Why Architecture Determines the Future of AI Innovation - with Aaron Levie of Box

Welcome, everyone, to the AI and Business Podcast. I'm Matthew DeMello, Editorial Director here at Emerge AI Research. Today's guest is Aaron Levy, CEO of Box. Box provides secure cloud content management and collaboration tools used by more than 100,000 organizations worldwide. Aaron joins Emerge CEO and Head of Research Daniel Fagella to discuss what it means to build an AI-first enterprise, from restructuring data workflows and federating unstructured content to creating modular architectures that can flex with rapid advances in AI models. The conversation also explores how companies can unlock ROI by making unsecured data searchable, reducing compliance overhead, and automating document-centric workflows across teams in regulated industries. But first, Emerge is very happy to announce that the AI in Business podcast is moving to video. You can search for AI in Business Vision to Value in Enterprise AI on YouTube, where you can see the video version of today's podcast. You can also find a link in the description to today's episode on your preferred podcast platform. That's youtube.com at Emerge AI Research. Again, that's youtube.com slash at sign E-M-E-R-J AI Research. Watch the full episode now and see how vision becomes value in enterprise AI. Also, are you driving AI transformation at your organization or maybe you're guiding critical decisions on AI investments, strategy, or deployment. If so, the AI in Business podcast wants to hear from you. Each year, Emerge AI Research features hundreds of executive thought leaders, everyone from the CIO of Goldman Sachs to the head of AI at Raytheon and AI pioneers like Joshua Bengio. With nearly a million annual listeners, AI in Business is the go-to destination for enterprise leaders navigating real-world AI adoption. You don't need to be an engineer or a technical expert to be on the show. If you're involved in AI implementation, decision-making, or strategy within your company, this is your opportunity to share your insights with a global audience of your peers. If you believe you can help other leaders move the needle on AI ROI, visit Emerge.com and fill out our Thought Leader submission form. That's Emerge.com and click on Be an Expert. You can also click the link in the description of today's show on your preferred podcast platform. That's Emerge.com slash ExpertOne. Again, that's Emerge.com slash ExpertOne. Without further ado, here's our conversation with Daniel and Aaron. So, Aaron, glad to be able to catch up. Hey, thanks for having me. Absolutely. We got a hell of a lot to dive into here, and I want to be able to start with the topic of enterprise AI infrastructure. I let you know off microphone. Audience-wise, we've seen this shift in the C-suite of the legacy firms from AI as a use case, where we're going to solve one problem, to what's the foundation I need to sit on? This is governance concerns, hardware, all kinds of things. When you look at companies transforming, and of course you guys are part of that mix, what are the components of infra that you think legacy firms should be thinking about? And if there's any order, I'd love to get your perspective. Yeah, it's a really interesting question at this particular moment. I'm going to abstract a little bit from infrastructure for a second and talk about maybe data. Because basically what AI did all of a sudden was I think it flipped the value of data on its head where we've always sort of had this general sense that data was valuable. But in practice, oftentimes for an enterprise, the more data you had, the harder it was to find information, the harder it was to synthesize critical business questions and insights. And so we've had this ongoing challenge. And sort of in the era of big data, we saw the first attempts at solving this. And then we saw with the data lakes and the lake houses, another attempt at sort of solving this, which is to flip the problem on its head, which is the more data you have, what if we actually could turn it into more value? So that's already been a trend that's been going on. Now all of a sudden AI emerges and we can now actually turn a much bigger set of data that we work with into insights. And that's the data that we run into, which is all of your unstructured information and your unstructured content. Yeah, PDFs and contracts. Exactly. And so the history of our space and our category is the more files, the more content, the more information you have, the harder it is to find things. The harder it is to get answers. The harder it is to find that needle in a haystack answer information within your corpus of data. So AI flips that on its head. Now we actually, with Gen.AI, you actually can get more value the more data that you have inside of an organization. But we often don't have architectures set up to actually go and deliver that. And there's a few different problems. One is your data is probably just sprawled out across dozens of different systems, maybe hundreds of different systems. There's totally different taxonomies and architectures and permission models across those systems. It's not very easy to sort of at this point in sort of state of the art to have an AI system that goes across all of your information sources. That's still just an unsolved problem until you have a federated index that has some way of kind of normalizing the data across different systems. You really can't have AI run across all of these disparate endpoints. Now, that might change at some point. And I think we're going to first see that maybe in the consumer space. We'll see how that gets brought into the enterprise. But I think the biggest question for most organizations that face, you know, in the enterprises is, you know, how is your data managed? How is your data organized? Do you have the right permissions mapped to the right access to information? Do you have your data in a spot where people can actually query it? Can you load AI models against it? You know, how are you actually organized in terms of managing your information? So that's kind of job number one. And then it's sort of layers of AI on top of that. Got it. So, yeah, I mean, setting the data table, you're talking about some things are new, which is, hey, maybe the more data we have, the easier things are, but also some things that are old, which is get your data house in order. This is 10 years back, Enterprise AI, early, early podcast. So that's still sort of the name of the game. It sort of opens the question for a completely different set of data. So if you're the chief data officer five or ten years ago, you were primarily thinking about the data coming in or out of your ERP system or in or out of your CRM system or in or out of your – Some analytics or payments or something quantifiable, right? That was sort of what your remit was. Now all of a sudden with AI, your remit is your contracts, your marketing assets, your financial documents, your intellectual property, the research and development data. So all of a sudden, if you're the chief data officer, digital officer, analytics officer, you might have like a literal 10x increase in the amount of information you have to be thinking about for extracting value and intelligence from your enterprise information. So it's a completely different ballgame in terms of now getting value from your data in the enterprise. Yeah, yeah. So certainly right that the remit of the chief data officer, also kind of a cooler position now with a lot more tools, a lot more capability of AI. And frankly, we just see more wielded influence by these folks, right? A head of AI seven years ago in a legacy firm was kind of like a pie in the sky, ivory tower gig, right? Or at best you're doing some limited ML models and some predictive analytics for a decision-making system. So it's now completely flipped on its head. You're waiting 18 months for your procurement data to get cleaned up manually by humans. Really, these were bad times. I was around. So now it's clearly much more accessible, way more power. And we're going to get into how legacy firms transform. But you're bringing up sort of an initial element of kind of if we want to start waking up the value of data and sort of turning AI into something and having an infrastructure kind of set up. Some of that is where those initial workflows can kind of fit in and where we can derive some value. You guys have a ton. I mean, you know, new kind of different flows in a little bit of what people might conceive of like an agentic sense within box and whatever. I'm sure there's a bunch of that. But when you look out, you know, you work a lot in regulated industries. That's a lot of our audience, by the way. Lifestyle FinServe, hell of a lot of them tuned in. Where do you just see often where the simplest ability to sort of wake up the value of this stuff is where you get those early wows and you kind of hook your customers Like what is that for you Yeah, there's sort of two big categories. So the first is generative AI is kind of giving you the first time you can talk to your unstructured data. And I think this is becoming more commonplace, but two years ago, it was kind of this breakthrough insight, which was, you know, And today, if you want to find – let's say you're in a company and you want to find the answer to an HR question. You first have to kind of guess like what is the – what's the query I should give a search engine to find the document that will have the information? So I'm like HR leave policy and I'm praying that that's a paragraph in a document that will then get me to the particular paragraph that I'm looking for. So the first thing we did was just like, well, actually, if you build a retrieval augmented generation system on a large amount of documents or content, now you just ask the question. It fans out, finds the most relevant chunks of text that answer that question. And then AI is basically the reasoning engine on finding that answer. So the first big breakthrough use case was like, well, what if we just let you talk to your data? So now you go to an HR portal and you say, what's our leave policy? And it goes across all the documents, hundreds or thousands of documents, and it finds the most relevant information. But the same thing is true if you're in a sales portal where you have a bunch of sales and marketing materials where you want to have a new sales rep coming in. They want to ask a question, you know, how do I best position this product for this customer? And instantly it's going to combine information from five or ten different documents to present you just that tailored answer exactly as an expert human would in your business. So the first one is I'm just going to drill into this a little bit. So you're saying use case one where you're getting these wows. And again, I think all of our audiences are thinking about where can I just get a pop where I can really see some productivity here. You know, accessibility is now so much greater and it's a big focus for our audience. It's just like, what can I get my hands on? Talk to your data. You talked about sales. You talked about looking at FAQ for HR related stuff. I imagine this could be contracts, whatever. Generally, I would imagine you really want those to be kind of hanging within that domain. I'm not going to say silo in the old school sense like we want to carve this data off. But clearly you want that model for sales not to be necessarily trained on HR and on legal, right? And so how do you think about breaking that out? How are you seeing smart firms set those slices up? Yeah, well, so for us, this was kind of the initial aha, which was – there's one slight caveat to AI right now, which is if you gave an AI system a million documents or 10 million documents or 100 million documents and you asked a sales question or an HR question or an R&D question, it's going to get, in some cases, it's going to pull in data that you don't actually want to answer that particular topical question. Because, you know, the AI doesn't actually have as much context as you do as a human. So it's going to look for what's the best match across that question and it's going to find anything that it can. And to the point where it might pull in information that's actually not the most relevant or most valid. So what we designed was basically a solution to just this problem. And you kind of said topics or domains or whatnot. For us, we have a feature called hubs where you create a hub and you link any information into that hub that's relevant to that topic. So you go to a sales hub. You go to an HR hub. You go to a marketing hub. and then you were asking questions on a particular data set that's well aligned to the topic that that person is coming in to ask questions about. So all of a sudden you kind of get rid of 99% of the kind of hallucination or getting confused with the wrong information types of problems that we've seen in AI because it's topic-based and it's sort of tailored to a particular use case in that organization. So that was sort of how we solved the problem. I think there's been different approaches. but we sort of first said, okay, you have to do some amount of pre-curation of information so that way we're looking at the most authoritative content on a topic. And then once you have that organized, anybody can ask any question they want and get basically the right answer back for what they're looking for. Yeah, and obviously we've seen this start off within a company before it goes outside of a company, right? Because if we have that 1%, let's say, who knows, maybe it's 10%. honestly if i have something that's wrong 10 of the time but it saves me 95 of the time every time it gets it right i don't care i'm using it every day um but but even with that degree you know with if it was one in ten it wasn't right we wouldn't necessarily show it to the customer that first wave of chat bots you know six years ago that flopped really quick because we were we were doing stuff outside we couldn't even handle inside right so you're seeing this start within talk to your data within your own teams we're looking at hr we're looking at sales we're looking at could be procurement whatever the case may be so that's a great kind of low-hanging fruit area What's the other spot where you're seeing clients in demos or actually bringing stuff to deployment and saying, okay, I get it. Like what is that for you? Yeah, so first one is talk to your data. Big breakthrough use case. The second one that just immediately became obvious was once you can have AI read a contract, look at an invoice, watch a video, listen to audio, then you can pull out the most relevant information from that data that sort of ties that information to a workflow. So, I want to have an AI, in this case an AI agent, go and look across every single contract in my business, pull out the most relevant sort of structured data from those contracts, and then let me automate or orchestrate a workflow across those contracts or across those invoices or across those digital assets or across those marketing materials. And so the thing that kind of just started to explode overnight was customers coming to us and saying, okay, I have all these contracts inside of these repositories. I don't know what's inside those contracts or I don't know what's inside those invoices or I don't know what information is actually inside of my digital assets and my marketing materials. So I want to be able to understand that and then I want to be able to orchestrate a workflow and then I want to be able to have dashboards and sort of low or no code applications. So we've put together that kind of suite of capabilities with the real breakthrough being once you can have AI sort of read that document, you can then automate really any business process that deals with content. Yeah. Just out of curiosity here. So over the last – I don't know. God, it's been a long time. We've seen sort of the document search and discovery space was way clunkier eight years ago, seven years ago. Obviously, it's a lot smoother now. we saw a lot of the early traction in terms of why people would jump on actually have to do less with efficiency and more with like compliance and risk related stuff. So, you know, you could say, hey, you know, here's a way to make this workflow astronomically faster by searching. And even if they could prove it, it's like it was like a little bit of a shrug in the legacy enterprise. But if you could say, hey, how much, you know, the LIBOR laws changed over there in banking. Jeez, you know, how much risk is sitting on in your contract? Right. And then that would be what would snap adoption. Are you sort of expecting that to still be the case in this wave? Probably I'm a little bit bipolar on this and I'll support kind of both scenarios. I think there's a sort of records management dimension, which is I do have to know what my liabilities are. I do have to know what sort of critical information is inside those contracts. That's a governance use case. That's a records management use case. Critically important, banking, life sciences, health care, et cetera. But I've had now quite a number of conversations with business executives. It's now even kind of a CEO-level conversation, which is actually if I could get insights from my data, would I be able to actually build better products? Would I be able to sell to customers in a better way? Would I be able to kind of accelerate the productivity in my organization because I could take a step that used to take five hours and take it to two seconds? Yeah. And I'm not doing that to save money. I'm actually doing that so I can go and deliver a better outcome for that customer faster. So a lot of our conversation has actually moved from kind of governance and CYA to actually, wait a second, maybe if I could accelerate that process or automate that workflow, I would have business insights for my data that would just turbocharge a part of my business that I just felt was maybe more stagnant previously. Yeah, we hope to see more of that, to be frank. I mean, the sort of conservative nature sometimes of larger firms sort of makes very predictable regulatory stuff sometimes be what makes them move. But we're hoping more and more of it is like, could this be fundamentally better? Yeah. Well, I think as we saw with cloud, as we saw with kind of consumerization of IT, it will start in pockets. It will start in teams. And it will start in smaller businesses. We'll see that, wow, there are these disruptive use cases that are emerging. and why is it that that smaller mid-sized firm is moving faster than the large company? And then the large company kind of wakes up and they say, hey actually maybe we haven taken advantage of AI as fast as we could and that kind of gets them to up their game from an innovation standpoint Yeah I think it really necessity being the mother of invention and enterprise necessity being the mother of even the smallest amount of adoption from our vantage point here. With that said, you talked about younger firms moving quicker, adopting this stuff. You guys are in that herd, and we've seen everybody in sort of B2B SaaS do the AI rebrand thing. Not everybody's actually turned it into real new features, new products, whatever, but everybody's had to, at least by visage, you know, kind of get in line with the new AI wave. You guys are doing a hell of a lot of it. You're rolling out new products with new customers. When you think about what it takes for a company that didn't start with maybe AI in the very middle of its product offering, I'm sure you were doing it somewhere, right? But to level that up and be more market ready, to be able to win market share as this wave hits. What has been crucial for you strategically to make that happen? Yeah, so almost back to the first question, actually, architecture sort of drives all of this. So our architecture that we, you know, realistically, we kind of lucked into in the founding of the company. We didn't really know that much about software architecture, but, you know, we kind of took a first principles approach to how we wanted to build the platform. And so we kind of lucked into a very modular services oriented approach as we were building the foundations of the platform. And so the way that we thought about it was we're going to have a single file system, a single platform that stores all the data. And then we're going to have layers or modules or stovepipes or whatever, whichever analogy you prefer of services that work on that data. And so we've been building up for now well over a decade and a half in the enterprise context these modules for security and user permissions and workflow automation and processing documents to extract text. So that way you can search it or being able to do rendering of that content in your browser, which means that we have lots of different sort of permutations of that content. So I say all that, which is to basically say that the moment AI emerged or Gen AI, we had all of these capabilities or technologies that we could instantly leverage and to accelerate our ability to innovate in AI. So for instance, if you're going to do a large amount – if you're going to do retrieval augmented generation, what technologies do you need? You need to be able to store the data. You need to be able to process the data, extract the text from documents, put it in a search index, and then be able to have a vector-based search system to be able to find the right chunks of text and send it to an AI model. All of that work, if you were starting from scratch, would take months, quarters, years. It would take dozens of engineers. We had every component that I just laid out. All right. So born lucky here. Hey, the way we built our initial product and let us automatically do AI. Yeah. Got it. like strategically, you still had to think about how do you bring this to life, right? So for you, you look at what you got and you had some good stuff, right? You had good customers and you had good architecture. You then had to ask really serious questions. What do we build, goddammit? And who are we building it for? And how do we initially roll it out? How did you think about that? Because those are tough questions. Whether your architecture is good or bad, that could still sink your ship. So how did you go about that, Aaron? Totally. So partly I just wanted to make sure that we were self-aware enough to not take all the credit. All good. It's not just strategy. Yes. We're very good at strategy, but technically we bet it from strategy from many years ago. So that made everything accelerate. So then what we said is, okay, we're going to have effectively an abstraction layer that we're going to build that connects the content inbox and the tools inbox and the services inbox to external AI models. and we're going to build this in an open and sort of platform agnostic way. We don't want to put all of our eggs in any one of the major AI players' baskets. So we want it to be fully abstracted from any AI model. And we want to make sure that we have sort of this middleware system that, you know, a request comes in, a user request, an API request comes in. It knows how to process the services within the box architecture. And then it knows how to send that data to the AI model really as a reasoning engine in this sort of ephemeral capacity and then gets an answer back. And so we spent the past two years basically building out this platform architecture that just gives us an immense amount of flexibility and speed. So the moment that just a day ago, Gemini Flash 2.0 was announced by Google, we're already testing it internally. So within 24 hours, we can be up and running with a new AI model. We can see how well it performs. We can benchmark it against the other models that we have internally. that just gets us so much more from a velocity standpoint. Yeah, yeah. So it sounds like, you know, philosophically, that's how you approached it. So you're sitting on good architecture. You took this sort of agnostic approach of this API layer, et cetera. And now you guys are running with this. You're thinking a lot about what the future of enterprise AI is going to look like. And so are a hell of a lot of other SaaS firms that are trying to level up. So are a lot of AI native firms that are trying to level up. There's some of them that have had to kind of rebrand around Gen AI after they spent years with AI. But it's going to be a hell of a lot of competition and a lot of value, which you've talked about. Those of you, the folks tuned in, if you don't follow Aaron on Twitter, I recommend it. He's got some good ideas about where this stuff is going in the enterprise world, which Twitter often ignores, frankly, Aaron. It's a lot of the, it's the cool kid Twitter stuff of like the latest model. But it's like, where does this matter to like develop drugs or prevent fraud or like things like that? So you cover that stuff and you're thinking a lot about it. When you ask yourself, what's going to separate winners and losers of SaaS companies leveling up AI capability to try to meet the market and win market share in this era where this starts to become the norm? Waking up the value data becomes the norm. What for you just short and sweet is like the separator of winners and losers in that fight? I think it'll be determined by the companies that treat AI as an extension or an add-on of their product versus AI as something that's sort of built into the foundation and core of their platform. The thing that we stepped back about a year and a half, two years ago, is we said if we were to start the company from scratch, so if Box were founded in this case in 2023, would we think about AI as sort of this other thing that you sort of pay for separately and you kind of like bring it to the platform? Or would you basically be pitching your company as like, no, no, like AI is just in the core foundation of what we do. And it's sort of, it's in everything, sometimes more or less. But we obviously came to the conclusion that obviously if we were starting in 2023, we would not be pitching boxes like we do like the basic stuff, but then we also have a smart version, the whole thing needs to be intelligent. The whole platform needs to be intelligent. So I think the winners from the losers will be ones that actually take an AI first approach to building their software, to building their company, and not thinking about this as some kind of secondary kind of afterthought capability where we can just add it on for some part of our customer base or use cases. And we've seen this before, right? Cloud, you know, the companies that dove first into cloud and said, you know what, we're actually gonna have a cloud first architecture, they had just a humongous advantage in innovation and selling to customers. Yeah. The companies that dove first into mobile, that had a mobile first approach, and they didn't treat mobile as sort of the secondary way to interact with their software, we're going to see the exact same trend with AI. Yeah. You're kind of hearkening to these previous waves. And I think to myself, man, there's a lot of analogies there. You talked about, hey, even with cloud, it began in pockets. It began with the lighter and the faster moving companies. just in terms of clicking for the audience, because many of them have been around through these waves, but they're not quite thinking about exactly how they rolled out and how they're going to lead to successes. What are a few of those analogies to conceptually click for people? How, whether it's cloud or mobile or whatever, how those past waves will show themselves in similar phases or maybe you see different phases with AI. What are those similarities or differences to make it click for the audience? Well, I think the biggest difference, first starting maybe with the difference, is the speed at which AI has kind of just, you know, entered the zeitgeist. So when we saw mobile kind of emerge, it, you know, let's say the iPhone came out in like 06, 07. It was, you know, good kind of three or four years for general consensus that, you know, the BlackBerry trend was over, the flip phone trend was over. This was the new platform. So about four years or so for that to really kind of lock in. And you fundamentally had to buy a hardware device to participate. So you actually had to like go out and get a physical technology for you to enter the game of mobile. Well, with AI, with a couple billion or a few billion people on the internet, 100% of that sort of, you know, the internet's user base can already instantly start to use AI on a daily basis if they so choose So the accessibility of this technology which means that every employee of every organization is already doing AI in their personal lives They already on Chatsbytee asking questions They already on Gemini asking questions. They're already on Perplexity asking questions. So unlike mobile, where you could kind of sort of see this, you know, okay, it's going to take a few years. People have to buy the device first, and then they have to download apps, and then it's going to enter the enterprise. You know, basically overnight, once Chatsbytee got created, 100% of your employee base is expecting that AI is going to enter the workplace and it can now enter through the existing interfaces and the existing devices that they already have. So I think that sort of ratchets up the expectations. And then it doesn't help that you've got NVIDIA just blowing out every number, which means that that puts the pressure on even more to say sort of what is your AI strategy? How are we going to adopt this? So I think it basically behooves whether you're a CEO, a CTO, a CIO, anyone in the technology space to make sure that you are kind of reinventing your business, your team, or your operations in an AI-first way. Man, well, Aaron, I hope you saying it will make it click. It's been a long time. Beating that war drum has been a long, long time. But at some point, necessity will be the mother of adoption. And it will probably happen, like you say. Enough people are at home getting massive amounts done to order their kids the latest soccer cleats and get their groceries done or whatever other things they're doing. And they're going to be in a certain workflow and just say, I'm not going to deal with this anymore. I mean, I mean, there's always signal in like, what are the college students today doing? And then when they enter the workforce, what is their expectation going to be? Oh, yeah, there is there is simply no way that a college student going through, you know, college today using Chachabit to do their their homework and study is going to enter them the office and be like, wait a second, I don't have a I don't have an AI system to talk to to help me accelerate my work. Like, are you guys like running? It will be like entering the workforce, you know, 15 years ago and you have to do faxes. So we're going to see a demographic shift. We're going to see this hit us in our consumer lives. That's going to bleed into our professional lives. This is happening. There's a lot of signal in what college kids are doing is super quotable. That's a great one. I'm going to end on a little bit of a fun question here. You've gestured about sort of AI being a bit of a religion now. And if you go on Twitter, it almost is these days, Aaron. So I'm in agreement with you. You've also made jest that, You know, if you're an AI startup, you don't have your own nuclear power plan. Are you really an AI startup? And, you know, you're making light of, in a funny way, something that is the case, which is that the race for this stuff is dead serious. And more than a handful of years ago, we've had Ben Gio on and Stuart Russell and some of these folks who, you know, Ben Gio famously up until very recently really still just sort of saw AI as a set of tools and whatever. And now is really, you know, as an eminent scientist and a relatively well thought out guy who I don't think has a religious around AI is asking about loss of control and where kind of general intelligence might go. Not something you talk a ton about, but I know our audience every now and again will put it in surveys. Like, what does this mean? Or like, do other smart business leaders think about this? Do you have a take at all around strong AI and its emergence? Do you kind of laugh it off? Do you have a take on it? Would love your perspective on this. Well, I think the race for AI and let's say maybe AGI is, you know, maybe the most serious technology race of the 21st century. So it would be hard to imagine a more important technology race for any country or company or team. And I think you're going to see this will be the defining geopolitical dynamic of the 21st century is, you know, which companies have the energy sources for AI, Which countries have the manufacturing kind of prowess to actually deliver the chips needed for AI? Which country has the engineers and the talent to build these AI systems? Because this is going to tie to our defense industry. It's going to be how you win or lose wars in the future. It's going to define the manufacturing capacity of your country with advanced manufacturing robotics. It's going to define the logistics networks of the future with self-driving. And it's going to define healthcare outcomes, economic outcomes. So it is the singular most important technology trend that will shape the 21st century for any country that sort of leans in or out. And then I think there's a lot of really great thinkers on, okay, so what are the risks? How much can we control AI? Do we have the right guardrails in place? I lean a little bit more toward we're still very early in this evolution. So we probably still want to focus on progress and driving more progress as opposed to kind of clamping things down. Certainly my opinion could change on that in five or ten years from now or sooner if there's evidence that these risks are emerging. And I have a lot of respect for the people that do feel like there's more risk. But I'm more on the side of right now you actually just want to go, go, go in a thoughtful and safe way. But this is a race and we have so much potential for what this technology can do. There's no doubt about that. And I think even the folks who are detractors in terms of more near-term fears are fully aware that there's so much good to be done here. But with that said, what would be like a line hypothetically that could be crossed where you'd say, you know what, okay, maybe we think about it. You know, is it AI takes over a factory and starts building something? Sometimes people say, oh, if it can recognize itself in a mirror or if it can build itself, if it can iteratively construct itself. We're already seeing a good deal of that stuff already. What would be a line where you'd say, OK, maybe at this point we maybe start thinking about what we need to do to reel this stuff in or consider the control problem, if you will? Yeah, great question. I haven't actually thought enough about what that red line would be, probably because I think it would have evolved. If I had defined it three years ago, I'm sure there are things in the 01 model that we can do that I would have said that's the red line. But now we have 01 and we've got the right guardrails and we kind of know that, no, actually you can apply this to only productive use cases. But certainly some element of AI doing many, many derivatives of its own work without any sort of user input telling it to do something where there's some evidence that that work is sort of unstoppable and you can't turn it off. You can't shut it down. But we don't have any evidence of that being possible today. And I think that many of the scenarios that people imagine that happening right now are still a little bit on the science fiction side. But I remain totally curious for any evidence that shows otherwise. But right now I'm in the mode of let's get the productivity gains and continue to advance this technology. I'm with you. Yeah, I think the frog just gets a little bit warmer is how it goes. And hopefully you'll be right here and that for the most part, it'll be kind of innocuous as we go and we'll get a chance to connect. No, the boiling frog is absolutely the singular risk of everything that I just said. We'll look back on this video and be like, well, obviously. What the heck? Yeah, exactly. But we don't have crystal balls. Anyway, Aaron, look, I know that's all we have for time, but it's been a real pleasure. Thanks so much for being here on the show. Thanks, man. Appreciate it. Wrapping up today's episode, I think there are three critical takeaways for enterprise technology and data leaders driving AI transformation across their organizations, tuning into today's conversation with Aaron and Daniel. First, data architecture is the true foundation of enterprise AI. Without unified, well-governed, and accessible data systems, even the most advanced models can't deliver meaningful insight or automation. Second, the shift from structured to unstructured data but requires rethinking how information is indexed, secured, and queried. Building modular systems and domain-specific data hubs helps reduce noise, improve accuracy, and unlock productivity gains across teams. Finally, AI can no longer be treated as an add-on. The organizations that integrate intelligence into every layer of their operations rather than isolating it to single-use cases will lead in speed, innovation, and long-term ROI. Interested in putting your AI product in front of household names in the Fortune 500? Connect directly with enterprise leaders at market-leading companies. Emerge can position your brand where enterprise decision makers turn for insight, research, and guidance. Visit Emerge.com slash sponsor for more information. Again, that's Emerge.com slash S-P-O-N-S-O-R. Thank you. Bye.