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 an EDM show here before? Raise your hand. Okay, we have a couple. Yeah, we probably were the same one.
We were talking about that before. I think a lot of folks on our team have been here mostly for that.
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 a backhanded compliment.
No, it's a full on compliment.
Okay. It's like, do you want to be a prolific tweeter? I don't know. I basically just tweet what we're running into at Box and what we're seeing when we're building agents and working in AI, what I'm hearing from customers and the market, and whatever is in the zeitgeist that resonates. So it's a very boring framework: a couple times a day, whatever I think is happening in the world.
A couple times a day, but it's a bunch of paragraphs and it's all spot on, very in tune with what's going on in the ecosystem. I'm in awe.
Thank you. I think most people know how to write, so you can all try this at home. Just try and type a couple paragraphs and see what happens. It's funny because I got a C++ probably in English in high school, but everybody thinks I write a lot. I was like, you could either write an email or write a tweet, so might as well write a tweet.
Do you use AI for your tweets at all? Okay. All right. Well, I think you're definitely one of the most knowledgeable.
Some of my tweets make it very obvious because I'll look at it later and be like, 'Oh my god, there's up to nine mistakes in this tweet between grammar and spelling.' So it's definitely proof of human work.
Yeah, you almost have to add that in on purpose so people know it's not AI.
Yep. In this case it happens fully organically.
I think you're one of the smartest people in terms of what it takes for enterprises to actually adopt AI. I imagine this is because you work with a number of them daily. What advice would you have for enterprises who are thinking about this?
Yeah. I'm super lucky because since the founding of the company, we get to be in the tech industry in Silicon Valley, seeing the breakthroughs in AI. But to translate that to a regular knowledge worker or enterprise, there has to be a bridge: simple interfaces, usable by end users, packaging infrastructure for use cases. AI is the same: incredible model progress, but most enterprises need a bridge to access it in a way relevant to their use cases. I spend a lot of time with large enterprises and see them adopt AI. The big question is whether we can get the same gains in knowledge work as we see in AI coding. We all use agents for coding, but the rest of knowledge work in marketing, sales, finance, etc. is the challenge. The main question on every enterprise's mind is when and how to deploy full agentic workflows across knowledge work.
Yeah, on that, what is the answer to that question? What is working?
I purposely left it rhetorical because we're figuring it out. The tricky thing is that AI coding agents have every great property: models hyper-trained on code, work is verifiable, users are hyper-technical. But in knowledge work, the work is not as verifiable, users are not as technical, and they don't have access to all the data. Agents inherit those same limitations. So the gap is how to bridge that. We're seeing highly domain-specific agent startups emerge, and enterprises themselves can build internal agents. We're in for many years of diffusion, and the doomers are wrong about takeoff because of the slow rate of diffusion into organizations.
How are you guys at Box thinking of tackling that with Box Agent? It seems horizontal, but you're building on a solid data layer. How are you thinking about bridging the gap?
We're in a quasi-interesting territory because we're horizontal. Our domain understanding is enterprise content at a horizontal level: file systems, how agents work with content, creating a pipeline to make data ready for agents in a secure, well-governed way. We have a document conversion pipeline, security and access controls. Box Agent's job is to be an expert at working with enterprise content, and it needs context of the workflow. It can plug into broader agents via MCP, like Harvey or Claude for legal, taking on domain-specific understanding.
