# Exo lets AI agents self-improve their harness using an exokernel design

> Original title: What if AI agents could safely improve their own harness at runtime? UC Berkeley PhD candidate and senior researcher @AlexKrentsel from @...

- Source: Arena
- Published: 2026-10-29T17:56:25.000Z
- Bad Influence: https://badinfluence.news/items/lxisih4q3nfpbg03flj42pij4
- Original: https://x.com/arena/status/2115865351397511204

## Summary

UC Berkeley PhD candidate and Google Research senior researcher Alex Krentsel presented Exo, an agent architecture that separates a protected mechanism layer from an evolvable policy layer. Arena says Exo was 31-62% cheaper than Codex and Pi on open-world RuneScape tasks.

## Post

What if AI agents could safely improve their own harness at runtime? UC Berkeley PhD candidate and senior researcher @AlexKrentsel from @GoogleResearch recently presented his work on self-improving AI agents at Arena HQ.

This architecture, Exo, borrows the exokernel idea from operating systems, separating an agent into a protected mechanism layer (state, history, secrets, and a "guardian") and a fully evolvable policy layer.

Alex walks through the design, results on benchmarks and open-world RuneScape tasks, where Exo was 31-62% cheaper than Codex and Pi, and more. 

Big thanks to @AlexKrentsel for joining us to present his work with @istoica05, @martin_casado, @ankrgyl, Sylvia Ratnasamy, @BerkeleySky, and @netsyslab. 

https://www.youtube.com/watch?v=Jglgg0m8EIs