Betting on the Cloud Before it was Cool: The Rise of Box w/ Aaron Levie

24:48What are you talking about? Like, this is my fiefdom. I'm not able to collaborate. It was a cultural thing. Yeah. So you needed like the business climate to change. You needed the technology atmosphere to be working in our favor. You needed the Moore's law cost curve to keep accelerating in our favor. So the cost of storage had to go down. The cost of compute had to go down. There was a brief moment of panic just to show you like how much we've benefited from 20 years of compounding technology. There was literally really a brief moment of panic I had somewhere around like 2006 or 2007, where somebody like asked a question of, let's say you had all the specifics are wrong about this question, but let's say you had 10 million users on your platform or 20 million or 50 million or whatever.

25:34Like, how would you store all the data? And I was like doing some math. I know like, like, oh, my God, I think it's like, like, you'd have to have like 50 data centers worth of storage. like and you know at the time it would have literally cost tens of billions of dollars of storage if you just if you took their their their you know kind of math you know question seriously and it's like quite literally there's not enough storage in the world for the idea that we're trying to pursue and so we were kind of like in total just like just pray hail mary mode that it all works out and it literally all worked out like like the like the you know again some of my numbers might be off by maybe 50 % or 100%, but it almost doesn't matter.

26:15Let's say when we first started, the hard drives that we were using were in the 100 gigabyte to 200 gigabyte range, plus or minus. Today, you can basically get hard drives, 25 terabytes plus. People are working on 50 terabyte hard drives. So that's about 100x improvement at minimum of efficiency. So in the exact same amount of space, that much kind of improvement. So multiple 100x improvement, if we just add a couple more years to this timeline. And it's the only way that our business model became viable was the cost of storage had to plummet. The performance on compute had to increase. Bandwidth had to get faster.

26:55There were so many ideas early on which were like, well, people are never going to want their files in the cloud because they're going to go on an airplane and they're not going to have access to them. And for years, that was very clearly a real risk. Internet started to get a little faster. And now, let's just imagine five years from now, everything Starlink, like boom, entire categories of why this would not be a viable business just disappear. And so those were the big mega trends that helped. And then the newest one that will be probably a bigger accelerant than any other one prior is going to be AI.

27:27And just now what AI can do with all of this data, which just adds one more vector of why you need your data in the cloud, why you need modern platforms for working with this information. And when, you know, storage costs was going down a ton and managing storage, That was a positive vector. But was there also like a negative competition vector too? Because now it's so much cheaper that like, you know, you may not need as much of a service. What was the kind of challenges that came with it too? I think we actually, I think in a pretty strategic way, got ahead of that. So once we saw a couple of years of that trend playing out, you know, Amazon had launched in 2006, 2007 or whatnot with AWS.

28:03Once we saw, you know, two to three years of their costs coming down, and it was very obvious that like all of these costs would eventually show up to the consumer as lower priced services. We just got fully ahead of it. And we announced maybe 2010 plus or minus a year unlimited storage in our business here. And basically we just said, you know what, we do not want customers worrying about do I have half a terabyte of data or three terabytes of data? And what's the cost difference? Because basically we're like, well, every year it's just going to get 30 % cheaper for us to store this data? Like why have everybody have to worry about every single year?

28:40What is the optimal amount of data to store? So we got ahead of that by just saying, let's take storage off the table. Let's shift the value proposition fully to the things you do above the storage. So what do you do above storage? You do security, you do workflow, you do data governance, you do collaboration. And it was actually very enlightening and important. So I'm actually extremely happy that the cost curve was obvious early on because what it told us is what is the viable business model in a space where the infrastructure is being commoditized? Well, it's all the things that you need in a world of abundance.

29:13So if I said you could store unlimited data as a company, that's fantastic. But now all of a sudden, like I need like workflows to manage the data. I need security permissions of who can access it all. I need to be able to plug it into lots of other software. All of that stuff has no sort of deflationary effect on infrastructure. It's all sort of just, you know, it's just software value that you build over time. How's the competition you think about over the years? Obviously, you guys have the Box Dropbox, which, you know, very similar names probably didn't help. Yeah, exactly. How was that psychologically?

29:46Like, how do you guys deal with it? Yeah, so I think one thing that has been absolutely pervasive since the starting of the company is a deep paranoia about just are we moving fast enough? Are we competitive enough? Are we well positioned? Are we differentiated? Early on in the Dropbox thing, we had to kind of figure out, we had already pivoted away from consumer to fully in the enterprise. We had to both first assess was that the right pivot because Dropbox was seeing a lot of success in the consumer. We quickly evaluated that actually, yes, it was like really the right pivot because now you want to be even more sort of differentiated from your competition.

30:22It emboldened us to go even deeper in the enterprise, having some of that early kind of consumer prosumer competition. So it kind of really helped us clarify, we have to be the enterprise leader in this next gen data space. You know, the bigger threat is always the bigger incumbents that have already access to the customers. They have distribution, they have a kind of a pricing power where they can fold things into one ELA bundle. So we had to say sort of what competes with better distribution and a commercial model that everything's included. Well, that means you have to have literally better functionality.

30:53You have to solve the customer problem better. You have to take advantage of the things that the big incumbent probably won't do, like being more neutral and integrate with all of their competition. So it's also clarifying to really understand what you need to do differently. And again, maybe another lesson embedded in that is like you can't – there's like no way to be fluffy about differentiation and strategy. And this is like not just kind of a hope and prayer kind of part of business. like you need to be so clear on what your asymmetric advantages are and then make sure that you put every resource possible into those asymmetries so they are just fully maximized.

31:32So for us, once we knew that, you know, what the rough contours of the competitive landscape were, we basically said like every part of our strategy needs to reinforce how we are competitive and not dilute or sort of, you know, be neutral to that. And that changed over the years? It did a little bit, but a lot of the cores remain. So if I look at why a customer buys us today, I think a lot of the adjectives would be pretty similar from 15 years ago. Simplicity, but enterprise-grade kind of capabilities. Scalability, deep security, openness and interoperability. Like, those things have remained basically the same all the way along.

32:13AI is adding, you know, new capabilities that we think we can differentiate on. But the core foundation of what we started the company with has persisted and retained its value.

32:47you no matter where you are on your journey. From maximizing runway to earning yield on your cash, Brex was designed to make every dollar count so you can focus on what really matters, building your dream. So whether you're just getting started or you're ready to take on the next big thing, Brex has your back. Check it out and see why the world's most innovative startups like Anthropics, Scale AI, and Robinhood trust Brex. So now let's get into the AI part. Yeah. You know, obviously, like the everything is kind of shifting. Right. And it's really hard to have like a view like five, 10 years from now of the speed that things are moving.

33:23But if you kind of had to try to predict where the points of value creation will still exist in a world, you know, kind of AI, what would that be? So I'll start macro and then maybe just zoom into box. I think that AI will transform almost everything about software and work. Let me just say that as an underscore. But a lot of the core principles of strategy differentiation, business model design, at least in the software space, I think persist and retain their timelessness. So in a world of AI where we can just, let's just say, let's say you get the infinite ability to deploy intelligence inside of an organization to do things.

34:13I think in that world, companies that are well set up to help orchestrate the workflows of those AI agents, companies that have the data in which those AI agents operate and the data that they operate on. The systems that can keep the data governance and the protection between the users and the AI agents matters a lot. Companies that already have a user interface that is sort of natural for the human user to come into, where an AI agent can pop into as well, matters a lot. So if you kind of distill that, it's like what companies can provide the best context for an AI agent to operate and have the right workflows and interface components for the company to deploy those AI agents.

34:58Like that, I think, will persist going forward. And so I think that puts a lot of traditional SaaS vendors actually in a good position because they often have the users, the data, and the workflows. So I'm bullish on the sales forces of the world and service nows of the world and the workdays of the world. And obviously the box, you know, box specifically because of the amount of data that we work with and the ability to deploy agents on that data with the right security and governance. Now, where it gets fun and where there's plenty of opportunity is I think AI has expanded the potential use cases for software by, let's just say, an order of magnitude, which means that incumbents, you know, you should only really by default bet on for doing the AI things that are in their defined categories.

35:43And that means that there's almost, you know, still multiple X opportunities for new startups to emerge where there's not necessarily a natural incumbent. where it's a crack between two products that a new AI agent makes sense to kind of fill. It's a non-software service, probably a professional service of some sort, and a software incumbent never went after, that all of a sudden agents can now go do, which now creates new software companies. So I'm both extremely bullish on the startup ecosystem and actually bullish on a significant portion of then the SaaS products that I think are in a good position to deploy agents within their software boundaries.

36:21So you saw the cost of hardware or storage basically go down like a ton, right? What do you think is going to happen in the industry of the cost of software goes in a similar speed down? So there's obviously a hobby idea, which is in a world of AI agents, let's say AI agents themselves get a lot cheaper. You could build as much software as you ever want, which means that any weekend, you or I could go create an ERP system if we wanted. in the limit, right? So then what should that do? I think the kind of microeconomic hat says, you know, the cost of a service goes down, that's going to drive more pricing competition and so on.

37:03I like that argument. I think it's very plausible. And so, you know, should your incremental renewal in 10 years from now be cheaper on an inflation-adjusted basis than it is today because AI agents helped build all that software? I think we can assume yes. Why? I think it won't be as pronounced as maybe one would maybe do if you just didn't understand software. And you just said, like, well, shoot, my inputs have come down by two orders of magnitude. And I can now sell that thing, obviously, then for like, you know, like 95 % less fewer dollars, like that should obviously radically re-architect the entire industry.

37:39The reason why that won't happen is that by the time you're like the Ford Motor Company and you're buying an ERP system that's going to run all of your global supply chain and financials, you want a company that understands your business, that you know you can call in the middle of the night when there's an issue, that is wired into every component of your system, that is like 1 ,000 % reliable. It's got 99.99999 % reliability. And when you stack up all of the needs of the Ford Motor Company and you say, so are they going to go and swap out their ERP vendor going from paying maybe, you know, $100 million or whatever the right number would be to$5 million or$10 million because, you know, that company still wants to make some amount of money.

38:19Is that delta worth the operational difficulty of doing that, the business risk of doing that, the long-term kind of relationship with that vendor of doing that? I think you're going to find that there's going to be a large number of times where it's not worth it. And what you're actually paying for, that extra premium on top of just the bits being written in a server, is you're paying for the overall deployment, partnership, management, service delivery, functionality that's going to be sustained that an Oracle or SAP or whatever is providing. So I think what the impact of this is more likely going to be the case is we're going to get more software for more categories.

39:02It's a great time to be a small business because small businesses will now get the equivalent of the highest tier types of technology for their business, you will have increased competition. You will have some deflationary pricing pressure. But I don't think it'll be as wholesale of a shift in the economics of software as some have written about. Do you think that like, so for example, one of the things I kind of look at software is that there's this whole category of customizable software, which is like SAP is like, it's a product, but you can basically do almost whatever you want on top of it.

39:33So a lot of software is like you have the 80 and then the 20 is you kind of have to adapt a little bit. And then if you pay a lot of money for that, you think that might change that? Like companies might want like now 100 percent of their needs base versus just buying something off the shelf. So you're saying take the 20 percent and make that 100? Yeah. You know, the reason why I'm not I'm not 100 percent bullish on that is because because actually a lot of times your software, your software is actually helping you solve a lot of internal workflows that you don't always appreciate. and it's sort of like reducing the decision set by like, again, multiple orders of magnitude.

40:09Like, do I really as a company, am I really going to like reinvent HR if I do HR from first principles and then go and define that in a new piece of software that I custom write? I'm less bullish on that than just saying, you know what, like HR is like not the thing that we need to reinvent every aspect of, payroll, all these kinds of things. So I actually prefer that that is just like in a system that knows how to do that. It's done that a trillion. It has done it's done a trillion transactions over the last year. And, you know, that's why I'm just going to like pay a few hundred thousand dollars or whatever for workday or if I'm a much bigger company, you know, obviously, you know, relatively more.

40:49And so I think there's a lot of software where like the company is not going to get that much business impact by over customizing it. And so that's not really the place to go focus even as a corporation. Jeffrey Moore has this idea of core versus context, which is basically when anytime you're thinking about resource deployment, you basically want to know what is the core of your value proposition, your business model, the DNA of the company versus what's the context. If you were 99th percentile in this one area of hyper-tuned for your business, it's still not going to drive a customer's purchasing behavior one way or another.

41:25And in anything that's in context, you basically want to outsource. You do not need to – and outsource, again, in the software world, just means pay a vendor, handle it for you. You do not need to have an IT team specialized in that one specific domain. It's just not worth it. The company is not going to generate more revenue for that. Most SaaS is in the context category for a given customer. Most customers do not want to be in the business of building a file system, doing workflows around files, managing all that. So they could vibe code that if they wanted. Like, let's just, again, in the limit, they could.

42:00But still, like, it is just not, like, they're not going to get more customers because they built their own software for that particular part of the stack of the problem. And so the same thing is generally true for HR systems, CRM systems, ERP, which is the whole reason why the enterprise software industry exists is because most people have very similar needs, at least by industry, for those kind of core tasks. Now, I think what we're going to see with either Vibe coding or just, you know, let's say AI generated code is that the core of the company, that's where you could now do way more. And so this is the exciting thing.

42:34And this is where I'd be spending way more time for, you know, as any average enterprise is like, holy crap, your core. Now, if you were only doing the basics of software customization for your core, now you can do that 10x more. If I'm an insurance company, I have to find a way where I can process insurance claims faster, deliver higher customer experience faster, have some kind of actuarial model that is sort of better than my competition. And that's actually where I'd go apply as much engineering effort as possible to build the best software for that particular use case. and not necessarily kind of outsource that from a vendor standpoint.

43:12If I'm in banking, like, and my ability to create the best customer experience, the best dashboards of my product, the best proactive ways of messaging customers with new financial services instruments they need to think about, like, that's where I'd spend my time, right? So not, you know, expense management. Like, it's like, just like handle that by just signing up for a service. So I think in a world of unlimited AI agent coding labor, you now can do the core in much more interesting ways, which occasionally will be an ERP system, to be clear. Your company might be so bespoke that you do need that wired up for your particular workflows.

43:51But I just don't think that's going to be every enterprise out there. Well, ERP is a big category, too. Some parts of the ERP may be core, the other maybe not, right? That's exactly right. So I think maybe we'll just get better differentiating which is which. And so I know that it's like, I'm not making like really hard trade-offs here, but like, I think the conclusion is your SaaS might be more valuable because it's the natural place for agents. And you still might create way more new code in the world for all these, you know, context things, which means that the cursors and replets, you know, their value goes up because we actually can just deliver software to more areas of the business than ever before.

44:26So one of the questions, I guess, as operators, you always think about is this company design question. How do you get communication across your employee base? Who needs to be into meetings? There's a lot of managing a big company that's just information architecture. Obviously, with AI, this seems to change a lot. I guess you guys are probably really well positioned to help lead the way there. How do you think about the problem? Well, okay. This is probably the longest version of the whole thing. So I'll try and compress it, because this is the multi-billion dollar, if we do this right, opportunity.

45:02If you think about all of the information that an enterprise has and it creates, every meeting you're in, every Zoom call you've had, every project plan, every board presentation, every internal memo, all of that is effectively digital corporate memory that has been created. And up until, let's say, a year ago, if we just snap an arbitrary line, You created those things. You shared them. People saw them. They get stored somewhere. They're still in some server somewhere. Nobody is ever going to see it again. Every email you've ever written to the company, every project plan I've ever shared with 20 people, there's an instant half-life that basically by three months later, you're going to get one person to see it every six months at that point.

45:45And then the ultimate end state of this is that your most valuable information is not only never seen again, it's basically its only value is probably for some like retention policy in some IT system. So like the most valuable information I've ever produced in the company is sitting in some governance policy in some kind of archive mode within our corporate box instance, let's just say. Okay, AI just completely inverts that. So all of a sudden, anything you've created that is valuable in any context in your business, all of a sudden can be used over and over and over again in perpetuity forever.

46:21And then it's just about how do you get employees access to the right versions of that? How do you make sure it's authoritative versus they don't need to see every one of the memos that was the draft memo or the draft document that wasn't the thing you decided on? So how do you effectively create an authoritative set of corporate knowledge that is available to the right people at the right time that they can interact with either by asking questions or sort of proactively serving them information? And that's our whole strategy at box, but it's more globally like one of the biggest benefits of AI, which is all of a sudden, you have a living, breathing memory of corporate knowledge for any employee in the company, which means that the brand new employee that started yesterday could ask a question effectively to the smartest person or infinite set of people in the company and get an answer back instantly.

47:08When in real life, that was never possible before. You cannot be, by and large, a first year out of college engineer and go and just have an open Slack channel with our CTO and ask them every single question all day long, 24-7. AI lets you do that. AI basically says, I can take everything ever written down by our architecture, our project plans, our product strategy meetings, our QBRs, and I can ask any question I want at any time. What's going on with this part of the architecture? I ran into a bug over here. When's the delivery date for this thing? And so you're not going to have 900 engineers be able to ask that question of our CTO or head of engineering all day long.

47:49Now you can with AI. That's the power of this. So it completely flips the communication model, the knowledge model in the enterprise. This idea of companies going AI first means that there is going to be this transformation of access to information ultimately via AI. So all very exciting kind of times for us. And what do you think, I guess, like there's this idea that like your tools end up shaping you, right? Like, you know, we built cars and then we built all the cities around it to be designed around cars. Now that we're going to have this new tool, right? Like, how do you organizationally change?

48:22What is the, in your view, or your hypothesis, at least, of like a better organizational design or process, given that we have this tool now that we didn't have before? Well, yeah. So this is where, because of my lack of imagination, I'm going to, if we just put this in a time capsule, and maybe we could like check back in five years or 10 years, whatever date you think is appropriate. If we put this conversation in a time capsule, here's what I would say to my future self. I think that when you ask somebody for their org chart, a company, large company, mid-size or large company, I think it's going to look similar.

48:54I think you're going to have the head of marketing and the head of marketing will have other heads of various marketing functions and so on and so forth across the organization. What will be different is that when you get down to the content strategy person, when you get to the person that writes the white papers, when you get to the person that does the webinars, when you get to the person that generates the ad copy for the online campaigns, they are going to have 50 AI agents working for them, dramatically expanding the capacity of what they can do and capability of what they can do. And then humans across the kind of chain of decision making or whatnot will be responsible for collating, curating, incorporating, orchestrating that work to whatever the ultimate value is that they're trying to create for customers.

49:45And so I think it'll be, you know, and again, hashtag complete, maybe lack of imagination. But like the thing I would bet on is that many things that we are used to in the enterprise look the same. The output levels that we have is what 10X is. And so when that person normally would be going and saying, hey, let's have a marketing strategy, they are going to infinitely run 20 simulations of every marketing strategy that they could do. They're going to go and be able to deploy that across 100 different markets and in every single language that's relevant with hyper-tuned ad copy for every single type of customer that they want to go and do.

50:28And so in that world, that person has now the output of, let's say, 10 or 20 people in the previous world. But as a result of that, that company is now doing more, they're selling more. And so something else in the organization now emerges as, okay, now we have to hire more sales reps, or we have to hire more engineers to build those features. So that's how I think this evolves. Again, totally willing to be absolutely, you know, absurdly wrong on this. But partly just by being in this space now 20 years, watching kind of corporate evolution, you know, if you remember, like, 15 years ago, Zappos had this idea of, like, we're going to have a holacracy, everything's flat, and it's all like these tribal things.

51:06I think, you know, some of these kind of very radical or the Dow era of crypto, I think a lot of these radical ideas just don't pan out, maybe just because we just like actually have a relatively well-optimized form of capitalism in the modern era. And then agents and AI ends up being just an accelerant to everybody's actual work within the organization. How do you think about the staging for that? Meaning like, okay, I understand that that's the vision in like five years. But given the tools that we have right now, right? And what we can do with AI right now, is there anything that you see as like, well, this clearly doesn't make sense anymore because, you know, we can just have AI do it.

51:46Or like, there's some shifts that you're already seeing kind of in the shorter term. Yeah. So, so we sort of have this internal concept of, if you just imagine like a two by two, one axis, let's say X axis is sort of frequency or repeatability of a task. Y axis is like level of critical thinking or like just intellectual horsepower you need for that task. And, and you just plot all of work, just like take all of work and you put it onto that sort of two by two, you do see some areas where like, okay, if an agent could perfectly do X thing, then obviously that would just mean like a function has meaningfully shifted.

52:20So frontline customer support is like the quintessential example of this. If we can answer 90 % of customers' basic questions, how do I download this thing? Why can't I log in here? How does this upload thing work? How do I change my password here? If we can deflect all of that with AI, by definition, we're going to have fewer people doing those types of emails back and forth with customers. And now even pre-AI, a customer could have just gone to our documentation and somehow found the answer. But obviously, that's just like, you know, inefficient for a lot of customers. I don't even like going through product documentation.

52:54But AI will obviously, you know, deliver those answers. So now the company has a choice. And I'll just tell you what we do internally when we have these kind of choices. We then say, okay, we're saving some dollars in this type of inquiry, type, we could free up those dollars, put them actually into relatively a similar part of the org under the customer success operation. But now I want to move those dollars into humans to do more customer success management, which tends to be a more strategic, more outbound, more proactive form of customer success because you're going into a customer, you're saying, hey, here are other use cases you could use box for, or hey, you know, here's a better way to get more value from our product.

53:35We are always in short supply of those types of individuals, mostly because of just our annual budgeting process and amount of available talent. And so now this gives us an opportunity to take dollars from one area that we're automating, put it back into the business, possibly even retrain a number of folks in the process, and actually then deliver better value for our customers on both sides. We're now being more proactive and the inbound requests are being answered more quickly. So I think when you plot out where AI efficiency comes from or AI kind of augmented work, there's going to be way more of those situations, which is, OK, I've automated this one thing, but on the other end pops out either a new need or a new opportunity that I can now go apply either AI or humans to.

54:14And that's mostly what we're seeing. But, you know, like maybe one more example that we're only in the early midst of. But so we've, as I mentioned, taken all of our corporate knowledge. We make it available to employees. We have a product called Box Hubs, where you go into a hub and you ask a question, and those hubs are for everything. We a marketing hub, a competitive intelligence hub, et cetera. And so if we can make our sales reps 5%, 10 % more productive, they're in front of a customer, they ask a question, they have a better answer for them. They have a better tuned message and an email campaign for a customer.

54:47If we can make our sales reps more productive, the immediate next area that we will take those extra dollars that were generated because of that productivity, guess where that's going to go? Back into sales. We're not going to just treat that as now additional profit. We will just go and reinvest it back into sales because sales productivity is like the holy grail of software. Anytime you can get sales productivity up, you should plow more capital into sales because that's obviously now a higher efficacy part of your business model. And so there's a lot of things that get written that I totally disagree with, which is like, oh my God, this AI agent now made the sales team's job 30 % better or 30 % faster.

55:29We're going to have 30 % fewer sales reps. I read that and say, I think we might be able to have 20 or 30 % more sales reps when you get that efficiency gain. And so by making all of these functions more productive, more efficient, I think you're actually seeing more demand in the economy for a lot of these types of areas of work. I guess stay on your sales and marketing example. One of the things that I noticed a little bit of my personal behavior is I'm a lot more buying than getting sold, you know, post-JJPT, meaning like, you know, I kind of go more and research through the AI kind of before I even engage with like a company.

56:07I'm curious to hear your thoughts on like, how do you think behavior is going to shift and how is that going to change like the process of selling and marketing to customers and enterprise? Yeah, I think that you have this now unlimited capability of research, which absolutely means that you're in a much more informed environment on the customer side. You know, we've always tried to assume a world of, let's say, the customer had perfect information, right? How would you then be differentiating? How would you be researching? Or how would you help them, you know, kind of, you know, find the value proposition?

56:36But I'm not, I don't have any, you know, kind of crazy, you know, thoughts on where the future of that goes. I think if anything, I mean, this is what's amazing. If anything, it might put even a higher premium on the sales rep and the in-person component and their ability to build relationships. So, you know, anytime you think about a world of abundance, you should always think about what then becomes the new area of differentiation or what's the outcome of that abundance downstream. So in an area where every customer all day long, you know, wired into their brain, had like unlimited information.

57:12Now, how do you stand out? It's like, okay, you got to go spend time with them. You have to, you have to, you know, give them the full tour of your, of your product. You have to, you have to probably pre-install a version of your product so they see it working in their environment. Like think about all the work that then will go into that. That will be the new form of tasks that we go and do. And so, so that's why like, like, I just, I just could not more vehemently disagree with most people online on these things. Everybody online looks at how work happens today. They look at AI efficiency. They're like, okay, 20 % efficiency, which means 20 % of jobs.

57:45And it's like the least imaginative way to think about what happens next. Anytime that you have AI enter an organization, they'll make one part efficient. That efficiency will lead to demand or set of things in another part of the organization that you then have to go and fulfill. Or your competition will also get that efficiency, which means you have to find a way to outrun your competition by doing more, which means you're going to reinvest dollars back into the business. So most of these claims of the sort of broad scale job changes or whatever, I think will be proven wrong in the long run.

58:16So I guess what you're saying is that, look, hey, if you're a company in a competitive environment, you cut some costs. If you don't reinvest, your competitor will, and then they'll end up winning against you. So you're going to kind of have to reinvest anyway, which is a weight of basically the efficiency going back to the consumer, right? Or the enterprise. Yeah, I think ultimately the, I mean, I couldn't put it better, like the efficiency gain will be captured ultimately by the consumer in the form of more product value, lower costs, more long tail of things they can now do in that particular service category.

58:53but I don't see a world where like you'd have to basically imagine one of two outcomes. Either somehow it was like the world allows companies to just run at all like 75 % profit margin and then like nobody like Bezos comes in and tries to just wipe that out. Okay, so that's almost impractical or the world is just totally happy with only today's level of service and so yes, we will compete out that profit margin but we don't innovate any further from where we're at today. We all are kind of doing the same things and AI just replaces the work that we're doing. That's implausible because I know of healthcare issues in the world.

59:27I know of education issues in the world. I know of consumer product ideas that I would love to be able to purchase if they existed. There's an unlimited appetite for more things in the world and there will always be. And so as long as that happens, that means you're gonna constantly have new companies. You're gonna have new services those companies wanna offer and AI becomes just another tool for them to go and do those things. Maybe one more point that, you know, maybe you guys would be in the best position to be able to kind of highlight for folks at some point. But my view would be, think about the small business that is five employees today.

1:00:00Think about that small business. What is their ability to grow at an exponential rate 10 years ago? You're basically almost always dependent on like, can I hire that world-class marketing person? Do I have enough money to bring in that ad agency? Do I have a totally cracked 10x engineer that can build that feature, you're always dependent on that. So of the millions of small businesses that get started every single year, whatever the right metric is, how many of those people, how many of those businesses actually have access to any one of those things that can help them grow their Shopify business faster, that can build the next feature for their customer that will generate more revenue?

1:00:36All of a sudden, you've whittled down probably 97 % of businesses. AI. Let's imagine this perfect AGI world of like an AI agent for everything that does everything perfectly. Okay. I go and I just launched like my local cafe and I'm like, I need more customers or my local barbershop, or I build a widget that I want to sell online with Shopify. And I'm like, okay, run the ad campaign for me. Help me with the supply chain, build this feature for customers. You really have to have no imagination to not think through what happens next. That company grows a little bit incrementally faster. They have to hire more front, you know, frontline people to help with that growth.

1:01:12That company grows, that feature ends up generating more revenue. They now need more engineers to then handle all the bugs and scalability things that then begin to emerge. The widget manufacturer grows and now they need more supply chain support. Like what we should see, again, if basically, if AI is as good as the people that are fearful of AI believe, then you will see in five or 10 years from now, all of these downstream positive impacts of growth that we would not be able to just anticipate by putting this into a little spreadsheet. And that's, I think, what you'll start to see more and more over time, is the access to unlimited capability by every business on the planet will create a self-reinforcing cycle where they grow more, they deploy more people that deploy more agents that help them grow more, and that just cycle repeats.

1:02:00And do you think that this adds to a scale benefit or it removes from a scale benefit? And what I mean by that is like, you know, like let's get softer, for example. One version of the world would be, look, you had software companies in certain categories. Now, if the cost of software goes down a ton, you can go into a lot more categories, leverage your existing Salesforce, bundle, et cetera, et cetera. But at the same time, new companies can be created much cheaper, but also copying these new companies can be much easier. So is scale helping more or kind of the same or reducing? Just because of how good I think a lot of incumbents are, they've found a way to kind kind of neutralize when these asymmetric advantages emerge.

1:02:39Mobile was one early on, which was like, wait a second, now I have a distribution channel on the phone where I don't have to go through the kind of classic gates. And then, you know, companies like Facebook figured that out, co-opted it through acquisition in many cases. And then we're able to kind of recreate that same velocity. So I think when you have really good incumbents, they find a way of neutralizing some of these advantages. But when you zoom out over a 20-year period or 10-year period, what you end up seeing is that there's just simply way more than startups that do reach scale. So weirdly, it actually is that the incumbents get bigger because there's more for them to do.

1:03:18And there's a way longer tail service area for startups as well to emerge. So I think I would just bet on both happening. And I think that we might be in an era where the mental model is not startup versus incumbent. It's just simply what white space have we never had software for. And it's just a race for filling out all that white space. And it's sort of hard right now to predict in any one category, whether it's the incumbent or the startup, as much as we're just going to get a lift on everything. So just to close it out, Aaron, when looking back to your trajectory, I would love if you could just point out in your view with hindsight 2020.

1:04:01what you got the luckiest about? Yeah. What you got right the most. And I think, you know, you said not by luck. You actually saw it and got it right. And kind of what were the biggest mistake? Oh, boy. So I think luckiest were probably the areas of like, just like the conditions of the environment that I was in. I got lucky, you know, where I was raised. I got lucky by who my parents were or my friends that were supportive or, you know, really interested in ideas. I had a core kind of, you know, group of college friends that I was the only one doing like totally random project, but they were like totally, you know, chill with it and very supportive of these ideas.

1:04:42So those were all kind of like very lucky conditions. And then like, you know, born at the right moment, you know, all those things actually do kind of weirdly matter. Like I was in college at the time that cloud and mobile were just starting off, which meant I didn't have like any other responsibility. I had nothing else to like drain my mental energy in different directions. And so those are like factors I had no control over that I totally lucked out in. Okay. Areas that I think it's Jim Collins, I could be wrong, has this idea of return on luck. And so it's kind of like ROI, but for luck. Like so, you know, if you have these lucky conditions, like do you exploit them to the maximum degree?

1:05:23And so those are areas where we tried to exploit a lot of these conditions of, okay, early in cloud, early in mobile, how do we exploit those? And so then the things that we did well, given that, were, I think, a very strong architectural discipline early on. We didn't let ourselves sort of overbuild in lots of different directions that made it then kind of very hard to fully take advantage of the momentum of cloud. some companies were like, oh, we'll have a cloud offering and an on-prem offering. Well, obviously, when you do both of those, you're then drained resource-wise. And many of our competition at the time that went that dual path, they don't exist anymore because they just never got escape velocity in one path or the other.

1:06:06Maybe one other lucky thing, Mark Cuban was an early investor. Very good advice early on on not hedging your bets, which I think helped then inform being more disciplined on architecture because it was so clear that you don't want to have dual paths. So architecture, team, culture, design, really retaining a strong ethos around the culture of the company. And then being disciplined on strategy on, okay, once we make the decision to go enterprise, you fully exploit that to the maximum degree. You don't do like half enterprise and then you're kind of like doing some other stuff on the side. You do enterprise.

1:06:42You do cloud. You go very deep in the areas once you know that they're working and not get distracted by whatever the flashy thing is. There's a difference between ADD of being excited for new innovation and doubling down on the core and ADD of I'm going to have 19 business lines. I'm going to enter these spaces. We've always been pretty good at, okay, double down on the thing that's working. With the help, to be very clear as a shout out, my co-founders were very helpful on actually constraining me at times. So I probably would have left alone. I probably would have overexpanded. And so my kind of founding team was very good at keeping me on the rails on that.

1:07:22And then decisions that I look back on are usually things around, could we have moved faster in the pivot from consumer to enterprise? Would we be one year ahead in compounding if we had done that one year sooner? Certain product decisions where maybe I would have pivoted or shut something down sooner early on. And I think any times where we've let any kind of culture decision linger too long, where you just have to make the right call quickly, I think those have been very important. I probably would have focused on getting to cash flow positive earlier. We burned a lot of cash, which was very helpful to sort of establishing ourselves in the market.

1:07:57But could we have done that with a better eye toward the cash flow part on just optimizing either spend or expense? I probably would have done that. But those would be some of the core things. Awesome. Aaron, this is amazing. I really appreciate you doing this. Good to see you. Appreciate it. Good to see you. 100%. Thanks to our friends at Atomic Growth for helping with production and distribution.