is one of the one of the big things that people talk about. And my response is like,
Oz: Please, prayer circle, prayer hands.
Carly Rector: Yeah, like there's some genuinely good potential impacts of how AI could impact things. There's also many, many bad ones, but like there are some potential good ones that I do already see in some ways, and some of them are people understanding what engineering is, people understanding the value of creativity and ⁓ human interactions.
Oz: Yeah.
Carly Rector: Yes, but also like humans reviewing code has never been the best way to make sure that code is good. ⁓ Yeah. So, and I did when I was at Amazon do both like the actual reviewing and the trainer for other reviews of people who basically reviewed, ⁓
Oz: I mean, let's put a pin in people value and creativity. Everyone writes more code than I can review.
Carly Rector: Yes, I mean they're still working on that one, but like again, like seeing some people start to be like, ⁓ huh, there's something here, isn't there? It's like, yeah, there is. ⁓ Which again, would have been nice if they noticed that or if they acknowledged now that it has existed for a long time and they've just been ignoring it. But I will take acknowledge it over not acknowledge it.
Oz: Mm-hmm.
Carly Rector: technical assessments for hiring. People went through an automated coding assessment. The automated system graded whether you passed or failed. And for some gray area where it couldn't decide It kicked it to human reviewers to go in and look at the code and decide if you passed or not. ⁓ So we were reviewing the code, and a lot of that was done ⁓ originally.
Oz: ⁓
Carly Rector: Thank you for listening the Fast and Broken podcast. I'm Carly Rector, and if you'd like to support my work, you can subscribe to my newsletter at carlyrector.substack.com. ⁓ it was done just like on paper. The recruiters would come and be like, here's a printed out assessment. Please read this code that this person submitted. And I had to like explain to these HR people that like, I'm not a computer. Like I cannot read the code on this paper and even tell you if it compiles or not. Like that's not how that works. And like I can generally like read it and try and get the gist of what they were going for. But like if you want to like test it, you should run the tests by a computer because that's
Oz: I'm Oz Brown. If you ⁓ want say hi Oz, feel free to do so on LinkedIn. You can find me under Olivia quote unquote Oz Brown ⁓ or you can check out America's Next Top Model for the original Wii console where I play about half the supporting cast.
Carly Rector: Yeah, so it's gonna be real easy to check out, I bet.
Oz: ⁓ just head down the nearest. Yeah, the nearest five dollar bin in a Walmart 12 years ago.
Carly Rector: That's gonna be way better.
Oz: They wanted you to like take a red pen, like circle the missing semicolons.
Carly Rector: Game stop. ⁓ I mean, not specifically, but like in general, like grade this code on whether it was good or not. ⁓ and so, and I had to like be the guidance and be like, okay, you're not grading syntactic correctness. First of all, because that's a stupid thing to grade. But second of all, because you physically cannot grade that as a human looking at code on paper. I'm sure there are some humans could do that, but also that's just like, that's not a productive thing to do. ⁓ Yes, and if you enjoyed this podcast leave us a like or a review it lets us know people are listening and helps more people find the podcast. See you next week
Oz: Okay. That's a party trick, not a tech skill.
Carly Rector: Yeah. And on every team that I ever been on. People have been waiting for like weeks for code reviews because people have just not gotten around to it. And then have gotten a code review that was just like, looks fine or you go into the code review and then you're like, this is so fundamentally wrong that you shouldn't even have written any of this code. Like I should have stopped you beforehand. And like, those are the two modes where like you get ⁓
Oz: Mm-mm-mm. I'm so sorry. Come see me after class.
Carly Rector: Yes, like those are really the two modes of human code review. And like, or like I have picked a bunch of stuff that is pointless. That also happens. Like, so like none of those cases are good, like are helpful really. Like there's a very, very narrow slice of things where like a human looking at the code is better than say doing a design review before you write anything or having tests that you run to make sure that it gets the right answers or.
Oz: Mm-hmm. Mmm.
Carly Rector: Having static analysis tools that make sure that all of your formatting follows certain standards and that there aren't any security issues that you can catch via static analysis tools. All of that's going to work way better than a human looking at the code. So that's one of the things where it's like, yes, yeah, you could say that AI generating more code than humans can review is a problem. Or you could say that we should not have been counting on humans to review code as our measure of quality anyway, and now we just gotta figure it out because there's too much ⁓ there, which is what I would say. So like, that's one example. And so that's one of the things that came from this software engineers say AI is rotting their brains article, ⁓ which is basically a bunch of quotes from software engineers, mostly at large companies about how they are using AI base coding tools, either because they wanted to try it out or because they're being forced to and how... it's not great in various ways. like many of them are real in like, yes, if you're forced to like review tens of thousands of lines of AI generated code, that's bad. the thing that I said in my comment to this is like, it's exciting to me to use Claude code because if you imagine before I was spending like 10 % of my time doing like the fun high level planning strategic stuff and then 90 % of my time writing code. Now I don't have to do that 90 % mostly and I get to do a lot more of the fun stuff and a lot less of the code. But if you're in a job where, for example, you're not allowed to do any of the strategic stuff or like that's not your role, all that you've gotten is like most of your work taken away. And the only thing I get to do like a tiny bit left of whatever the grunt work is left that is not that and not any more fun stuff. So that sounds bad. Like I understand all that. I think those are a lot of real.
Oz: Mm. Right, right.
Carly Rector: real issues, especially with how it's being implemented in big companies. ⁓ But also, I don't think the conclusion of this article is kind of just like it's fundamentally bad for software engineers, and I don't know that that's true. I think it's just...
Oz: I mean, here's the thing. just, I just have better things to keep in my mind than Java syntax. Like, like anything like. And anything.
Carly Rector: Yes. Yeah, I saw one, one take that someone had that I thought was really pointed is that like there are different kinds of software engineers. There are software engineers for whom the code was the point. Like they enjoyed just like putting together the code and like writing cool, elegant function or like, et cetera. And those people are really upset that they're not able to do it that anymore because of the AI coding tools. Or there are people who really didn't, care less about the actual writing of the code part and are there for like
Oz: Mm-hmm.
Carly Rector: solving of problems and being able to just like put together something that does a thing and therefore are quite happy to have that part taken away from them because they didn't really care about that in the first place, which I think is true to some extent.
Oz: I feel a little of both of those, like... I guess, I guess, yeah, professionally, I prefer the high level thinking stuff, but like I do, there's something a little bit soothing about just busting something out in Python. Like what, you know, just like, just like, I just, I just need a little, little JSON parser, you know, or like,
Carly Rector: Mm-hmm. Yeah.
Oz: a little thing that ⁓ forum posts into RenPy code so that ⁓ write visual novels more easily. That was fun. ⁓
Carly Rector: Mm-hmm. Yeah. Yeah, and maybe it is just that I've done it enough that that doesn't have any novelty for me at this point. ⁓ But who knows? So like, yeah, I think there's definitely a range of reactions to it. ⁓
Oz: Yeah, yeah.
Carly Rector: So that was more or less my takeaway of this article is like, I don't think it's rotting my brain. I think it's making me exercise my brain more than I was before because I don't have to do any of the stupid grunt work stuff that I don't really want to be doing anyway. ⁓ But I also see how that is not the case for many people. ⁓ And I just think it's harder to identify exactly what the cause is there than they are necessarily implying.
Oz: I think that over the next five to 10 years, unless bubble pops catastrophically, and who knows, obviously it's going to really, really change what engineering looks like, especially in big tech. And I think it could change what coding looks like in big tech in a lot of different ways. And I hope...
Carly Rector: Yep. Mm-hmm.
Oz: I hope that it happens in a way. where there's more appreciation for, like we're talking about, like the high level stuff and the thought that goes into putting together systems and understanding how things are going to interact with each other at scale. And that the demystification of the code itself lends to a greater appreciation for engineering. And I know that's...
Carly Rector: you Mm-hmm.
Oz: Probably a little optimistic and idealistic, but it is fun to dream.
Carly Rector: Yeah, I hope that I can see some of that happening like already to like, again, I'm not in a big company. like, caveat with the fact that I think big companies are way worse at this, especially because you have to still do like stupid sprint processes and things like that. And also there's a lot of companies where even worse than Amazon about not letting engineers think about anything about the within the code that they're writing, making it hard. And I'm working with startups, but
Oz: You
Carly Rector: I'm definitely seeing people already who have picked up the vibe coding and they have run into walls and then they have come to me and said, Carly, what's happening here? And then I will explain something that I know off the top of my head because I'm a good software engineer. I'm like, ⁓ yeah, that pattern is problematic for blah blah blah reasons. You should probably do it this way. And you can see the realization on their face. It's like, ⁓ there's stuff here. So.
Oz: Yeah, yeah. I mean, yeah, you just, you... It's such a black box for a lot of people and I just don't think that that serves the people inside or outside of the box.
Carly Rector: Yeah, like I've always, always been a big advocate that there should be much less of a wall between product and an engineering or whatever you want to call it. Like the, where the product people are like, no, I could not possibly understand any of the technical details and, and the technical people are like, I don't care about the product. just get requirements thrown over the wall to me. Like that has always seemed like a terrible way to operate for me, but
Oz: Absolutely. Right, I know, I know, so...
Carly Rector: One thing that I definitely, definitely think that AI is going to do is break down that wall because to the extent that you could ever just toss something over the wall, which was not high, you could just toss it over the wall to an AI now. ⁓ but you can't for a lot of things. You have to have some, ⁓ I was saying, I think last week even on the podcast, understanding of both. the problem that you are trying to solve and the nature of the systems that you're trying to use to solve it and like how you can fit those things together in the best way. ⁓ And I think to the point of another thing that I talked about recently last week, we still have a half finished sub-tech post about the roles like forward deployed engineer are starting to combine those more of, which is basically an engineer that goes into a company and talks to the customers and builds something directly without their being.
Oz: Yes. Mm-hmm.
Carly Rector: a PM or anything in front of them. that's according to Andreessen Horowitz, the hottest job in startups as of last year and still continues to be very hot. I think both OpenAI and Anthropic are building like their own companies that are basically ⁓ just deployment full of engineers that figure out how to use AI ⁓ for customers product ways. ⁓
Oz: Dope.
Carly Rector: I think there's a lot of that already happening and I do expect that trend to happen because I just think that's what makes sense because the piece of work that is left is figuring out what the problem is and then how to translate that to a shape of a solution such that you can tell it to an LM maybe but like there's still work there so I think I think definitely people are already starting to see that in my opinion. ⁓
Oz: And point out to the LLM when it has a terrible idea, because it will sometimes.
Carly Rector: Yes, that too. Um, so yeah, that, that hits some of the stuff in our agenda. Let's like briefly try and talk about the actual bullet points here. Um, one is this ramp report that showed Anthropic is beating OpenAI in business adoption for the first time. As of April, the graph is very exciting for Anthropic. It's like.
Oz: Yeah!
Carly Rector: gone up way by like, it has changed from like the February number, which was like 4.4 % anthropic and 18 % open AI to 34.4 % anthropic, 32.3 % open AI in April. I saved, February, February, 2025. So over the last like year, it has gone up dramatically and open AI has gone up and then slightly down a little bit, which is. super interesting. And a lot of that is the launch of Claude Code in general, which happened, I think, May 2025. And then big hype around Claude Code that happened around like December, January this year for reasons that no one is entirely sure about.
Oz: Yeah, makes sense.
Carly Rector: It was just like people had free time over the holidays and started playing with it and or Opus 4.5 finally pushed it over some threshold to be really good. But like I didn't think there was a step change there. Other people were like, yeah, there was definitely a step change. It finally became very good. Like in December. I'm like, don't I ⁓ as someone who's been it this whole time, it definitely got better, but like not it wasn't functional before and now it is now. ⁓ yes, ⁓ so. ⁓
Oz: Not that much better.
Carly Rector: Um, but yeah, and then obviously all of the stuff around the Department of Defense thing and people just thinking that Anthropic was better in general as a company swapping over has happened also. uh, so yeah, that's interesting. Um, they caveat of this post with like, this doesn't mean that Anthropic is winning the race. Now, like one thing that you can take from this chart is that it's actually very easy for an underdog to take over and that could still happen. Like maybe.
Oz: Mm-hmm.
Carly Rector: DeepSeek comes out of nowhere and wins in another year. Like, we don't know.
Oz: ⁓ Remind me about Deep Seek. Is Deep Seek the Chinese one? Okay.
Carly Rector: Yes, yeah, DeepStake is the Chinese one that launched and then kind of wasn't that exciting really and why is it still at like 0.1 % on this chart the whole time?
Oz: Was it more energy efficient or was that overhyped?
Carly Rector: Um, it was, uh, in the sense that it was like much, much more everything efficient in terms of like, said they trained it with, for only like $5 million or something like that, which there was some questions about that. And there was also some accusations that they were distilling the big AI labs models. And that's why they could trade it for so, so cheap. Um,
Oz: Mm. Ah, oh no, someone's... He tried to...
Carly Rector: Yes, one of those... I definitely saw an OpenAI article that was like, ⁓ no, OpenAI accuses other people of stealing their intellectual property! Who would do that?
Oz: I can't believe someone would do that to sweet, darling little open AI who's never done anything wrong and it's his birthday.
Carly Rector: Uh-huh. so yeah, but yeah, things could change at any time, but it's, Anthropics doing pretty well. And it's mostly because of Claude code. Um, we have on this thing, like why Claude code is better, why Claude code leak wasn't that interesting, which has to do with some of the stuff that we were talking about a little bit in the sense that like the Claude code harness A is distinct from Claude, the model. which is I think most people who saw that leak didn't understand at all and they were like, Claude's source code is leaked, ⁓ like doomed the company now or whatever. It's like no, it's different, ⁓ it's just the harness. That is an important part of it for the reasons we discussed, but it is not the core part of the model. ⁓ other part of it is because it is just like ⁓ a node harness, like technically all of the source code was on everyone's computers already, like ⁓
Oz: Yes.
Carly Rector: version of it because that's how you run the code. And so like it wasn't super secret even before then, like you could have tried to reverse engineer it, it would just kind of been harder. Like you didn't have like the raw human readable code and that was what leaked. ⁓
Oz: may not like be as technically interesting but I do think it's funny. Yeah.
Carly Rector: It is funny, that's true. And it was interesting to see people do analysis of the 500,000 lines of code, exactly what was in there. And I do think it gave an advantage to the competing tools. Because the other part of it is obviously what Claude Code does is not a secret. You can just use it and then see. Unless you work at OpenAI, which they have banned from using Claude Code.
Oz: Yeah. Yeah, yeah, yeah.
Carly Rector: because they were like, people at OpenAI are using it and just figuring out what it does and copying it, so we're going to block them. ⁓ But yes. ⁓ So you can just use it and see. So there's just like a lot of reverse engineering or even just like seeing what features it has and copying them ⁓ that was going on already. And then this makes that easier for them to do. But it was always just like, Claude Code has a few months head start on some of the other tools.
Oz: ⁓ no!
Carly Rector: and not like it's a fundamentally secret, secret approach that they have. I do think one of the reasons that it's better is that they are better at building for customers generally. So one of the things I have talked about on this podcast before is that Claude Code was not intended to be a consumer product to start with. It was an internal tool for Anthropic. And they were just like, actually, this seems to work really well for everyone.
Oz: Mm-hmm. That's right. That's right, it was.
Carly Rector: 95 % of anthropic people adopted it just naturally without us asking them to really quickly because it was actually genuinely really useful for them. Maybe we should launch this to people. And they were like, sure, let's launch it. So they launched it as like this CLI tool that wasn't even that user friendly, but was really useful. ⁓ So, and people found it really useful and started using it. ⁓ And then like the guy who does a lot of the work on it ⁓ is on X all the time and sees people tweeting about it and will be like turning around, be like, yeah, that's a good, good feature idea. Turn it around. It's out now, tomorrow. ⁓ level of work on just Claude Code. ⁓ So just being really responsive to like what actually makes it more useful for people instead of trying to build it as something that is, that looks good as a product, I think is.
Oz: Mm-hmm. Right.
Carly Rector: helping them a lot, which is one of the things that I feel like is just encouraging in the world.
Oz: Encouraging like from a perspective of, yeah, dang, sort of seems like when you focus on making a good product, you make a good product, but like still a little bit concerning, like from the bubble perspective. ⁓ Just in terms of like, yeah, dang, you have made something real good. Does it make you any money? ⁓ it does? OK, OK.
Carly Rector: Mm-hmm. It makes them a lot of money. Yeah, that's where a huge amount of the revenue is coming from. Now let me do a quick Google search. that's one of the reasons why their revenue gone up a ton. ⁓ Plod Code alone brings in 2.5 billion, a feature that has more than doubled since the beginning of 2026. ⁓
Oz: Now how does that compare to their compute costs?
Carly Rector: ⁓ Still, I I think they've been doing a lot of calibration on this. So this is a whole thing otherwise, which they have been trying to like calibrate their usage limits for people. And they have in fact been dropping their usage limits, both because it's expensive and also because they just simply do not have the capacity for it. And people have gotten really mad about it. That has not actually been a problem for me. I also talked about this a little bit last.
Oz: Mm.
Carly Rector: last time, is that people are like, oh, Claude's going to die now. And like the RAMP article actually talks about this a little bit too. It's like, can't I keep this up when it's getting worse and charging people more and et cetera. And it's like, okay, but I, again, as a person who's been using it the whole time, it was worse for me, worse for me in like November last year.
Oz: Mm-hmm.
Carly Rector: when it was chewing through tokens and I hit my five hour usage limits like all the time ⁓ and to actually like stop working or pay for a la carte tokens to keep working. And that has not happened to me since then. ⁓ I could look at my usage limits now let's see. My max usage for the week ⁓ is ⁓ 4%.
Oz: Mm-hmm. Mm-hmm.
Carly Rector: And that resets on Sunday morning. So that's basically, I have used 4 % of my limits for the week. And again, I am using it all the time. So it's fine from my perspective. It's not fine if you're like running open claw or you're like having a Ralph Wiggum loop that just tries the same thing a hundred times in a row until it gets it right. Like there are stupid things that you could do.
Oz: Okay. Okay. right. I haven't heard that term of Ralph Wiggum Loop. ⁓ What is... Okay, now I must ask. Tell me of the Ralph Wiggum Loop.
Carly Rector: It's a real popular one now. Ralph Wiggum loop is Ralph Wiggum from Simpsons. ⁓ but basically just like write a thing that keeps saying, did you do the thing? Like Claude will come back with an answer and it will just say, did you solve the problem? Like repeatedly ⁓ So just like keep trying until it gets done. ⁓ so you just ⁓ loop it it gives you the right answer.
Oz: Right? Yeah. I see. So like... So like the miracle sort algorithm, but real. OK.
Carly Rector: Kinda, yeah. So people do that and they think it's really cool and I'm like, that's stupid. ⁓ I mean, like, I guess it does work in the sense that like, if you tell an LLM to retry things enough times, it's more likely to get there, but. is drastically inefficient and doesn't work for many things where like if you have not defined your success criteria correctly or thoroughly, that's just fundamentally not going to work and also chews through tokens. Anyway, my point was that Anthropic has actually been doing a lot of stuff to bring their inference costs in ⁓ both because they just don't have that much capacity. ⁓
Oz: Mm-hmm.
Carly Rector: and just because they want to be able to make money. I think they're similar to... OpenAI is also like this. are actually doing okay on inference costs ⁓ in terms of their...
Oz: Okay.
Carly Rector: getting approximately-ish, probably not fully, but approximately enough money to cover what it costs to actually serve the traffic. ⁓ And of the costs are coming from training, ⁓ because that's a huge chunk of work you have to do up front before you have the model. ⁓ And they have to keep putting more money into that to train the more bigger models.
Oz: Mmm.
Carly Rector: I don't know the actual details. I would assume that Anthropic is in a place where they're at least within like a couple of times of their inference costs via revenue. And like I've talked about before, like again, I'm at 4 % of my usage. So if they cut my usage by ⁓ like to 20 % what it currently is, that would be fine with me ⁓ because I'm not using much. So ⁓ like,
Oz: using 20 % of 20%.
Carly Rector: Yeah, mean that might be a little harder because I have gotten up to like 7 % of my weekly limits before. But like everyone who's like if Anthropic comes out here and like doubles the price or triples the price or cuts the usage limits drastically everyone's gonna be in trouble. It's like well not everyone because I would be fine. So like some people are gonna be in trouble for sure, but everyone who's like this is all completely unsustainable, at least some of it is sustainable, I can sustain it.
Oz: Mm-hmm. So it's a skill issue.
Carly Rector: Skill issue, yes, on the subject of usage limits compute costs ⁓ is ⁓ SpaceX deals that they did, which I think is super interesting. So the specific announcement headline on this Anthropic blog is higher usage limits for Claude and a compute deal with SpaceX. ⁓ because the important part, they think, for their customers is that they are doubling usage limits based on this deal because they just got a whole bunch more capacity, ⁓ capacity ⁓ from the Colossus data center. So the specifics of this deal are Elon Musk made a deal with Anthropic to lease them capacity the giant Colossus data center that he built for XAI, ⁓ which is of SpaceX now because... ⁓
Oz: Hmm.
Carly Rector: for silly reasons, SpaceX has acquired XAI and XAI acquired X. So Twitter and XAI are all under SpaceX now. It's the AI Rockets Twitter company now. And it's going to go public soon as that. So the important part about knowing that XAI is under there is that this is the data center that he built to train and run Grok. so the funny part about
Oz: Okay. All right.
Carly Rector: being like, I'm going to lease all this capacity to Anthropic now, is that implies Grok is not using it. Because Grok is not popular. Which makes sense, but it is interesting that Elon Musk would admit to that. ⁓
Oz: Okay, yeah, no, that scans. ⁓ We're not into the fascism maximizer 9,000.
Carly Rector: ⁓ And he tried to walk it back a little bit. He's like, ⁓ we're still definitely using it and like we'll be training new models soon and we'll build out new capacity for that blah blah blah. the other article I linked here on that topic was ⁓ did XAI just concede the race? Which, because it just like, are you giving up? Because you're handing your chips to Anthropic, which is a competitor. ⁓ So...
Oz: Mm-hmm. Yeah, I saw that one first.
Carly Rector: ⁓ kinda feels a little bit like giving up. ⁓ but, and then the other big reason for that, of course, ⁓ that people called out is the enemy of my enemy is my friend, which is that Elon Musk is in this huge lawsuit with OpenAI and hates them a lot, ⁓ and would like to do anything he can to kill them at this point for petty reasons, like that he wanted to be in charge of OpenAI originally, which came out a while ago as part of this lawsuit. ⁓
Oz: Hmm. ⁓ did he? Yeah, that scans.
Carly Rector: Yes, the quote that I heard in I was listening to today that was insane is like in some email with Sam Altman, I think it Sam Altman who was talking about this, they asked Elon in the hypothetical world where Elon was CEO of OpenAI and he died, who would take it over? And Elon was like, I haven't given it that much thought, but I don't know, maybe my children. ⁓ And everyone's like, yeah, sounds great. Let's make the... AGI future AI machine, hereditary oligarchy. That sounds perfect, yes. ⁓ Anyway, that's fun But anyway, they didn't let him do that and he's mad about it. ⁓ What OpenAI says is reason for this lawsuit. ⁓ Elon says it's because OpenAI a charity and he invested whatever million dollars in them.
Oz: Yeah, great. Dune.
Carly Rector: that they just took as a non-profit and then became a for-profit company, etc. which did happen. But the most interesting part of this lawsuit is not any of the actual details of the lawsuit, but all of the early OpenAI communications that have come out about it. So the importance of our heart being there that Elon Musk is extremely butt-mad that he's not in charge of OpenAI and therefore is willing to work with their competitor to kill them.
Oz: one of the least professional humans to ever be the richest man in existence.
Carly Rector: ⁓ huh. Yep. SpaceX is leasing a whole bunch of their data center capacity to Anthropic. also in this platformer article about did they concede the AI race, they point out that Musk himself has repeatedly called Anthropic a key threat, complaining that, among other things, Claude, quote, hates whites and Asians, especially Chinese, heterosexuals, and men. ⁓ Actual tweet from Elon Musk some number of months ago. ⁓ But he has a more recent tweet that says met with people in Anthropic and ⁓ evil detector didn't go off. He thinks all of them are doing good work and genuinely care about doing the right thing for the future or whatever. ⁓
Oz: ⁓ the father of Mecha Hitler thinks we're doing the right thing for the future. I'm so reassured.
Carly Rector: Hahaha! Yeah, mean a lot of the Musk, early Musk stuff was crazy because some of the quotes that I've seen also is that the reason that he wanted to found OpenAI in the first place is that he thought Demis Hassabis, who's the CEO of DeepMind that was then acquired by Google and is now the Google AI, ⁓ was basically a super villain who didn't about humanity and was gonna be the evil overlord of AI and he needed a counter evil overlord company.
Oz: Okay, okay, they are sure, I guess.
Carly Rector: ⁓ huh. So yeah, that's fun. I saw this deal and I was like, that's interesting for a number of reasons. But if he is conceding that XAI is not a real competitor here, I'm not going to complain. It's like the quote at the of this ⁓ platformer is, In the meantime, the most irresponsible of the leading AI companies is now effectively selling itself off for parts. It may be the best thing Musk has done for humanity in ages. Which is, like, not deliberately. He's still doing it for incredibly petty stupid reasons, but if he's doing it for petty stupid reasons and he manages to kill Grok anyway, that's fine, I guess. Yeah.
Oz: Let's break that clock twice a day.
Carly Rector: Um, but that does pull in another circular funding deal to segue a little bit to talking about circular funding deals, which is just the whole mess of things, which we have talked about previously in this podcast in October. I looked it up. Uh, is this Bloomberg circular finance chart about how everyone in AI AI is giving money to each other. And it's just, it's real, it's just real bad. They have an interactive version on the Bloomberg site where you can go and see all of it. ⁓ and share. So we can explore the deals. For example, Anthropic is.
Oz: Mm-hmm.
Carly Rector: backed by Google and Amazon, it has also announced plans to raise money from and do business with Microsoft and Nvidia. So like the most common pattern that you will see on this chart is a company that provides computing capacity investing in an AI company that will then buy computing capacity back from them, So that's the sort of pink and blue investment hardware lines. the services and investment lines are kind of similar too. ⁓
Oz: Right. Mm-hmm.
Carly Rector: the most directly egregious circularness of it because it is just like I actually even even like two years ago, ⁓ I heard VCs actually complaining about this because what what is bad about this from a VC's perspective is that if Microsoft invests in open AI, they can do it at a drastically inflated valuation because the valuation doesn't actually matter to them. They're just getting their money back. anyway because they're just buying cloud compute from them. Whereas the VCs have to actually make money off the equity of the company and the later exit, etc. So, oh no, all the money is fake! And it's fake in a way that doesn't advantage us this time! What's that about?
Oz: right. ⁓ no, we made all the rules fake and unfair.
Carly Rector: So yeah, the big sort of strategic is what they call it, investors, which is like big companies that invest because it's useful for them and not because they're trying to make money. ⁓ inflate the valuations even more than they would already be inflated, just the hype cycle. So, ⁓ is the complaint of VCs here. And so that's the sort of like the direct circles that are happening here a lot. ⁓ But then there's obviously the ones where it's just like the money goes around various places, because all of these are interlinked with each other. ⁓
Oz: Mm-hmm. Mm-hmm.
Carly Rector: one of the things that I think it's important to understand about this is that this is particularly bad, but I don't think this is unique to the AI space. ⁓ in that ⁓ there's a lot of business to business deals that are fake money and that nevertheless get treated real money by say all the press around it.
Oz: That's true. Yes.
Carly Rector: first of all, and second of all, it becomes real when these companies IPO and go into the real stock market and then they can sell the stock for real money. So, it's... I don't know.
Oz: Mm-hmm. the onanistic hand of the free market.
Carly Rector: Yeah. So I've talked about this before also on this podcast, but I think a really critical bubble moment we will have is when OpenAI tries to go public. Cause they, they keep saying that they're gonna, and there's been some reporting that their CFO said, no, we shouldn't and we're not ready for it. And that Sam Altman was overriding her, which is not, not great. ⁓
Oz: ⁓ yeah, that's right!
Carly Rector: But also there's been a bunch reporting that like, because as we saw on the ramp chart, they're not doing that great lately. Their usage is dropping compared to Anthropic. And their revenue and Anthropics is about equivalent now per latest reporting. They're both at like 25 billion annualized, I think. ⁓ And so people are like, wait, but if they're about the same and Anthropic is growing faster, why is OpenAI like three times the valuation of Anthropic right now? ⁓
Oz: Mm. Mm-hmm.
Carly Rector: The answer is that all numbers are fake and it was all high- ⁓ But that's the short answer. ⁓
Oz: Welcome to capitalism where the rules are made up and the points matter ⁓ so much.
Carly Rector: Yes, ⁓ but it was more hype driven. But like also people are noticing this now, which means like in private secondary markets, there's a bunch of reporting I've seen that like, people are trying to find buyers for their open AI shares at the current valuation and not being able to find them. Whereas everyone is scrambling to trying to get into Anthropic. And Anthropic is maybe raising at a trillion dollar valuation before they go public. I saw a little bit of reporting on that. So
Oz: Mm.
Carly Rector: OpenAI had talked about going public at a trillion dollar valuation or more this year. ⁓ and I don't think that's going to work out for them if they try and do that, because it sounds like people aren't going to buy even at the current price and are trying to get out of it. But it might, like what might happen is that they do it ⁓ and there's
Oz: Mm-hmm.
Carly Rector: consumers who are still waiting to get in that all of the early investors can offload their shares into the market and then it crashes. And that might be the bubble pop, but... Or they could not do it and Anthropic could go first and that could work fine. I don't... Like there's various ways that this could play out, but I think the opening IPO will be a very pivotal moment here.
Oz: Yeah. Yeah.
Carly Rector: notably like there was some reporting that Nvidia did not want to invest that much more in OpenAI. Like they had again said that they would put in like some crazy number like a hundred billion dollars or up to a hundred billion and then they said actually we're not doing a hundred billion we're doing like ten or whatever and everyone was like um oh nvidia losing faith in open ai and they're like no no no we said up to like we didn't promise anything but also um it
Oz: Mm-hmm. right.
Carly Rector: the other comment, I think it was a Jensen Huang comment that was like, this is our last opportunity to invest in OpenAI. ⁓ And it's like, why would that be the case? If they go IPO, you could buy their stock at any time. Or even if they do another raise after this, before they go public, you could invest at any time. ⁓
Oz: Yeah. What do you mean your last chance? This will be our last chance to hand someone money.
Carly Rector: Yeah, and what the implication there is that they think this is the peak, and therefore it would not make any sense to invest after this. Or it only kind of makes sense to do it now even. Which I think is probably true, and I a lot of people are thinking that might be true. But maybe they'll re-hype it up, I never know. So, but yeah, all the numbers are fake. It's very depressing.
Oz: Mm-hmm. ⁓ Man, if only a German philosopher had written about this. ⁓
Oz: Mm-hmm. Mm-hmm.
Carly Rector: Which, again, the first step to solving a problem is acknowledging you have one, etc. people are at least noting that this is not great, including Bloomberg, notably not a socialist bastion usually. ⁓ is biggest one here, obviously. It's the biggest ⁓ circle this chart right now, and also the one with the most links here. And their stock has been booming lately, even. Like, I think they have 4.8 trillion valuation on here, and I think they're coming up on six ⁓ with stock rallying recently, $6 trillion, Because... ⁓
Oz: Mm-hmm.
Carly Rector: Everyone all the investors are being like well We don't know who's gonna win the the AI app race except that we know it's gonna be Nvidia because look at this chart Doesn't matter who wins Nvidia is gonna come out on top we think so So everyone's trying to get into Nvidia and they are notably a public company so everyone can just buy their stock
Oz: Yeah. Yeah. All right. Let's see how that works.
Carly Rector: Yeah, I mean, they're also the what they like to call picks and shovels company of AI, which is like if it is a gold rush analogy, if you've never heard that saying before. It's a startup thing, but it's basically like in the gold rush, you don't want to be the person who's digging for gold. You want to be the person selling picks and shovels. ⁓ then it doesn't matter if anyone actually finds gold. You're doing fine. You got your
Oz: I have not. ⁓
Carly Rector: You got your money, more or less. So, is here, picks and shovels of the AI boom, which is people have to buy their chips. Doesn't matter if they do anything successful with them.
Oz: Boy, what a fun way to live. ⁓ Historically, ⁓ Gold Rush noted as real cool and fun. ⁓
Carly Rector: Mm-hmm. Mm-hmm. Real great for the economy.
Oz: Good for the economy, good for the environment, good for the people involved, good for the people not involved.
Carly Rector: Yeah. So yeah, that's that. And there's some other picks and shovels companies on here, but basically a lot of the hardware chip companies are getting that treatment. ⁓ Or any of the more foundational tools to it, just like the companies doing data labeling are also sort of in there, like that kind of stuff.
Oz: Yeah, I makes sense. Yeah.
Carly Rector: the things where it's just like, is foundational to the whole thing and it doesn't depend on anyone particular successful bet.
Oz: Right, right, right, because everyone's gonna need these things.
Carly Rector: Yeah. And I guess slightly positively, you could use chips for lots of other things too. So some of the what happens when the bubble pops takes that I've seen have been like, but it's not necessarily bad if we have a bunch of chips in data centers, because there's a lot of things that you can compute, ⁓ which it better than some other bubbles. ⁓
Oz: Yeah, yeah. and afford graphics cards again, that'd be great.
Carly Rector: Yeah. Everyone build their cloud gaming companies. They've never been able to make work until now.
Oz: ⁓ yeah, this time I swear, I swear the cloud gaming company, just give me one more shot at it. Making a viable business model.
Carly Rector: What's wrong? So yeah, so our final topic is your topic. Do want to give a little summary?
Oz: Yeah, kind of ⁓ pursuant to what we've been about with the technological problems versus systemic problems there has been a big discrepancy in how various, either laws or, know, of service have been enforced on large versus small creators. And kind of the two, of that I've been seeing the most of is one, ⁓ the way censorship been coming down on smaller creators, mostly driven by these payment processors, know, Visa and Stripe and such, disallowing, you know, adult on storefronts, but not necessarily from large production companies. ⁓ then sort of, ⁓ isn't it funny how copyright law doesn't matter anymore? ⁓ Like, you know, remember ⁓ they were putting grandmas in prison for ⁓ Napster and now we just like, yeah, just kidding. ⁓ But the payment processor obscenity stuff, ⁓ man, now that Kickstarter is also like really cracking down ⁓ ⁓
Carly Rector: Mm-hmm. Mm-hmm.
Oz: you know, adult material. ⁓ kind of depressing. Like I saw a tweet from someone on the Baldur's Gate creative team being like, hey, shouldn't, doesn't our game not meet Steam community guidelines? Kind of interesting that you're tossing off a whole lot of, you know, small and, you know, marginalized queer creators, but not, not Baldur's Gate. Baldur's Gate's not a problem, huh? And, you know, I think, I think,
Carly Rector: Yeah.
Oz: some platforms have tried to be more judicious about it than others. know, HIO has been like quite imperfect, ⁓ but at least are like showing some sympathy and sensitivity for creators. But I think it's really scary, especially when, you know, a lot of legislature is also rolling back protections for minorities and queer folks that like,
Carly Rector: Mm-hmm.
Oz: You see things with any queer themes being called quote unquote adult. If there is just. Any whiff of anything remotely sexual about it, ⁓ things are getting tossed off for stuff way more tame than ⁓ the victorious secret, display that I would walk past every day going to the mall job.
Carly Rector: Yeah, think the Baldur's Gate comment thing I think is really interesting and pointed, which is just like, yeah, you will toss people off your platform if they don't matter to you. And the big ones are fine because you can't afford to lose them. ⁓ Sort of on that topic, I think the most egregious sort of platform moderation case of this. recently was X not getting kicked off the app store when it was full of deepfake porn. Yes. Yeah, like, because it's like, it has more and worse stuff on it than like so many things, so many things. And like Apple is historically like really strict about that stuff. ⁓ And no, no, no comment there because they're not going to pick a fight with Elon Musk. The only reason, so.
Oz: Yeah! Right, what do you mean? Yeah, yeah. Yeah, it's just kind of going mask off. mean, like, it's always been this way, and especially for certain populations, it has always, always been this way. But just how many things go rules for thee, but not for me. That lives were ruined, people were killed over piracy. ⁓
Carly Rector: Mm-hmm.
Oz: And now Disney wanted to invest in Sora, you know, Soraia. I mean, they didn't, like it fell through, but like just the fact that they were even talking about that is insane. like the number one rule in all, you know, that you hear people talk about all the time is don't mess with the mouse. just how draconianly protective Disney has always been of their intellectual property. And still just don't know.
Carly Rector: Yeah. Yes. Yeah. Yeah, like that, when that was announced, that was just so wild to me because, like, Sam Altman thinks that he can get away with everything, but he appears to kind of be able to, for some reason that I don't understand. Like, less so now. Like, he's hit a few losses now and therefore is being very confused and struggling about it. But like, got a few L's for Sam Altman on the board, but like, he was on a real run of W's there and the-
Oz: Mm-hmm. We've got some L's on the board.
Carly Rector: convincing Disney that they wanted to him money for giving him their intellectual property. The ⁓ of the deal there were never completely revealed. And I did see some reporting that it ⁓ going to involve some licensing fee being paid from OpenAI to Disney in addition to Disney just investing in OpenAI. But that was just wild. ⁓ we'll only slightly get into this podcast, but like the Elon Musk open AI trial has been a lot of digging up receipts, like it's internet drama only it's for a a trillion dollar almost company and, the richest guy in the world. Yeah.
Oz: Yeah. Yeah. And just, it just, it just doesn't, it just doesn't matter anymore.
Carly Rector: Yeah, or like it does matter for some people and it's doesn't matter if you're an AI company and you can just be like, but this is the future.
Oz: I feel like more things in human history function like forum drama than we care to admit to ourselves. But this is future.
Carly Rector: And yeah, on the topic of does copyright even matter anymore, I always think about Mark Andreessen's comments to the Copyright Office, like a couple years ago, about like the whole question of can you use copyrighted data for AI training, basically. And his comments were more or less like, it has to be allowed because the business model doesn't work otherwise. And it will kill American innovation if you don't make this be the case. Yeah, I think that's definitely true and unfortunate. And I think more people should have gone through forum drama in their formative years to have a better idea of how to handle it.
Oz: Hahaha! Get on Reddit, start some flame wars, pay for it. Anyways, it's like...
Carly Rector: Sincerely, like...
Oz: We should be having high schoolers signing up for something awful. My business model doesn't work if everyone I make eye contact doesn't hand me a crisp Benji.
Carly Rector: intro freshman year course, have a flame war and come out of it and then try and make friends with everyone in your class afterwards, I guess. ⁓ Yes, exactly. ⁓ But because he's Marc Andreessen and he's the head of the biggest venture capital firm in the country, he could just be like, but the businesses won't work if this doesn't happen. And people listen to him for some reason, it is like, okay, but like, there are many businesses run by creators, or even just like small studios are journalists or like
Oz: Yes, I like that. mod a large unruly community or a small insane community.
Carly Rector: Yeah, just like organize any group just at all I guess because like again the thing is most of these guys have done ⁓ nothing nothing even remotely like that they don't even like plan ⁓ outings like ⁓ their their wives or secretaries that ⁓ so they just ⁓ many many businesses that don't work if you can just take their intellectual property for nothing that's also a business that doesn't work so but we see which one matters and it's the one that can donate hundreds of of dollars to a political campaign so
Oz: They don't plan outings. What?
Carly Rector: Mm-hmm.
Oz: they united those citizens, didn't they?
Carly Rector: Yep, yeah, it's bad. So yes, I think it should be mandatory you have to organize a social group before you're allowed to make any consequential decisions for the world economy. ⁓ But I think that's hard to enforce. Yeah. Yeah.
Oz: ⁓ Carly, I think it might be broken and I think it might be broken fast. You ran ahead and how are you enforcing this? Oh, oh, thinking, yeah, this is a new department and went under the attorney general to like enforce LARP participation among venture capitalists. Mm-hmm.
Carly Rector: ⁓ theoretically there are more things to say about that, Yeah, definitely send them all out on LARP weekend first. I mean. The whole issue of payment processors dictating content standards, I think we have talked about the podcast before, but it's just like such a wild artifact of how we have ended up in this situation.
Oz: Right, right. What do you mean, Visa gets to decide whether or not I get to look at boobies? that's not Visa's business.
Carly Rector: Yes. Yeah, for people who maybe as familiar, this is a common pattern among platforms where they will be like, we as a platform need to impose more content standards, especially on like pornographic or ⁓ Hello and welcome to the Fast and Broken podcast where we take a critical look at the fast moving world of tech and online culture. I'm Carly Rector, a software engineer and startup founder who has been using Claude code since its release and cursor for even longer than that. And I've said that before, we're going to get into it again, sexual for it's some by some usually unclear definition content because our payment processors allow it to be for Adult content and payment processors meaning Stripe in the case of Kickstarter But Stripe is doing it because they are on top of like credit companies and things like that specifically ⁓
Oz: Right.
Carly Rector: ⁓
Oz: ⁓ Well, I'm Oz Brown. I am a ⁓ support enthusiast who has somehow, despite my efforts, also become a software engineer. And I have been your software engineer.
Carly Rector: you have, very exciting. ⁓ Oz is a friend and mentee question mark and past employee when I was running my startup that I founded, per my intro. So, yeah, which for anyone who doesn't have the context shut down like two years ago now a little over two years ago now. So Oz is back at Amazon ish.
Oz: Yes.
Carly Rector: And so because the credit companies don't want their money being used for porn, they just get to dictate the content standards for any website in the world. ⁓
Oz: Yeah, I am.
Carly Rector: I am doing other stuff, but we've got some background. In to like, being friends from the internet, ⁓
Oz: So I was reading a little more on this, months ago, believe it was ⁓ Denise Dreamwith was talking about this, that basically, you have payment providers that handle high risk and low risk transactions. And so the ones we're mostly used to, Stripe and everyone that Stripes sitting on top of, That's, that's fair. That's fair. And I've also been using Claude Code for longer than some others, ⁓ due to, your encouragement.
Carly Rector: Yeah, which we can talk about a little bit. So the topics we're going to go over today, we're to talk a little bit about AI coding tools, as alluded to by that intro. how Code is doing, our software engineers rotting their brains, some other developments in the world of ⁓ everyone is Claude Code now, or maybe they're stopping for various reasons. ⁓ Then we're to talk a little bit about Cool.
Oz: they're expecting to process a certain volume of fraudulent activity. whether this is based data or this is just kind of how the industry is, like quote unquote adult entertainment has always been considered like high risk. ⁓
Carly Rector: the ongoing saga of circular finance deals in AI. in particular, I do want to talk a little bit about the anthropic SpaceX deal, because I think that one is interesting for its own ways in addition to the fun circular finance nature of it.
Oz: Mmm.
Carly Rector: Yeah, this is starting to sound familiar. I probably also read this post at some point. And I do think there's some data around that. I remember hearing anecdotal stuff from even just back in the day when you had to subscribe to cable channels and things like that. Dudes coming in and being like, ⁓ no, I didn't subscribe to porn.
Oz: Yeah, you probably did. ⁓ I'm very excited to your take on that one. ⁓ Yeah, yeah.
Carly Rector: Yes. And then we're talk a little bit about ⁓ Kickstarter's of clarified, clarified content policies and content and copyright policies online in general and the uneven enforcement of them in some cases and how that sort of affects the online creator and content ecosystem, Channel or whatever who charged that to my credit card. No, I didn't I didn't do that No, so like I think there's there's definitely more of that in that area ⁓
Oz: Who charged that to my hotel room? Yeah, yeah. But just the problem is like what's getting defined as that is getting tighter and tighter. And, you know, I mean, I'm a bit more liberal in, you know, what I...
Carly Rector: Let's jump into AI coding tools. I think the first thing I want to do is actually do a little bit of definitions here because I feel like people who don't use them don't really know even like exactly what AI coding tools are, or I will particularly like narrow this to agentic coding tools, which agentic is a word that can mean
Oz: you know, I support sex workers, I support sex work. I'll just say that on the podcast with my professional name attached. ⁓ That, you know, I think that stuff should be recognized and protected and it's, you know, not necessarily a payment provider's business. you know, what within the law a person wants to do with their money. And also there's kind of the double problem of like what...
Carly Rector: I'm just... Mm-hmm. whatever people want it to be, perhaps as much or even more so than AI does. But the way Anthropic usually talks about something being agentic is it's an LLM using tools in a loop.
Oz: gets defined as acceptable and unacceptable eroticism, which is ⁓ obviously way outside the scope of this podcast.
Carly Rector: So what that means is it's not that you ask it to generate something and it just gives you some text and it's done. It's it generates something and part of that something is deciding to take some action and then it can get the results of that action back and take more actions such that it can do multiple things and just sort of proceed and do a whole bunch of work on its own without you having to prompt it every single time. ⁓ So in the case of agentic coding tools that might mean you ask it to do something and it not only like generates code but it also actually edits the Yes. But yeah, we've definitely talked about supporting sex workers on this podcast before. And in fact, it reminds me of another related topic that I think we talked about a little bit before, which is ⁓ open AI doing erotic chat bots. of speaking of double standards, which they have they have put on pause for now. ⁓ So that was the whole thing because
Oz: ooo Mm-hmm.
Carly Rector: First of all, Sam Altman said like two years ago in an interview that they wouldn't do erotic chatbots because that seemed like a hack to juice user engagement or whatever that they wouldn't need to do except as a last resort or something like that. ⁓ And then they announced that they were gonna do that in like December or something like that. So everyone's like, you're at your last resort already, huh buddy? ⁓ But also one of those things where it's just like, okay, like so. file and puts the code in there, and then runs some ID linter checks on it to make sure that it's syntactically correct. And then it runs some tests. And then it finds some errors. And it goes back and fixes the code. And then finally, it comes back to you and says, OK, I'm done. ⁓
Oz: I accept it's a lot, all right. My favorite part is when it clarifies requirements. So I'm like, yeah, you're right, Claude Code. That is actually what I meant, not the thing I asked you.
Carly Rector: Yes. Sam Altman, OpenAI gets to make erotic content because again, they're not real sex workers. So you get to build the content scraping company and then use everyone's work and generate that and you get paid for it, but the people who actually do it, no, they're terrible, dirty people who cannot use Visa cards or whatever. That's cool. So like... Yes,
Oz: Yep. Yep. Yeah, yeah, amazing. Right, right. That is a great point of like, well, what what are you trading your erotic bots on?
Carly Rector: one way to think about that is like the LLM ⁓ a bunch of knowledge about how to do things like how like the actual syntax of code and like knowing a bunch of design patterns and libraries. but what it doesn't ⁓ know how to do ⁓ is exactly how to take actions or like Mm-hmm. I mean, for sure, a lot of the existing chatbots are trained on Archive of Our Own. That is explicit. They have said as much, even in some of their official publications. consistent workflows. So if you imagine, even if you had some really smart person working on something, you generally don't just, like, give them a task. like, go, without giving them any more information than that. Or, some to work within. Like, ⁓ an SOP or, like, guidance about, like, we usually do it this way, like, following these steps. Like, you write the code, then you run the tests. Or... ⁓ we like these kinds of standards, et cetera, or like, if you don't understand, you should clarify. And so basically what Claude Code, the product tool is, building all of that kind of like common sense workflow stuff into a wrapper around the LLM that makes it just work better. ⁓ So for example, they added fairly late into Claude Code's ⁓ actually, think maybe five or six months ago. ⁓ the Ask User Question tool, which specifically lets it like... give you a structured set of questions where it's like pick A, B, or C. I don't know if you've even seen it do this at all. Because it only, yeah, because it didn't used to do that. Like it would ask you questions sometimes, but it would just like give you a list of questions, free text, you would have to type them back. Whereas now it will just like give you structured questions with multiple choice responses, and you can click them and go next question, etc. Making it really easy for it to be like, okay, you gave me some instructions, but it was unclear in like these different ways. And
Oz: ⁓ yeah, I hope so. you're barista, a lonely single barista, or ⁓ you're botanist, or you ⁓ run flower shop new in town and you've broken your leg ⁓ and there's one bed. ⁓
Carly Rector: Mm-hmm. Yes. Yup. Yes.
Oz: Martha Wells visionary, she knew that the robots were gonna be watching pulp garbage to learn to be people.
Carly Rector: these are the questions that I want to clarify. And like, here's what I think the recommended thing is basically like allowing you to clarify in a more structured way. ⁓ Yeah, the one thing I tried to put in here, but it was paywalled for me, was a Forbes article that was released on Janitor.ai, which is a AI erotic roleplay platform. And it is run by three dudes. this article was like the three dudes who are in charge of like the erotic roleplay website that so many women use, et cetera. So there's, to be on topic here, there's a lot of aspects here of
Oz: You know, as you're saying that, what it kind of reminds me of is, made a joke at the beginning about customer support enthusiasts, but I did get ⁓ my start doing customer support work and in particular ⁓ support over forums, over the Xbox forums. And since a lot of times we'd only have like one customer a day, ⁓ Mm. Okay.
Carly Rector: appropriating some things and then monetizing them because if you're a cool guy running a company with AI in the name then it's fine and you are a woman who is actually just doing the thing then it's evil yeah
Oz: per issue, you'd really want to give them a good answer so that it's not like ask them a one line question and then come back another line. And so we talked a lot about how do you format something to the customer where they will give you back the maximum useful information? Because of course, anyone who's ever written an email or a message with a bunch of different questions in it, you do not get all of them answered. ⁓ we a lot of time discussing how do you maximize the odds of getting the important.
Carly Rector: Okay.
Oz: Right. that. ⁓ boy.
Carly Rector: Mm-hmm. You do not.
Oz: information back to move forward on an issue. it's kind of interesting that ⁓ that's also kind of being built into these ⁓ agentic systems.
Carly Rector: Yeah. But yeah, I think there's a lot of really concerning things that follow that pattern of like, the deciding factor is not the content, but the entity or person that is presenting it, and therefore it is really a judgment on that person and not on the content.
Oz: Yeah, yeah.
Carly Rector: to sort of segue to our first bullet point here, which is like ⁓ the Claude code leak the Claude code harness itself is a separate thing from Claude the LLM ⁓ And there's a lot that is in the harness actually because it turns out that all that kind of stuff is what makes it work really good. And there is something like 500,000 lines of code and just like the harness for Claude code, which is probably I think so. Let me let me make sure that's right.
Oz: 500k in the harness?
Carly Rector: 500,000 lines of code. Which I think is bloated, probably, because they are having Claude write all of their code for Claude code. But there's just a lot of stuff in there like prompts and tool use guidelines and just a bunch of things in there that are
Oz: Yeah! ⁓ Mm-hmm. Mm-hmm.
Carly Rector: best practices for how to work with code, basically. ⁓
Oz: I mean, I think one frustrating thing about being an engineer is that people don't want to believe that tech requires engineering. You know what I mean? Like they really do want to think that we're wizards and like, it's almost like, you know, my wife's an artist. And so people are like, ⁓ they, know, they think that like drawing is magical. And like, it's like, no, it's a lot of thought and skill. like, so people, you know, think that like tech.
Carly Rector: Yes.
Oz: is magic and like it's not like no it's all just thinking.
Carly Rector: Yeah, it's like for a long time, I have felt that it's difficult to explain to people what a software engineer is and what they do. Like kind of both in the non-tech world and just like to non-engineer tech people even. ⁓ Because like non-tech people just don't understand at all.
Oz: Mm-hmm.
Carly Rector: like people just don't have the concept of like what it is involved. I think they do a little bit more now in terms of like you write code and stuff, but even still, think will imagine like HTML, and not like the amount of logical putting together of systems that is the actual engineering part of it. And how little that has to do with code in a lot of ways. And. Like I took an algorithms class in school where we didn't write code. Most of it we didn't even do on a computer. We did it on paper because it was just like doing algorithmic proofs, Cause it wasn't a matter of just like how, how did the computer runs it. It was more a conceptual. this one is especially depressing to me because ⁓ there's so much like protect the children bullshit around it and it's just people who have done much more research and are much more eloquent than me on the topic have talked about why protect the children is
Oz: I know. it is never about projecting children. It is.
Carly Rector: ⁓ But that like there are definitely a certain set of conservatives who, for example, don't just want to stop children from seeing porn. They want to stop everyone from seeing porn and that they will do that just by making it really inconvenient for everyone under the guise of protecting the children, for example. like that's. thing. And as people are figuring out that just because AI can write code doesn't mean that it can engineer, people are having to sort of discover that more. ⁓ And so I think that a lot of Claude Code and a lot of similar people have to build agentic harnesses for things is them figuring out that just having the raw capability doesn't do it for you. ⁓ You need to actually do some systems engineering there. ⁓
Oz: Yeah. Yes.
Carly Rector: a real motivation that some of them have said out loud, literally, but yet there are still a lot of Democrats who are like, well, I'm not going to be the person who says don't protect the children, so I'm on board this too. And it would be really cool if anyone had any amount of spine, but, you know.
Oz: one thing I kind ⁓ like about vibe coding as a concept, like I'm not as against vibe coding as lot of people, at the same time that ⁓ people are finding out that a computer can't do engineering, I think there are other people ⁓ are finding out that they can. Yeah, for real. Well, it's fun to have dreams.
Carly Rector: Mm-hmm. Yes, sure. And I'm actually quite pro vibe coding ⁓ to many people. Like, not only because like for my entire life, basically, but definitely my entire adult life, I have felt that more people could be software engineers than are. Yeah.
Oz: Hey And I think in conclusion, ⁓ all technology was a mistake. Rocks weren't meant to think. And we have done the devil's work.
Carly Rector: That's it. Yeah. ⁓
Oz: ⁓ yeah. Bring back carrier pigeons.
Carly Rector: Like especially women, obviously, but just like in general, because it's framed as like you're sitting at the computer and you're just like typing basically. You're like the hacker or whatever, and you're putting in too many semi-colons. It's like, no, that's not, that's not the fun part though. Like the fun part is like you're building a thing that does something and you're figuring out how to solve problems. And like that happens to incidentally involve code because that's how you like tell the computer how to do it. But that's not necessarily the most important part of it. ⁓ I don't think all technology was a mistake. Some of the internet is pretty cool, but...
Oz: Yeah.
Carly Rector: and Vibe coding allows people who don't do the code to do some of that other part of like having the fun part of actually making things. ⁓
Oz: Yes, yeah, and I think, know, vibe coding isn't a bad way to learn coding if you're going back and studying it also, you know, like...
Carly Rector: Mm-hmm. Yeah, and like I think there's definitely some pitfalls of it and a lot of them are like people getting too over excited about it and being like, yeah, I could just launch an app, I don't need software engineers anymore or like let's just vibe code everything now and you don't have to take a step back and be like, is this a problem that we actually needed to solve it all like let alone with a vibe coded app? So like, but like which I have some empathy for because it's like people are getting suddenly unfathomable power. Like they just get really excited about it. It's like that's fair, but still. And like there's some risks of obviously security things. You don't want to use it for things where it's important. But I also am not as a big stickler on that for a lot of things. Cause if you're just building an app for yourself that runs locally, it does, it genuinely does not matter. Vibe code your heart out, like, et cetera. ⁓ But, and then yeah, like.
Oz: Yeah, yeah.
Carly Rector: really concerning because it's like it's not like tech was great a decade ago, or even like 15 years ago, I do think it has just gotten way worse ⁓ it'll Yeah, and just like people when people run into issues with it, they do have to figure out like, ⁓ I actually like need to learn some basic engineering concepts here more or less to figure out how to go past like the limits that they hit there, which I think is really interesting as well. So.
Oz: Yeah, it's gone mask off. Yeah, yeah, no, I agree. just like demystification of it all, you ⁓ You can be a software engineer and Claude can help you, or ⁓ maybe you should just learn software engineering. can do that too.
Carly Rector: in a lot of ways, like again, you can just be like, ⁓ is XAI posting a bunch of deep fake porn on Twitter? Is that against my terms of service? Yeah. Am I going to do anything about it? No. Is anyone going do anything about it? No. Because there's nothing anyone can do about it. Like, okay, ⁓ like just where we live now. ⁓ And depressing. It's bad. ⁓ Yeah.
Oz: No. Yeah. Yeah.
Carly Rector: Yeah. But yes, I think the biggest thing there is like. Like, mean, part of it is, again, that like I'm a very senior software engineer and I have gotten to the point where like I work at very large scale systems. I'd like, if you're say a principal engineer at Amazon. A lot of things that they would say about principal engineers at Amazon is that often some of them just did not write code for months or occasionally years at a time. Cause like that's not the most efficient use of your time at that level. ⁓ other people can write code. Other people cannot think at what the level of a hundred person organization is basically. ⁓
Oz: Other people can write code. Not everyone can go into a half hour meeting with that person who wrote code and was like, yeah, here's how this is going to break something you've never heard of.
Carly Rector: Yes, or like plan the six month roadmap for like the entire organization, technically based on like an understanding of what the goals of the business are or whatever. And like that kind of stuff. So that is sort of like the angle that I come at it from, which is like, I have not felt like code was the point in quite a while more or less. ⁓ So like my, my experience with Claude code has been like,
Oz: I do have a hard time when people who have assured me that I'm a rube ⁓ notice what I noticed 10 years ⁓ and are like, damn, that's crazy. I'm like, ⁓ is it, are we going to like reconsider everything else you've been wrong about and I've been right about? No, we're not going to do that.
Carly Rector: this is amazing and I'm not sure I could go back to not using it ever because like the gap between what like I could come up with something that in my head that I know exactly what it is and that I could make it it happen has just narrowed so drastically because that was so big in the first place. And I think like that that gap depends on your feelings about it because it's like for me I could be like, ⁓ I figured out exactly how the system should work say at Amazon and then Yeah. Between that and actually getting the system launched involved spending another couple of weeks writing up a design document, logically breaking it out into pieces that other people than me could do and flesh out further ⁓ in further depth that even more people could do because it's just like there was just too much work there that I was not gonna do all of it. It was a team's worth of work and therefore I had to like... define it and break it out into the like such that a whole team of engineers could do individual pieces of it at the level that they were able to do and work on that over the course of like five sprints or whatever. And then it got launched and that's just a lot of work. And now I can just be like, Claude, I vaguely described this spec to you. And then it's like, great, I created a plan and you want me to do it? And I'm like, yes, and then it does it. ⁓ So like.
Oz: Mm-hmm.
Carly Rector: the level of speed, like people talk about, is this like a 10 % improvement? Is this a 10x improvement? Is this a 100x improvement? Like genuinely for some of the stuff that I'm doing, I do think it's a 100x improvement because otherwise it would have taken months and I can do it in a day. Which is crazy. And that's why every time everyone's like, well, I don't know if coding tools are actually that, that much improvement. like, no, you don't understand. ⁓
Oz: Yeah, that's crazy.
Carly Rector: ⁓
Oz: just wish there was some humility in tech, you No. ⁓
Carly Rector: No. I mean, yes, I also wish that, but I guess I have stopped having any hope for that to be the case.
Oz: And think that's, you know, I think that's where a lot of, but I don't even know what the word I'm looking for is, but like, kind of like the tension between there's some stuff that these tools are really, really good at. And then there's everything else that is being forced to adapt to it that like, it's like, well, no, it's, you know, it's bad there. You know, it's not good at doing that thing or it's extremely energy intensive or it only works because it's, you know, Denethor, yes I do wish that.
Carly Rector: Yes.
Oz: trained on a bunch of stolen artwork. ⁓ know, again, like, wife is an artist, artist. ⁓ if ⁓ she was a digital artist, ⁓ we might have some more contentious feelings about, ⁓ you know, AI use the home. But luckily, ⁓ computers can't stab you yet.
Carly Rector: Mm-hmm. yet. I mean, they can probably, but not very well.
Oz: Actually, I guess that's sort of radiation therapy as a computer stabbing you.
Carly Rector: Yeah, was going to say there's definitely medical robots for that kind of, not tattooing obviously, but.
Oz: Not tattooing. Yet.
Carly Rector: Yeah, only because there's not that much money in tattooing probably, but ⁓ yeah, but yeah, like there's, the theme of this podcast is that there's a lot of tension here and that I think there are some good ways to use AI and bad ways to use AI and we should talk about both of those things instead of being either on the hype train or the throw it all in the trash forever train ⁓ because.
Oz: I am on the technically impressed and practically concerned train.
Carly Rector: Yeah, like, the other part of it is like, the, could do something in a day that I could do in months is like cool, first of all, and fun because I can get things done, but also as someone who likes to think that I'm working on things that are valuable, like that feels really valuable because that means you can do so many more valuable things with that. So, and like,
Oz: Mmm. Yeah, yeah. I mean, Carly, if the people, there were more people like you doing things and less people unlike you doing things.
Carly Rector: Yeah, and well, so the thing is, I think that's the problem with a lot of the, at least like the scientist side of the AI hype thing is like scientists and engineers are all people who work on stuff where they're like, this is all really valuable, but it's really hard. And if I could just do more of it, it would be better. And so, and often they are very narrowly focused on that and don't imagine that there are other fields in which that is not the case. ⁓
Oz: Mm-hmm. Right, right. And you know, and and and I industry loves to downplay its its harmful effects, especially within itself. you know how hard it is to get someone to see the flaws in the things that puts food on their table, you know.
Carly Rector: Mm-hmm. Yeah.
Oz: You know, my, my, the irony of my dad, who's extremely concerned with global warming, but like, you know, I grew up when he was working for oil companies and like thought everything I said about environmentalism was like, you know, weird hippie stuff. And now he has ⁓ a ton of climate anxiety and he's right.
Carly Rector: Yeah. Yep.
Oz: And so it can be hard when we're engineers and we're in tech and you're getting inundated with these constant messages of all these ways that it's improving things and innovating and yada, yada. And it's like, well, yes, it is doing that, but it's enabling scammers ⁓ and it's allowing unscrupulous companies to
Carly Rector: Uh-huh. The amount of ⁓ discovering that tech people are doing of stuff where it's just like, this is literally the foundations of the humanities. you had not been dismissing the entire value of a humanities education your entire life, you would have known this. you're not the first person just discovering this now.
Oz: lie about where their product's coming from ⁓ ⁓ causing a ton of environmental damage and it's really annoying on Google searches. ⁓
Carly Rector: Yep.
Oz: And so it's like, yeah, know, Claude code does make a lot of things a lot easier. And also I understand ⁓ why I have a lot of friends who are really angry.
Carly Rector: There's a lot of that stuff, but at least they're discovering it now, and we can have these conversations a little bit more.
Oz: Hubris!
Carly Rector: Yes, and like I also understand both of those things, which it makes it hard, but
Oz: I wish, my wish is that more times when someone discovers or newly learns to value like a large branch of knowledge or skill that they previously hadn't, that like their first instinct wasn't like, I am so much smarter now. Like their first instinct was like, ⁓ shit, ⁓ was something before. Maybe I need to ⁓ reevaluate the.
Carly Rector: I hope the things are separable to some extent, I guess, is the idea. don't think AI is to die forever, partially because it is actually very useful for things and is proving so in real ways. yeah, so given that... ⁓ Yeah.
Oz: Well, the technology's not gonna go away. the value of my contribution here, no, it's always now I'm extra right.
Carly Rector: I would like us to be having a conversation about like, okay, what if we didn't do some of the bad things quite as much ⁓ and not the like, it should just all die forever because it's not gonna die forever. So like that's the theme of the podcast, but more or less. ⁓ Yeah. Yes. On that topic, I don't think I've talked about this one on the podcast. I've ranted about it on LinkedIn a little bit. The whole thing about AI people talking about how taste is the differentiator now.
Oz: Well, I think, think,
Carly Rector: Like, anyone can build a thing, but having good taste about what the right thing is to build and the aesthetics of it and the interaction of it is important. And I read one whole long blog post about, like, okay, how do you develop taste? What does taste mean talking about looking at really good music producers ⁓ and art studies things like that. I'm like, yeah, because this is a whole thing. ⁓ Guys, like...
Oz: systemic problems aren't solved by individual actions. And I think that's true on a personal level. And it's also true on a technological level, right? Like while we have a system that is rewarding a race to the bottom and, capital as the be all end all success metric, whatever the technology is, it's going to be used in a deleterious way because ⁓ the people who ⁓ control ⁓ Okay.
Carly Rector: Mm-hmm.
Oz: Oh word? human creativity matters and a breadth of knowledge and experience? That's crazy.
Carly Rector: Yes.
Oz: have absolutely no reason to be responsible with it.
Carly Rector: Yeah, and that goes to something else that I can maybe like bring us a little bit more back on the list of things that we were supposed to talk about on this list. But like, which is like a lot of the problems that AI is causing are like not new problems. They're just drastically accelerated problems. And like, if you got rid of AI, they would still be there, but less so. like, maybe that's still that's
Oz: ⁓ yeah, yeah, yeah, yeah. Correct.
Carly Rector: But, that's also an AI-positive thing, is that people do have to start caring about that stuff A little bit more. still fine and better and it's like brings it to a less so level that's acceptable because it was before. But like maybe we should just figure out how to solve the underlying problem to some extent. And like the more the most like simple version of this one is with one of the things that came up in this 404 media article about software engineers rotting their brains, ⁓ which is the problem of like AI generates more code than humans can review, which is one of the one of the big things that people talk about. And my response is like,
Oz: Please, prayer circle, prayer hands.
Carly Rector: Yeah, like there's some genuinely good potential impacts of how AI could impact things. There's also many, many bad ones, but like there are some potential good ones that I do already see in some ways, and some of them are people understanding what engineering is, people understanding the value of creativity and ⁓ human interactions.
Oz: Yeah.
Carly Rector: Yes, but also like humans reviewing code has never been the best way to make sure that code is good. ⁓ Yeah. So, and I did when I was at Amazon do both like the actual reviewing and the trainer for other reviews of people who basically reviewed, ⁓
Oz: I mean, let's put a pin in people value and creativity. Everyone writes more code than I can review.
Carly Rector: Yes, I mean they're still working on that one, but like again, like seeing some people start to be like, ⁓ huh, there's something here, isn't there? It's like, yeah, there is. ⁓ Which again, would have been nice if they noticed that or if they acknowledged now that it has existed for a long time and they've just been ignoring it. But I will take acknowledge it over not acknowledge it.
Oz: Mm-hmm.
Carly Rector: technical assessments for hiring. People went through an automated coding assessment. The automated system graded whether you passed or failed. And for some gray area where it couldn't decide It kicked it to human reviewers to go in and look at the code and decide if you passed or not. ⁓ So we were reviewing the code, and a lot of that was done ⁓ originally.
Oz: ⁓
Carly Rector: Thank you for listening the Fast and Broken podcast. I'm Carly Rector, and if you'd like to support my work, you can subscribe to my newsletter at carlyrector.substack.com. ⁓ it was done just like on paper. The recruiters would come and be like, here's a printed out assessment. Please read this code that this person submitted. And I had to like explain to these HR people that like, I'm not a computer. Like I cannot read the code on this paper and even tell you if it compiles or not. Like that's not how that works. And like I can generally like read it and try and get the gist of what they were going for. But like if you want to like test it, you should run the tests by a computer because that's
Oz: I'm Oz Brown. If you ⁓ want say hi Oz, feel free to do so on LinkedIn. You can find me under Olivia quote unquote Oz Brown ⁓ or you can check out America's Next Top Model for the original Wii console where I play about half the supporting cast.
Carly Rector: Yeah, so it's gonna be real easy to check out, I bet.
Oz: ⁓ just head down the nearest. Yeah, the nearest five dollar bin in a Walmart 12 years ago.
Carly Rector: That's gonna be way better.
Oz: They wanted you to like take a red pen, like circle the missing semicolons.
Carly Rector: Game stop. ⁓ I mean, not specifically, but like in general, like grade this code on whether it was good or not. ⁓ and so, and I had to like be the guidance and be like, okay, you're not grading syntactic correctness. First of all, because that's a stupid thing to grade. But second of all, because you physically cannot grade that as a human looking at code on paper. I'm sure there are some humans could do that, but also that's just like, that's not a productive thing to do. ⁓ Yes, and if you enjoyed this podcast leave us a like or a review it lets us know people are listening and helps more people find the podcast. See you next week
Oz: Okay. That's a party trick, not a tech skill.
Carly Rector: Yeah. And on every team that I ever been on. People have been waiting for like weeks for code reviews because people have just not gotten around to it. And then have gotten a code review that was just like, looks fine or you go into the code review and then you're like, this is so fundamentally wrong that you shouldn't even have written any of this code. Like I should have stopped you beforehand. And like, those are the two modes where like you get ⁓
Oz: Mm-mm-mm. I'm so sorry. Come see me after class.
Carly Rector: Yes, like those are really the two modes of human code review. And like, or like I have picked a bunch of stuff that is pointless. That also happens. Like, so like none of those cases are good, like are helpful really. Like there's a very, very narrow slice of things where like a human looking at the code is better than say doing a design review before you write anything or having tests that you run to make sure that it gets the right answers or.
Oz: Mm-hmm. Mmm.
Carly Rector: Having static analysis tools that make sure that all of your formatting follows certain standards and that there aren't any security issues that you can catch via static analysis tools. All of that's going to work way better than a human looking at the code. So that's one of the things where it's like, yes, yeah, you could say that AI generating more code than humans can review is a problem. Or you could say that we should not have been counting on humans to review code as our measure of quality anyway, and now we just gotta figure it out because there's too much ⁓ there, which is what I would say. So like, that's one example. And so that's one of the things that came from this software engineers say AI is rotting their brains article, ⁓ which is basically a bunch of quotes from software engineers, mostly at large companies about how they are using AI base coding tools, either because they wanted to try it out or because they're being forced to and how... it's not great in various ways. like many of them are real in like, yes, if you're forced to like review tens of thousands of lines of AI generated code, that's bad. the thing that I said in my comment to this is like, it's exciting to me to use Claude code because if you imagine before I was spending like 10 % of my time doing like the fun high level planning strategic stuff and then 90 % of my time writing code. Now I don't have to do that 90 % mostly and I get to do a lot more of the fun stuff and a lot less of the code. But if you're in a job where, for example, you're not allowed to do any of the strategic stuff or like that's not your role, all that you've gotten is like most of your work taken away. And the only thing I get to do like a tiny bit left of whatever the grunt work is left that is not that and not any more fun stuff. So that sounds bad. Like I understand all that. I think those are a lot of real.
Oz: Mm. Right, right.
Carly Rector: real issues, especially with how it's being implemented in big companies. ⁓ But also, I don't think the conclusion of this article is kind of just like it's fundamentally bad for software engineers, and I don't know that that's true. I think it's just...
Oz: I mean, here's the thing. just, I just have better things to keep in my mind than Java syntax. Like, like anything like. And anything.
Carly Rector: Yes. Yeah, I saw one, one take that someone had that I thought was really pointed is that like there are different kinds of software engineers. There are software engineers for whom the code was the point. Like they enjoyed just like putting together the code and like writing cool, elegant function or like, et cetera. And those people are really upset that they're not able to do it that anymore because of the AI coding tools. Or there are people who really didn't, care less about the actual writing of the code part and are there for like
Oz: Mm-hmm.
Carly Rector: solving of problems and being able to just like put together something that does a thing and therefore are quite happy to have that part taken away from them because they didn't really care about that in the first place, which I think is true to some extent.
Oz: I feel a little of both of those, like... I guess, I guess, yeah, professionally, I prefer the high level thinking stuff, but like I do, there's something a little bit soothing about just busting something out in Python. Like what, you know, just like, just like, I just, I just need a little, little JSON parser, you know, or like,
Carly Rector: Mm-hmm. Yeah.
Oz: a little thing that ⁓ forum posts into RenPy code so that ⁓ write visual novels more easily. That was fun. ⁓
Carly Rector: Mm-hmm. Yeah. Yeah, and maybe it is just that I've done it enough that that doesn't have any novelty for me at this point. ⁓ But who knows? So like, yeah, I think there's definitely a range of reactions to it. ⁓
Oz: Yeah, yeah.
Carly Rector: So that was more or less my takeaway of this article is like, I don't think it's rotting my brain. I think it's making me exercise my brain more than I was before because I don't have to do any of the stupid grunt work stuff that I don't really want to be doing anyway. ⁓ But I also see how that is not the case for many people. ⁓ And I just think it's harder to identify exactly what the cause is there than they are necessarily implying.
Oz: I think that over the next five to 10 years, unless bubble pops catastrophically, and who knows, obviously it's going to really, really change what engineering looks like, especially in big tech. And I think it could change what coding looks like in big tech in a lot of different ways. And I hope...
Carly Rector: Yep. Mm-hmm.
Oz: I hope that it happens in a way. where there's more appreciation for, like we're talking about, like the high level stuff and the thought that goes into putting together systems and understanding how things are going to interact with each other at scale. And that the demystification of the code itself lends to a greater appreciation for engineering. And I know that's...
Carly Rector: you Mm-hmm.
Oz: Probably a little optimistic and idealistic, but it is fun to dream.
Carly Rector: Yeah, I hope that I can see some of that happening like already to like, again, I'm not in a big company. like, caveat with the fact that I think big companies are way worse at this, especially because you have to still do like stupid sprint processes and things like that. And also there's a lot of companies where even worse than Amazon about not letting engineers think about anything about the within the code that they're writing, making it hard. And I'm working with startups, but
Oz: You
Carly Rector: I'm definitely seeing people already who have picked up the vibe coding and they have run into walls and then they have come to me and said, Carly, what's happening here? And then I will explain something that I know off the top of my head because I'm a good software engineer. I'm like, ⁓ yeah, that pattern is problematic for blah blah blah reasons. You should probably do it this way. And you can see the realization on their face. It's like, ⁓ there's stuff here. So.
Oz: Yeah, yeah. I mean, yeah, you just, you... It's such a black box for a lot of people and I just don't think that that serves the people inside or outside of the box.
Carly Rector: Yeah, like I've always, always been a big advocate that there should be much less of a wall between product and an engineering or whatever you want to call it. Like the, where the product people are like, no, I could not possibly understand any of the technical details and, and the technical people are like, I don't care about the product. just get requirements thrown over the wall to me. Like that has always seemed like a terrible way to operate for me, but
Oz: Absolutely. Right, I know, I know, so...
Carly Rector: One thing that I definitely, definitely think that AI is going to do is break down that wall because to the extent that you could ever just toss something over the wall, which was not high, you could just toss it over the wall to an AI now. ⁓ but you can't for a lot of things. You have to have some, ⁓ I was saying, I think last week even on the podcast, understanding of both. the problem that you are trying to solve and the nature of the systems that you're trying to use to solve it and like how you can fit those things together in the best way. ⁓ And I think to the point of another thing that I talked about recently last week, we still have a half finished sub-tech post about the roles like forward deployed engineer are starting to combine those more of, which is basically an engineer that goes into a company and talks to the customers and builds something directly without their being.
Oz: Yes. Mm-hmm.
Carly Rector: a PM or anything in front of them. that's according to Andreessen Horowitz, the hottest job in startups as of last year and still continues to be very hot. I think both OpenAI and Anthropic are building like their own companies that are basically ⁓ just deployment full of engineers that figure out how to use AI ⁓ for customers product ways. ⁓
Oz: Dope.
Carly Rector: I think there's a lot of that already happening and I do expect that trend to happen because I just think that's what makes sense because the piece of work that is left is figuring out what the problem is and then how to translate that to a shape of a solution such that you can tell it to an LM maybe but like there's still work there so I think I think definitely people are already starting to see that in my opinion. ⁓
Oz: And point out to the LLM when it has a terrible idea, because it will sometimes.
Carly Rector: Yes, that too. Um, so yeah, that, that hits some of the stuff in our agenda. Let's like briefly try and talk about the actual bullet points here. Um, one is this ramp report that showed Anthropic is beating OpenAI in business adoption for the first time. As of April, the graph is very exciting for Anthropic. It's like.
Oz: Yeah!
Carly Rector: gone up way by like, it has changed from like the February number, which was like 4.4 % anthropic and 18 % open AI to 34.4 % anthropic, 32.3 % open AI in April. I saved, February, February, 2025. So over the last like year, it has gone up dramatically and open AI has gone up and then slightly down a little bit, which is. super interesting. And a lot of that is the launch of Claude Code in general, which happened, I think, May 2025. And then big hype around Claude Code that happened around like December, January this year for reasons that no one is entirely sure about.
Oz: Yeah, makes sense.
Carly Rector: It was just like people had free time over the holidays and started playing with it and or Opus 4.5 finally pushed it over some threshold to be really good. But like I didn't think there was a step change there. Other people were like, yeah, there was definitely a step change. It finally became very good. Like in December. I'm like, don't I ⁓ as someone who's been it this whole time, it definitely got better, but like not it wasn't functional before and now it is now. ⁓ yes, ⁓ so. ⁓
Oz: Not that much better.
Carly Rector: Um, but yeah, and then obviously all of the stuff around the Department of Defense thing and people just thinking that Anthropic was better in general as a company swapping over has happened also. uh, so yeah, that's interesting. Um, they caveat of this post with like, this doesn't mean that Anthropic is winning the race. Now, like one thing that you can take from this chart is that it's actually very easy for an underdog to take over and that could still happen. Like maybe.
Oz: Mm-hmm.
Carly Rector: DeepSeek comes out of nowhere and wins in another year. Like, we don't know.
Oz: ⁓ Remind me about Deep Seek. Is Deep Seek the Chinese one? Okay.
Carly Rector: Yes, yeah, DeepStake is the Chinese one that launched and then kind of wasn't that exciting really and why is it still at like 0.1 % on this chart the whole time?
Oz: Was it more energy efficient or was that overhyped?
Carly Rector: Um, it was, uh, in the sense that it was like much, much more everything efficient in terms of like, said they trained it with, for only like $5 million or something like that, which there was some questions about that. And there was also some accusations that they were distilling the big AI labs models. And that's why they could trade it for so, so cheap. Um,
Oz: Mm. Ah, oh no, someone's... He tried to...
Carly Rector: Yes, one of those... I definitely saw an OpenAI article that was like, ⁓ no, OpenAI accuses other people of stealing their intellectual property! Who would do that?
Oz: I can't believe someone would do that to sweet, darling little open AI who's never done anything wrong and it's his birthday.
Carly Rector: Uh-huh. so yeah, but yeah, things could change at any time, but it's, Anthropics doing pretty well. And it's mostly because of Claude code. Um, we have on this thing, like why Claude code is better, why Claude code leak wasn't that interesting, which has to do with some of the stuff that we were talking about a little bit in the sense that like the Claude code harness A is distinct from Claude, the model. which is I think most people who saw that leak didn't understand at all and they were like, Claude's source code is leaked, ⁓ like doomed the company now or whatever. It's like no, it's different, ⁓ it's just the harness. That is an important part of it for the reasons we discussed, but it is not the core part of the model. ⁓ other part of it is because it is just like ⁓ a node harness, like technically all of the source code was on everyone's computers already, like ⁓
Oz: Yes.
Carly Rector: version of it because that's how you run the code. And so like it wasn't super secret even before then, like you could have tried to reverse engineer it, it would just kind of been harder. Like you didn't have like the raw human readable code and that was what leaked. ⁓
Oz: may not like be as technically interesting but I do think it's funny. Yeah.
Carly Rector: It is funny, that's true. And it was interesting to see people do analysis of the 500,000 lines of code, exactly what was in there. And I do think it gave an advantage to the competing tools. Because the other part of it is obviously what Claude Code does is not a secret. You can just use it and then see. Unless you work at OpenAI, which they have banned from using Claude Code.
Oz: Yeah. Yeah, yeah, yeah.
Carly Rector: because they were like, people at OpenAI are using it and just figuring out what it does and copying it, so we're going to block them. ⁓ But yes. ⁓ So you can just use it and see. So there's just like a lot of reverse engineering or even just like seeing what features it has and copying them ⁓ that was going on already. And then this makes that easier for them to do. But it was always just like, Claude Code has a few months head start on some of the other tools.
Oz: ⁓ no!
Carly Rector: and not like it's a fundamentally secret, secret approach that they have. I do think one of the reasons that it's better is that they are better at building for customers generally. So one of the things I have talked about on this podcast before is that Claude Code was not intended to be a consumer product to start with. It was an internal tool for Anthropic. And they were just like, actually, this seems to work really well for everyone.
Oz: Mm-hmm. That's right. That's right, it was.
Carly Rector: 95 % of anthropic people adopted it just naturally without us asking them to really quickly because it was actually genuinely really useful for them. Maybe we should launch this to people. And they were like, sure, let's launch it. So they launched it as like this CLI tool that wasn't even that user friendly, but was really useful. ⁓ So, and people found it really useful and started using it. ⁓ And then like the guy who does a lot of the work on it ⁓ is on X all the time and sees people tweeting about it and will be like turning around, be like, yeah, that's a good, good feature idea. Turn it around. It's out now, tomorrow. ⁓ level of work on just Claude Code. ⁓ So just being really responsive to like what actually makes it more useful for people instead of trying to build it as something that is, that looks good as a product, I think is.
Oz: Mm-hmm. Right.
Carly Rector: helping them a lot, which is one of the things that I feel like is just encouraging in the world.
Oz: Encouraging like from a perspective of, yeah, dang, sort of seems like when you focus on making a good product, you make a good product, but like still a little bit concerning, like from the bubble perspective. ⁓ Just in terms of like, yeah, dang, you have made something real good. Does it make you any money? ⁓ it does? OK, OK.
Carly Rector: Mm-hmm. It makes them a lot of money. Yeah, that's where a huge amount of the revenue is coming from. Now let me do a quick Google search. that's one of the reasons why their revenue gone up a ton. ⁓ Plod Code alone brings in 2.5 billion, a feature that has more than doubled since the beginning of 2026. ⁓
Oz: Now how does that compare to their compute costs?
Carly Rector: ⁓ Still, I I think they've been doing a lot of calibration on this. So this is a whole thing otherwise, which they have been trying to like calibrate their usage limits for people. And they have in fact been dropping their usage limits, both because it's expensive and also because they just simply do not have the capacity for it. And people have gotten really mad about it. That has not actually been a problem for me. I also talked about this a little bit last.
Oz: Mm.
Carly Rector: last time, is that people are like, oh, Claude's going to die now. And like the RAMP article actually talks about this a little bit too. It's like, can't I keep this up when it's getting worse and charging people more and et cetera. And it's like, okay, but I, again, as a person who's been using it the whole time, it was worse for me, worse for me in like November last year.
Oz: Mm-hmm.
Carly Rector: when it was chewing through tokens and I hit my five hour usage limits like all the time ⁓ and to actually like stop working or pay for a la carte tokens to keep working. And that has not happened to me since then. ⁓ I could look at my usage limits now let's see. My max usage for the week ⁓ is ⁓ 4%.
Oz: Mm-hmm. Mm-hmm.
Carly Rector: And that resets on Sunday morning. So that's basically, I have used 4 % of my limits for the week. And again, I am using it all the time. So it's fine from my perspective. It's not fine if you're like running open claw or you're like having a Ralph Wiggum loop that just tries the same thing a hundred times in a row until it gets it right. Like there are stupid things that you could do.
Oz: Okay. Okay. right. I haven't heard that term of Ralph Wiggum Loop. ⁓ What is... Okay, now I must ask. Tell me of the Ralph Wiggum Loop.
Carly Rector: It's a real popular one now. Ralph Wiggum loop is Ralph Wiggum from Simpsons. ⁓ but basically just like write a thing that keeps saying, did you do the thing? Like Claude will come back with an answer and it will just say, did you solve the problem? Like repeatedly ⁓ So just like keep trying until it gets done. ⁓ so you just ⁓ loop it it gives you the right answer.
Oz: Right? Yeah. I see. So like... So like the miracle sort algorithm, but real. OK.
Carly Rector: Kinda, yeah. So people do that and they think it's really cool and I'm like, that's stupid. ⁓ I mean, like, I guess it does work in the sense that like, if you tell an LLM to retry things enough times, it's more likely to get there, but. is drastically inefficient and doesn't work for many things where like if you have not defined your success criteria correctly or thoroughly, that's just fundamentally not going to work and also chews through tokens. Anyway, my point was that Anthropic has actually been doing a lot of stuff to bring their inference costs in ⁓ both because they just don't have that much capacity. ⁓
Oz: Mm-hmm.
Carly Rector: and just because they want to be able to make money. I think they're similar to... OpenAI is also like this. are actually doing okay on inference costs ⁓ in terms of their...
Oz: Okay.
Carly Rector: getting approximately-ish, probably not fully, but approximately enough money to cover what it costs to actually serve the traffic. ⁓ And of the costs are coming from training, ⁓ because that's a huge chunk of work you have to do up front before you have the model. ⁓ And they have to keep putting more money into that to train the more bigger models.
Oz: Mmm.
Carly Rector: I don't know the actual details. I would assume that Anthropic is in a place where they're at least within like a couple of times of their inference costs via revenue. And like I've talked about before, like again, I'm at 4 % of my usage. So if they cut my usage by ⁓ like to 20 % what it currently is, that would be fine with me ⁓ because I'm not using much. So ⁓ like,
Oz: using 20 % of 20%.
Carly Rector: Yeah, mean that might be a little harder because I have gotten up to like 7 % of my weekly limits before. But like everyone who's like if Anthropic comes out here and like doubles the price or triples the price or cuts the usage limits drastically everyone's gonna be in trouble. It's like well not everyone because I would be fine. So like some people are gonna be in trouble for sure, but everyone who's like this is all completely unsustainable, at least some of it is sustainable, I can sustain it.
Oz: Mm-hmm. So it's a skill issue.
Carly Rector: Skill issue, yes, on the subject of usage limits compute costs ⁓ is ⁓ SpaceX deals that they did, which I think is super interesting. So the specific announcement headline on this Anthropic blog is higher usage limits for Claude and a compute deal with SpaceX. ⁓ because the important part, they think, for their customers is that they are doubling usage limits based on this deal because they just got a whole bunch more capacity, ⁓ capacity ⁓ from the Colossus data center. So the specifics of this deal are Elon Musk made a deal with Anthropic to lease them capacity the giant Colossus data center that he built for XAI, ⁓ which is of SpaceX now because... ⁓
Oz: Hmm.
Carly Rector: for silly reasons, SpaceX has acquired XAI and XAI acquired X. So Twitter and XAI are all under SpaceX now. It's the AI Rockets Twitter company now. And it's going to go public soon as that. So the important part about knowing that XAI is under there is that this is the data center that he built to train and run Grok. so the funny part about
Oz: Okay. All right.
Carly Rector: being like, I'm going to lease all this capacity to Anthropic now, is that implies Grok is not using it. Because Grok is not popular. Which makes sense, but it is interesting that Elon Musk would admit to that. ⁓
Oz: Okay, yeah, no, that scans. ⁓ We're not into the fascism maximizer 9,000.
Carly Rector: ⁓ And he tried to walk it back a little bit. He's like, ⁓ we're still definitely using it and like we'll be training new models soon and we'll build out new capacity for that blah blah blah. the other article I linked here on that topic was ⁓ did XAI just concede the race? Which, because it just like, are you giving up? Because you're handing your chips to Anthropic, which is a competitor. ⁓ So...
Oz: Mm-hmm. Yeah, I saw that one first.
Carly Rector: ⁓ kinda feels a little bit like giving up. ⁓ but, and then the other big reason for that, of course, ⁓ that people called out is the enemy of my enemy is my friend, which is that Elon Musk is in this huge lawsuit with OpenAI and hates them a lot, ⁓ and would like to do anything he can to kill them at this point for petty reasons, like that he wanted to be in charge of OpenAI originally, which came out a while ago as part of this lawsuit. ⁓
Oz: Hmm. ⁓ did he? Yeah, that scans.
Carly Rector: Yes, the quote that I heard in I was listening to today that was insane is like in some email with Sam Altman, I think it Sam Altman who was talking about this, they asked Elon in the hypothetical world where Elon was CEO of OpenAI and he died, who would take it over? And Elon was like, I haven't given it that much thought, but I don't know, maybe my children. ⁓ And everyone's like, yeah, sounds great. Let's make the... AGI future AI machine, hereditary oligarchy. That sounds perfect, yes. ⁓ Anyway, that's fun But anyway, they didn't let him do that and he's mad about it. ⁓ What OpenAI says is reason for this lawsuit. ⁓ Elon says it's because OpenAI a charity and he invested whatever million dollars in them.
Oz: Yeah, great. Dune.
Carly Rector: that they just took as a non-profit and then became a for-profit company, etc. which did happen. But the most interesting part of this lawsuit is not any of the actual details of the lawsuit, but all of the early OpenAI communications that have come out about it. So the importance of our heart being there that Elon Musk is extremely butt-mad that he's not in charge of OpenAI and therefore is willing to work with their competitor to kill them.
Oz: one of the least professional humans to ever be the richest man in existence.
Carly Rector: ⁓ huh. Yep. SpaceX is leasing a whole bunch of their data center capacity to Anthropic. also in this platformer article about did they concede the AI race, they point out that Musk himself has repeatedly called Anthropic a key threat, complaining that, among other things, Claude, quote, hates whites and Asians, especially Chinese, heterosexuals, and men. ⁓ Actual tweet from Elon Musk some number of months ago. ⁓ But he has a more recent tweet that says met with people in Anthropic and ⁓ evil detector didn't go off. He thinks all of them are doing good work and genuinely care about doing the right thing for the future or whatever. ⁓
Oz: ⁓ the father of Mecha Hitler thinks we're doing the right thing for the future. I'm so reassured.
Carly Rector: Hahaha! Yeah, mean a lot of the Musk, early Musk stuff was crazy because some of the quotes that I've seen also is that the reason that he wanted to found OpenAI in the first place is that he thought Demis Hassabis, who's the CEO of DeepMind that was then acquired by Google and is now the Google AI, ⁓ was basically a super villain who didn't about humanity and was gonna be the evil overlord of AI and he needed a counter evil overlord company.
Oz: Okay, okay, they are sure, I guess.
Carly Rector: ⁓ huh. So yeah, that's fun. I saw this deal and I was like, that's interesting for a number of reasons. But if he is conceding that XAI is not a real competitor here, I'm not going to complain. It's like the quote at the of this ⁓ platformer is, In the meantime, the most irresponsible of the leading AI companies is now effectively selling itself off for parts. It may be the best thing Musk has done for humanity in ages. Which is, like, not deliberately. He's still doing it for incredibly petty stupid reasons, but if he's doing it for petty stupid reasons and he manages to kill Grok anyway, that's fine, I guess. Yeah.
Oz: Let's break that clock twice a day.
Carly Rector: Um, but that does pull in another circular funding deal to segue a little bit to talking about circular funding deals, which is just the whole mess of things, which we have talked about previously in this podcast in October. I looked it up. Uh, is this Bloomberg circular finance chart about how everyone in AI AI is giving money to each other. And it's just, it's real, it's just real bad. They have an interactive version on the Bloomberg site where you can go and see all of it. ⁓ and share. So we can explore the deals. For example, Anthropic is.
Oz: Mm-hmm.
Carly Rector: backed by Google and Amazon, it has also announced plans to raise money from and do business with Microsoft and Nvidia. So like the most common pattern that you will see on this chart is a company that provides computing capacity investing in an AI company that will then buy computing capacity back from them, So that's the sort of pink and blue investment hardware lines. the services and investment lines are kind of similar too. ⁓
Oz: Right. Mm-hmm.
Carly Rector: the most directly egregious circularness of it because it is just like I actually even even like two years ago, ⁓ I heard VCs actually complaining about this because what what is bad about this from a VC's perspective is that if Microsoft invests in open AI, they can do it at a drastically inflated valuation because the valuation doesn't actually matter to them. They're just getting their money back. anyway because they're just buying cloud compute from them. Whereas the VCs have to actually make money off the equity of the company and the later exit, etc. So, oh no, all the money is fake! And it's fake in a way that doesn't advantage us this time! What's that about?
Oz: right. ⁓ no, we made all the rules fake and unfair.
Carly Rector: So yeah, the big sort of strategic is what they call it, investors, which is like big companies that invest because it's useful for them and not because they're trying to make money. ⁓ inflate the valuations even more than they would already be inflated, just the hype cycle. So, ⁓ is the complaint of VCs here. And so that's the sort of like the direct circles that are happening here a lot. ⁓ But then there's obviously the ones where it's just like the money goes around various places, because all of these are interlinked with each other. ⁓
Oz: Mm-hmm. Mm-hmm.
Carly Rector: one of the things that I think it's important to understand about this is that this is particularly bad, but I don't think this is unique to the AI space. ⁓ in that ⁓ there's a lot of business to business deals that are fake money and that nevertheless get treated real money by say all the press around it.
Oz: That's true. Yes.
Carly Rector: first of all, and second of all, it becomes real when these companies IPO and go into the real stock market and then they can sell the stock for real money. So, it's... I don't know.
Oz: Mm-hmm. the onanistic hand of the free market.
Carly Rector: Yeah. So I've talked about this before also on this podcast, but I think a really critical bubble moment we will have is when OpenAI tries to go public. Cause they, they keep saying that they're gonna, and there's been some reporting that their CFO said, no, we shouldn't and we're not ready for it. And that Sam Altman was overriding her, which is not, not great. ⁓
Oz: ⁓ yeah, that's right!
Carly Rector: But also there's been a bunch reporting that like, because as we saw on the ramp chart, they're not doing that great lately. Their usage is dropping compared to Anthropic. And their revenue and Anthropics is about equivalent now per latest reporting. They're both at like 25 billion annualized, I think. ⁓ And so people are like, wait, but if they're about the same and Anthropic is growing faster, why is OpenAI like three times the valuation of Anthropic right now? ⁓
Oz: Mm. Mm-hmm.
Carly Rector: The answer is that all numbers are fake and it was all high- ⁓ But that's the short answer. ⁓
Oz: Welcome to capitalism where the rules are made up and the points matter ⁓ so much.
Carly Rector: Yes, ⁓ but it was more hype driven. But like also people are noticing this now, which means like in private secondary markets, there's a bunch of reporting I've seen that like, people are trying to find buyers for their open AI shares at the current valuation and not being able to find them. Whereas everyone is scrambling to trying to get into Anthropic. And Anthropic is maybe raising at a trillion dollar valuation before they go public. I saw a little bit of reporting on that. So
Oz: Mm.
Carly Rector: OpenAI had talked about going public at a trillion dollar valuation or more this year. ⁓ and I don't think that's going to work out for them if they try and do that, because it sounds like people aren't going to buy even at the current price and are trying to get out of it. But it might, like what might happen is that they do it ⁓ and there's
Oz: Mm-hmm.
Carly Rector: consumers who are still waiting to get in that all of the early investors can offload their shares into the market and then it crashes. And that might be the bubble pop, but... Or they could not do it and Anthropic could go first and that could work fine. I don't... Like there's various ways that this could play out, but I think the opening IPO will be a very pivotal moment here.
Oz: Yeah. Yeah.
Carly Rector: notably like there was some reporting that Nvidia did not want to invest that much more in OpenAI. Like they had again said that they would put in like some crazy number like a hundred billion dollars or up to a hundred billion and then they said actually we're not doing a hundred billion we're doing like ten or whatever and everyone was like um oh nvidia losing faith in open ai and they're like no no no we said up to like we didn't promise anything but also um it
Oz: Mm-hmm. right.
Carly Rector: the other comment, I think it was a Jensen Huang comment that was like, this is our last opportunity to invest in OpenAI. ⁓ And it's like, why would that be the case? If they go IPO, you could buy their stock at any time. Or even if they do another raise after this, before they go public, you could invest at any time. ⁓
Oz: Yeah. What do you mean your last chance? This will be our last chance to hand someone money.
Carly Rector: Yeah, and what the implication there is that they think this is the peak, and therefore it would not make any sense to invest after this. Or it only kind of makes sense to do it now even. Which I think is probably true, and I a lot of people are thinking that might be true. But maybe they'll re-hype it up, I never know. So, but yeah, all the numbers are fake. It's very depressing.
Oz: Mm-hmm. ⁓ Man, if only a German philosopher had written about this. ⁓
