SaaStr 809: Why Enterprise AI Adoption Is Moving 5-10X Faster Than Cloud with Box's CEO and Co-Founder, IBM's VP for AI and SaaStr's CEO and Founder

0:01Welcome to the official Sastr podcast where you can hear some of the best Sastr speakers. This is where the cloud meets. Up today on the Sastra Podcast. If you've already opted into being in this ecosystem, you're probably thinking like, okay, I'm building a company. I'm going to hire a sales team. I'm going to do the marketing thing. Lowering the reduction of how easy it is to build these things doesn't make it any easier to build one of these things. The thinking through the long term of building a company and building a franchise business over the long run is just very different. And I mean, IBM has sort of proven just like what that takes over, you know, obviously decades.

0:41And so I do think that we will have to separate the sheer sort of like, holy crap, it seems like there's 100 copycats of every single space from the companies that are just like, I'm in this for the long run. And we are going to keep cranking day in and day out, iterating, building a better product, better serving our customers. the things that transcend AI will become probably then even more important. Hey, everybody, get excited. We just hosted 10 ,000 of you at the Sastr Annual AI Summit in the SFB area. It was insane. It was off the charts compared to last year. It was a deep dive on everything new, everything AI, everything go-to-market.

1:18And we're getting ready because Sastr AI is coming to London in December. It's Christmas with Sastr. On December 2nd and 3rd, we're bringing Sastr AI to the heart of Europe. This is your chance to connect with thousands of SaaS and AI executives, founders, and investors, all sharing the secrets to scaling in the age of AI. If you're a founder, a VC, a revenue leader, SaaSter AI in London is where the future of B2B meets the power of AI. And we just announced tickets and sponsorships, so don't wait. SaaSterLondon.com to grab your tickets. SaaSter AI in London, where B2B meets AI and the next wave of innovation begins.

1:53See you there.

1:59Hey, everybody. This episode is brought to you by Adio, the AI native CRM. Just connect your email and Adio instantly builds a powerful CRM. With every company contact and interaction you've ever had, get 15 % off your first year. That's 15 % off your first year at attio.com slash saster. That's adio.com slash saster. All right. All right. So, all right. So thank you, everyone, for joining. Thank you, guys. So Jason decided to make a surprise visit, and he's going to... We just found out 45 seconds ago Jason was going to join us. So I found out last night. Thanks, everybody. All right. So, Aaron, we're going to start off.

2:43Yeah. AI agents is probably the hottest topic in the entire software industry. You guys have been early adopters of it. Could you just walk us through, I mean, what you've been looking at, what you've been doing and share some stories about that uh maybe like raise your hand if you're building an ai agent okay good we're at the right conference so a lot of ai agents and i think for us for for everybody else probably in software there was this moment like two years ago where the conversation was okay are we all going to have like a chat interface on our software it was the chat should be team moment of AI.

3:23And everybody was in this mad rush of like, should we put a little chat bot? Should it pop up while you're in the software? Should you change the core user experience? That was the primary conversation. And it was a user experience conversation. It was more of like an HCI type conversation. But underneath the hood was the really big idea was forget focusing on just, it's a change to the user experience of software. Think about AI as actually a change to the labor model of using the software. So what can you do in a world where you can effectively create digital workers that can go automate anything within the confines of your software?

4:05So that's kind of been, I think, what we've all been up to for the past year plus is how do you turn these models, especially now with reasoning capabilities, especially with tool use and better context windows for managing objectives, what if we could turn these things into agents that you can go deploy a hundred times, a thousand times, a million times inside of your software instance? So for us, that's been allowing for AI agents that can do things like read your documents, summarize information, extract data from documents, automate workflows. We have a great partnership with IBM, with WatsonX, where we are bringing model choice to our customers, powering things like the latest LLAMA models, integrating with things like Orchestrate.

4:44So the whole idea is what if you can have agents across your enterprise software talk to other agents in other enterprise software. And I think for all of us as an ecosystem, this is going to completely change what interoperability looks like, what the future of building software looks like, and obviously a lot to unpack. Yeah, so you brought up one interesting area. I love your thoughts on this. Just for more of the audience's understanding, agents versus assistants. Yes. Because we saw the era of assistants and everything was assistants at that point. What makes agents different? Yeah, I think, I mean, this is, everybody has their own probably different nomenclature and vocabulary for this.

5:24We kind of think about it as models. You know, we've always had, we've had AI models or some version of ML models and then kind of some of the initial transformer-based models for the past, let's say, five plus years. Then we had these assistants, things like ChatGPT, et cetera, are really kind of these assistants to go back and forth with. And now we're obviously seeing this era of agents, which really fundamentally go do work for you. And that work could take a minute or it could take an hour or it could take a hundred hours for the agent. And then in human time, that could mean days of work or months of work being compressed.

5:58And you'll deploy these agents across your software stack to go execute work within your software environment. And so deep research from OpenAI, perfect example of an agent. I want you to go do a bunch of research across the web, come back, provide an aggregated answer for me. and the powerful thing unlike a traditional just query with a model is it's not just going to think for one second come back with an answer it's actually going to go do basically what a human would I'm going to read all the documents I'm going to check my assumptions I'm going to change my course of actions based on new data that I learned in the review process I'm going to correlate information and then go and actually collate it in some sort of ultimate asset at the end of that And that's now what is becoming possible with agents.

6:46We've seen things like operator from OpenAI, where an agent can actually control a browser. That becomes incredibly powerful. Imagine if you could deploy a thousand agents in browsers, and all of a sudden in the cloud, you have a thousand browsers running with AI agents doing work for you. I want to go scrape data from different interfaces. I want to go do QA testing on software. I want to have an agent that acts as a security researcher finding different vulnerabilities. So we can now effectively think about our software as having labor attached to it, which for all of us building enterprise software is like a completely mind bending concept where we're no longer just limited by the number of people that we sell to in an organization.

7:27So if I'm building legal software, I can only sell to the 10 lawyers inside of a company. Now, all of a sudden, my software comes with infinite legal capacity and it opens up all new kind of addressable market for our businesses i'm sure we're you're seeing that throughout the conference can i ask a just really tactical question maybe to both when you talk to customers because aaron you talk to so many customers right half your week is customers probably 30 percent raj at ipm do real world customers do they want agents do they are they or do they just want ai to solve their problem what i worry i know this sounds specific and maybe the trains left the station on agents right but do customers really want a thousand agents or do they just want are they still learning what ai is and i want to talk to my documents i want to solve my problem do they really want do businesses want a thousand agents we're building them don't get me wrong but i would say they process this beauty of agents and what we're seeing is there'll be thousands of agents working in the background but typically one interface.

8:25In the background. And there'll be one interface for the client. Yeah. And so that's where the beauty lies is because you're able to have disparate systems start talking to each other, but you only have one interface. I'll give you a perfect example. IBM went down this agent journey back in 2023 and we've actually been using it internally. And last week we announced, we saved over$3.5 billion internally from cost savings because HR functions are there. procurement functions are there so every different area that you go down and it's only the same is exactly the same interface so i don't even know in the background if i'm talking to success factors or what other products i'm talking to i'm actually just talking to the my watson x bot essentially for sure yeah i think i think you you know you're bringing up a very prerogative question and we should probably remove you from the stage for for the fair enough this is an ai agent conversation okay i apologize but uh no it's very question like this is uh this is kind of classic clayton christensen jobs to be done yeah what what is your product being hired for it's a good way to always think about you know sort of solution a product market fit solution fit and the underlying kind of theory of of clayton christensen was like we don't go buy products we buy effectively we hire solutions to solve our problems and so what is the problem that an enterprise is running into well, if they're reviewing legal documents, they want to be able to just get lower risk in their customer contract negotiation.

9:56They want to have better renewal rates. They want more accelerated deal cycles. They don't want any agent for that. They want that as the outcome. And so obviously, collectively for all of us, the closer we get to conveying those ultimate outcomes, the closer we get to pricing towards those outcomes, I think that will be the natural direction this space goes in the agents at least for lack of a better term that we've landed on at least describes what's happening behind the scene that's causing that deal cycle to go faster behind the scenes you hit it on the head yeah typically you don't know what you know and i know but you most people don't know what cloud you're running on for example it's all happening in the back end yeah magically yeah but you know it's same way we should look at agents because all this is going to be happening in the back end.

10:45What's really interesting is, and Box was actually one of our first partners to join us on this, but we opened up an agent catalog last week where companies that are creating agents can submit to the IBM catalog and then IBM sales teams can sell it. So even if you're a five-person shop, you submit your AI agent that you built specific for a use case. Within this catalog, you're able to take that and then the IBM sellers are able to sell it for you. So I think that's important, especially for the groups that we have here. Because if you're an earlier stage company, you're looking for that traction.

11:18We're enabling that. And obviously, Box doesn't need to help, but Aaron's been a great partner on this AI journey. And the moment we talked about it, he's like, I'm in. Let's go. You've got this IBM Salesforce, which is legions in size, right? And they are now recently, I should know this, but I don't. They're off selling agents. And how many? We kicked it off last week. Okay. And I'm just, I'm very, how many do they sell and how do they get smart enough about each agent to sell it? So it becomes a use case specific. So if you look at the way that IBM's mentality is, is not to go industry specific.

11:53We go horizontal. We want folks that have the industry expertise to leverage that. So, you know, as the seller is looking at various AI use cases and clients are starting to ask for it, that's a great opportunity to say, well, within our agent catalog, we have this. will even make it self-service at one point. So even the sellers don't need to touch it. If a customer wants to go on there and just download it from the catalog, essentially, they could start leveraging it. But what an interesting time though to be in software, right? Just because think about what you just kind of said. So IBM is going to be going to a bank or an energy company or a life sciences company.

12:29I'm making this up, so correct it. And basically being, they're going to be bringing digital labor to that organization. it's no longer a software sale it is not you know let me go find your process and i'm going to sell software for that process it is i might actually bring the process to your organization in the form of this digital labor so it's going to actually kind of shift the deck chairs on what what's actually happening who's what like i think we're all going to increasingly look almost like professional services companies probably to some extent because you're bringing now solutions to the customer as opposed to just, A, use our software.

13:07Hopefully you have the people to implement it. Hopefully you have the people to even use it at your own risk. I think this is a fundamental change literally in what software is going to be procured for. You know, another interesting thing I'm seeing, which I didn't expect, is if your AI is very good, broadly speaking AI. So let's imagine these agents are going to be great. What is really interesting is it can make folks... IBM has an incredible depth of customer relationships that people don't even get. I mean, incredibly deep and broad. Box is my favorite customer, by the way. Mine as well. You can be...

13:47Older companies can be much better with AI. Yeah, absolutely. Because maybe there's a few parts of IBM, not Watson, but a few parts that maybe are a little older. But if the AI is great, if the agents, if I can walk in with the box agent, you can tell me what it's like. All of a sudden, that team can be much more agile. I'll give you a specific example if you're curious. But like one one that Zendesk is much better than a year ago. Yeah. But I've invested a lot in contact centers. I'm on board of this company called Gorgeous, which is like Zendesk for e-commerce. Just like you have HubSpot. We have HubSpot.

14:22Shopify is Klaviyo, which is a billion, right? If you go over to that world, the Shopify world, it's not Zendesk. It's gorgeous. Over there, Zendesk used to be old. The plugin was a little creepier for Shopify. It couldn't really interact with FedEx, so you returned. So they won 20 ,000 customers, right? Zendesk is probably still creaky under the hood, but the AI is bitching. The AI is bitching. So Zendesk in one year - That's a technical term. Yeah, it's a technical term. It's an NGS. It comes from the CSAT Bible of bitching. We need a bitching benchmark, AI benchmark. We know what we need is a bitching agent.

14:54okay yeah so they're they're still a better in this vertical right but the gap is narrowed so if i'm the ibm sales team in one month i might actually have much better solutions to sell with these agents yeah like overnight it could change it's it's kind of crazy we always have to kind of recheck in with uh to our assumptions and and biases on what the state of the art is on the technology because things that literally three or six months ago were not possible all of a sudden become possible. And so we like the rate of change in this space means that first of all, it's like the most stressed out I've ever been because of good sign.

15:31Yeah. I mean, if that correlates, the opposite is worse today. That is, um, I mean, I kind of enjoyed what it, when I wasn't stressed, but, um, uh, but, but my stress is absolutely a corollary to how interesting the technology industry is. So, so it's a good sign for all of us because it means it's a very dynamic time and lots of stuff is happening. But to this exact point, if you go back a year ago, the amount of work you would have to do to pack into, let's say, a non-reasoning model for giving it exactly the right sort of context, instructions, kind of hacking and tool use, versus today, just having 03 go do that, or having the new Lama 4 go and do that with reasoning capabilities, you instantly have now caught up probably an entire year of development from other companies with just a single model upgrade now is happening.

16:23And so it does mean that it's like a nonstop level of change that we're seeing in this space. I was going to say, like last year, I think the big buzzword was rag. Yeah. And now we've seen the shift for agents, but that happens so fast. So fast. Right. And the interesting part about this is, as we look at all these tens of thousands of agents. We're also providing the capability of if companies, I saw like 50 % raise their hand that they were building agents. We actually released agent builders last week as well. And then on top of that, we already have pre-built agents. So what we used internally are Ask HR, Ask Procurement, et cetera.

17:01We've actually released that and productized it. So now folks can, you can actually just purchase it versus spending time to build it. So we're opening up the aperture because as Aaron and said it's moving so fast that if we don't keep up with the market, that's why he's losing sleep. I'm sleeping great, by the way. I don't know if the tech is like 1am. Yeah, it should be stressful, right? The rate of change is so high, isn't it? Even just take something as simple as let's say GitHub Copilot, right? Basically invented the current paradigm of AI coding, was obviously the de facto standard, and let me just do a dangerous survey how many people are using cursor or windsurf or replit right now okay so like last year it was close to zero last year that would have been zero and now all of a sudden we're already seeing a kind of a tide change in the direction of a new set of of startups that are attacking the space and so i think right now we're in an environment where nobody can take any of their position for granted it's a fast-moving space i do think and and we'd love to even hear what you're seeing from startup plan what you're thinking about from the customers you work with, I do think you're starting to see a pattern of how to have moats begin to emerge, this sort of relationship between the agent, the data that it works off of, the level of sort of domain specificity of that agent, obviously the capability of the model.

18:26You start to get this dynamic where you could have a reinforcing or sort of virtuous cycle of how do you build a moat around what the agent is doing. and uh and so because i think one existential question for everybody whether you're big or small is how do you have a sustaining advantage if so much of the value is packed in the model and to your point about this end desk or gorgeous you know dynamic like how do you make sure that you can have a have a position of of strength that you build on top of and so we're starting to you know have these big questions of like okay what does it mean what is the new network effect in a world of AI.

19:01And it's some degree of that user working with an agent to create something. By creating something, that data feeds back into the agent's future capabilities and you get this sort of flywheel that ends up being pretty positive. But I don't know what you're seeing in the larger company side. From the enterprise standpoint, one thing that's very interesting is a lot of these enterprise companies are actually sitting on a ton of data. And they've been sitting on this data for many, many years, slowly trying to cleanse it, trying to figure it out. With the adoption of AI, even companies using the internal data in order to get solid outcomes, we're seeing that over and over again, right?

19:40So, I mean, the beauty of this is now you have all of this data that you've been using. And also, you know, a lot of the models are built off of public data. But how many of these models are actually built off of enterprise internal data? Right. Very limited. Right. Once we unlock that, I think the pace of change that we're going to see is going to go even faster. That's what we're seeing today. Yeah. There's a kind of a fun question that you can start to ask your customers, which is sort of like, what data are you sitting on that is proprietary to you? And so thus, if you had AI that worked on your data, what would you have as a differentiated outcome relative to your competition?

20:15And very quickly, you realize more companies are actually in the data business or information business than they maybe initially had thought about. where they start to realize like, okay, actually I do have some proprietary information or insight about the real estate market or about my talent clients, if I'm a talent agency, or obviously if you're in financial services, you're sitting on a wealth of data. So the question is if I can bring AI to that data, what can I do with that in a unique way relative to my competition? And you can kind of like go really far with this, which is right now, as an example, if you look at like a company's income statement or balance sheet there's nothing that approximates the value of their data yeah i still get confused about that goodwill line but i'm maybe throw some some you know data in the goodwill line i don't know if a lot of companies do that the question would be the question would be in 10 years from now will we see this sort of emerge as a thing which is like what how good is your data as an enterprise how much data do you have yeah are you actually able to turn it into business value.

21:21And AI will be the thing that basically unlocks that because what AI wants more than anything is more data to work on. And that's what it thrives on. So we will enter this completely new shift and actually like, what are the assets in an enterprise in a world where you can deploy AI agents to have completely disproportionate outcomes in your organization? And I think data is going to be one of these. but and it's but we're you asked the question i'm just i'm learning from you on this one i'm not seeing modification i am i i haven't seen any i can't think of a single i mean you talked to more than i can't think of a single company i work with or no no that feels like there's a stronger moat today yeah if that was the question you were asking you tell me some examples to me i actually think about box as a case study actually we do this new podcast i did with rory from scale i did the other day and i said if there's anyone that should be benefiting this from it i said box and let me tell you why because i've been in this document space for a while as you remember right you are a document og yeah and i'm like i've been and i've been a box user at least since 06 i'm like so i got i've got like forever content okay i could never talk to it before ai so everyone that's been in documents has wanted to talk to their documents now i can and and not only can i talk to my documents because it's a what's aaron can tell us now or in six months that also means i can do stuff i never thought i could do with my back like the initial case study is okay tell me everything in box where i have the wrong indemnity provision you start there and then all of a sudden you can talk tell me what contracts i shouldn't have signed tell me what contracts had lacked authority tell me where like the designs are off on the measurement on and then something or just tell me where i could have an upsell opportunity based on those contracts yeah who's got pricing that that maybe you know you know you'd be you you have a better position for that uplift yeah the the value your moat bigger we think so but But I'm, you know...

23:16I'm just asking you to learn. Well, I don't need to turn this into a Wall Street conference, but yes, the... That happens in SaaS or something. Inadvertly. We're very... We're big fans of our remote. But the... But I think the big question is what is the value inside of your data for any enterprise? And then if you're building software, find ways to get the customer to tap into that. That's what's going to build this flywheel. Like, I'm actually very bullish. Right now, you know, there's like a little bit of a meme of like these AI coding tools. You can just swap them out. Yeah. I think that's like a temporary thing that you'll see on Twitter or X or whatever, I think the mode is actually fantastic because if you're the place where you have agents that understand your code base, they understand your development methodologies, they understand your documentation, they understand your specs, that is something that becomes this ongoing flywheel that just gets bigger and bigger and bigger.

24:03That will be true of every single segment in healthcare. Imagine if you're the AI system that understands a particular set of patients and the particular sort of practice areas of that healthcare provider, you're not going to want to swap out that AI system. So for as fast as everything is moving and as disruptive as everything is right now, I think this is a window in the next couple of years where I think we'll have the defining platforms get built and effectively cemented, assuming that they stay current, right? If you're asleep at the wheel in this space, it's over, right? You can lose your entire position probably in a two-year period at this point.

24:40But it's also a point where you could cement your leadership position, I think, in a very unique and different way where that wasn't possible. Let's go back 20 years to cloud. There was a five-year period where people could just kind of futz around. Maybe I'll do a hyperscaler. Maybe I won't do a hyperscaler. And somehow, 10 years later, 15 years later, basically, you've got three, four, or five major platforms all at massive scale people had time to kind of figure it out yep ai is probably not going to look like that like it's gonna it's gonna get locked in and compound the adoption's going so fast but i get i want to believe that i'm with you right i've got your back but let's and i'm not i'm gonna be quiet i'm not gonna ask myself it just shows up on our panel my last question but i'm not an mcp expert but let's take it out a couple months right Right.

25:37So you've got this data lock, right? But if my MCP is great, maybe I just grab those box files and that data and I abstract away box and sure box has the data and the healthcare company. But if I can talk to it through a myriad of different levels and MCPs and everything, is it really a moat? I would actually argue that MCP is a thing that only enhances the underlying moat because data wants to be free. That doesn't mean that there's no moat around managing the data and making it useful, it wants to be incorporated in some other workflow. So for us, the more places where customers can use their data is only a net positive.

26:14We want to connect to every MCP environment that's possible as a tool. And that just makes the continued feedback loop of having more data, at least in our environment, more valuable because you know that it's going to be accessible from all these other systems. So I think of that as kind of a, the analogy would be like did apis reduce people's modes no it actually in many cases made companies more sticky over time yeah because they could be embedded into even more software which made that even a greater level of of sort of concentration so so i i i mean i i see no reason why mcps wouldn't be just analogous to all of the the apis that we built over the past couple of decades yeah i was also just say that one other aspect with the mo if you're able to establish the moat.

27:00Let's say there is a moat that's established within AI. There's a significant monetization opportunity for all of these companies because then you have the ability, once you're showcasing and you have that moat, once you're showcasing the AI capabilities, we're seeing significant upsell in the marketplace where companies are able to raise their prices because of these AI capabilities, but they're continuously adopting. And so I don't think there's going to be only, you said two years. I think it's less than two years that we see this evolution and it's going like a thousand X faster than the cloud adoption.

27:31Yeah. Cause everyone's using it cause a thousand X kids start using it, you know, and probably in high school doing homework and, and so on. So like everyone's using it. It's in the mindset already. And that's where the adoption is happening so fast. I mean, I don't, I don't think it's that crazy to say that. I mean, this is easily the fastest adopted technology in history. If you look at, at chat to VTs, you know, any other published metrics somewhere, what 500 million active users on something on, that order of magnitude in two, two and a half years, two and a quarter years. Yeah. There's never in history, there's never been something like that because first of all, in every other technology transformation in history, you didn't have enough people wired up on the internet to put those kinds of numbers.

28:13So it's just like de facto, it's obviously correct. And so, so think about to Raj's point, how many people right now are being wired completely differently. I don't know if anybody has gone to a college, knows somebody in college, maybe just graduated college. I was speaking at a college class recently and it's like, these, these kids are like, I want to be a little bit like old man yells at cloud kind of thing, but you get off my lawn, but it's like, you're also kind of like amazed by how it works, but like they do not work like us. They're much more efficient. They're much more efficient. Maybe they don't know anything as a result.

28:49I don't know. We'll figure that out as they enter the workforce, but they work so differently and it's like, they're going to come into our enterprises and be like, wait a second. like you take two weeks to come up with a project plan for some marketing strategy. Like literally that thing was just spit out by Claude in five seconds. Why would you meet about that? Like they're going to wonder why do we work the way we work? And obviously we're going to tell them like this technology didn't exist. Like give us a break, but like think about an entire onslaught of an, of a new generation that is just like, why would you spend time reinventing the wheel?