Uploaded March 2026 | Updated September 2026, 2 weeks ago
swyx and guest host Jeff Huber (CEO of Chroma, past guest!) chat with Aaron Levie, CEO of Box, on how AI agents will transform enterprise knowledge work and why companies must adapt workflows to make agents effective. He that argues enterprise files contain critical context that agents can continuously use to answer questions and create new value, but deploying autonomous agents requires new infrastructure for permissions, governance, security, and agent identities, especially for collaborative file access.
We cover:
- prompt injection, unresolved regulation and liability
- why “easy mode” (agents acting as the user) won’t scale to autonomous enterprise agents - context engineering and agentic search constraints, including limited context windows, messy data, and model judgment on when to stop searching.
- Box’s internal evals, agent-read/write use cases, the multi-year enterprise rollout
- Aaron's founder-CEO approach running a 3000 person public company but with the energy of a small AI native startup and “build in public” feedback loop.
00:00 Adapting Work for Agents
01:29 Why Every Agent Needs a Box
04:38 Agent Governance and Identity
11:28 Why Coding Agents Took Off First
21:42 Context Engineering and Search Limits
31:29 Inside Agent Evals
33:23 Industries and Datasets
35:22 Building the Agent Team
38:50 Read Write Agent Workflows
41:54 Docs Graphs and Founder Mode
55:38 Token FOMO Culture
56:31 Production Function Secrets
01:01:08 Film Roots to Box
01:03:38 AI Future of Movies
01:06:47 Media DevRel and Engineering
swyx and guest host Jeff Huber (CEO of Chroma, past guest!) chat with Aaron Levie, CEO of Box, on how AI agents will transform enterprise knowledge work and why companies must adapt workflows to make agents effective. He that argues enterprise files contain critical context that agents can continuously use to answer questions and create new value, but deploying autonomous agents requires new infrastructure for permissions, governance, security, and agent identities, especially for collaborative file access.
We cover:
- prompt injection, unresolved regulation and liability
- why “easy mode” (agents acting as the user) won’t scale to autonomous enterprise agents - context engineering and agentic search constraints, including limited context windows, messy data, and model judgment on when to stop searching.
- Box’s internal evals, agent-read/write use cases, the multi-year enterprise rollout
- Aaron's founder-CEO approach running a 3000 person public company but with the energy of a small AI native startup and “build in public” feedback loop.
00:00 Adapting Work for Agents
01:29 Why Every Agent Needs a Box
04:38 Agent Governance and Identity
11:28 Why Coding Agents Took Off First
21:42 Context Engineering and Search Limits
31:29 Inside Agent Evals
33:23 Industries and Datasets
35:22 Building the Agent Team
38:50 Read Write Agent Workflows
41:54 Docs Graphs and Founder Mode
55:38 Token FOMO Culture
56:31 Production Function Secrets
01:01:08 Film Roots to Box
01:03:38 AI Future of Movies
01:06:47 Media DevRel and Engineering
