Aaron Levie on Why Open AI Wins
Summary
Aaron Levby, co-founder and CEO of Box, joins The a16z Show to argue that open-weight AI drives ecosystem innovation and that the US should support domestic open models to counter Chinese competition. He discusses the ethics of model distillation, asserting that training on AI outputs is not fundamentally different from training on public data. Levby also shares Box’s experience with the latest frontier models, noting that Opus 5 offers superior value for enterprise knowledge work, and explains how model routing is becoming the default strategy for companies seeking to optimize costs and pe…
Topics discussed
Intro: Aaron Levby and the open weight AI debate
Astrology acquisition and open source status
Jensen Huang's open weights letter and US strategy
Distillation ethics and the line between training data
China's AI capabilities and the case for open innovation
US dependence on Chinese open models and national security
Jensen Huang interview analysis and US infrastructure build-out
Reflections on the Jensen Huang and Dario Amodei podcast
Hypothetical: US open-sourcing a frontier model
Economics of AI: Inference costs vs. model licensing
Why closed labs like Anthropic avoid open sourcing
Long-term monetization and safety arguments for closed models
Claude Opus 5 performance in enterprise knowledge work
Opus 5 vs. Fable: Cost, coding, and safety guardrails
Government risk frameworks and model capability testing
AI's impact on Box's engineering roadmap and hiring
AI coding agents and project timeline estimation
Model routing and the value of the applied AI layer
Closing remarks and upcoming AI news
Podcast outro and subscription details
Legal disclaimer
Listen ad-free on Castria