welcome everyone to the information's ti tv my name is akash was free chat it is tuesday may 5th on today's show the information has exclusive reporting about some big product changes at amazon our amazon reporter will join the show to share with us what she knows We'll then turn to our reporter at the courthouse, Rocket Drew, who is covering the Elon Musk OpenAI trial for a recap on Greg Brockman's testimony. We'll also unpack Coinbase's layoffs and we'll then get into our exclusive reporting on the hidden costs behind XAI's fast and cheap data center build out. And we'll wrap with a look at AI coding, how startups are differentiating themselves in an increasingly competitive landscape. It's going to be a fun show, so let's get right on into it. Amazon is planning some big product changes for its core e-commerce website, some of which take a page from Google's playbook in search. My colleague Catherine Perloff wrote about that in a story out today. I want to bring her on to talk all about it. Catherine, welcome to the show. It's great to have you here. Hi, Kosh. So what product changes is Amazon planning at their company? Well, it's a little, it's still hypothetical, but, you know, so I spoke to the VP of core shopping at Amazon, a key executive who kind of controls partly how the website looks. And one idea she said they've been sort of tossing around is creating sort of a more hybrid search experience. So up until now, there has been the Rufus chatbot, which Amazon formally introduced in 2024, where users could have kind of a more chatbot-like experience about shopping for products. And then the more traditional search engine where, you know, a shopper could just put in terms and then they get sort of a list of items instead of a more like paragraph conversational like response. And what Amazon is thinking about is could there be a more combined experience where there is a paragraph describing, you know, some of the product features and then below that the results. So, you know, this might remind you a little bit of Google AI overviews, which, you know, they rolled out a few years ago where they kind of provided this AI summary and then links to the site below that. And, you know, this is just one experience Amazon might offer. It's not necessarily going to be that way all the time. You know, Amazon has a different business than Google where they're selling. They want people to buy things and not just people to buy ads. But yeah, it is interesting sort of how the Amazon website might evolve with AI. So let me understand this. So when I go to search for products on Amazon, I know I have the search bar there and it takes me to all the results. And so is the idea here that what's under consideration is if I search for household items or shoes or whatever, then I will get an AI overview type of thing with links to results and then the listings? Or what is under consideration? Yeah, so maybe there would be like a summary. Maybe you're saying like household products – this is an example I'm coming up with – like to get stains off the counter. And maybe then there would be a rufous-generated summary of sort of like, you know, what are different chemicals or different, you know, what the market looks like for these types of products. And then below that summary, there might be actual product listings for, you know, different types of wipes or counter cleaner, sprays, etc. So, you know, and that's different where basically kind of that summarized answer that was more like, you know, conversational, someone talking to you was sort of bifurcated in the Rufus chat window, which a shopper had to sort of toggle over and choose to use. and then there was a separate kind of main shopping experience. And this is sort of, you know, could maybe be opting in by default in certain instances, shoppers to a more conversational experience while not fully taking away the ability to browse specific items. So I got to be honest, like, I don't know if I'm going to like this. I mean, you know, I kind of like when I search for the thing and I just want to see the options right away. and now they're introducing, they're possibly going to introduce this AI overview thing. I mean, why do you think they're doing this? And is the only reason because Google did it? I mean, we don't even know how successful the AI overview thing has been, right? Well, I think Amazon is aware and their business is different than Google. So one thing Amanda Doerr, who I talked to, mentioned was like, for something, searching for something like mail. Her role, Amanda's role is what? VP of course shopping at Amazon. Okay, okay. That, you know, you, you know, when you're searching for milk, you don't really want a paragraph about, like, the difference between 2% and whole. You know the difference. So they wouldn't, you know, there's definitely a lot of types of searches where the kind of standard format where you just see a list of products is the best option. There are other types of searches where you're like, you know, what are the best hiking boots, is an example she gave, for the season. Um, and in, in those cases, you know, maybe you want some more information about like what makes a good hiking boot. So I think Amazon is trying to figure out what are the types of products where more information is helpful. And I think there certainly are some of those types of products. Like I know when I'm looking for say like new earphones, um, I usually have to like go search on Wirecutter or some other site and try to figure out like what people like about, you know, AirPods. So I do think that there are like items where people do need more information and there are items like milk where people really don't. And I think Amazon is trying to kind of thread that needle and figure out like a little bit of like a wire cutter ask, like where there are recommendations that can be useful. That's where it might lean into these types of searches and stuff like that. Exactly, exactly. And, you know, it kind of helps Amazon in a way go up the funnel, which is kind of marketing jargon. But, you know, kind of that's sort of been almost Google's historical role, helping you with the product discovery process. And then you go to Amazon when you know what you want. But if you can go to Amazon to learn about products, you can spend more time on Amazon. Maybe they can sell you more ads. There's probably lots of good things that can happen there for them. So, right, right. Now, you mentioned your conversation with Amanda Dora, the VP of Core Shopping, I think you said was her role. What questions did you come out of the conversation with in terms of how you think this could be rolled out or, you know, the company strategy at large? Yeah, I mean, I am still sort of, yeah, curious sort of, again, how they thread the needle. She also mentioned that, like, and we've reported on this in the past, they've been tweaking the way that, like, the regular search engine works to sort of be more responsive for natural language. And, like, they've, you know, now I believe, like, 60% of searches use autocomplete, and they've been trying to use LLMs to make autocomplete smarter. So it is sort of like, are the lines going to blur? So that is definitely a question I have. Like, you know, as regular search becomes more natural language-y, like, you know, how does it become different from a chatbot? And maybe it's in the response, but then, you know, does everyone start asking questions to the Amazon search bar? How does that sort of change the experience? Another, you know, question I have is, like, they have been really interested in this new buy-for-me feature, which is helping shoppers find items that are not on Amazon's own site, but they, you know, will scrape the web or use other features to find out about these items. And then you can buy that item that isn't on Amazon's site. They'll direct you to the other site and you can, you know, find it and buy it there. And I'm also very curious, like, to what degree does Amazon want to be sort of like the first stop on the Internet? kind of, again, taking that role that Google's often had and being the place where you learn everything and you do product discovery, even if that means buying a product that's not on Amazon and like how important that is for them and, you know, where that kind of puts them, yeah, in the whole e-commerce process. Right, and last question for you, Catherine. So we don't know exactly if or when they are planning to roll out this type of feature. We just know that they are considering it at this point, right? Yes, we don't know the exact timeline, And though I will say, you know, yesterday, actually, there was some observers who saw that, like, when they used the regular Amazon search bar, it would, instead of just populating with a list of results, it would go into a more, like, chat, conversational, Rufus-like experience. So that was, you know, and, you know, we asked Amazon about it. They're like, you know, we're testing things, whatever. So I think it shows they're kind of already thinking of more hybrid modes like you know people have been seeing them in the wild um right you know so it seems like things are underway an exact product roadmap we don't exactly know but yes right great well katherine i want to thank you for coming on that is katherine perloff our amazon reporter here at the information the information's rocket drew is covering the elon musk open ai trial from the courtroom all week long. Here is his latest update for us on OpenAI President Greg Brockman's testimony in court on Monday. Greg Brockman had a pretty rough day in court. A lawyer for Elon Musk had the chance to ask Brockman, who was a co-founder of OpenAI and its current president, some tough questions about his personal finances including his stake in OpenAI itself but also his his stakes in companies that have done business deals with OpenAI That includes Stripe Cerebris CoreWeave and Helion Energy As far as his stake in OpenAI, Musk's lawyer asked him right out of the gate, and he confirmed that his personal stake in OpenAI is worth in the neighborhood of $30 billion. That puts Brockman in kind of an uncomfortable position in the lawsuit, because Musk is arguing that Brockman has been unjustly enriched by his involvement with OpenAI, and that OpenAI has breached a charitable trust. It's sort of broken some charitable obligations or commitments by operating more as a for-profit than as a non-profit. In its founding documents, OpenAI says that it's not organized for the private benefit of any individual, which raises questions about Brockman's $30 billion stake. Now, in response, OpenAI's lawyers pointed out that most OpenAI employees do have equity, that Brockman's equity stake was decided without Brockman's input, and that OpenAI's charitable wing, the OpenAI Foundation, is very well funded. Now, Elon Musk's lawyer did not find that argument very compelling. He compared it to a bank robber taking a million dollars from a bank, but then saying, well, you have millions of dollars left. So Musk's lawyer in particular had a field day with records of an old diary that Brockman kept. And in this diary, Brockman is contemplating different paths forward for an open AI for-profit subsidiary, and he muses about which path, a path with Elon or without, will more quickly take him to a billion dollars. And in court, Brockman suggested that a million dollars would sort of be sufficient for him. So Musk's lawyer countered, well, if a billion dollars is sufficient, why not take the difference, call it $29 billion, and return it to the non-profit? That turned into a pretty uncomfortable line of questioning for Brockman. It didn't seem like he really had an answer to that prepared, but he did fall back on saying that the OpenAI nonprofit is very well resourced. So it was a pretty tough day. I think both sides, the OpenAI side and Elon side, have come out scathed by questioning from the other side's lawyers. So we'll see if OpenAI's lawyers questioning Brockman tomorrow are able to restore and rebuild his case and see if he comes out looking any better after that questioning so that's what's up next in the muskvian trial coinbase said that it plans to cut 14 of its workforce in order to manage costs amid the market downturn and improve efficiency using ai the announcement comes ahead of its first quarter earnings on thursday our crypto reporter yuechi yang joins me now for her reaction Yuichi, welcome back to the show. So Coinbase is doing layoffs. We've seen layoffs across the crypto sector. What was your reaction to this one? Yeah, so Coinbase today said they're laying off 14% of jobs, and that's affecting about 700 people at the company. It is a big, sizable layoff. But as a percentage of the total number of employees, it's actually smaller than the layoffs that Coinbase did in the prior bear market in 2022 and 2023. And at that time, Coinbase did two major rounds of layoffs. The first one was 18%, followed by another one that was 20%. And then when CEO Brian Armstrong announced the second layoff back then, he said that in hindsight, Coinbase should have cut further in the first round. So we are waiting to see if this layoff is it for the rest of the year or if there might be even more layoffs coming out depending on market conditions. Okay, so this is kind of interesting because we have these two crypto bear markets that have come sort of a couple, I don't know, a couple years apart maybe. Was that bear market worse than this bear market? Yeah, that's a great question. I think that bear market in 2022 was characterized by dramatic blow up of big companies like FTX and a series of crypto lending platforms. And I think in the current market, it's more like a quiet down market. You don't see these dramatic collapses, but then at the same time, retail investors are just losing interest in crypto. and a big part of the crypto economy is down when you look at the token prices for a lot of these tokens that are not Bitcoin, not Ethereum, but the smaller ones, they just really haven't recovered from even the last bear market. So I think the crypto industry is going through this fundamental change. It's unclear whether some of these tokens will ever make a comeback. and so that's also going to affect Coinbase business model as well. Well, and the reason I'm asking that question is because I think it really sets us up to answer the next question really well, which is whether or not these cuts are bear market cuts or if they are AI cuts, which is the reason that Brian Armstrong cited in his note to employees. I mean, we live in a world where CEOs can say whatever they want in terms of why they're making cuts, but what's your assessment as to whether or not this is actually a true AI related cut or if it's just an excuse for the crypto bear market? It's a bear market cut. And I'll show it. Okay. All right. That's it. See you later. The number shows it because in the last quarter, the fourth quarter of 2025, Coinbase revenue was down about 20 percent it posted a net loss of about 670 million dollars um so coinbase hasn't reported the first quarter result yet as you mentioned it's going to report earnings on thursday but some of its peers like robin hood has already reported and robin who said its crypto revenue was down by almost half in the first quarter so we know that the numbers are not going to look good. And some of the other crypto exchanges like Gemini, like Crypto.com have already started layoff. So the market condition is definitely the most important factor for the layoff that we're seeing here. AI is also an issue. Coinbase CEO Brian Armstrong is very vocal about encouraging employees to use AI. In today's tweet announcement, he said that there will no longer be any pure manager, meaning that every leader in the company will also be an individual contributor. And he also envisioned this future where there will be one person teams at Coinbase where people will act multiple functions, including being an engineer, being a designer, being a product person with the help of AI. Do you have any sense from people at the company, around the company, whether or not they are in favor of this structural change? I'm not talking about the layoffs. I'm talking about this change from having pure managers to everyone being an IC, an individual contributor, everyone having a little more ownership. I'm sort of thinking about this. And I'm asking this question because it's not just Coinbase, right? I mean, this is something that we will see across tech companies. I wonder if you've talked to any people at all about whether or not employees like this or if they kind of liked the, you know, being part of a team maybe. Yeah, I think Coinbase CEO himself started to talk about how he, even he as a CEO has returned to the code base of Coinbase last year and started to do some coding himself again using AI tools. So it is a top-down approach. Everyone needs to get their hands dirty. And I think from employees' perspective, this layoff probably isn't too much of a surprise. Everyone knows the market has been in poor condition. Do they think there might be even more? You just said the bear market, they did two rounds of cuts, and then they said we could have cut even more. Is that a concern? Do you expect maybe there to be more cuts? I think it's definitely going to be a concern, and it depends on market conditions in part, which we don't know how it will play out. And I think this is something that Coinbase will be asked about when they do report earnings to whether they will commit to a certain level of profitability or to be able to make sure that the company can outlast the number of cycles that we're seeing. I want to pivot to talk a little bit about another piece of crypto news this week. So Andreessen Horowitz has raised its latest crypto fund. It is just over $2 billion. What was your reaction to this fund that they raised? I think the size stood out for me. It's much smaller than the prior fund they raised in 2022. It's about half the size. So back then, they raised a $4.5 billion fund. That was a record size. and it happened right before the crypto crashes occurred in 2022. I think even in the blog post that they announced, the new fund today, they alluded to the fact that the market cycle is currently on a quieter part. And they mentioned that they wanted to invest in founders that are building products for everyday uses, even though right now they don't get as much attention. So I think they're pretty aware of the fact that a lot of the interest from investors have moved on from crypto to AI, but they seem to suggest that that's not necessarily a bad thing because it allows them to stay focused and to also capture and identify good founders at a time when they don't get attention. and and it strikes me that that perhaps is a little bit of a contrast to the way that han ventures positioned their latest fund that they also announced this week i believe i mean they still said that they focused on crypto they said they sort of opening the aperture a little bit to take advantage of where AI and agents could intersect with crypto But it sounds like Andreessen is saying we're going much deeper into crypto. We're finding the best crypto founders first. That's what the goal is here. I think for Andreessen, the intersection between crypto and AI is also an important focus. In their blog, they did allude to the fact that agentic payments will become more of a thing and the internet will be run by agents in the future. They didn't really spell out specific examples of how this intersection between crypto and AI will look like. It's kind of like a high level, a big picture start. So we look forward to seeing how exactly that will pan out. Great. Well, Yueqi, I want to thank you for coming on. That is Yuechi Yang, our crypto reporter, here at The Information. SpaceX has been highlighting XAI's speed and cost advantage in its data center build-out compared to rival AI labs, but exclusive reporting from The Information found those advantages come with some hidden costs. My colleagues Theo Waite and Valida Pau published the story today. Theo joins me now to share with us what he knows. Theo, welcome back to the show. It's great to have you here. Hey, Akash. So tell us what we know about XAI's data center build-out efforts. So, you know, XAI is this company that, you know, pitches themselves as doing things very differently from other AI companies and other data center builders. And, you know, the first, you know, aspect of that is that they even chose to build their own data centers in the first place. Like if you look at OpenAI, Anthropic, other, you know, large AI labs, they're largely letting other companies do the building. xai is building their own and in spacex's x1 um s1 sorry after acquiring xai they talk quite a bit about this metric called um price per megawatt in data center construction and the draft of the s1 that we've seen says that spacex is or sorry xai's which is now part of spacex uh xai's cost per megawatt is 2.7 million dollars per megawatt compared to a industry standard of uh 12.3 so cheaper cheaper access to compute like 25 of the cost of what other people cost of what other people pay to build data centers but xai didn't define that term precisely it's unclear exactly how they measured it there could be some you know creative um decisions being made about what to include in in that number. But the point is XAI is saying we know how to do this a lot cheaper than our competitors because we're crafty. And that comes out in a lot of ways. Okay, so crafty. So expand on that a bit more. What is behind XAI's craftiness here? So they're using a lot of kind of like temporary and improvised measures, you could say, to get their data centers online over the past couple of years. That includes using these mobile gas generators to power the data centers, which has been quite controversial because they produce air pollution and are kind of meant for temporary use in remote areas, and they're running them close to a city. The other kind of thing they've been doing is relying on these mobile chillers, which basically look like semi-truck trailers, and they just have over 100 of them kind of parked outside the data center to cool the servers in there. And those are also a temporary measure. And overall, you have all of these temporary, unusual setups that kind of add up to a lot more potential places where things can go wrong. And my understanding is that's contributed to the data center sites having a lot more outages than would be standard at a data center that took a lot more time to stand up. So they put together these improvised solutions to get these data centers up and running faster. And you've also, I mean, you've covered a number of these. I mean, what, there's Colossus 1 and then Colossus 2, I think, is it being built at this point now? Do I have that right or is it built? There's one building of it that's, like, mostly done, and then there's another building of it that is not done at all. Okay, okay. But the point is, I mean, these are the XAI data centers. And so what they're touting here in the SpaceX IPO is that we can get these up and running faster because we have these improvised solutions. We've got generators. We've got trucks. And what your reporting shows is that sometimes these improvised solutions aren't actually as reliable, understandably, as if they were to have taken a little more of a, you know, I guess, traditional approach to build. Well, to be clear, they're not touting the temporary measures in the IPO. That's based more on my reporting. The IPO pitch is that everything is amazing and perfect, obviously. Right. Okay, so they get the data centers up and running. Do they get access to the compute? Another part of your story that was interesting, though, was you talked a little bit about the utilization of compute and how well they are using all the compute that they've built capacity for. Talk a little bit about that for us. Yeah, so credit goes to my colleague Anissa actually for getting this memo. But there was a SpaceX executive who's been installed at XAI named Michael Nichols who kind of sent this memo about this internal metric basically showing how efficiently XAI is using the chips that it has, that it has spent a ton of money to get in its data centers and said that this metric is embarrassingly low and XAI needs to figure out how to use it better. And in the process, a ton of the leadership on the data center team and the computing team has left the company. So it's made that push. The other part of your story that was interesting is, are there any, well, you spelt it out in the story, and I'm hoping you can explain to us here. I mean, what are the safety implications behind doing things the way they have? Yeah, so, you know, in the process of reporting on the story, I found out about a few different incidents where, you know, XAI employees have either, you know, been involved in accidents or made other complaints about working conditions. employees of XAI or contractors working on the site. In January of this year, there was a worker who was using this device called a fusion machine that is basically this really powerful industrial machine that fuses pipes together. He got his finger stuck in the machine and it crushed it and it was amputated. Following that incident, the contractor that works for XAI said that they implemented additional safety measures, including basically having multiple people verify that nobody's hand is in this device when it's operational. And then last month, there were also some complaints at the site about generators being used inside that made employees sick. And then last fall, there was an incident where a couple of people got electrocuted. So I mean, With this kind of thing, it can be hard to say what is just a typical accident in the course of construction because construction is inherently a dangerous field. but with some of the, you know, with both the finger loss thing and the air quality issue like XAI or its contractors have, you know, committed to doing things differently following that, which would, you know, indicate something is going on that they had to address. So data centers are, you know, very large, ambitious projects and trying to do them really quickly can have, you know, downsides. Right. Big picture for you, Theo, before we let you go. I mean, as you think about questions that you have to push the story ahead, what are sort of the next topics that you're interested in pursuing on your beat? Yeah, I mean, I think that like with XAI, right, there is still this question of how on earth they're going to make money from these data centers. Like XAI's revenue is just not significant. and it's a gigantic source of burn for the combined company now. There's still kinds of, they had this deal they announced with Cursor where Cursor is going to use some of the compute from XAIS data centers, but there's a ton of ambiguities around that. I think it's just still going to be a gigantic question for investors in the SpaceX IPO that, you know, it's not... The actual revenue side of all this. I mean, you can bring down the cost as much as you want, but if nobody's buying the products, then that's another issue. Exactly, exactly. Well, Theo, I want to thank you for coming on. That is Theo Waite, our Elon Musk reporter, here at The Information. The SpaceX cursor deal for a potential acquisition surprised a lot of people. And in a world where AI coding is fiercely competitive, the big question is how do startups differentiate themselves in this environment as the big are getting bigger? I want to bring on Zach Lloyd, founder and CEO at AI coding company Warp, for his point of view on this. Zach, welcome back to the show. It's great to have you here. Thanks for having me, Akash. Great to be here. So I want to talk about Warp but tell me very quickly what was your reaction to the SpaceX Kersher deal announcement partnership whatever we want to call it at this point I mean a couple of things I guess one, it really speaks to just like how AI coding has taken off as a market. That's a incredible price for a AI coding company. And I mean, cursor makes sense, very, very big. From like my perspective, it's interesting to see another one of these AI coding companies sort of partner or maybe be acquired by a model lab. So you see sort of consolidation amongst the model labs as the big players in the space right now. And so explain to us a little bit about how Warp fits into the AI coding ecosystem. Are you a direct competitor of Cursor? Do you guys use each other's technology? How does it all work? So we are, in a sense, we're a competitor. But what Warp is doing at this point is trying to position itself as like a great place to run any coding agent. So we're complimentary in a lot of ways. So you can use you can use cursor, you can use codex, you can use cloud code, you can use Warp's own agent within within Warp. And the way that I think about our company and our positioning is like a you know, we're trying to sit one level above any particular coding agent and be a great place for people to run any of them, which matters because every, I don't know, month, week, what is the state of the art here is changing, and people really want flexibility for how they do that. So this is a tool where people can use this platform, and they can use both Codex and Claude and Kersher, Claude Codex, they can use all these tools inside the Warp platform. They can use all of them, and where this really, it's nice for individual developers, it really matters for companies as they are trying to figure out, like how do you set up systems for deploying, especially cloud agents at scale, you don't necessarily want to be locked in to one model or one provider. And so what we're trying to do is provide optionality to companies where they can adopt sort of one piece of infrastructure that lets them sort of switch between those different providers and optimize things like costs, which are an increasingly big concern for companies, get governance around like how those agents are working, audit trails, secure them all. And so we think that the best position for us is to be something that cuts across them, which is one reason. So last week we open-sourced Warp, and it's a big part of this strategy of trying to be a neutral ecosystem player in what is a very, very competitive market where we're going against XAI and Anthropic and OpenAI. We would rather not compete with them. We would rather be something that enables them. Why did you make the change to open source? Yeah, the general philosophy here is like, we thought we could actually move faster and build a better product if the community was participating. The other really, really big thing, like why now of it is that the agents and agents are doing most of the heavy lifting in our repo, they're actually like writing the code and doing the planning and verifying the changes. agents have become so powerful and so good that we felt like we could actually sort of accelerate the product development meaningfully by bringing the community in and having them do a bunch of the sort of like provide like what's what's the right features to build and are they working right so it's it's kind of a new way of working where there's humans but if you look at the repo it's all being built by agents it's very cool you know i want to ask you about the the vibe coding craze at large and the current state of it. I mean, on one hand, we have companies like Apple that are sending their coders back to boot camp, you know, to relearn the latest and greatest ways in which software engineers should be doing their job. You also have the whole token maxing movement, right, where companies will encourage their employees to basically use AI as much as they can. And I imagine Vibe Coding is at the center of that. But on the flip side of that, you know, you have these security flaws. And, you know, I'm sure talking back saying now, we saw that tide turn as well. You know, I think it was Uber's CTO that said, you know, we've kind of gone through our budget for it. And, you know, now we don't have any budget left for our AI endeavors in certain pockets. Anyway, the details are what they are. Do you think Vibe Coding, it's had its moment. Do you think there might be a pullback at all at these bigger tech companies where they say, well, maybe we should actually do things the old-fashioned way because it's a little safer in some cases? I don't think there's going to be a pullback to manually writing code by hand. I think those days have come and gone. I think Vibe Coding, The way that I think of it is like you're just sort of like, you know, driving in the back seat while the agent does all of the coding. You don't know what's going on. I don't think that's a good strategy either. So I would say those are the two extremes. And what you're going to see at companies is something in the middle, which I would call like agentic engineering, where, you know, you really have to reimagine what is the right way to build software in this world of new agentic capabilities. and it's not by hand and it's not by like totally like closing your eyes and let the agent do it. It's by adopting a new set of best practices and controls and securing it. And so I think you're going to see a middle ground. I do think actually costs, security, centralized auditing and governance, this is like how you're going to do this at scale. And we're moving past the point where real companies are going to just be like, go nuts on some coding tool on your laptop. They don't like that era is going to end pretty rapidly. Now, I want to ask you very quickly about open source models, actually, given that you are now firmly in the ecosystem. So one of the questions I've had is, you know, we've seen, for example, meta traditionally their models have been open source. And we've seen a lot of success with open source models, broadly speaking, especially I'm thinking about the models from China. is it at all more difficult to assess the effectiveness of open source models on these benchmarking leaderboards compared to closed source models? I mean, I'm not as close to the ground as you are in this space, but is there any more challenge evaluating the effectiveness there? Really? I mean, there is just general challenge with these public benchmarks and that the sort of training sets have kind of absorbed all of the benchmarks. And so you see these models kind of overfitting to the benchmarks at this point. But you can generally see how open-weight models are doing compared to the Frontier. It looks something like they're three to six months behind. It's kind of like how I would characterize it. The interesting question is like, is there going to be a point where three to six months behind is like good enough? and what I think will happen is that it will depend on what you're doing, which kind of sounds obvious, but like for some tasks, you will need to be at the frontier. You'll need to pay the frontier token rate, which is like, you know, 100x, the open weight token rate. And for some tasks, you won't. And so I see that happening and it leads me to believe there is going to be a future where it's a mixed bag, where optimizing the cost of how you're using these models really matters a lot. And, you know, I expect the open weight models to continue to sort of trail a little bit behind the frontier, but generally kind of keep pace three to six months behind. And last question for you. I mean, remind us then with your pivot to open source, this is a playbook that many companies have used throughout the history of Silicon Valley. There obviously is a way to build very viable businesses while still building open source technology. But tell me a little bit about how you're thinking about your own business model then in this era of open source and how you're planning to really scale with this technology? Great question. So the idea behind open source, like I said earlier, was it's going to improve the product. The other thing that it will do for us is help build the ecosystem and help us grow our user base. The part though of Warp that revenue generates for us is like on our server. And it's really about like agent orchestration and cloud infrastructure for agents and also like our own agent harness. All of that is not. Bring that down a level for it. Would you bring it down just a level for us? So so harness orchestration. What is all? Yeah. So let me more simply, our business is really around deploying agents for businesses to do like long, hard tasks, to do automations and to like have a agent that works super well, like a premium version of the agent. All of that stuff is still proprietary for us right now. Whether it will be forever, I'm not sure. But the part that we open sourced was actually always free. It's like the app that developers download and use every day. And so open sourcing that just helps get the community more excited. It'll help let them improve the product. And it's cool because they can actually build into it now so easily with agents. And so for us, it doesn't feel like a huge risk. It feels like something with a ton of potential benefit for increasing distribution and ecosystem, which are, I think, two of the ways that in a world where software doesn't cost anything to build, you can actually build a lasting advantage as a software company. Right. Well, Zach, I want to thank you for coming on. That is Zach Lloyd from Warp here on TI TV. That does it for today's show. A reminder, we are on the stream Monday through Friday at 10 a.m. Pacific, 1 p.m. Eastern. If you can't make it then, episodes are available on theinformation.com, our YouTube channel, or wherever you get your podcasts. Make sure to follow us on social media, on X, Instagram, and TikTok. I'm already excited for our next show tomorrow. Have a great rest of your Tuesday. Bye-bye for now. Thank you.
