The Rise of the AI Economy and Digital Labor with Salesforce CEO Marc Benioff

This is the biggest, most exciting transformation that any of us have ever gone through, and we are just at the beginning of it. It's a step out of enterprise software and the traditional applications world and into digital labor. You're really looking at three to $12 trillion market opportunity, maybe bigger. We've given birth to some new species. It's kind of a scary thing. We've only seen it sci-fi. When you look back at previous technological waves, what feels different about AI? Speed. This is happening much, much faster. The huge amount of innovation in AI has been open source. Everything that was true in AI a year ago or six months ago is not true today. If you weren't paying attention, you got left behind already. Every company and all of us are all gonna have to shift into an incredible new world, and it's wide open. Mark, welcome to the show. Thank you so much for being here. Oh yeah, it's great to be with you guys. It's great to see you again. Yeah. First question we have is, you've been in tech for a long time and you've seen a few big tech waves. You've been at Salesforce. You founded over 25 years ago. Before that, you were at Oracle. You've seen the rise and fall of the sort of dot-com bubble. You've seen cloud. You've seen crypto. Now you're seeing AI. Do you think the hype around AI is justified at the moment, or do you think we're maybe getting a little carried away? Oh no, this is the biggest, most exciting transformation that any of us have ever gone through, and we are just at the beginning of it. And when you look at it on a financial basis, it's a step out of enterprise software and the traditional applications world and into digital labor, where you're really looking at three to $12 trillion market opportunity, maybe bigger. And this is just an opportunity where every company and all of us are all gonna have to shift into an incredible new world. And yeah, it's all driven by AI and what the future of AI looks like, and it's wide open. Yeah, yeah. When you look back at previous technological waves, what feels different about AI? Speed. This is happening much, much faster, and that's always been true for AI, and I can kind of give you my assessment of why that is. Yeah, please. So I think when you look at the last few decades of AI, things have gone very fast, and we're moving into this incredible technology, and where AI is right now is amazing. But one of the reasons that it is going fast and has gone fast, and especially in the last, call it 10 years, as we kind of shifted from deep learning and these predictive models and machine intelligence models, and then we kind of crossed into prompt engineering and then generative AI, and now we're moving into agentics and soon robotics, that that moment of, you know, that we have all these next generation models, we call them large language models now, but of course we've always had models. And then we also have new reasoning models as well. So that's a huge breakthrough that's really come together in the last 12 months in a huge, incredible new way. And we're also all anticipating these kind of world models, which are the ones where we're gonna have all this kind of incredible new sensory stuff. So it's going fast. Why is it going so fast? One of the reasons why it's going fast versus things that we've seen like in other industries, like you look at maybe other technology companies or even biomed companies or pharmaceutical companies is because the huge amount of innovation in AI has been open source. And we've really seen that, especially starting with deep learning. And as that came out of a lot of the big universities, all of a sudden it was like very clear to everybody exactly what was going on and what we could do. The vision, the early vision models, these kind of very beginning, just as we kind of stepped into prompt engineering, as we stepped into the LLMs, that this open source phenomenon, this is really different. And so it lets us go much, much faster. And even in the reasoning models, it's just letting us go faster. Like everything that was true in AI a year ago or six months ago is not true today. It's like, if you weren't paying attention, you got left behind already. So it's like, oh, you have to really pay attention to what's going on. And there's a lot of nuances, but it's going fast because it's all open source. And a lot of these AI researchers are really into the open source phenomenon. They were in universities, they were professors, they were post grads, they were coming into this and posting their research and it's out there. And that is what is cool. And so you have speed, you have a lot of speed. And it's also turning a lot of these things into commodities. So, as we're going up the chain and you can see it for sure in the LLMs in the last 48 months, that it's kind of, they're all one feature or two features ahead of each other, but they're all about at the same level. And it's a big commodity, or then you can have a big breakthrough like we saw with DeepSeek, where all of a sudden, oh, this is a real change from the traditional transformer model to a new kind of model that's much cheaper to deploy. And all of a sudden I'm like deploying at 90% less cost. And then my friends were very competent initially are like, this isn't true. And then, oh no, it's actually true. And we didn't realize what they had done, but they open sourced the whole thing into an MIT license, which means everybody can just use it. Really cool. And that is not how any other software has really come along in the last, call it, you know, three, four, five decades. And you could see it like on your phone. It's not like iOS and, you know, it's an open source model that everyone is like working on. It's like proprietary and proprietary teams that are working in those things, the technology is just gonna go slower. So, you know, the iPhone still kind of looks like it did when we got our first iPhone. You know, and, you know, that's not true on AI. So we're going faster. Yeah. Now we couldn't agree more on the research side, Hassan and I both used to live like one street down from the University of Toronto, which is where a lot of the big AI research happened. And like every three months, like, you know, we'd meet these grads and researchers and Hassan was working at a tech incubator at the University of Toronto. Every three to six months, we were just seeing something completely massive. And I think the pace is like not just kept up. It feels like it's going up significantly right now. So couldn't agree more with you on that assessment. Yeah. These Canadian universities have also added a huge amount to the AI set, Waterloo, all these places. Like you can really see, like, it's incredible what came out of Canada. And yeah, I heard it's the 51st state now. Is that true? Okay. Let's see, like the Carni, you have a new prime minister. Yeah, I've known Mark Carni for a long time. I'm like, it's like amazing to see him, like is now the head of Canada. Yeah. Incredible. Yeah, it's a crazy election. It was a crazy election. We did not see that coming for sure. Yeah. I don't, nobody did. Yeah. He didn't. For sure. He didn't see it. Yeah. Absolutely. So Salesforce has been building with AI for quite a while. Einstein's been around for a few years. We have AgentForce now. Can you tell us and the listeners a little bit about what Salesforce is building with AI and what customers should be excited about? Right. There's a lot to be excited about, but mostly right now what I'm excited about is the customer success. So when we look at what's really going on with customers and what we're getting excited about, we're looking at some of these stories that are coming out. This product's been kind of out there now for five or six months. And so we're really starting to start to see these scaled stories. And one of the first stories which continues to unfold is OpenTable, where it was augmenting their employees and now it's like extending to their consumer experiences. And this idea, you can even see it at Salesforce, you know, at help.salesforce.com, that instead of speaking directly to a customer service rep or trying to deal with FAQs, that I'm interacting kind of how I did with a large language model, but with my customer support department. And I can ask it questions and help resolve issues, maybe increase my licenses or, you know, change my subscriptions. And it's amazing. So for Salesforce, for example, we've already done more than 500,000 conversations with customers directly on help.salesforce.com. And that has been very exciting. And then we have customers who are deploying this, not just OpenTable, but other amazing customers. Like one really cool story is probably a product you know, Smartsheet, which is this kind of cool cloud-based spreadsheet. And they've put it right inside their product. So not only are the end users getting support and help, but you can increase and decrease your licenses and all of this. Another really great story is Remarkable. You know, the pad that you can draw on, you know, you can go there, you can start to work with the agent to get help. Like I did it myself. Like I need to replace my stylus. How do I do that? And airlines, you know, I just got back from a trip to Singapore, Singapore Air, who we've worked with for like eight years on working on their sales org, on their service org, on their marketing org. They're building an agentic layer over the whole company. And, you know, that's gonna let people get a level of service and support and even sales that they were not able to do it before. And Finnair, who has 11 million consumers, is also deploying. So that's cool. These are stories that are happening much faster than I would have anticipated. I feel I'm a much more in a race. I think that, you know, we have these things happening all over the world. This is technology that requires, you know, a level, still a level of implementation. They're using our data cloud. They have to get their data right. You know, they're shaping the responses of these agents. They're installing these guardrails. There's a semantic layer. There's an action layer. There's a data layer. And we're super inspired by it. And we even introduced, you probably don't follow what's going on with Tableau, which is one of our products in data analytics and business intelligence. It's a company we bought like eight years ago, but we have completely rebuilt the product. Last week at the Tableau user conference in San Diego, we launched the new version and it's a big deal. And I'll explain to you why, because it's not only the Tableau that everybody loves, the familiarity, the kind of functionality, but we've put it on top of our data cloud, which, you know, when we bought the company, they never had that idea that there was a kind of a client server, almost relationship where you had the shared data repository. So that's awesome because you're getting all the data from your company and all of our apps are in that data cloud and it's federated out to all these other data repositories in your company, like a Snowflake or a Databricks or a Redshift or a BigQuery or something like that. But then not only do you have the Tableau that you love, the data cloud integration, there's also an agentic layer on Tableau and agent forces there. So you have a triad or a trinity. You'll have the app, the data and the agent. And why that's important is I was with a customer yesterday. He was using a large language model to do some very sophisticated simulations. I won't go into the technical details, but it was like really fascinating and cool and you relate to it. But then I was like saying to myself, this isn't exactly what they wanna be doing because what they're doing is where they're working with this large language model, which is kind of like a text in and text out and maybe some, all right, let's say it's multimodal and they're like throwing stuff into it and then there's stuff coming out and they're so excited about the output and they can't believe it and they've had these wow moments. And then there's a hype cycle with, you're in this exercise and of course you're getting these amazing things and at some moment you'll be deflated and going, well, that answer isn't very good. Ask another, tune your prompt. But here's the thing, with Tableau though, as the front end to the large language model and you're interfacing, you're also seeing the visualizations with it. And that is really cool. And I think like for this one specific customer, I was saying to myself, they would be going faster and getting more excited and they were super excited about what they were doing with the large language model if they were using this Tableau, we call it Tableau Next because the repository would be getting filled with the data and instead they're having to load all the stuff into the large language model. And then the second piece is the visualizations are happening. If you know how Tableau works, these kinds of visualizations that can come out of it are amazing. So then the combination of these two things, you're just gonna go faster because the reality is that we're still the ones writing the prompts, at least for now. And we're still the ones who are like asking the questions, at least for now. So we're like saying, hey, I'm running my business and here's my quarter and here's my products and here's my geos and how's my business doing and where should I focus and what should I do and all of these things. But just having that then stream of text back or even a couple of whatever images isn't what I want. What I really want is I want that incredible visualization. So that idea, it would just cross a lot of bridges, right? Which is that, yes, I have conversational capabilities in my business. I can have digital labor. I can augment my employees. I can have more productivity. But in this case, I'm kind of doing something where I'm getting a level of enlightenment I cannot really get with any other capability. So all of these things are important. It's not like just one is important. Number one is employees do wanna be augmented with these things. Like we're augmenting our stuff personally and we wanna augment ourselves professionally. Two is we want more productivity where, hey, the computer's just resolving some of this stuff on its own. The person wants to change their password, improve their subscription rate. We don't need a human in the loop on that. And then the third piece is I'm running something, managing something that's incredibly sophisticated. I have a huge management team. We're on this. I really want the software to be running side by side with me. And I need a lot of information visually in text and images and video in every possible way. And I need real enterprise software expertise, security, sharing. So all of these things are important and can work. So this is a really cool moment in enterprise software as we kind of start to come out with the next iteration. It's awesome, Mark. I worked at Snowflake early on a few years ago and just like the data app and the agent-force combination you're talking about, that just sounds phenomenal. And if you can get the real-time visualization, that just, I think, makes the life of an enterprise user so much easier if you can see that impact, the data feeding in. And that's amazing to hear the changes you've made at the tableau and how it's kind of refining. Yeah, we probably ingested about one and a half trillion records from Snowflake last week. You probably remember in the early days of Snowflake, we kind of provided our API outs. Yeah, I worked on the integration with your Salesforce team early on, actually. Yeah, and that was an incredible investment we made. I think we made 1.6 billion on our Snowflake investment when it went public. And that idea was we helped Snowflake get going by bringing data into the platform. But today, Snowflake is giving us way more data back because they've been able to aggregate data from so many different data sources. And now our data cloud is federated to Snowflake. So our apps are able to benefit from that integration, bidirectional integration. And then when we're running tableau, right? So here we're running tableau on our data cloud, but it's federated off to Snowflake and grabbing that stuff and dropping it back in. So that's different than where we were, wouldn't you say? Like even just two or three years ago. And that's a whole different, it's not that people weren't using tableau on top of Snowflake before, but this is a much more sophisticated and much more exciting, you know, next generation of enterprise software. 100%. I was looking at like one of your release notes from the spring 25. And I saw you guys released this reasoning engine called the Atlas reasoning engine and some library of prebuilt skills. I thought that's like really cool, very powerful. And we're just curious to know like any customer success stories of folks who are already using it or like what powers that gives like an enterprise user on day one, if they just like open up their Salesforce and they're like, holy ----, we have this reasoning engine now available. Reasoning is the path to accuracy. So in large language models, we realize that because they're word models, they inherently have some level of inaccuracy, or I guess, well, you know, we've called them hallucinations or where things just don't really make sense because it's this word, that word, this word, and then you get your answer. And sometimes it works. And then sometimes it does not work. And we've all had that experience, right? Is that a fair characterization? Now we couple the large language model, the reasoning model. Reasoning model is different. Is the reasoning model is kind of takes on the prompt and then gets the answer from the LLM. Then it starts to try to diagnose and figure out exactly what its answer is gonna be. And then as it comes up with its answer, it then keeps that knowledge and it doesn't have some static dataset that's driving it like an LLM. It has something that's constantly improving. Now, I don't know if that's the path to AGI or not. Some people think it is, that these reasoning models are now running and they're stacking and building their own kind of, you know, almost like synthetic data sources. It's not exactly a synthetic data source, but it kind of, you know, you're stacking each answer up. So you're getting more and more knowledge. These two things together, the LLM with its kind of its dataset combined with the reasoning model with its dataset that's more dynamic, this is giving us a much higher level of accuracy. So when we work with a large, like we're working with a large healthcare company or we're largely working with a large media company, actually the largest media company is right here, it's Disney. So with Disney, you know, we're seeing very good accuracy rates like in the 90s because we're also coupling it with the reasoning model. And before that, I would say that the accuracy rates were more in the 60s. So it's a big leap in terms of capability when you couple LLMs and reasoning models together. That's amazing. Mark, I was reading one of your op-eds in the Time Magazine, and I think you coined the stem, the unlimited H. And in your like minds, I'm curious to know like, what do you think like with this new LLM wave, what feels unlimited? Is that like it's unlocking new creativity for us? It's unlocked more bandwidth, compassion, just productivity, what feels unlimited? Like I'm sure like the readers, like our listeners would love to hear that. It's not compassion because, you know, computers don't suffer, so they can't really be compassionate. So if they don't suffer, they're not gonna really have compassion. I heard that from a friend of mine many years ago. And I would say that when we talk about the unlimited H, it's kind of a scary idea. At least it's scary to me. Maybe it's not scary to a lot of people, but the idea that the computer and, you know, the intelligence gets more and more, and we move into these worlds where, you know, more and more of the coding, maybe even all the coding is being done simultaneously by the computer. And then all of a sudden, you know, the computer is able to interface with robotic systems or manufacturing systems and can make its own robots and rockets and goes into space and is self-replicating and building its own worlds, you know, in the future. And it's like some AI is out there now, you know, running some factory on some planet, building some robots, and maybe we're tapped into it somehow through some brain machine interface. And it's like, that's the unlimited H where it's just like, it's gone, that we've given birth to some new species. It's kind of a scary thing. I think it's like, it's, we've only seen it sci-fi, you know, and is that what's really happening right now? I think that there's a lot of people who think that. Our futurist is sitting over here on this couch. He was involved in writing these movies like Minority Report and War Games, or we saw the movie like Her Terminator. That's the unlimited H, Ed. I don't know if I'm ready for it or not, but it does seem like we should talk about it because it seems like we're all starting to experience what could be about to happen. I think the conversation is so critical, right? Like it helps us understand and prepare. Well, we know we're not there completely because did you realize that Zane said, oh, I'm going to start the recording now? Yeah, yeah, exactly, he still asked you to click. There was no AI that said, I'm going to start the recording now. Oh, I see Mark is here. I'm going to start the recording now. No, we still need Zane. Zane still has a role. He goes, no, I'm going to start the recording now. So that's how we know we're not quite there yet, right? That's not the unlimited H because we're limited still by Zane. He's one of the limits, but when we don't have Zane, then we become unlimited. Unlimited H. No, so Mark, like I think one thing you had mentioned, and as you can talk about this unlimited H and the coding changing, I think in your recent quarterly financial results at Salesforce, you talked about like you're seeing significant productivity gain at an engineering level and Salesforce may not hire a significant amount of engineers this year or the next year. I just curious to know, like, is that just you're implementing now LLMs across the board in your product org, engineering org, and that's where like the gains are coming and that's where you see agents becoming a key part of your organization now as an internal team as well? Right, so this is the CEO, Mark, now is, you know, number one saying, hey, did you see Cursor? Did you see Windsurf? Like we did it ourself. You probably know one of the first coding engines was ours, CodeGen, which was an LLM just doing coding. So I would say, number one, we have our own internal products called CodeGenie. I would say that number one is, yeah, these engineers should be getting more productive right now. This is a moment of productivity. So we were really saying 30%, our head engineering came and he said he wants 50% more engineering capability this year. There's a, you know, it's an organizational issue. We're not running, it's not five people in a startup. We're like a big organization. It's gonna require some management, change management. And yeah, we wanna target 50% productivity this year and not just there, but in support, we talked about also, right? And services that we provide, professional services for our customers. So all of these things together, we want all of these things to have more productivity, but where you also know that I'm hiring more salespeople this year because I still need to get to more customers and that this level, like think about the kind of explanations and capabilities that we're really talking about and the nuances and the newness of all of these things. There's still a lot of face-to-face in the B2B software world that I'm in. Mark, you've also talked about, you know, as a CEO learning how to manage AIs, like it's not, it's no longer just about managing employees. You have to learn how to manage a team of AIs going forward. Can you maybe give us a little bit of color around that as well? Yeah, there's no question that that's true. And we already talked about help.salesforce.com as a critical part of my support organization. I was just managing our sales team and sales org, and we're talking about our sales agents and our SDR capabilities as well. And another thing is like when I'm writing the business plan myself for the fiscal year, you know, you should know that this is the third year in a row that I'm not only just sitting with another executive writing the business plan for the year, but I'm also using an AI. So it's the three of us. It's me, this other executive, and the AI. And I will ask the AI questions like, hey, how does this compare to what my competitor is doing? Give me a letter grade on this idea. You know, am I considering all aspects of this, you know, market? Where am I missing out? And in some areas, even it doesn't know or isn't able to put things together. But in a lot of areas, it is. Like, I'll tell you that because I've been using the AI for the last three years, there's definitely things that I found that I would not have found. But even like yesterday, I came up with something where I'm like, this is interesting. I hadn't really thought about this before. This is a good idea. We should be focused on this. And it was not in any of the AI responses. I went back and looked, and I'm like, the AI still doesn't, it knows what it knows, but it doesn't completely know what it doesn't know. And coming up with something completely new or totally out of the box, it's good in some of those areas. I'm sure you've done it. But in some of those areas, luckily we still have a role. So Zain, you're still hired, and Hasan and I are still like, you know. Absolutely. Still adding value to society for now. So that's good news. I just have one final quick question. So Hasan and I know we're at the intersection of tech and media. You are at a very unique place because obviously you founded Salesforce, but you also own Time Magazine. With the amount of AI-generated content that's gonna be out there, it looks like very soon there's gonna be more AI-generated content on the internet than human-generated content. How do you think about the future of media and what our relationship as humans to media is gonna look like? Well, I think it's changing a lot. And then at some level, it's not changing. I own Time Magazine a Tuesday of last week, about one week ago, we got a call, urgent call. Please come to the Oval Office. President wants to talk Time Magazine. Oh, wow. So our editors and our photographers and our videographers all got on the plane from New York, flew down there to Washington, D.C., rolled into the Oval Office, and we had the photographer who did the 2015 cover and took a picture, did a long interview, long-form interview on Friday, published the cover on Friday, published the long-form interview, and there's nothing else like it in media last week. Yeah. There's no other press conference or anything else that's at all like what we produced. It was not AI produced. Yeah, yeah. And it was important piece of journalism. If you didn't read it, you should read it. It's like, this is worthwhile. And that's important that there is that level of transparency that we're doing the long-form, that we wrote an article about it, but we also did the long-form interview. And AI was not on this one. Yeah. I don't know if we used AI to touch up the photo or not. I don't think so. I think we just ran the photo as is. And I think it was just like, here we go, you know? So that's it. I think in putting together a big list like the Time 100 and all those things, you know, we can put together our list and then say, did we forget somebody? Who else would you put on that list? How would you look at this? Is this comprehensive? Those are things the AI is very good at, right? Yeah. Because the AI is able to look at the whole world. And maybe our writers who are kind of in their own New York world, you know, don't know about something that's happening in Canada. Yeah. Or India or some other country. And they might say, hey, did you know Mark Carney is not on your Time 100 list? He wasn't, by the way. So the Time 100 most influential event, which was, you know, last week, Thursday in New York and a big Time 100 summit. And we published the Time 100 list. And now Mark Carney is the head of Canada. He's not on the list. He should be on that list. Maybe the Time 100, maybe the AI would have seen it. I don't know. Yeah. Hot stuff. Yeah, absolutely. Kind of funny. It's two funny stories, right? Yeah. So it's just new ways to look at these things. So Mark, like two more questions and we can wrap up. Like, you're always very curious, you know, like we get a lot of folks on the podcast and want to know like, how do you use AI on a day-to-day basis? Considering you're like CEO of like one of the most well-known company of our age, current age, like how do you kind of use AI on a day-to-day basis? What are some of your favourite apps which you're using? Well, I have all of the AI apps on my phone. So, and I'm coming between all of them. I don't know what your experience is, but my experience is they're all actually a little bit different. Some have different strengths and some have different weaknesses. And, you know, where I'm using it is asking questions about my business, like we talked about, but even like I have a friend of mine who just had surgery on her hand two weeks ago, and I can see that her healing is not going as well as it should be going. And I was wondering, it's an older woman, she's in her mid to late seventies. And I'm like, well, should she be using the hyperbaric chamber? Because that can provide a lot of healing in this environment, you know, and post-surgically, and what are the studies on that? And what are the rates and how often? And yeah, I'm using it like that. I'm using it to ask questions. I'm also just asking it a lot of basic questions. I'm probably going to it maybe more than I am going to a normal search engine. So that's important. And I'm using it for the kind of things that I'm sure a lot of people use it for like editing and for summarization, or if I get a huge amount of content and I wanna have something a little summarized, that's important to me. But I noticed that it's still, you know, there's gaps and I'll give you one of the areas is Apple has CarPlay. I use it a lot and it has a summarization of like text messages now, and it's just not as good as it could be. So I think there's still gaps. And I think that where we'll be, this isn't a very aggressive statement, but where we'll be, even just two or three years from now are gonna be light years away from where we are now, that this thing is really moving with speed. And I went to a conference last week with a lot of AI folks, and wow, they all have a lot of anxiety around where AI is going. So that was extremely interesting. So maybe that could be like one thing we can wrap up with, like what are some things that you are very excited about, which considering you also went to this conference, you think are coming up in the next 12 months? Well, I think it's kind of what we talked about, the unlimited age. Like we can talk about augmenting our employees, making them more productive, providing new levels of light and with new next generation tools. But when some button, when we do get to that moment where Zane is not telling it to start at the recording, but it knows it's guiding us, it's maybe prompting us, it's giving us subtitles on the bottom of the screen. It's telling you to ask these questions based on what I'm saying, which is what's not happening right now. I don't think you have it set up like that. You know, and AI is more pervasive. You know, I'm in my office here and there is no robots running around. I have no AGI here. I, you know, we're still kind of in the un, we're still in the limited age. Yeah. We're still in a co-pilot age. We're not in the unlimited age. What's that? Like a co-pilot age right now. It's like an assistant to us. Oh God, that's the clippy age, the co-pilot age. Yeah. We're out of that age. We're definitely into the agentic age. So thank God we left that co-pilot clippy age behind. Yeah. That's the agentic age, I like that. Yeah, we're in the agentic age and we'll move into the robotic age at some point. You've all seen like there's hundreds of companies working on these humanoid robots and those potentially could be very cool with the models that they're building. And you've probably seen like the Aloha model out of Stanford and others and this idea that they can do like, hey, you're getting home and your dinner's gonna be made by this humanoid robot before you get there and house will be clean and your laundry will be folded. Yeah, absolutely. Not yet though, I guarantee neither one of you, all of your laundry is gonna be folded when you get home tonight, right? Yeah, yeah, we still gotta go do that, 100%. You are still in the limited age, but I hope that we move into the unlimited age. It'll be extremely, I mean, it's gonna be different. Yeah, absolutely. Like, we're looking at this stuff every single day and the pace of progress is pretty crazy. I don't know if you saw the AI 2027 paper. I was reading that about how AI can unfold over like the next 10 years and how it's looking with the US, the China, the arms race and having like an unaligned AI and so on. So I think it's an incredibly both exciting and scary time to be in. With that, Mark, thank you so much. This has been incredibly informative and educational for us. Any last words before we wrap this up? Yeah, well, we don't wanna move into the unlimited, unaligned age. Like, you just opened that door just briefly. You're looking at that paper and you're talking about, oh, unaligned AI. Yeah, unaligned AI and an unlimited age may not be the best possible combination for humanity. Yeah, no, absolutely. I couldn't agree more with that. Thank you, Mark, so much for this. Really appreciate this. Great to see you guys. Thanks so much for doing this. Thank you so much for this. All the best. Thank you so much, Mark.