Box's Aaron Levie: On Reinventing Yourself in the AI Age and Enterprise Diffusion

Key ideas

Levie argues the gap between AI models and enterprise workflows is where value will accrue, and that diffusion beyond coding will be slower than Silicon Valley expects.

From a machine transcript of the audio. Quotes link to the moment they are said.

1. 1

   The bridge from model to workflow

   Levie argues that even a very intelligent model still has to connect to data systems, human checkpoints, legacy software and changed processes inside a company, and that research labs are unlikely to take all of it on. He compares this to cloud infrastructure, which he says did not prevent Snowflake or Databricks from existing. He says openly he is biased, with a concentrated bet on applications, and expects value across the stack and not in one or two labs.

2. 2

   Coding is a poor diffusion benchmark

   Levie lists why coding spread fast: its value sits in text, labs train and test on it daily, users can fix their own setups, and the work is highly paid. A sales rep, he says, is still limited by whether customers reply and have budget, and knowledge work lacks an equivalent of GitHub and has stricter access controls. He concludes that rollout will take longer than the Valley thinks, and that as building gets faster, reaching the customer probably matters most.

3. 3

   Incumbents need an agent and open APIs

   Levie tells established software companies they cannot pick one of two moves: build an agent that clearly beats outside agents on their own product, and expose the product to outside agents through APIs. He says the second opens new usage, pointing to how much more he uses Salesforce now that he reaches it through AI tools. He adds that talking to customers surfaces agent use cases nobody else has thought to build.

4. 4

   Expect background work, not chat

   Levie says chat will stay in every product the way a search box does, but that most enterprise work is process running behind the scenes, including work too costly to ever give to people. He bets that in five years about 90% of enterprise tokens will come from things no user kicked off, with people reviewing results. He thinks dashboards, queues and task lists for those workflows favor vertical products.

5. 5

   Reinvent through data and demonstration

   Levie says that after ChatGPT Box carved out a team with its best people and met daily, and that keeping company documents in one system makes agents easier to run on them. He describes using a list of heavy token users to inspect what is useful, not as a contest, and having one employee demo a workflow to a whole team. He says Box is very uncomfortable with dropping code review, because customers entrust it with their data security, and expects always to give up a little productivity as a result.

In Aaron’s words

What’s available here

• Original source

  Linked. Opens on YouTube.

• Player on this page

  Available. YouTube’s own player, loaded only when you press play.

• Full text or transcript

  Not reviewed. No license to show a text version here.

• Our summary

  Draft. Written by us with AI from a machine transcript of the video’s audio (Whisper) and checked against it. Machine transcripts can mishear; each quote links to the moment so you can hear it. Not yet rights-reviewed.

• Ask questions with AI

  Not reviewed. Off until there is licensed text to work from.