Box CEO Aaron Levie remembers the moment he realized ChatGPT was a game changer

Aaron Levie has been in the tech business for a long time—the company he cofounded, Box, is older than the iPhone. Like many techies, he’s long envisioned a future where a magical AI would be able to do amazing things.

But as the big breakthrough neared, even he didn’t see it coming.

“I have some friends who are deep into AI, and we were chatting like three months before ChatGPT, and they were like, ‘Are you using LLMs for anything?’ And I was like, ‘Yeah, we’re doing some basic text classification and a security thing’—and that was the end of the conversation,” he recounted in an interview with Fortune last week.

“And then three months later I was like, ‘Oh, my God, this is going to revolutionize every aspect of our business.’ It was just simply like the model passed a threshold where you could all of a sudden see its horizontal generalizability. And I just always couldn’t see it before.”

It’s a revealing anecdote that helps explain the frenzy for AI in business right now (and maybe even some of the wild valuations). Just as consumers immediately grasped the value of conversing and getting info with ChatGPT, the tech industry’s eyes suddenly opened to the potential of how large language models could unlock a whole new set of powerful capabilities.

For Box, Levie says, LLMs basically provide a way to tap into all the data a company has. Until now, that data needed to be “structured”—that is, tagged and classified with metadata into a database so that tech systems could make sense of the information. As a result, much, if not most, of a company’s data (for example, PDF image files and audio recordings), was out of reach. With LLMs, Levie says, that problem is instantly resolved—it’s almost as if all the world’s language barriers were suddenly erased and everything anyone has ever said becomes instantly intelligible to you.

One of the first manifestations of this advance for Box is something called Hubs. It’s a visual dashboard that can “see” documents scattered in different places throughout an organization (conversations in Slack, PDFs on your hard drive, and HR training materials in the cloud) and automatically group them into relevant categories (at a law firm, for example, think by clients, or by contract type). From that jumble of unstructured data, a worker can instantly extract whatever bits of data they need.

To hear Levie say it, it’s a fulfillment of what people envisioned 50 or 60 years ago, in the age of mainframe computers; the idea that workers could access a “gigantic brain” and quickly retrieve information and answers.

But I had another question for Levie: Given how rapidly AI is evolving—to the point where even longtime CEOs like himself were surprised by the sudden power of LLMs—how can a company today figure out an AI strategy without worry that by the time it’s implemented, the landscape will have already changed?

The answer is understanding how history plays out in the tech industry. “You know the incentive structure of all the players, so you can kind of plot it out, like, ‘Okay, we probably don’t want to be overly reliant on one technology platform, because there’s going to be many that are all going to be fighting for lowering the prices and getting all the customers on their platform,’” Levie said.

“Once you have the contours of how the industry is going to play out, you can divide and conquer.”

And with that, here’s what else is happening in tech news today.

Alexei Oreskovic

Want to send thoughts or suggestions to Data Sheet? Drop a line here.

This story was originally featured on Fortune.com