This video is an official livestream of the Brazil at Silicon Valley 2026 event, focusing on the sessions held on April 8th. The content covers discussions on AI innovation, entrepreneurship, venture opportunities, and the future of technology, featuring various speakers from prominent tech companies and academic institutions.
Heat. Heat. Heat. Buddy, hey. Heat. Heat. Heat. Heat. Oh yeah. Hallelujah. Heat. Heat. Heat. Heat. N. Heat. Heat. N. Hello, my name is Billy from Reound Music. We are so happy that we jumped over 3,000 subscribers. Thank you for your trust and your support. We want to celebrate our success with a free giveaway. You will get the music you hear in the background for free, including your personalized premium license. All you have to do visit our website freesoundmusic.eu. Choose the tab premium license. Fill in the form with your name and address. We need it to personalize your premium license. Link to this video. You have to choose a payment method, but we will not charge you anything. Choose PayPal. Please type into the message box 3,000. We will prepare your personal premium license and the download link to this uncompressed audio file of this rack time track in maximum 48 hours. All the best. Have a great time on YouTube. And now let's listen to the entire song. Heat. Heat. Hey. Oh. Okay. Heat. Heat. Hey. Hey. Hey. I loved being at Brazil Silicon Valley and we're seeing the talent meeting the big ideas, the big problems and solutions at scale. Would you join Kville? forology. Good morning, BSV. So, did everyone survive all the happy hours? Probably not everyone, right? Today, it will be a long day and very interesting day. So, it's hard to believe that we got to the final day of the panels of BSV 2026, but not the final days of BSV because this year we innovate. We are also a startup here. We have the experience this year. So, hope you have registered yourselves to the two more days of experience. We are going to have tomorrow the tour to Berkeley and the master class. We are going to have the master class at Stanford, the visit at Google and on Friday we are going to close with the visit at Apple and you know it feels like it was yesterday that we kick things off right we in the first meeting recruiting recruiting the volunteers the students talking about the topics the participant the speakers a lot was going on in the last year in in order to organize BSV 2026. And over the past days, we have exchange ideas, challenge assumptions, and immerse ourselves in one of the most innovative ecosystems in the world and it feels like it's a lot to explore. So yesterday, I took notes of some some lessons, powerful lessons of our speakers. So let's go through them. Use Brazil as a stepping stone to go global. Solve real problems. Selling AI as a replacement for humans is a losing proposition. People want to remain relevant. You should have genuine passion for your company's product. Don't get discouraged. Don't give up early. 11 sessions yesterday, countless conversations, but enough looking back. So I will hand over to Bruno to go through today's agenda. >> All right. Thank you, Lisa. Good morning, Brazil Silicon Valley. Welcome. Hope everyone got enough sleep. We have an incredible day ahead. And you know every year we try to think of our we try to think to ourselves what what will actually make the 16 to 20 plus hours from Brazil to Silicon Valley worth it right in terms of content. And the good news is we have incredible incredible lineup ahead today. So Scott Brady will kick us off by tackling what's actually next in AI. Next we have Jean Moda from Crew AI who will then challenge everything you think you know about Brazil's infrastructure opportunity. We have SJ Levie and Astroteller from Axe the Moonshot Factory who will push us to think about what governments need to do and stop doing to let innovation thrive. Then Mike Kger will take us from building Instagram to what Anthropic is doing to reshape intelligence. Alvin Gley will force us to ask who's really winning the AI race and who's really getting behind. And then at noon, don't skip the breakouts. These will be three intimate sessions on AI iteration, expanding Brazil's talent pipeline, and Brazilian founders building in the US. I can promise you that some of the most intimate, best conversations today will happen in those smaller rooms. After lunch, Darren Cook will break down how to find venture opportunity where others are too scared to look. Macho and Jessica Leon will tackle what labs simply cannot build and who will. Piladman will show us what global research actually learns from local markets. And finally, David Thcker and Hbertunis will close us out with the big picture on AI and where all of this goes. These are Silicon Valley legends and emerging talents operating at absolute frontier. We hope you're excited. Stay sharp, ask hard questions, and let's make today count. Thank you. >> Thank you. >> Hello everyone. Super excited to be here today. I'm Elis Kotoski. I'm doing my MBA and my master's in environmental sciences at Stanford. And I was also the sponsors lead this year at BSV. And before we kick off our amazing panel today, we have to look at the foundation that made this possible. This conference simply would not happen without the commitment and support of our sponsors. They aren't just donors, they are true partners in innovation. To show our deepest gratitude, I'd like to invite our sponsors to the stage. Let's give them the warm, happy welcome they deserve. First, Julia Dominguez from Google. Adriano Marinho from Cellcoin. Marina Garcia from Kubu. Please stay on the side. We'll take a picture. Uh Bianca Juna from Jumpstart. Uh, Fernando Mirandes from Pedonet, >> Fernando Bosco from Palia. Teresa Borges from UI soft bank and Leonardo Mas from Vizio. Please come to the stage so we can take a picture together. Look at this. Don't do that. Hi everyone. I'm Larissa. I'm the head of audience. For a long time, you all were names in a spreadsheet for me. So, I'm really happy to see you all here. Uh, and I will invite the institutional partners to receive uh also the recognizment. I have my own. >> Hi everyone. Good morning. My name is Karolini Mariano and I'm the VP of marketing for BSB 2026 this year. And today we're so happy to uh give a small thanks to our institutional partners. So uh Lisa, I'll let you start. >> So I will first invite Lucas Mises from Thank you for another year of partnership. >> Thank you. uh Fernandanda Calloy from LA first year of partnership. >> Thank you so much Caro Rodriguez from Good And we have two other partners that are not here now but I will mention and we will deliver later. One is Daniel Marquez from Senos Celisio and the other one is Adita from Jeron Falcoins. Thank you. Now I would like to invite to the stage the marketing institutional partners that have helped us connect um Brazil to the Silicon Valley today. So I'll start with uh Moning agency and Eduardo Vieiraa. Next I would like to invite Mariana Telli from Consul Brazil in San Francisco. Um, next I would like to invite Karina Mida from Endeavor. Um, Annie Fadu from Majena Pubic. And um, last but not least, Luan from Cuckoo. All right. Now, uh I would like to invite the marketing partners to take a photo. And are did you do your photo already? >> No, not yet. So, let's >> Okay, >> let's start with marketing, please. Okay. Uh Thank you Guys, thank you so much. Thank you everyone. I'm going to break the protocol here for 60 seconds. We just thank our sponsors and partners. They deserve every word. But there are two people in this room that I need to call out. The first one is Chiago Medos. Chago came as audience and just start building our network platform. The second one is Johan Finger from Mona Agency already a BSV partner already done more than enough. But 3 days before the event we launched the platform we were midnight fixing the bugs before each one of you could find them. That's the portrait of BSV people who show up before you ask. And to honor these two, we would like to give them a BSV vest, a symbol that they are not just supporters of their of this community, they are part of our team. Please join us. Back up. Back up. Good morning everyone. My name is Elisa Pereira. I am a MSX student and a researcher at the Gishto Economy Lab at Stanford. I wake up every day thinking what is next in AI and I had the privilege to be Scott's student and also a mentee at GSCB. Scott is a serial entrepreneur who has founded three publicly traded companies and being CTO and CEO for the last 16 years. He has partnered with Eric Shimit at innovation endeavors doing deep tech investments in n areas around AI. Okay. have a quick few questions for the room. How many of you have opened Instagram today? Yep, a lot of people. How about used Claude? A lot of people. The person responsible for both of those products is about to walk onto this stage right now. My name is Leisia, third year in the BSV organizing team, returning to content this year. I'm a junior at Stanford studying economics with a focus on market design. And I've been doing some work myself on AI agents as economic actors and what that means for the products and the markets being built on top of them. So yes, I have a selfish reason for being very excited about this panel. Right now, we've spent two days asking, "What does it mean to build beyond human scale?" The two people about to join us have been living the answer. Mike Kger grew up in S. Paulo. He first joined us in 2019, and we're so so glad to have him back. He came to Stanford, studied symbolic systems, and then did something most people only dream about. He built o sorry, he built something over a billion people use, Instagram. In 2024, he joined Anthropic as chief product officer and has since moved into leading Anthropic Labs, the part of Anthropic focused on what comes next and what should even exist. Our second guest is Jade Lie, general partner at CO2, the firm that led Anthropic's most recent funding round. Before CO2, Jade was in and Horowits and Playground, spending over a decade backing founders from seed to growth. What you have on this stage right now is rare. Someone building at the frontier of AI and someone funding it who have already chosen to build the future together. So, please welcome to the stage Mike Kger and Jade Lie. Awesome. Well, I feel like we were just talking about this in the back of the room and how crazy this week has been. I don't know if you slept at all, but yesterday was a big day. Maybe tell us a little bit more about Project Glasswing. >> Yeah, absolutely. Great to be here. We're going to do this in Portuguese, right Jade? Um uh so project glasswing um is an initiative that we put together at Anthropic. So I've been at Anthropic for for two years now. Um and uh it's paired with our um latest model which is called Claude Mythos which we have in preview. And as we were preparing that model for release, we realized that it was not just great at writing code, doing agentic tasks, but also very very good at finding um and fixing um vulnerabilities in software. And um as we thought about how do we bring that to market, we realized that just releasing it would unleash potentially a lot of um not just people solving their problems with uh these models, but also potentially exploiting it as well. So we took a step back and said what's the best way of actually getting the world more ready for this? Um and so we'll put together uh project glasswing which really brought together this is really cool to see like really a group of um the leading companies that have sort of cyber defense at their core you know whether it's a crowdstrike whether it's Google who's doing a lot of the work there whether it's Microsoft whether it's AWS Cisco yeah Palo Alto Networks um like all the way down the stack um and basically said all right we're going to we're not going to release this model publicly um but we're going to effectively arm these cyber defenders with uh access to the model um you know a lot of usage credit so they can fully use it to secure their software. Um knowing that you know there are other model providers out there, there are other models out there like these capabilities will become more um sort of distributed you know in the next months and so can we basically give people more of a head start against that. Um that was the the sort of foundation of of project Glass Wing. Um, you know, on one hand it's like the model's awesome, like we use it, and I can't wait to get models like that out to to the public, but it felt like the right uh way of of sort of meeting the moment um in terms of how we release this particular model. >> Yeah, I mean, the capabilities is really interesting. I think um one of the things at CO2 we love tracking a lot is the MITER chart. I don't know if everyone's seen the miter chart um but it's sort of on the y- axis is sort of the time horizon to complete the task and then the x- axis is is sort of the model release time and you sort of see the capability slope and it sort of hit exponentially. Um, but one of the things about the most this this chart that's like extremely clear is really how fast you guys are releasing these models. It's compressing the the cycle time from years to weeks and it's sort of like the self-reinforcing mechanism. You can sort of feel a little bit um and and and we're sort of seeing this in product like I think from a user like we're we're using a ton of claude. Um like the product velocity you're shipping out is pretty insane. like how are you designing this sort of like self-reinforcing sort of dyn sort of mechanism into product development? >> Yeah, it's been an interesting journey because uh you know I got there in 2024. At the time we had about 30 product engineers total. Uh we had cloud.ai which was our sort of web version of cloud. We had no mobile apps. Uh we definitely didn't have cloud code or co-work or any of these different pieces. Um and I think the thing that I realized was you know we built claude the first version of the website before the models were very good at code you know this is still like cloud 3 days >> and then we trying to build like a whole like infrastructure on top of it but the foundation wasn't actually that AI friendly and so a lot of the work last year it was kind of painful because the second half of last year we didn't ship as much as you know I would have liked us to but it was because we were laying down the foundation for this year so that was sort of for example basically all of the cloud products now at their foundation use cloud code. The co-work uses cloud code. Even cloud AI when it's doing more of the like spreadsheets and all these other works is using um our agent SDK which is basically cloud code under the hood. But to do that whole migration was a big project but it unlocked a lot of velocity this year. And so you know what it probably looks like from the outside is like they were pretty slow and now they're releasing like a thing every single day but it was really the culmination of like a lot of of laying down that ground. And what's interesting is I'm hearing more and more from companies that they are doing that. So one like I won't name the company but like very popular um sort of like AI native product like you've heard of it. It's like a a big product. Um, I was talking to somebody there and they basically spent January and February pens down on any new feature development and rewriting all of the software, you know, for like being more like both the product being more agent native, but also being one that could be iterated on more easily with cloud code. And, you know, everything I ever learned when I was studying computer science is like you don't rewrite systems. You know, if you ever read like the mythical man month, there's this whole thing around like the second system syndrome where everybody's tempted to be like, well, we, you know, there's all these warts in the first version. We'll get it right the second version. And it's like six months later and you've just like recreated different problems and lost the all the like knowledge of the first >> attachment you have to to the prior one. >> Exactly. You've done that. I actually think that that is now broken and you can actually do it in uh like it actually might make sense to rewrite large parts of your stack with these models. Um, and and definitely within labs, which is what I work on now, there's products that we've like re like learned a bunch and rewritten, you know, two or three times before we even think about launching. And so that's a weird change and that's not a that's not a a calculus that I think most people are ready for, but we did it a lot inside Anthropic and I'm now seeing many of our customers attempt more of this. And you know it it it has its you know pitfalls as well but I think overall if you're finding that your limiter like is not the ideas and it's not the models it's your existing infrastructure existing product like it might make sense to sort of scaffold your way around it and replace it with something that feels like it was you know coded in in 2026. >> Yeah. Yeah. That's super interesting. And you mentioned a little bit about cloud co-work. I think we're starting to see these new product surface areas like you're like you're talking about and it's starting to consume more than nontechnical workflows like I think like cloud code everyone's very familiar it was very high quality model but now like cloud co-works capturing all like the marketing finance sales all these different sort of um workflows like do you feel like basically the world's all every workflow in the enterprise is ultimately it's just going to boil down as code and and really the the thing to really focus on is like UI UX like the products like the interface like how do you actually get like tune it to the to the workflow? >> Yeah. First of all, I have to give Jade credit. She is the most prolific feedback giver of anybody I know. Like she will write me like 10page things like here's why co-work sucks. Like here's like the thing thing that what you need to fix. It's just great. Like I I always love the the feedback. Um but I think it's yeah it was really interesting. We had an insight maybe halfway through last year. you know, a lot of our focus has been on enterprise and on coding. And I don't think we had even in our heads like fully combined how much those two things are actually related until um one of our engineers Jeremy was working on a project to um basically give claude like even within cloud AI the ability to write and run code not just to solve like to make apps but to solve other problems. And we realized something really interesting like the way we've been approaching for example spreadsheets or PowerPoint was all right well Claude can you know use the computer and move the mouse and click things around and that's like okay but it's like pretty slow. Um Claude's not that good at it you know reliably. Um and then it also sort of involves you having to like just watch like the computer use. I was talking to a founder of a company that does uh UI automation and they were like uh the companies that they would deploy it would report that they actually didn't save any time because the workers would then just watch like the their AI use. It was like so entertaining to do and I was like okay. Um so our the alternative idea that Jeremy came up was like what if we had um uh you know Claude write code in order to generate presentations and then we formed like kind of a strike team around this. This is actually where skills came from because we realized out of the box claude is like okay at making a a PowerPoint but then you give it a skill around here's how like you know decks should look like here's the design aesthetic here's the kind of code you should write and then make that a package that is reusable all of a sudden it was producing really high quality um presentations and that was the real unlock where we realized a lot of it I think does boil down to code whether it's uh you know doing research across different sources that's just a series of like agentic uh you know retrievals and parsing, generating presentations, modeling in Excel. You know, we've done a lot of work with Excel now. Um, so yeah, I think a lot of it, not everything, but a lot of it does boil down to procedures that can be coded. And then it's about do you have the right data connectors? Like that's a huge piece that we still need to tackle. And then have you created a product shape that people understand like the the difference in like co-workers been adopted like crazy. It's actually the underlying capabilities were there months before, but they weren't in the right shape. And then in the right shape, people just got it. >> Yeah. like the right ergonomics of it to to get it going. I have to thank you for that because the amount of time I don't have to like stay up overnight to like put presentations and IC memos together and everything thanks to co-work. >> Um but uh well I guess in the room today you have a lot of builders here and I'm sure they're all wondering if you were to build Instagram today in the AI age like what would that look like? Yeah, I think there's maybe like two parts to that question, which is what would building it have been like and then like how would we build it differently like in terms of the actual product. Um I did an experiment a couple months ago where I was curious cuz the timeline of Instagram is um Kevin and I got together um and we were working on a product called Bourbon which was kind of the the precursor to Instagram and it had some shared roots but it was it was different. Um, and then we worked on that for about 6 months together. And he had been working on it sort of a solo foundary for even longer, like probably another 6 months or a year. And then we pivoted to Instagram and then 3 months later launched V1 because we were able to build on so much that we had already built. >> Um, I did a a a test. I was like, how quickly could Claude recreate Bourbon in its final form like before we shipped. Um, and it was basically two hours. I was like, gosh, like that year. But then it was actually a really interesting process of reflection because you know we didn't know what the end point would have been like at the beginning. We had to do the building and iterating and showing it to people and realizing that it wasn't right. Somebody the metaphor somebody gave to me is like the difference between like Netflix shows that drop all of their episodes all at once and then ones where you get to like see it unfold week over week. And I think that that a lot of true about product development like the the journey is as much uh part of the process as just like knowing what's done at the end. And even today, if you ask cloud for like, hey, I want like a location check app with these features. That's not you're not making a lot of decisions, you know, I was um listening to Ted Chang, the sci-fi writer, and he has this really beautiful like piece about creativity where he was talking about like like creativity is about a lot of choices. And he's like, "Look, if you somehow gave a prompt to the AI that like in which you were expressing like a thousand choices and then it outputed something that was basically the output of those thousand choices, like yeah, maybe that would like reflect as much as I think human creativity is in like a book that is produced." I thought it was like an interesting thing. But I think the same thing is true on product development, right? You can every incremental bit that you add into the prompt is going to be more and more your product. But I think the only way you'll get to the insights of what to even put in there is through the experience of actually building. So I think right now like a lot of the building would have been much faster. But I think there is still a natural human timeline of like all the choices and findings and things that you need to invest in. Like I think it's actually really beautiful. It means that there's still a lot of human in entrepreneurship that matters. So I think that that's huge. Then the product itself is interesting because like the you know Instagram B1 had like very little smarts, right? It was the feed was chronological. So sort by time easy. The popular page was uh as people try to reverse this algorithm so much. It was actually like pretty simple. It was like it was a decay of likes over time was basically it was like it was actually not very fancy. It was it was basically that. So I was like as your things got likes then the likes would kind of drop off. Um that was it. That was not machine learning. That was like very very simple. But over time we added more capabilities. And um it's like interesting to think about like what would have been valuable to have in there in the first place. Like maybe some a powered editing, but like actually it was fun to kind of mess with your photos. That was part of the joy of it. Probably the most important thing would have been helping you find your community because that's like ultimately what made Instagram work was that people found their subgroups of car enthusiasts or you know people in a different geography or people interested in a certain kind of photography. Um and that took eventually us building a machine learning team. Actually, you'll hear from Hodrigo next, I think, was like really instrumental in bringing uh like machine learning to Instagram. Um, but you know, it's more of a sort of add-on. I think the core doesn't need a lot of AI. >> Yeah. Yeah. Well, so now your role has shifted from CPO of Anthropic to now leading Anthropic Labs. Um, what changes like what what's what's different? What's the same? What is your vision for Anthropic Labs like? >> Yeah, this really came from, you know, I was I was checking in. One thing I really admire about anthropic is that there's a culture in there of, you know, you are not your role and if there's like a different thing that you could be doing inside the company that is more sort of attuned to where you are, you know, you should go do that. And so, uh, I saw this uh, um, uh, Sam Mcandish, who's one of our one of the co-founders there, has like hold held four very different roles than even the time that I've been ananthropic. you know he was CTO then he gave up and he was like working more as an individual contributor and now he's chief architect and he's like inhabited all these different roles so there's like precedent for it and so late last year I realized the or has grown it was only like 400 people you know started at 30 you know it was like 10x in two years um we had like a portfolio of products a lot of external customers my day was effectively like meetings after meetings it was very hard to do like 0ero to one creative work and like at my heart I'm a builder right and there's never been a better time to build with these tools. And so the disconnect between like what I was doing and spending my time on and what I love doing was just growing. And it was only going to get wider. Like if you fast forward a year, it's like, well, if we keep doing our jobs right, we'll have more customers, so more external meetings and more products, which means more like context switching and a bigger team, which means more management. It's like none of these are going in the direction of the things that I want to do. And so um I realized what I wanted to go back is basically starting new things from scratch. Um and at the time we were rebooting labs. So Labs v1 which was back in 2024 uh generated cloud code. So you know like if it only had done that that would have been you know a huge uh success. Also MCP so model context protocol came from labs. Also cloud skills which I just talked about is like what powers a lot of our office work. Um it's like an amazing set of things. It really shows that in a lab that is working as quickly as enthropic having a team that is focused not on the next product rev but like what's coming down 3 to 6 months what's like not really possible with the current models but should be if you improve the models like those are all questions we ask within labs and so the two came together really well we were about to reboot labs um and I was looking for like my next thing and then those two things combined um really well but it really like in terms of the vision for labs it's really like pick where we are today and then push some sort of um thing to its extreme. So you know right now we we talked about the meter charts like right now you know the models are capable of hours of work but most people use them for minutes of work still how do we bring that sort of ability to do hours of work to more people that's one you know extreme you know right now you have like one you know a bunch of conversations with cloud what would it mean to have like one perpetual conversation with cloud like how would it get to know you better so all of these different extremes is really what we're pushing and hopefully showing that the models are not there yet because then we can be sort of informing the research cycles that will then get us there in the next release. >> Yeah. How do you cultivate um a culture at labs that allows for such experimentation and creativity to like sort of brew a little bit to have like the cloud codes and the MCP and everything to to have the room and space to discover in some way um and not be sort of pulled into the product uh and research sort of timelines in some way. I think it's it's actually really related to to the investment side, which is you want to give people flexibility and and the ability where like they they're not like they're not seeing success or failure of any individual product as sort of existential to their like time at anthropic, right? But you also want to have accountability. So balancing those two things is really important. So we run a process where every two weeks uh every product that's being worked on in labs basically gets a like funding or like defunding decision. And it's just stressful for the teams. But I think it's a good stress, right? Like it's it's accountability around what did you do the last two weeks? What are you working on? Like have we learned everything that we need to learn already by this product? Maybe we've learned it and we it's time to say great, let's take those lessons. Let's broadcast it to the company. Let's go work on something else. The thing I mostly feel is the opportunity cost. Like we have very talented people on labs. Making sure that they are always focused on the next thing that could be the next cloud code is really important. And if something turns out to not be that useful learning, let's like wind it down and go on there. And um but uh it was really valuable actually. We rebooted labs and within four weeks we had already defunded two projects that had like started up and were exciting and then kind of hit a wall and like it looked like they'd kind of run their course. I think having set that early as a precedent was really important >> and maybe just keeping a catalog of it and maybe the timing when timing is right to then pull it back out again >> which is a great example actually in the first version of labs we had a product that was uh using computer use so like cloud using your computer um to do a couple different things. One was teach people how to do something on their computer like Photoshop. Photoshop's pretty complicated like what if you could say how do I you know edit a mask in Photoshop and it could go and show you the first time. Um, we realized the problem at the time was one, the models weren't good enough yet. And two, they would eventually solve the problem, but often not in a very linear way. They'd like click around, open menus, like maybe Google for it, come back, and you're like, this is not an educational like it solves it, but in in a way. So, we actually wound down that project and say not right now. But then when the models improve, we actually now just actually two weeks ago launched computer use and co-work. And that was, you know, almost a year after we had done it in lab. >> Yeah. Yeah. No, that's awesome. Well, I want to bring it back to your roots uh from Sa Paulo and you have a lot of Brazilian founders here today and a lot of uh company leaders. Like what's some advice you would give them in terms of where should founders be building in Brazil and and what should companies be positioning themselves to do to to make the most out of this AI wave? >> Yeah, I think there's a couple of things that make me really excited about sort of builders in in Brazil and internationally, but in Brazil specifically. Um, one is it's never been more possible to get further without you know having to go off and fund raise like go through a lot of like the hiring like a lot of these pieces become important but can start uh much sooner right you can have that first prototype with cloud code you actually launch a first version with cloud code um or some of these other tools so I think that's very exciting from a like ecosystem perspective to have more proof it also puts much more uh sort of power in the entrepreneurs's hands where they can fund raise off of hey I've actually got my first users like I actually have a thing built out not just I um you know I have this great idea. That's number one. And number two is it goes back to your question around sort of what the role of the builder is in this world. Like I think understanding a market, understanding industry has never been more important. And what I get excited is when I hear from entrepreneurs in Brazil that are, you know, deep on the medical side or the education side or the legal side or understanding how government like could work better. And those like very deep regional specific insights are ones that are like the furthest from the model capabilities today and the actually the most defensible still. And so that gives me a lot of excitement for the next couple years of how does things get built. They get built more, you know, nimly. They get built more cheaply. They can launch sooner. And the meanwhile, they get built with a lot more deep insights as to what makes things a regionally specific thing that actually can scale. So it'll be very interesting to see the next. >> Yeah, I know. It's super exciting. And maybe to to end it off, maybe even on the enterprises side, like large um uh companies in Brazil, like what what can they be doing to position themselves better working closely with Anthropic probably and um what what else could they be doing to help drive the economy forward? >> Yeah, I think there's a few things like one is being really open to experimentation and tools, which goes against the sort of like typical IT procurement of like we settled on this one technology and it's going to be our like stack for the next 10 years. Like it just doesn't work in AI and so >> lean into the rate of change. Yeah, leaning into it and being able to say like, "All right, well, maybe not everybody's going to get co-work tomorrow. Maybe we're going to pilot it in this department, but you know, meanwhile, we're also going to try a cloud code pilot." And like I think those are the ones that we've seen really succeed is like that experimentation culture involves letting go of a little bit of control, but know that, you know, I think that's how you stay um stay competitive and really like lean into that things are changing quickly. Like they're going to continue to change quickly. You have to adapt um and use your advantages, but also be open to this like the rate of change. But I think it's especially important for for these larger enterprises. >> Perfect. Awesome. I think we're we're there. Um, thank you so much. >> Thank you. >> We now have a coffee break. We'll be back at 10:30. Heat. Heat. You guys Heat. Heat. Heat. Heat. Heat. Heat. Hey, hey, hey. Hey, hey, hey. Hey, hey, hey. Hey. Heat. Hey, Heat. Heat. Heat. N. Heat. Heat. Hey, hey. coming back. Come on. Hey, Heat. Heat. Heat. Heat. N. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Hey. Hey. Hey. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat. Heat
