Closing Bell Overtime · Oct 19, 2026
With Aaron Levie, CEO of Box
Photo: TechCrunch, CC BY 4.0 via Wikipedia
OpenAI has had as many as 10,000 AI agents working together on a single task, Aaron Levie, CEO of Box, told Melissa Lee and Michael Santoli on CNBC's Closing Bell Overtime.
Most talk about AI agents still pictures one assistant doing one job. Aaron described groups of agents dividing up a hard problem, and said the first uses will be in drug discovery, trading and, on the risk side, cyberattacks.
"This wasn't really practical to talk about six or twelve months ago — the models themselves were just not smart enough." — Aaron Levie
Aaron runs Box, which he positions as the governed layer that controls the company data agents work with.
The full segment is covered here so you can skip it.
Here are the 4 predictions that matter.
Key Takeaways
Groups of AI agents can now work together on one problem, which the models could not do six to twelve months ago
Cyberattacks may be one of the first places agent swarms appear
Life sciences, software automation and trading are the likely first high-value uses
Consumers will not pay directly for agents, but enterprises will, once the value is clear
1. What an Agent Swarm Is
Aaron defined an "agent swarm" as a group of AI agents coordinating on one shared problem. OpenAI's examples have run up to 10,000 at once, and the approach is already being used in math research.
Aaron Levie points to OpenAI's agents collaborating on a single task
"OpenAI using 10,000 agents to work on some collective task... the agents are actually able to collaborate together." — Aaron Levie
He said this was not possible until recently, because earlier models could not coordinate well enough.
