Why Enterprise AI Adoption Is Slower Than You Think — Aaron Levie (Box) + Harrison Chase

Hello. Thank you so much for being here. Thank you. I think my last time I was in this building, it was like a rave. Has anybody been to like an EDM show here before? Raise your hand. Okay, we have a couple. Yeah, we probably were the same one.

I was like, this is a very familiar venue. So, okay. I was doing some research on you before looking at your tweets. You're a prolific tweeter. How do you do it?

It's funny. Prolific tweeter is sort of one of these like backhanded compliments.

Yeah. And I think this is, I mean, just to point back to even the first question, I'm super lucky because this has been true basically since the founding of the company. You guys, you know, live, I think, the same world. Probably many, many folks live in this world as well. You know, we get to be in the tech industry in Silicon Valley, sort of seeing the breakthroughs of what's happening in AI today, but previously it was for us it was cloud computing initially and our whole business model. You could almost net the core business model that we had when we started Box, which was there's these breakthroughs happening in tech and in this case it was the early days of cloud and cloud infrastructure and all that, but to translate that to the regular knowledge worker or a regular enterprise, there has to be some bridge. You know, you have to have simple interfaces, they have to be usable by end users, there's got to be a way to package up this infrastructure in a way that makes sense for use cases. And I think AI we're seeing the same thing, which is we have these incredible breakthroughs happening in AI. Model progress is absolutely insane. Infrastructure and framework and the kind of surrounding scaffolding is kind of progressing at an incredible rate, but most enterprises in the world have to have some kind of bridge of how do I get access to this in a way that's relevant to my use cases and to my organization. And so I get to spend a lot of time with, especially large enterprises, but businesses of all sizes across industries and we get to kind of see what's happening in their organizations as they go and adopt AI and as they try and bring all this amazing technology to their organization. I think it's probably what everybody here is experiencing whether you're building a product for the enterprise or you're in an enterprise yourself and maybe it can be netted out to we have this insane progress we're seeing in things like AI coding in particular. You know, we're all using agents for design, for developing code, and we all see every day the full potential of what happens when an agent can kind of run loose on a large task and a large project. And then we look at the rest of knowledge work in marketing and sales and finance and research, and the big question is can we get those similar gains in all those other parts of the organization in these business processes that are in domain-specific or industry-specific areas. And that's the main question that I think is on the mind of every single enterprise, right? We all as technologists are using this in our daily lives, our colleagues that are maybe not in tech are using it in their daily lives but more for a chat sort of basic interaction. So the big question is when and how do we get these full agentic workflows deployed across the rest of knowledge work. That's kind of the topic that takes up the mindshare of I think most enterprises today and again some of it is then building this internally, figuring out how to do that, some of it is working with partners and vendors and providers to make that happen.

I purposely left it rhetorical because I think we're figuring it out. But the tricky thing is, and I don't know how much this will resonate or not because I think being in tech, so many people here, it's like you have to kind of step out for a second and think about this. Like when you think about the full amazing power of AI coding agents, it has sort of like every great property going for it, right? The models themselves are hyper-trained on code because that's a plentiful content type on the internet. The work itself is verifiable, so you kind of know if the code worked and so that helps both in the RL process and the actual when you're using the agent to do some work. The users are hyper-technical by definition and so like when the agent does something and it goes wrong, you know how to fix it or if it says I need access to a data source, you generally have the ability to get the right MCP server involved in that workflow. You know that like when you're in a coding agent and it's like hey I need to download this CLI from this vendor and this package, you have the wherewithal to say yes or no based on that situation. As a brief aside, my dad has gotten into coding but he's not historically a coder. He is kind of good at Excel and data science stuff and I was just watching him use Claude Code and he was just like yes download that package and yes, and like for sure his computer now is full of malware. Like there's just no chance. Every npm attack we've seen the past like 3 days, he's got every one of those. So but like that's what... and sorry dad wherever you are, but this is what a lot of teams adjacent to engineers are going to experience. And then imagine if you're even more adjacent and you're in marketing and you're in sales, all of that kind of technical complexity. So that's a bunch of issues. And then you actually have a kind of a pesky problem that is super so boring to talk about, but it's just so real. When you're an engineer, you generally via the codebase have access to most of the data and the context to work within for whatever you're working on. And like permission structures on codebases, you know, generally are fairly wide open because you just need access to the codebase to do whatever you're doing. In knowledge work, that's just not the case. Like every single individual in every single team within an enterprise has different permutations of what data they can access and so by definition agents also have those same permutations which just multiplies the complexity of when an agent goes and does a task. So think about all those things that we just said, right? So the work is not as verifiable as code. The models are still working on getting better at these non-coding domain tasks. The users are not as technical. So kind of getting the full universe of things they need to plug in to make these agents work is obviously harder. And they don't have access to all the data they need often to do the work that they need. And so agents again inherit that same level of lack of access, which means that as a knowledge worker we can just go around and we can go and say hey, you know, Sally, can you give me access to that thing, but an agent doesn't have that same ability to navigate that complex paradigm of like well what does an agent do when it runs into a limit of what it has access to. So this is the gap and so then the question is how are we all going to as an industry and within each industry going to bridge that gap. And so I think you're starting to see some of the makings of examples of where this might work. So first of all, like highly domain and vertical-specific agent startups are emerging and that's one great thing because now you can basically say you know what we can deliver this in a really kind of purpose-built way so we can dramatically reduce all of the issues on how do you get the context to the agent. Well we know the domain so we already have a leg up on that. How do you get the right data to the agent? Well here are the set of MCP connectors that we're already going to work with. There's usually some version of like a forward-deployed engineer in that workflow. So already you have people that are going to be wired to that vertical that can help support that. So I think on the startup side this creates an incredible opportunity if you're doing domain and vertical-specific agentic work. And then on the enterprise side it's another sort of advantage which is you understand your enterprise better than any outside vendor. So it also creates an opportunity for how do you go and bridge these agents and this technology into your organization. And that's another set of roles and interesting opportunities that we're starting to see emerge which is like okay does your technology team have like an AI agent sort of internal forward-deployed engineer and how are they setting these things up. So that's a little bit of the early things that we're seeing, but I think we have to be in for many many years of diffusion of this technology. And we're going to have this huge cognitive dissonance between the extreme capability of AI coding agents that just are continuing to go hyperbolic versus what we're going to do in the rest of knowledge work because of all those issues that I mentioned. And it's just we're just going to be in for years and years of deployment and change management and adoption. And that's I think what we all have to kind of collectively sign up for. And oh by the way, as just a brief asterisk, it's also why the doomers are generally going to be wrong on the takeoff because of the sort of rate of diffusion of how this is going to actually enter organizations.

Yeah, so I think we're a little bit in this quasi-interesting territory because we're horizontal. So we don't have the level of domain specifics of insurance and banking and legal etc. The thing that we have domain understanding on is enterprise content at a horizontal level and how to work with things like file systems and how to handle all the ways that agents and models need to understand working with our file system and working with enterprise content and how do you create a pipeline of making sure your data is sort of ready for agents in a secure way, in a well-governed way with the right access controls and the right sort of security and governance on that. And so we do a lot of things even well before we even think about AI to get enterprise content or unstructured data ready for AI. So we have a whole document conversion pipeline that turns your content into a variety of file formats that agents work well with. We do a lot in the security and access controls area. So that's a lot of the kind of core plumbing. And then with the Box Agent, its really its job is to be an expert at working with enterprise content and then it needs obviously the context of that workflow. So it's going to need either skills or custom agent knowledge which is the form of like what is your business process? What are you trying to automate? And that looks like I think anybody else's agent that you'd go build or we'll plug into a broader agent. So via our MCP server we'll plug into a Harvey or a Claude for legal and then that will obviously take on the domain-specific understanding.

Yeah, so I think there's a bunch of things that we've benefited from just maybe sometimes we got lucky like we made an architectural decision 10 years ago that we obviously made it for people but we equally get the same kind of gains. Within a world of agents, we've been like most companies, you all have lots of these like deeply religious battles on things like should we which JavaScript framework to go with and which database to use and that's obviously the classic engineering battle that would happen. We had years and years of those same kind of incredibly intense battles on things like how do you design a file system and how should the permissions architecture work and so we benefited from incredibly intense sort of processes and ways of making decisions to ensure that we had the most amount of flexibility. Again more for people use cases sometimes for application use cases but now we're benefiting from the fact that agents get all of those benefits as well. So within Box as an example, there's not two ways you can store files. There's only one way you can store files. There's only one canonical ID for every object in the system. There's only one governance model. So we versus many of our kind of maybe legacy competitors where there's like three different ways that data could show up in the system. It's a single file system. It's a single governance model, a single permissions architecture. So agents get the benefit of all that. Then there's some areas where we've realized, oh, like we probably do have to evolve the platform. Our search engine needs to take in different signals because the agent can now consume a hundred times the search results as a human. So if it can consume a hundred times the search results as a person then you might need to rank things differently or maybe ranking is less important than just context in the search results because then the agent can do its own ranking. So we've had to do a bunch of tweaks to our search engine as one example. We've had to do a bunch of things where in our document pipeline, we have to convert data in an agent-ready format. So we have an API now that will take basically any content type and turn it into markdown. So some things that just make it so our agent or any external agent can work with this data much better.

We have sort of like what we wish would happen and then we're also realist. And by being in enterprise software, anybody in enterprise software, you kind of know that ultimately the customer is going to win. There's very few companies that can kind of hold out from overall customer trends and whatever that kind of mega trend is. So I think you know and this is probably something that every single one of the software players has to face right now which is when does your agentic experience show up versus when are you headless and even headless takes on two forms because it could be our AI being headless or it could just be our regular APIs being headless. And so that has its own kind of set of dimensions to it. But I think the way that we kind of think about it is we have to get insanely good in our harness at things where working with content is the most important thing, where being token efficient is the most important thing, where having the right choice of models is the most important thing. We've got to be the best at that when it deals with enterprise content. So we have, for instance, an agent that just does document extraction. And we'd like to think that that agent doing document extraction is going to beat any kind of horizontal system that you could send that problem to because we're just hyper-wired to that. We have our own evals for that. We only work on that problem. Conversely, if you're just working with Claude Code or some other or ChatGPT and you want to have content show up inside of that experience, then we know that that's going to be 100% headless and that'll be really driven by the customer's sort of interaction and whatever they decide to roll out to their organization and we are completely indifferent sort of philosophically to either of those approaches that the customer wants to do. I think that's what we all are going to have to sign up for if you're building software is you should just expect that you should create a world-class experience within your product and you should create world-class APIs that any external system can interact with. And then by volume it might be that by volume of your system usage it's mostly going to be headless and then but for people kind of doing a bunch of work it will often be that they'll be in your interface or your agent for that.

And that was actually a pretty big symbolic moment. So you have basically the world's largest SaaS company, you know, fully endorsing that headless is the future. I've already used their headless MCP a ton of times for ways that previously, interestingly, and this is the bull case for all of this. I've used their MCP server now for tasks that were both more complex and much higher volume than what I ever would have had an internal analyst go do. And this is sort of the Jevons paradox kind of whole thing of this which is there's lots of things where when you're inside of a workplace and you kind of know like hey, that's a person that's going to have to go spend 5 hours on this project or three hours on this project. Is that really the most important thing for them to go do versus everything else that they're going to work on? Whereas an agent, I don't care other than the token costs. So I give tasks to a variety of agents all the time that I wouldn't have asked a person to do, but it ends up doing then useful work for me. And so Salesforce MCP server is a great example where I'll just hand off a thing of like some market research and it'll go and do maybe 20 minutes of work that would have been the equivalent of hours of human work that I just absolutely would not have ever had somebody else go and do. And that's the example of now the new surplus of what we're going to start to see is you'll just use these systems 10x or 100x more depending on how valuable that data is or where it shows up as kind of being value-accretive in the organization.