In this OPTO Sessions interview, I sat down with Box CEO and co-founder Aaron Levie to get specific about where enterprise AI is actually going: from flashy demos to real dollars. The core idea is simple but massive—about 90% of corporate information is unstructured (contracts, invoices, PDFs, decks), and historically computers couldn’t really “talk” to that data. Aaron explains how AI agents finally unlock it: read documents at scale, extract key fields, and then let teams query and automate workflows on top of that intelligence. We dug into what Box shipped—Box Extract for large-scale extraction, Box Automate for workflows, and how agents can be built fast (Aaron says under two minutes) and deployed across roles like sales, marketing, product, and legal. We also covered why Box is going model-agnostic (OpenAI, Anthropic, etc.), how they benchmark models by task, and why governance matters: agents inherit Box permissions, so they can’t access data you don’t have rights to. From an investing lens, Aaron’s take is that value accrues across the stack—chips, hyperscalers, model providers, and now the application layer. But adoption won’t be overnight: this is tool change plus behavior change. His best advice for this quarter is the same one I’d give—spend time using AI tools aggressively, because hands-on reps beat theorizing every time.