Box CEO: AI Agents Explained - Real Use Cases, Challenges & What’s Next | Aaron Levie
Summary
In this episode, Aaron Levie, CEO and co-founder of Box, shares an insightful narrative intertwining the company’s origins, AI strategy, and enterprise technology transformation. He begins by reflecting on Box’s founding story rooted in long-term friendships and the early challenges of shifting enterprise data storage away from physical media towards cloud-based solutions. Levie highlights the profound underutilization of enterprise data—95% of which remains unused—and explains how AI agents, combined with Retrieval-Augmented Generation (RAG) and model orchestration, can unlock significant productivity by automating complex, previously deprioritized workflows. Box’s strategy notably excludes building proprietary foundational models; instead, it focuses on integrating and orchestrating third-party AI models, simplifying risk and accelerating innovation. He discusses the technical challenges, including error compounding when chaining AI agents and the critical role of human oversight to maintain reliability. Data governance and security emerge as central concerns, emphasizing strict controls to avoid inadvertent data exposure amid AI integration. Levie envisions a future with interoperable AI agents communicating via protocols akin to APIs, fostering an ecosystem that extends workflows across applications. The episode also explores Box’s cultural approach to AI adoption, cultivating internal enthusiasm while managing expectations through founder-mode incubation and continuous leadership support. He contrasts AI’s additive role in Box’s business model with historical examples of disruptive pivots, stressing measured growth. The conversation further delves into public versus private company dynamics, arguing that many AI companies may benefit from avoiding the volatility of going public. Overall, Levie offers a pragmatic and forward-looking perspective on AI’s evolving enterprise impact, combining technical depth with strategic business insights.
Key Takeaways
- 1Enterprise data remains vastly underutilized, with Aaron Levie estimating that about 95% of it is never effectively used due to current software limitations and operational challenges. AI agents layered over existing enterprise content infrastructures can replicate human tasks and automate workflows, unlocking significant latent value within this data. By integrating Retrieval-Augmented Generation (RAG) and model orchestration, companies like Box can make this data actionable without the need to build foundational AI models themselves.
- 2Box deliberately avoids building or fine-tuning its own AI models, choosing instead to orchestrate a variety of third-party large language models (LLMs) through APIs, allowing rapid adaptation to the latest AI innovations. Their AI platform includes modular components like AI Studio for creating customizable primitive agents and a model abstraction layer supporting integration with major providers such as OpenAI and Anthropic. This strategy mitigates the significant cost, complexity, and operational risks of training proprietary models.
- 3Chaining multiple AI agents to perform sequential tasks introduces the problem of compounding probabilistic errors, which significantly undermines reliability in enterprise workflows. Each agent's output carries inherent uncertainty, and as these outputs feed into one another, errors propagate and magnify. Consequently, enterprise deployments require cautious design of agent interactions, often demanding human oversight or task segmentation to ensure accuracy.
- 4Human-in-the-loop processes remain essential as AI agents tackle complex documents and large datasets. While simple, limited-scope tasks can achieve near-perfect accuracy autonomously, broader or more nuanced tasks require human review to correct errors and validate outputs. This balance maintains trust, reduces operational risk, and facilitates progressive AI integration.
- 5Data governance and security are paramount concerns when deploying AI agents broadly within enterprises. AI’s capability to autonomously access and analyze data expands the risk of inadvertently exposing sensitive corporate information if access controls and policies are insufficient. Strict governance frameworks, auditability, and carefully scoped AI permissions are essential to prevent compliance violations and information leaks.
- 6Box envisions a future enterprise AI ecosystem characterized by interoperable AI agents communicating via a Model-Communicating Protocol (MCP), akin to how APIs enabled integration across software two decades ago. This architecture would allow specialized agents—such as Box’s internal agent, Salesforce agents, and contract management agents—to collaborate seamlessly, enabling orchestration of complex cross-application workflows.
- 7Successfully integrating AI into large organizations demands deliberate cultural transformation. Box pursued a balanced approach, incubating AI internally with a founder-mode mentality while fostering broad organizational buy-in and setting realistic expectations about the challenges ahead. Through leadership commitment, transparent communication, iterative experimentation, and internal evangelism, AI adoption expanded organically across teams, embedding AI tools into various roles from sales to engineering.
- 8AI-driven automation of mundane and repetitive tasks in software development, such as patching SDKs or managing library updates, represents immediate, high-impact use cases. Levie calls these 'low-hanging fruit' where AI can relieve engineering teams from time-consuming activities, allowing them to focus on strategic and creative work.
- 9Public market dynamics pose unique challenges for AI-focused companies. Levie asserts that many tech firms might never need to go public, as private markets now offer sufficient liquidity and less volatility. He describes an 'adverse selection' where companies going public often face volatility and investor pressures that may hinder long-term innovation, suggesting that staying private can be strategically advantageous.
- 10AI’s current impact on Box’s business model is additive rather than disruptive. The company continues to sell user seats as core revenue but layers AI enhancements that expand market opportunities and deliver new customer value. This measured integration contrasts with historical examples like Netflix’s disruption of Blockbuster, reflecting a strategic augmentation of existing products instead of a wholesale pivot.
Notable Quotes
"I would argue that probably for 95% of our data, people have things that they would love to be able to do with that data that they just don't ever do. The thing that I get excited about agents is, is the expansion of now what people can do with software and start to solve the use cases that we just never ended up prioritizing in most of these categories."
"I think actually it's good to have a heightened degree of scrutiny and regulatory pressure on being a public company. It is absolutely net positive for the average shareholder that we have all of these systems and governance controls in place. I think that's only a good thing."
"Which means that any shareholder can basically decide at any point, do I like the price now or do I want to hold it? And you don't really have to like, like there's a very real scenario where a lot of these companies that are doing this model don't have to go public ever. Like, and it just, and there's not a premium that they, that they give up in that process. In fact, there might be a premium to staying private because they don't have the volatility that we all deal with in the public market and, and so on. And almost an adverse selection to going public."
"So we started the company in college and we were, um, my co-founder and I, we were sophomores in college and, and we, we had, uh, a variety of friends. And the reason I'm bringing them up is because we eventually had kind of four co-founders. Um, but a variety of friends through high school, we'd all try different projects and startup ideas together. So, so we actually all went to high school together. A few of us went to middle school together. And, um, and so we know each other for already by, by sophomore year of college, 10 years, um, which is an incredible, you know, incredibly lucky to, to be able to have a, uh, a friend group that you've known for 10 years that, that you're still in touch with."
"And so maybe there was this moment where you could access your data from anywhere and you just put it in the, in the internet and it wasn't called the cloud, but, but, you know, you put it on the internet and you access it from any device. And, um, and that was the original idea."
"I've launched shitty stuff before, um, where you just see traffic and people are clicking things and clicking an ad and you're like, cool. This is the first time like users, I would like interact with them and they would be like, I like this. I don't like this."
"And so we were at, at home in Seattle. We pitched literally everybody. So if you were a VC in the Seattle area in, in 2005, you got an email from me. Um, or at least you passed it along to your associate and they didn't take a meeting with us. Um, and we had like three or four meetings rejected by everybody. Great process. Like I, I, I, I'm glad we got the rejections. It, it, it built, you know, this kind of grit, uh, in us very early on."
""We were not cloud from day one running on the cloud, but we were cloud to the customer. So we kind of rode that. We didn't, you know, so no pivot was involved. We were selling the idea of cloud to many companies.""
""The energy is like absolutely in the category of how many use cases can I, can I apply this to? How do I start to lean in more? We've got these 20 experiments running. We'd love to, you know, make these ones in production. Those ones not. And there's no resistance. The only resistance is purely on the, on a pure technical basis.""
""Without ChatGPT, none of this would have happened. Now maybe a different timeline, somebody else would have done ChatGPT and Gemini would have launched first and whatnot, so we could still have been on this timeline.""
""So crossing the chasm kind of breaks out different stages that, that technology kind of goes through. And, and the, the key is to, to think about technology. And in this case, and these are the stages are like these really early adopters, then early, he calls them these early pragmatists. And there's this, there's this sort of chasm, which is, there's a lot of things that we as early adopters do that never make it mainstream.""
""So, so then you've got these early, uh, pragmatists, which are, this is like the person who's actually, um, the actual person in a company that's like, okay, maybe I can adopt this thing for, give me some acceleration in this area. And, and that's like your first indication that maybe things are taking off. They, they want a little bit of an early advantage, but these are still the kind of crazy people in the company. They're not, they're not like, like the CIO of Goldman Sachs. Like who's like, it's like, in this case, actually the CTO of Goldman Sachs is actually this person.""
""AI coding is very clearly in the pragmatists. Uh, you know, get up, get up copilot now, I don't know, three years old, uh, cursors flying off your shelves when service flying off the shelf, you know, replet, all these guys are, are, are kind of being, you know, used everywhere. So you're, you're in that, in that kind of hyper growth cycle of that early pragmatist to the pragmatist. Um, and, uh, and so that one's totally safe. Like that's, there's no, like you can just extrapolate at this point. Like we don't have anything in our way. It's probably if anything, just pure like sales rep constrained on, on how this could grow faster.""
""So the way to think about Box is we're a platform that helps you store, share, manage, collaborate on data. We have a layer of, you know, security. We have a layer of file permissions. We have a layer of data governance. So that's what we've been working on for, you know, basically two decades. What we added was a layer, uh, which is just, we call it the AI platform. Uh, and it has kind of all the plumbing that you would expect you would need to do to be able to work on content in an AI, uh, in an AI, uh, uh, context.""
""We have hundreds of billions of files. So we're figuring out what we want to do for the kind of broad corpus, but we have a feature that lets you kind of target a certain set of data and then, and then effectively be able to do rag on that data. Then we have a layer that, that is kind of the model abstraction layer. So it, it, you know, connects to, to, you know, right now four or five major models, but you can kind of theoretically think about it as complete, you know, bring your own model architecture. Then we have an AI studio that lets you, you know, uh, uh, go in and create custom configurations of a model, a set of instructions and a set of tools. That's how you create basically today, primitive agents that will become much more, um, comprehensive over time.""