Aaron Levie

Aaron Levie is the CEO and cofounder of Box. His recent work centers on applying AI to enterprise content and directing Box’s agent effort. He treats access to information, permissions, and human oversight as central problems for agents working inside organizations, and credits CTO Ben Kus and the broader AI, search, infrastructure, and product team for the technical work.[2][1]

At the May 2023 announcement of Box AI, Levie presented generative AI as a way to extract value from enterprise content. The announced direction paired OpenAI models with information already stored in Box, with existing file permissions governing AI access. Planned capabilities included answering questions about documents, summarizing them, and generating content in Box Notes; initial access was to come through a Design Partner Program.[3]

In his March 2026 Latent Space interview, Levie distinguished agents acting on a person’s behalf from agents operating independently. The former can use that person’s access; the latter need their own workspace, limited permissions, oversight, and accountability. He also described an unresolved boundary: an agent may combine information shared by different people, raising questions about who should be allowed to see the resulting output. File permissions alone do not settle that question.[1]

Levie explains coding agents’ rapid adoption through the conditions surrounding software work: code is text, the relevant material is accessible, users are technically skilled, and builders receive close feedback from those users. Other enterprise workflows have more fragmented context and tighter access restrictions. His account emphasizes the work organizations must do to accommodate agents: “What’s happening is we are changing our work to make the agents effective.”[1]

In spring 2025, Levie and Box COO Olivia Nottebohm coauthored How To Be an AI-First Company. In Nottebohm’s account, their framework prioritizes agent investments by how often a process repeats and how intellectually difficult it is. She describes internal adoption as a wider leadership effort involving Levie, Ravi Malick, Jess Swank, herself, and the executive staff.[4]

Levie has also described the usefulness of agents through personal experience. A February 2025 Latent Space article quoted his reaction to OpenAI Deep Research: “In a few minutes it did what used to take a dozen hours.” The comparison concerned an unspecified task he had performed, rather than a general productivity benchmark.[9]

His public commentary extends from individual tasks to how software and organizations should adapt. An April 2026 article quoted his prescription to “make software that agents want.” A May 2026 article reproduced his argument that enterprise adoption requires IT upgrades, usable context, workflow modernization, and change management: “there’s no shortcut to getting that intelligence applied to a business process in a stable way.”[6][8]

Levie’s enthusiasm for faster task completion does not imply that he expects immediate reductions in workload. An April 2026 report attributed to him the observation that AI had not reduced work intensity in the teams he was discussing. That was an observation about those teams, not a measured claim about the labor market.[7]

These enterprise AI concerns follow an earlier career spent building cloud content software for businesses. Box’s founding and enterprise strategy provide the historical context for Levie’s later views about organizing information for AI.[2][5]

Levie cofounded Box in 2005 while in college and left his studies to build the company. Dylan Smith was a cofounder; in his 2018 interview with Simon London, Levie described a four-founder team and recalled receiving early funding from Mark Cuban before leaving college.[2][5]

Levie dates Box’s decision to focus entirely on enterprise customers to 2007. He recalled major enterprises becoming more accepting of cloud services around 2009–2010, after which Box chose to invest aggressively in growth. His advice about scaling came with conditions: recurring revenue, durable customer relationships, and network effects could justify substantial investment, but companies needed to establish demand and sound economics first.[5]

By 2018, Levie was already arguing that useful enterprise AI depended on information being stored and organized in a usable form. He associated Box’s centralized cloud architecture with that opportunity.[5]