Exo: Berkeley's self-improving agent architecture cuts costs 31-62% vs Codex
arena · x · 2026-10-30
Alex Krentsel, UC Berkeley PhD candidate and senior researcher at Google Research, presented Exo at Arena HQ — a self-improving agent architecture (with Ion Stoica and others).
- Borrowing the exokernel idea from OS design, Exo splits an agent into a protected mechanism layer (state, history, secrets, and a "guardian") and a fully evolvable policy layer, enabling safe runtime self-improvement of the agent harness.
- On benchmarks and open-world RuneScape tasks, Exo was 31-62% cheaper than Codex and Pi.
- Talk video is available.
More from coding & agent
- Henge: Open-Source ComfyUI Video Editing Interface Built on Custom MiniMax H3 LoRAs — Round_Awareness5490 · 2026-10-30
- Lean 4 formalization of an asymmetric metric-tensor physics paper hits GitHub — SubtleGradient · 2026-10-30
- YC-backed agent-native office suite launches free in ChatGPT and Codex — ycombinator · 2026-10-30
- Supabase never had an agent strategy — its CTO explains why it won anyway — socialwithaayan · 2026-10-30
- Understand Anything hits 88.9k stars: turn any codebase into a queryable knowledge graph — adnan_hashmi · 2026-10-30
- Scopewerk: Open-Source Rust App Puts a Hanmatek Scope on Your Desktop With MCP for AI Agents — debreuil · 2026-10-30