This Week in Startups - Box CEO Aaron Levie breaks down Box AI and generative AI’s impact on business | E1738

Episode Date: May 9, 2023

Aaron Levie joins Jason to discuss the launch of Box AI (19:13), AI as a platform shift, its potential effects on employment dynamics (25:53), and much more! (00:00) Aaron Levie joins Jason (4:12) Thoughts on the impact and pace of AI (6:54) Aaron's "Aha!" moment (10:02) Squarespace - Use offer code TWIST to save 10% off your first purchase of a website or domain at https://Squarespace.com/TWIST (11:33) Getting consent to use data from customers (13:50) Creating an AI model for early adopters (19:13) Aaron demos Box AI (24:33) MasterClass - Get up to 35% off for Mother’s Day at https://masterclass.com/startups (25:53) AI making the workforce more efficient (33:29) How AI technology affects morale (36:45) The non-obvious uses of AI (38:24) Hampton - Join the Hampton community today at http://joinhampton.com/twist (39:44) The Writers Guild strike in Hollywood (46:57) Copyright and citations (53:49) The challenges with Crypto FOLLOW Aaron: ⁠⁠https://twitter.co/levie FOLLOW Jason: ⁠https://linktr.ee/calacanis⁠ Subscribe to our YouTube to watch all full episodes: https://www.youtube.com/channel/UCkkhmBWfS7pILYIk0izkc3A?sub_confirmation=1 FOUNDERS! Subscribe to the Founder University podcast: https://podcasts.apple.com/au/podcast/founder-university/id1648407190

Yeah, AI's, AI's gotten pretty crazy. It just reminds me of, like, seeing the internet for the first time, and you're like, how does HTML work? And they're like, look, view source. And I'm like, okay. And they're like, okay, now put it over here and use this hot dog editor. I'm like, what? Hot dog editor.

You dreamweaver. And I'm like, okay, wait a second. And now, wait, you reload the page. Remember those conversations? Reload the page. That's really good. Yeah. It's definitely, it feels like mid-90s discovering the web and then mid-2000s of like,

We had Ajax. He had, like a bunch of s just came together. Yes. And it all just worked. This week at startups is brought to you by Squarespace. Turn your idea into a new website. Go to Squarespace.com slash twist for a free trial. When you're ready to launch, use offer code Twist to save 10% off your first purchase of a website or domain.

Masterclass. Learn from the world's best minds. Anytime, anywhere, and at your own pace. Get 15% off an annual membership to Masterclass at Masterclass.com slash startups. And Hampton, are you a startup founder or CEO seeking expert advice? Join Hampton, the private, highly vetted community for high growth founders. Get invaluable insights, connections, and support to accelerate your business at joinhampton.com slash twist today.

All right, everybody, really thrilled to have a long-term friend of mine, incredible entrepreneur, and part-time internet comedian. Aaron Levy is back on the program for his fourth appearance, just to tell you how long he's been around. Episode 224 in 2022, episode 389, 2013, episode 1173, February 21. And again, today, this puts you getting close to the five-time. Club. What's the record? Well, there are some journalists who have done like news roundtables on a regular basis. So taking them out, we do have a number of five timers.

Andy Ratcliffe has been on, I think, five times. Glenn, the CEO of Redfinn's been on maybe four or five times. But it's a rarefied error. That's a good club. All right. Sounds good. It's a pretty good club. And it's, but Brian Chesky joined the first timers club.

just yesterday. He did a great episode. And convicted felon from Fire Festival was on. Yeah. And that was our first convicted felon. If my PR team had known about that, I don't think I'd be on right now. So this is remarkable that we stuck this one in. We did.

We're doing the felon now. This is okay. All right. You know, Billy McFarland was like, I'm a super fan of yours. I've been tracking you since high school. and I am turning it around. I mean, I learned a lot of lessons and I'm like, what's he doing now?

Fire Festival too. Oh, okay, okay. He learned a lot of lessons. He learned a lot of lessons. Would you like to sponsor it? Okay. But he has $28 million in restitution. He's paid off $40K.

He's charging $1,800 an hour to do marketing for people. Yeah. So he's just basically like a marketing guy. Yeah. Who's two out of every, I think like two out of every three dollars he makes goes to the government. and restitution. He makes like 30 cents on whatever dollar he makes for the rest of his life. Wow. Okay. So 20, 30 years from now, it's paid off and then he'll be making profit.

Exactly. So you talk. I mean, it's kind of like a series D company in Silicon Valley today. Hey, oh, but I wanted to have you on because you and I were just talking about, my lord, AI is moving fast. Yeah. And Box, I was a, my poker game last night and somebody said, well, Box is the obvious best company to incorporate AI because I mentioned you were going to be on the pod. So you got a nice shout out there at the poker game. And I noticed you started sharing some demos. So I think just your general perception at the start here to level set of the impact of this technology and how quickly it's moving because it does seem like this was a, a very slow kind of grind and then very quickly became wildly impressive.

So your thoughts on that as a technologist. Yeah, I mean, I would characterize it almost exactly the same way, which is, and I think most people were really deep in AI, I would probably say the same thing. I mean, it's kind of like, you know, 10 years we've been, you know, incrementally making this progress. And, you know, you had the transformer paper five years ago. And it's like, well, you know, that's obviously a big unlock. And then you had GPT2 and three.

And, you know, I think we probably both. played around with the early versions of GPT, the GPT versions. And it was like, you know, super interesting in terms of, oh, that's kind of cool. A computer can do that. But there was like, I don't, I don't, I certainly didn't have an aha moment of like all the sudden this is going to be incorporated into all software. I thought it was like a really, you know, compelling, you know, demonstration of, of what we can now do with tech prediction and whatnot.

And then obviously chat chabit, kind of both the combination of, of its improved, you know, 3.5 model and just the right interface. you know, to capture the zeitgeist of, of what we can now do with, with AI and interact with AI. And so, um, so that combination, you know, obviously, um, you know, put us in a completely different conversation about where AI would be incorporated. And so, um, so we had, we'd been doing seven or eight years of work in AI, um, and, uh, that the challenge was every single use case the customer had, you know, I, I want to understand my contracts. I want to understand my film scripts. I want to understand my blueprint files. Every conversation.

we had was a different AI model that you had to narrowly, you know, kind of train and implement for that one use case. And so it never really scaled because it would be like, you know, be like if we were building on the web and every feature you wanted to build, you needed a different, you know, tech stack or CPU for that feature. Just would never scale out to all the different, you know, kind of scenarios we had. And, you know, with GPD 3.5 and now 4, it's like, it can just solve all the things. So it can do the contracts and the film scripts and the, uh, uh, and the press releases and the marketing documents and the blog post. And so it just understands everything.

And then you can implement it in these much more generalizable ways, which is a real breakthrough in terms of just how many use cases we can now solve with AI. So when you saw, you know, Gmail 2018, 2019, you know, predicting your next word or two, you know, okay, pretty clever. That's helpful. But it wasn't like an aha moment. What was your aha moment in the last couple of months where you said, okay, I need to, and I'm assuming this is the case, correct me if I'm wrong, I need to get the entire company dialed in 100% on this opportunity. Yeah.

I mean, the specific revelation was probably no different than anybody else on the internet is just you can give it literally any prompt. And, you know, at 99% likelihood, it's going to come back with a response. Now, accuracy aside for a second, and we can get into kind of how we solve that problem. just the fact that it will attempt to do everything was was sort of the mind-blowing moment. I remember, like, probably my first, like, big aha was, you know, you initially start with use cases, which are, like, these, like, very straightforward kind of facts, because, like, we're so used to, like, we go to Google and we type in a thing and then we get an answer. And so your first things are like, wow, like, it's really cool that I got that answer. So it was, like, really basic

questions of, like, you know, what's the better way to travel? you know, between, you know, X in my place and, and you're like, oh, wow, that's actually pretty cool. Like, I thought through, you know, you might have to wait at the airport for this long, and so maybe it's actually more efficient in this case to drive. And so that was pretty cool, but very fact-based. And then, like, when you start to give it things that are not fact-based, but actually things that it has to take two pieces of distinct information and kind of combine them and reason

through it. And I was like, that was really where I was starting to be, you know, my, my head started explode where I you know we were giving it like um so you know peter um Michael Porter uh sort of five forces framework which is this sort of competitive framework of how you think about suppliers and competitors etc and so so take that um take that concept and apply it to apply it to things that that no one has ever done a Michael Porter five forces forever like like in the history of the world and and it just did a really really good job like We were like, do Michael Porter's Five Forces on an airport?

And it was able to take, you know, one set of information it has and then combine it with a completely unrelated topic and basically completely exceed what even a human would probably have answered for that. And that was the big breakthrough of like, okay, so this is not just going to be this sort of question answer engine of raw facts. It's actually going to produce new information for people and then become really an assistant where you can just go to it and ask. ask at anything. And so that's, that was, you know, some more of the aha moments when you start to take and combine those, those, you know, two unrelated things. And you're actually leveraging the fact that it has a wealth of knowledge within it. And, um, and, you know, way more than any, what, what any person would be able to keep track of? Um, that's where you really get the power of it. And so then we said, well, what if we combine our data set with that? And, and, and then you start

to, you know, kind of create new use cases that just never before would have been possible. And that's, that's where we, you know, kind of, you know, sort of didn't like, through a full company pivot, but about as close as you do when you're 2,500 people. And we just said, okay, this is a big moment for us. We're going to go and incorporate this technology. If you can tell from the podcast lately, we've been doubling and tripling down on Founding University of launch. In fact, it's basically the future of our venture capital firm. And that's awesome because I'm working with a couple of hundred early stage founders really early and getting to see what tools they use. You know what tool they show up with most? They show up.

with Squarespace. They put up their first website instantly, quickly with Squarespace, and it's beautiful, and it makes them look like a million bucks. The thing you may not be aware of is that Squarespace, beyond the beautiful templates that make your company look like a million bucks, and that work on mobile. It's not just a pretty website. It is a powerful e-commerce platform now, and they have member areas. What's a member area? You know people like to sell content now and premium content? It's a big business. Well, they have that built in to Squarespace, and they don't take, you know, double-digit percentages of your revenue like those other platforms do. And they also have appointment scheduling. So, you know, if you're doing a business where you're a

consultant, you want to charge for your time, well, you have scheduling built into it as well. And this is the brilliance of Squarespace. It's going to look beautiful, as you know. So here's what I want you to do. Just head to Squarespace.com slash twist for a free trial. When you're ready to launch, use the offer code Twist. You save an extra 10% off your first purchase of a website or a domain. We love your Squarespace. You know how it is. When you're a technologist, everybody in your family, friends, your circle, your network, come to and say, hey, I got to get a website up. Can you find me a developer, a designer, a product manager? And you just say, you know what? Yes, I can find you all that and more at Squarespace.com slash twist.

How many customers, broadly speaking, do you have? How many documents are in the repository? And then how do you start this process of the very scary prospect, I think, for some customers, of, oh my God, am I feeding my data to this? And is it going to wind up jumping the fence? Am I training the box model or the open AMI model? And are they going to use some corporate secrets and a strategy document? And then somebody else says, hey, how does Michael Porter reinvent an airport and a new airport in New York?

And that was their proprietary stuff. So that's like got to be objection or concern number one. Yeah. Yeah. So let's start first with that. and then we can kind of talk about actually what we're doing. So first and foremost, just emphatically, first of all, every interaction you have with Box AI is explicitly, you have to decide to use it as either a user or the enterprise. So we're never having AI kind of do anything in the background without your explicit, you know, kind of consent on using the service.

That's the first thing. The second thing is that at least for all of the use cases that we're talking about right now, there's no training whatsoever. You're just using the AI model in its sort of stateless form as really a reasoning engine to be able to help us with natural language type tasks. And so there's no training of the AI model. There's no logging of the information that you gave the model so it can train later. All of that is done in this very kind of stateless, ephemeral way where we're just kind of asking the AI model to help us. with the user's query. And then probably the third big element is just we have a very,

very high standard for security, compliance, data privacy, just by virtue of our customer base, our very large enterprises across every industry, hospitals, federal government, et cetera. And so we really, really take security and privacy very seriously. So we're not trying to lean into any of the privacy elements from a risk standpoint. This is completely meant to be a very, very, a conservative approach to how we're going to treat, you know, data privacy and compliance.

Now, certainly there must be some early adopting folks and some CEO calling you saying, hey, I do run this organization. I would like you to create a language model for me across these 4,000 employees. And I, as the CEO, God, King, or a queen, would like to be able to do queries or my management team across our entire corpus of documents. So is that on the roadmap? And are Are you starting to get those, you know, sort of, I'm all in requests from your customers? We are, actually. And what's really interesting about this is it's not even, it's not the obvious sort of cut lines of, you know, kind of conservative businesses versus more aggressive businesses and, you know, sort of unregulated. we're seeing that most CIOs, most CEOs are recognizing that, you know, AI is a real platform shift.

This is much like, you know, mobile or the web in terms of we think about it as a platform shift. It's a scenario where, you know, this is now me talking, but our customers are kind of reinforcing this, which is it's a general productivity uplift where we can use this technology to, you know, make decisions more effectively, find information faster, spend. more time on the right areas of our business that only humans can do as opposed to computers could do better. And so we're seeing, you know, almost universal optimism about how to leverage this technology. Of course, there's a long list of, you know, items you have to just go through from a compliance security privacy standpoint. But in general, it's, you know, across our 110,000 plus

customers, obviously, you know, we spend more time talking to some of the larger ones, but there's a lot of, a lot of excitement around it. And, and, you know, there's a lot of nuances those. So for instance, this idea of training, you know, a model with your enterprise data, it's actually not clear that that is a problem that, you know, we can really easily solve at the moment because, because when you train a model, you know, with proprietary data, it's very hard to sort of figure out, well, which things should we leave out of the data set because they might, they might accidentally inform the model and then give up, you know, kind of sensitive information to, you know, the wrong employee as an example.

So, so we have to... Permissions are paramount here. And permissions are in the previous file, subfile organization are just not the paradigm that we have just shifted to. The paradigm is here's everything and... Well, well, so that's actually, so this is, I mean, humbly, this is where we think we play a a role, which is, which is actually you do want permissions on, on the date. that the user is going to want to be able to query against. And the AI model is really used as basically this brain to understand the information

that you're giving it, that the user already is allowed to access. And you're using that basically AI model to reason through what that user can already access. So at least in today's architecture landscape, and there'll be different approaches that different companies take. The point is not to train a model on your enterprise data, because again, then all your permissions are going to be leaked into that model. It's to separate the underlying access controls that people have to information from the AI model, which you use as a reasoning engine for that data. And that's at least for our kind of use case, which is a lot of sensitive data, but you do want to be able to work through it with natural language. That might not be as

relevant for maybe if you wanted to have a train against your code base where all of the engineers are allowed to see all of the code base, that's something where you might want to have then a kind of a specific enterprise train model. The perfect example would be legal and HR. The legal department, everybody being able to query the legal departments could be any kind of settlement agreements or confidential agreements to come up. And then somebody typing in, who are the most underpaid people at this company? And they're like, oh, it's me. Who's the most overpaid people at this company? Oh.

Yeah. So you can instantly imagine like past like 10 employees, you can't train, you know, your entire enterprise on, you know, your entire enterprise data set on a single model. So this is the, this is basically where our work has gone. It's only, you know, we're only four months into it. But it's, it's gone to this abstraction layer where we can keep the access controls in place, but use an AI model to reason through information that the user is allowed to access. and that's, we think that's going to be the breakthrough for how we implement this.

Makes total sense. The head of human capital has access to all of the human resources information, but a recruiter might have only access to 10% of the information, and they might do different queries about different data sets and summarize or whatever. So how about some demos here of, you know, the, I would suspect there are things that are really easy to implement and provide. massive value. You know,

like we like to do as product people, we put those on an X, Y chart, those would be like the quick wins. Yep. Big impact. Everybody's going to use it, and it's easy to implement.

What did you get to first? Yeah. We'll get to the harder ones. Sure. So we, I feel, this is fun going back to like the startup routes

of a forced live demo. So here's basically how it works. So I'm in, I'm in my box account right now, just looking at one file. and this is actually, it's a little meta because this is the press release for the announcement, but it's just an easy kind of public document. So this is the press release, and you'll be here, you know, you just click this Box AI icon.

We're still going to continue to play with the ways of this gets incorporated in the product, but for now it's this sort of chat interface on top of your content. And so now, you know, this would work whether it's a three-page document or a 500-page document, but again, for demo purposes, we want to keep this symbol. And you just say, you know, please summarize this announcement. And if it works pretty quickly, because we're in a dev environment, yep, it'll just go in and summarize the announcement. And then you say, what are three use cases for Box A.I? And so the key is it's saying, according to this document, here are three use cases.

So it's not attempting to kind of go into the large language model and say, what do you know about Box A.I? and can you respond to it? It's just saying from this document, what are you finding inside of this piece of information that could answer this query? So that's sort of the way that we've set the prompting up and that also then dramatically reduces kind of hallucinations

that you might see because it's not attempting to make anything up. It really is, we're really kind of forcing it to stay within the document. But there's one little, there's one kind of cool thing which is while we're telling it to only answer questions based on the document, We're giving it a little bit of free reign when you want to do more transformative type use cases.

So let's say you want to take this document and turn it into something. We're instructing the model to go along with that. So as an example, we'll just see if this works out to prove my point or not. Please write an email to an automotive company pitching Fox AI, and please include three auto-specific use cases. So in this query, you know, there's not, this information is not in the document, but we're sort of, you know, having to go along with that instruction to say, use this source information to answer this question still. We'll see if it, if it kind of comes up as a desired answer. And so, you know, it wrote this email to an auto company about, about use cases that they might have for Box AI. And so, so now you can imagine, you know, your sales rep and you're looking at, you know, a product marketing, you know, PD. and you really quickly want some ideas in front of a customer of like, so what are some use cases for this new technology? You know, Box AI would be this assistant for you to go and quickly, you know, answer that question.

And so any, basically any kind of, you know, tasks we give it, other than just using the large language model as the sort of database of answers, it will generally do on your content. And then, yeah, sorry. Well, I was going to say, you know, when you have a good demo like this, your mind starts to just come up with ideas. and I was immediately like, who in the organization has to approve this?

And I wonder if I connect this to my email and I can source 100 leads from the database who haven't been contacted. So you start to get into information you probably have, which is the org chart and the permission. So who, you know, you could be asking who handles the automotive category on our sales team in the United States?

Can you get them to approve this? And in our Salesforce or wherever that data is stored, can you get me all the sales leads in automotive that haven't been contacted this year. Yeah,

yeah, exactly. And so your mind can instantly kind of extrapolate now what's possible. Now, what's interesting is

this is where, this is where, you know, we in the valley and everybody, you know, I mean, you know,

figuratively or literally, you know, have to respect that the technology can do all that. But we're probably like five years away from, from,

you know, society actually understanding how that would all come together and having the technology actually integrated in this kind of capacity. And so this is where, and I remember when we were first rolling out

cloud to enterprises, it was 06, 07, 08, we were in front of enterprises, and we had our early adopters, and my hunch was like, okay, in like three years from now, obviously every single enterprise workload will be in the cloud.

It's so clearly obvious to the entire world, how much more efficient this is, how much more scalable it is, how much more secure it is. And, you know, fast forward, it's, it's now, you know, 15 years after that prediction, and we're still in the phase of, like, mass adoption of cloud computing. And so, and the reason for that is just like, you know, the world takes a lot of time to change on these really big tectonic shifts.

And so, you know, the use case you just, you know, mentioned, it's got to connect email and Salesforce and your data from box. That's going to be a lot of wiring together that we have to, that we have to, you know, come up with. And then what will be interesting is do the AIs talk to each other? So does the box AI talk to the Salesforce AI, which talks to your email AI, and they are coordinating, you know, that information exchange in a, you know, a secure way? And we are again, we're at just the starting point of how does that actually get architected. One of the reasons why criticism can feel obnoxiously aggressive is that sometimes people use criticism to sort of dominate or assert superiority.

And that is not helpful criticism. So state your intention to be helpful. That was Radical Kander author and friend of this week in startups Kim Scott. And she just did a masterclass session called Tackle the Hard Conversations with Radical Candor. If you're a business leader, you can learn so much for masterclass. There are amazing lessons from Bob Eiger on leadership, Chris Voss, on negotiations, and our friend Alexis Ohanian on startup investing.

And so much more. of their craft are a masterclass teaching you whatever you want to learn. Paying for an unlimited masterclass subscription is a total no-brainer. We just had an awesome insight from Kim in just 15 seconds. Imagine how much you're going to learn in 10 minutes or an hour or maybe two. If you invest that in the next couple of weeks, I highly recommend you check it out. Get unlimited access to every class. And as a twist listener, you can get up to 35% off for Mother's Day.

What a gift. Go to masterclass.com slash startups now. That's masterclass.com slash startups to get 35% off for Mother's Day. Yeah, this guide on the side concept feels really powerful in terms of just efficiency. When you look internally at your own company, what would you predict, if I had to ask you in a percentage basis that, you know, the average team member or the entire organization, how much more efficient will they be by the end of 2023? if you get everybody to adopt using chat GPT and other AIs. How much more efficient will your company be? Yeah, this is the big question, I think, for all of us.

And so if you kind of imagine, let's say, every knowledge worker for now just starting out with that demographic, you know, has an AI or multiple AI assistants that do relevant tasks for them. You know, Workday will have one for HR. Salesforce has one for sales. No, shin. Yeah, no, everybody, but everybody has kind of AI incorporated their workflows. I think that the way that we should think about it for now is, is how much time do you spend in your job trying to, you know, triage information sources, find answers to questions,

sort of do more information, you know, based tasks, you know, writing something, brainstorming something, you know, discovering something. And, you know, depending on the job, You know, if you're in, let's say, you know, customer support, that could be like 50% of your time because you're always going back to the knowledge base and, you know, trying to get an answer to something. If you're an engineer, you know, you'll go talk to engineers and they're spending, you know, two or three hours to find, like, who on the internet has optimized this particular SQL query before. And, and like, that's like, you know, theoretically, instantly solved inside of GPD4 or copilot. And that just becomes this unlock. And so I think depending on what your job looks like, you know, somewhere on that continuum is, you know, you might get back 50% of your time. You might get back, you know, 20, 30% of your time. And my, and my kind of like, like totally amateur hour macroeconomic, you know, kind of, you know, kind of view of that is that is that we won't really even notice what, what, you know, how to measure that because, because that that engineer or that customer sport rep will just be doing, they'll be doing more of,

they'll just get to the next task faster. They'll get down the roadmap faster. They'll get to the next sales call faster. Exactly. So it's not eliminating their position. It's augmenting them. And when I talked to Brian Chesky about this, you know what number he came up with?

What? 30%. And I had come up with 30% because I was like... Of efficiency improvement? Across all of Airbnb employees. And I just thought about that. Well, hiring has been in that 10 or 20% a year range. So on a meta prediction,

I think you could see organization. instead of trying to hire somebody, you know, to throw a body at a kind of management culture, I think the next two years of management culture, who knows how long will be, how do we get the AI to do this? Is there a prompt we're missing? Is there a different AI tools? Should I be using Po from Cora? Is this Bard going to do a better job? You'll be just AI shopping, chatbot shopping. And I think 30% is the right number. I think it's like, So I think that this is going to be, this will be extremely interesting to watch, you know, clearly, but I think there's a couple ways to cut it. So, so there are some things, there are some tasks in a business that are not kind of like infinite. They're like, they just have, like this one discrete thing has to get done. And so if you make, if you make that one discrete thing 30% more efficient, then you scale that out and then almost by definition, you would, you would have 30% less labor, you know, across the economy or across a particular business for that thing. Then there's a lot of tasks that are

not kind of finite. They're, they are like literally kind of, you know, bottomless and boundless in terms of ongoing. And so I think engineering is one of those things, which is we are always constrained. We are universally always constrained by the number of engineers we have. And so if I can make our product roadmap go 20 or 30 percent faster, that does not, that we are going to hire the same number of engineers that we can afford. We're just going to accelerate. We're just going to accelerate the word is going to be on a relative basis to what we would have been doing, 30% more productive. No different from 10 years ago, we were probably 30% less productive then because we had

engineers working on, you know, different open source libraries or different, you know, systems that now the cloud just does for free for us. But we didn't hire fewer engineers as a result of that efficiency. We actually, if anything, we hired more engineers because they could actually go and do more productive tasks as opposed to these sort of lower level, maybe less differentiating tasks. And so I think for areas of your business where there's no particular upper limit, there's no upper limit on how many sales reps you need, how many engineers you need,

you know, how many account customer success managers you need because those things are only constrained by how many customers you have or, you know, how innovative is your company and your roadmap. And so I think these just become accelerants into the future. And then there are going to be areas where there was a finite limit that you actually need, and we can make that area of the business more efficient. But I think there's more areas of businesses that are boundless and kind of limitless than those that are inherently finite. Yeah, I agree strongly because if you just think, I was trying to think of which category

would be one that could be like telephone operators, you know, like, okay, that's just completely been replaced or, you know, travel agents who book your tickets for you, right? And they type in the search instead of you doing, okay, obviously those went away. And then I just thought, well, okay, maybe customer service or success. And then I thought about it and I was like, you know what?

You're always trying to get a customer to be more successful with the product. Right. So instead of dealing with login issues or navigation issues, those will be done by the AI. Yes. But then you'll be getting to, like, here are some scenarios for you to use our product.

Here's some, you know, advanced scripting stuff. you can do with this product. You're going to just go to the higher level stuff. That makes people churn less. And yeah, it just feels like for there's going to be a group of people who resist this, but I don't think they should be scared. I think this is like an incredible opportunity. It feels to me like we just went from dial up to broadband.

All that did was increase joy, you know, and usage. This is going to speed up developers. I think the developers are going to be stoked to be more efficient and get more done. Yeah, there's no developer that I've talked to that has been using any form of AI for productivity that has been like, I really wish I could go, you know, do that research again of how people, you know, optimize this query or, or how they interact with, you know, this external library or API.

Like, like, no matter what everybody is like, if I could just see a quick example of, of, you know, how everybody on the entire internet solve this problem, I'm going to be able to leap forward faster. it doesn't mean you like copy and paste the code and you just you just implement it you know with without um you know without any additional labor um but the speed of of which now you can jump to the next uh task and the ideation you have is just so much faster and so i don't think there's any any returning you know from that um but i'm i'm way more in the optimistic uh camp of this just acts as a general productivity lever uh for the economy yeah it this feels to me like the way out of whatever economic situation we're in just like mobile broadband. It just kind of help the economy and founders get enthused. I mean, I can hear the enthusiasm in your voice. You're in the office on a Friday, you know, just getting it done.

And are you feeling that the enthusiasm level inside the org has gone up? Because listen, it's been a tough 18 months. Stocks going down, layoffs in big tech. Is this changing morale? Yeah. Yeah, we have, you know, we've been in kind of hunker down mode for, for three plus years. We went through an activist battle. And so we've kind of already had this grinded out, you know, kind of mindset.

So fortunately, that's been kind of baked in. But I think, you know, independent of our specific, you know, morale on that front, I think this is, this is sort of what we all live for in tech. is like you want there to be a platform shift at some interval, you know, 10 years, 15 years that just shakes things up. And you just, I mean, it's exciting that the companies who thought were monopolies now are, are, you know, they actually have, you know, real, you know, strategic crises that they have to go solve. There's an all new technology that is just really fun to play with. I mean, you know, most of the work that went into to box AI was, you know, these are like 1 a.m. sessions, you know, with, with the, the team or CTO on like, okay, can we, how do we solve this?

And how do we figure this thing? We hired a, we hired an intern at, at midnight on a Saturday. And just, just, you know, some, some kid at Stanford that was just like online and ready to rock right away. Love that. So it's just like, those are like the fun. I mean, I'm probably being like overly nostalgic in, in our, in our strategy here. But like, that's what makes start is fun is, is, you know, something comes out of left field.

You have to adapt to it. You have to figure out, you know, is this a tailwind or a headwind? How do you incorporate the technology? The other fun thing about AI specifically is that it has the service area of it is just so large where like every day you're either somebody externally is doing a discovery that opens up. You know, we saw the auto GPT thing just a month ago and you're like, okay, well, that's a whole new vector of innovation. Plugins. Data analysis.

Did you use the data processor yet? I have not used that one. Is that a pretty nuts. Like you upload a CSV file and you're like, it's like, okay, here's the columns in it. And you're like, okay, tell me about the data. And it's like, tell me three trends. And it's like, okay, it seems like, and I just uploaded like an Airbnb CSV of like all the different L.A. things. And it was like, yeah, it seems like Santa Monica has more than this place and this.

And the average price is this. And I'm like, whoa, I make some charts. And it's like, okay, here are some charts. And somebody did it internally. We were doing diligence on a startup. And we just took the diligence folder and took their. revenue and projections and everything. He said, make some charts and tell us about this. And it hallucinated. There were some bum charts that didn't make any sense. You're like, this makes no sense.

But then there were some that were like, oh, yeah, that would be something that an associate in a venture firm would spend two hours on and, you know, the researcher just did it. And it's like, okay, great. It's totally great. And the fun thing is actually like, you know, even the demo I just gave you are kind of like the obvious things that you would do with AI. But when you start to think about the non-obvious stuff and actually part of our problem is just gonna be how do we help

encourage people to leverage this in these non-obvious ways that's where it gets really fun so one scenario we gave it like our earning script from one quarter and we said if you were Warren Buffett how would you improve this earning script

and it's just like a random use case like nobody in the entire world would ever have asked that question of an earning script but all of a sudden it starts to amp up the cash flow message and the share repurchase message and the capital allocation message. And so now, you know, when you're preparing to go in front of a whole bunch of financial analysts and you want to, you know, figure out what's the right tone, what things should we lean into,

you have this, you know, instant expert that's seen the entire world of finance, everything ever written online about, you know, every earnings call, every, you know, ever, and you have that now instantly at your disposal. And so, you know, if you have a little bit of imagination on how you can start to incorporate AI, you know, per your due diligence use case or, um, or any way that you want to work with your data, the,

you know, literally there's just not, not, there's not a limit to what you can do with this stuff. The, the most interesting thing I've found is when you push it to become more creative and you say, give me five more ideas.

Give me five ideas that nobody's ever thought of before. Give me 10 ideas that are crazy. And like, you literally use words like this and it's like, okay, permission granted. And like, then the genie starts doing your wishes.

And you're like, whoa. When you're the founder of a high-growth startup, things can get chaotic. We all know that. You face a ton of questions. You've got problems all day long. And you don't know the answer to everything, especially if it's your first time as a startup founder. But you don't have to face these issues alone.

That's where Hampton comes in. Hampton is a highly vetted private community exclusively for founders and CEOs like us. Hampton's mission is to create the most valuable and engaged community for high-growth founders. Here's why some Hampton members have already called it life change. When you join Hampton, the membership team over there, carefully hampton, seven other members to join your core group. This group is like your own personal board of directors. They provide you with advice, critical feedback, and will aim to help you accelerate your growth.

Hampton members run some of the fastest growing startups in the world, and you probably use their products and services like morning brew, Blank Tree Coffee, dribble, and more. The connections, personal accomplishments, and a sense of belonging are what make the Hampton membership special. And, you know, community is super important. It can get very lonely out there as a startup founder. So here's a very simple call to action.

If you're ready to scale your business, join the Hampton community today at joinhampton. com slash twist. That's join H-A-M-P-T-O-N dot com slash twist today. Joinhampton.com slash twist today. I don't know if you've been following the writer strike which started this week in Hollywood. And on page two of like the. update to their members, the last, like, item before, like, coffee and donuts or something completely meaningless in their negotiation was, uh, we're asking for a ban of AI to write scripts,

to ingest previous scripts and to use it in the future to create derivative works. And, uh, then there was like what the, uh, studios said back to them. The studio was like, we will agree their counter was, we will agree to do a yearly meeting on new technologies. And I was like, yeah. You know what you need to do, writers? And I just took, I asked it to come up with, just for giggles and to, you know, put it on Twitter. I said, give me like five themes about Biden and five themes about Trump that a late night writer for a late night talk show could use to brainstorm jokes.

And it was like, yeah, Biden, you know, sometimes makes gaffes and he loves ice cream. And Trump uses weird words and he's a narcissist and he's addicted to Twitter. And I was like, okay, well, that nailed it. Then I said, oh, do a Twitter exchange between the two of them. that's funny. And it started making some, and they weren't zingers, like they weren't ready for prime time,

but they were good brainstorms. And so I guess the question is, like, did you get early access for your Twitter? Because you're hilarious on there. And like, what impact is this going to have on your joking on Twitter?

I think, I have not yet fedded any Twitter stuff on. And, you know, Twitter is his own vector of, of, you know, kind of landscape right now. But yeah,

on the writer, on the writer's right thing, here's the thing. I'm actually super sympathetic to any demographic that thinks that AI is going to have an impact on their job, simply because it is this foreign technology

coming out of nowhere. Nobody wants to feel vulnerable in everything they've learned their whole life. So actually, I'm like, I totally am of the view that these are super important conversations. There's not

like an obvious thing of like, you know, Silify Valley is really going to have to, you know, try and avoid the, the kind of like, like, oh, I, you know, learn to code. Yeah, exactly, exactly. It's just, it's like, it's like, we have to be thoughtful about, about these kinds of, of trends. The reason I'm optimistic is, is just because I think there's still, you know, there's, to even your point, like, they weren't zingers. Like, like, you know, and so, so I, I think there's a lot of stuff that, that, you know, is, is still in this, like, it's novel because it's new, but it, it, It is not actually going to, you know, really be able to go and replace the level of creativity that a human has for the vast majority of ways that, you know, we actually pay creative people for, you know, today. And I think there's a kind of a little bit of a divide. I think there's, there's the use case that you'll see from something like mid-journey, which is, which is, you know, really, really compelling, you know, images.

And I think the scenario there is that we just end up having more ubiquitous creativity from so many more people. but the experts get even better. And the idea generation increases, you know, tenfolds across the world because now you can be, you know, some random kid in high school and you could, you know, be ideating on a new product, you know, design that would never, you know, existed in the world, but now you have this AI engine that can help you with that. So I think creativity just explodes because of this.

And on the, you know, and more of the professional trade side, I'm just bullish long term on humans really, really like other humans to create things. And there's some, and it's just totally different to imagine going to a movie that, that AI generated from, you know, something that, that Quentin Tarantino made.

Of course. And so I just don't think that our brains are wired to care about, about, you know, if a human didn't make it, but, and the field is entertainment, I'm skeptical of that's going to replace, you know, what the humans are doing. If you look at CGI,

they thought, okay, this is the end of acting. This is the end. And it's like, yeah, Blade Runner and Star Wars use miniatures. Right. Amazing. And then you go back and you look at the miniatures and you're like, eh, it's kind of taking me out of the experience. Like, it doesn't look that good.

And then you see CGI today and you're like, you know what? We can make you a Star Wars series every three months for you and your kids. And so are less creative people? No, there's more. Right. Because now you can go down the long tail of content and you can make the Ashoka series about Anakin Skywalker's Padawan and it's like

okay great go for it actually I mean exactly to this right like this is actually I mean you remember I mean you know 06 oh 7 and 08 we thought like you know well we're just going to be watching YouTube and UGC is going to

disrupt all Hollywood and like all the tech people were like oh this is the end of Hollywood and it's like no like if anything it actually contrast the quality of the really good stuff even better and we get just as excited or more excited about the about the good content and so to your point I mean AI is an enhancement

to that creative task, but from everything I've seen, not a replacement of. It's going to do something great for startups, too. Like, when we started, what did you spend on your original logo? This is like a question you can tell when a founder started the company

and how much you spent, because in the, to that, aughts or whatever they call them, like you would actually pay to get a logo done in all likelihood. And what did you remember what you paid for the first box logo?

I actually designed it. So, So, so, so, it was free in our case, but, but yeah, I mean, that's a, you know, five, 10,000 bucks. Actually, I probably spent 20 hours just rounding each of the letters. So, so way, way too much time. It would be way better to have a headache. But you got quoted five to 10. Oh, easily. Yeah. Yeah. And you're like, well, I have more time than money. So I'll just do it myself on the weekend. And now, you know, then there was like this medium period where like these websites, dribble, be Hans, whatever, Fiver, you can find. You can find. mind all these folks and you can get a logo that's reasonable. There were design competitions like 99 designs and other ones for like 500 bucks, 250 bucks. And designers were really upset about that. But if you're doing a high end logo, like you're still can charge five or 10 grand.

There's people who will pay it. Like if box is going to do, did you ever do your logo over again and do like a whole brand treatment in the last 10 years? No, but I'm not trying to ruin your questions. Only because you're just that cheap. We're just that. We're either that cheap or we do that all in houses. Got it.

But the theme is completely accurate. Yeah. And things like Uber redid their whole path. I remember Travis redid all of Uber at some point. Hired an outside firm. It'd probably cost a half million dollars or something like that. We can afford to do it. We want to have a thoughtful discussion about it.

And yeah, I mean, AI can make a 7 out of 10, but we want a, you know, 9.5 or 10 out of 10. We're going to go that way. Right. I bet you all these writers are already using ChatsyPT to brainstorm ideas. Just like they took out old joke books or old. or old scripts and look for ideas or watch old Kurosawa movies and decide like, hey, this is an interesting character set

from Fortress. So let's do R2D2 and C3PO for the next generation. Yes. Yeah. I think that's my take. I think the one asteris that I'll add is just, I do the copyright piece is super interesting. Say more.

Well, they're just, just these AI models are trained on everything they can get their hands on. Seems unfair to you. You saw Barry Diller was like, you know, now is the moment you have to you know, fight if you are a content creator. And so, I think we're in for, for, you know, a couple years of really, really interesting, you know, case law, you know, coming out around how this all plays out.

Where do you stand on it? Somebody trains on every, you know, song ever written. And then they make new ones. That seems profoundly unfair to me. I mean, Getty images busted stable diffusion, like, instantly. That was like, dude, you literally blurred the Getty image. like, come on, bro. Like, that's just not cool.

Yeah. Oops. I, yeah. Where does it hash out for you? What do you think that would be fair? I, I, I, I, the problem is, is that, um, is, you know, fair, uh, is somewhat different from technical, technically possible. And so we're going to have to figure it, figure that out, you know, I think the, like,

intellectually, like, ideal outcome would be something where, you know, you know, data gets licensed to model training. Easy. And, and there's a, you know, that we all put a little tag in our website that's, you know, says you can train our public information or you can't. Robots.t. TXT for AI. Exactly. It's AI.txT.

And I think that, and then what is the new creative commons for that? And actually, then probably as a result of that outcome, you know, the models will be basically as good as they are today because there'll be enough information that is in more in that public domain. And maybe, you know, it's just not as good at Quentin Tarantino scripts. And that's, that's sort of the one gap. Or they license it. Yeah, I think there's an opportunity there.

So you start thinking about as an opportunity. Newspapers are struggling, right? Right. All kinds of content creators are struggling. You look at something like Reddit. You look at something like Quora. You want people to participate in those communities. So licensing Quora, licensing Reddit.

Yeah. And then you would have Google and Microsoft and Facebook while competing to who gets the Reddit corpus. Who gets the Twitter corpus? Maybe they give exclusives. Maybe they don't. I'm going to pour a little. little water on that one, unfortunately, because you can also imagine that the costs of running these models or training them will become so high that it all of a sudden drops the productivity

that you get from using them. If the price of a token were 10 times more expensive because it had only licensed data within it, a lot of the use cases I just demoed an example would be a lot harder. And so, um, to just, just do in any kind of affordable way. So then what would happen is you would end up having, um, you know, more of these public domain large language models would probably be the ones that end up taking off. Um, and so, so I mean, it's, it's like, it's a nice theory that, that everybody could start charging for this data, but then the models would ultimately then get trained on the stuff that was, was cheap, um, uh, because it would just be, you know, we, we can't have, uh, I, I mean, I don't think you could have the equivalent of, um, the, of,

the, you know, the Comcast negotiation wars with, with, you know, X cable, you know, station, uh, cable channel, like, for AI models, um, because the, the, the many to many problem on that one is just going to be at insane. So imagine, you know, somebody comes in, Reddit says, well, we're going to pull our data and you've got to, you got to pull

it out of your model. It's like, like, yeah, it's impossible. I mean, basically, if, if, if, if, if, gety images wins an injunction, like, like, damage is being caused. they could get like this injunction or preliminary injunction against stable diffusion. They've got to just turn the model off and start over. I mean, that would be wild. It's going to be totally wild.

So we're, I have my popcorn. It's nuts. We're going to stay out of that one and just, and we'll use whichever AI model is legal. I like industry. I like when the industry comes up with something like self-policing, like robots.t.t.

So I think your idea, a.a.a.a.comst, great start. And then, you know, a little bit of cash here and there and then citations. I noticed in, I have been grinding on about this, which like, I don't have a problem with you using my stuff, but can I get a link back? That's just courteous. And in the new version,

have you played with the web chat chippy 4? Yeah. It's kind of janky. It crashes constantly. But when you hit the drop-down, it shows you what web pages is crawling out on the web. It shows you what search it did, and it links to it. Yes. So I thought that was kind of dope. I was like, okay,

at least I can link back to that and they get some traffic or whatever. But I think this brings like a really interesting micropayments concept. This is where I think Google, I'm curious, your thoughts on like Google's vulnerability right now, because that was like the big topic last night at the poker game.

I feel like this could be like Google's way of actually cementing in like some revenue sharing with the people they index. Where like, hey, if I used you in an answer, I give you a fraction of a penny. Right. Yeah, super interesting. I mean, I mean, you could, you could certainly imagine a world where they strategically would,

would want to kind of make it very, very hard for any subscale player to, to, you know, commercially operate. And so, you know, if you were really, really like, you know, game theorying this out, like, you'd want to be on their policy team being like, oh, we've got to really lock down the copyright issue. Because they're, you know, then you'd have to be at their scale to go and actually do that. Yeah. But I still don't, my brain can't figure out how you do the kind of like sub penny, you know, type model because even in the example you just mentioned, which was, which is the web browsing thing, that one's easier to do because it's pulling out the web pages and then using the AI model to reason through them. So then you obviously know then what it's citing.

You know, if you interact with the AI model directly and it, and in a sort of black box fashion where it's not connected to the internet, I don't know how we, you know, eventually track, like, what portion of the model, you know, should... Came from... Condé Nass Traveler versus New York Times travel section. Exactly. Yeah, it is a... But, you know, sometimes these problems are opportunities.

All I can say is, and I think you'll agree with me, thank God this is the platform ship and not crypto and not VR. Because, oh, my Lord, the two most annoying people I've ever met are people with VR headsets trying to get you to put them on and crypto people not shipping products. Thank God. That's something. I've rested my case on crypto last year,

so I've been able to move on from the topic. But yeah, yeah. I mean, it's just unbelievable. Like this, speaking of hallucinations, like, has so much money been pumped into a space with so little to show for it? Yeah, I think the challenge with crypto is basically like the, you know, there's been tweets over the past couple years

which is like, what's the best, you know, demo that you can do of crypto? And then, like, somebody will record something or show something. And it always requires you to be bought into a philosophy as opposed to, as opposed to caring about the use case itself. So, so you're wowed by the fact that, you know, with no intermediary, I could do X thing. But X thing is not, like, wasn't like a new thing that you couldn't do before. It was just, it was just a thing that we already do, but now with no intermediary. And so the problem is, is, like, translating that to regular consumer. consumers, like, there's not enough bandwidth to tell people why they should care about that

philosophy because they just like, no, I just feel like, like, I can already do, I can already move money to people. I can already, you know, communicate. I can order a cup of coffee with my watch. Exactly. I can use Apple pay. Yeah. And this is better why? You really had to care about the underlying architecture to get bought in. And versus, you know, something like AI is like, like, you just show people like the thing and they're like, like, their mind explodes. And they're like, I like, what genius is. Give me more. Yeah, exactly. When can I get it?

My mom is, you know, she keeps asking me like, you know, can I, well, she's asking me to do the chat chbt queries. So, but like she keeps asking me for, for, can you ask chat chvd? Also the laser printer in the basement, Aaron? Yeah, exactly. Exactly. It's jammed again. That's correct.

So there's a pretty good test. If, you know, we, we have, you know, people texting people to ask chat chabot to, to run a request for them. So that's like when, you know, cross the chasm in technology. When is all this going to be available on Box? And how are you going to charge for it? Because I just got a bill from Notion. I had all of my team sign up for Chat, GP2.

I said, pay the $20, put it on your personal car, put on your corporate car, whatever. Just start playing. There's no like multiplayer buy mode or whatever. There is in the sandbox. I did that too, but not for the actual consumer product. But then Notion was like, hey, you're paying like this amount per month, but the AI is $20 more per person and $10 per more. I was like, I got to make that decision now.

Right. ChatGPT4 or the notion version of it. It's embedded. I like both. Do I just spend $250 on each individual in my organization a year and it's $500? I kind of am like, maybe I'll just buy both. Yeah. When is this going to be available?

How are you going to charge for it? Yeah, I think you actually got to the heart of the issue of why, you know, why you'll have competition driving prices down as opposed to up, you know, vis-a-vis that licensing question because the only reason notion it's charging you so much is because, you know, the amount of tokens that they're outputting, you know, from the AI model are, you know, pretty vast and also unpredictably broad, you know, depending on which user is using it. And so they have to charge more for the product. For us, you know, our goal is to incorporate some degree of the technology into the core of box. So that way as many people as possible

can leverage it. And then for things that are extremely high volume, you know, we'd probably have to have some additional monetization, but what we think about this more as a scale play of like how do we just completely, you know, transform the product overall, but TBD on some of the specifics on that. Yeah. When will it be available or when you get shipping? So it's rolling out right now. What I just demoed in, we're starting with a number of kind of private beta customers.

We want to get the user experience right. We want to, again, figure out that kind of pricing and performance side. And then I would say the coming months and kind of quarter to, it would be giant. All right. listen, continued success. Congratulations on, you know, getting back to the office. You got people back in that beautiful office in Redwood City? We do. We're in like a two to three day a week model. How are people enjoying it? What's the reaction? Well, you know, they're telling me it's good. And I mean, I think I can see happiness from a distance. But I, you know, there's a lot of

benefits to remote. We actually still have a large remote, you know, contingent box. But I do think having hubs, getting people together, you know, we're in New York, we're in Austin, we're in Chicago, we're in SF and Redwood City and then internationally. And so I think having convening, you know, a convening place where you can, you know, spitball ideas, you know, get mentored, you know, new and newer employees, you know, have a way of learning the craft and the trade. I think these are all important things. Does it need to be five days a week? Probably not. But I do think, you know, having people come together is really important. So we're getting the benefits of that. Yeah, I'm going back to an office. I'm actually looking for a space.

and Sam Mateo now because I'm like, I just don't want to be at home anymore. I want to hang out with founders more. I want to do like, I want to do this in person again. Yeah. So maybe, you know, be able to do some in person interviews, have lunch with the person after and just, you know, see more people. It's like, so dystopian. Yeah.

To be only remote. It's just. I think, I think that's the thing is like, is like, we got into this battle of like, oh, it's remote versus the office. And it's, you know, even from our remote people, they love that, I'm speaking, you know, generally, but they love that we have offices because when they then come into town, oh, their callings. there and they can see them. And so, and so it's, you know, it's, it's, it's not like these things have to be

these extreme polarizing topics that I think we've turned them into. So, yeah, the finance people in New York are like, everybody get back to the office. Right. You know, every day, you have to suffer. And then like, they're in Italy and Aspen for four months. And I'm like, exactly. I'm like, really? Okay. Yeah. Sure. I mean, I guess you can take a helicopter to work, so it's all good. All right, everybody, follow Aaron on the Twitter, L-E-V-I-E. Sign up for Box.com. If you're looking for a great job,

a great boss, and just a legendary company, go to Box.com, look for their careers page and you get a job. Yeah, I always like to build you up at the... I appreciate that. That's great. You're still hiring, right?

You got some job wrecks. We are. I mean, the boss credentials that you just gave me, I think were a little bit overstated, but appreciate it. I've known you for over a day. You've been on this pod for over a...

a decade now. You're a fun guy. You're thoughtful. You make great product. So it's a good place to work, folks. And you can learn, right? You can learn from a legend. All right. We'll see you all next time. You will learn in person. Yeah, exactly. We'll see everybody next time. Bye-bye.