Every Agent Needs a Box

A visual investor takeaway from Latent Space's conversation with Box CEO Aaron Levie on why enterprise agents need governed workspaces, identity, search, evals, and workflow redesign.

Latent Space talked with Aaron Levie, CEO of Box, about why enterprise agents need governed workspaces rather than open-ended access to company data. The episode was published on March 5, 2026 and runs 1:16:58. 1 2

Core thesis

Enterprise agents need a new content layer: identity, permissions, search, evaluation, and a place to read and write company context. Box's argument is that agent adoption depends less on a generic chatbot interface and more on governed access to the files, meetings, contracts, specs, and workflows that already define how a company operates. 2

Three investor takeaways

  1. Agent identity becomes infrastructure. Autonomous agents cannot simply inherit full user access, because enterprise workflows need liability, oversight, permissions, and boundaries between the human account and the agent account.
  2. Context engineering is the bottleneck. Levie frames the hard problem as search, ranking, pruning, and knowing when to stop, not just giving a model a giant context window.
  3. Workflow redesign is the GTM wedge. The episode argues that coding agents worked first because developers adapted their workflows around agents. Other knowledge-work categories will need data cleanup, new process design, and implementation help before agents become reliable.

Companies and stack mentioned

Box, OpenAI, Anthropic, Google, Chroma, Okta, Cursor, and Codex all come up in the episode's discussion of enterprise agents, model progress, identity, and developer workflows. 2

Why Granola and agent infra investors should care

Meeting intelligence becomes more valuable when it turns conversations into permissioned, searchable, evaluated context that can be written back into workflows. For an AI investor, the episode points to an investable cluster around agent identity, context engineering, enterprise search, evals, observability, and implementation services.

Related content

Comments

Sign in to comment.