Sashiko is an LLM-driven automated patch review system that runs against the Linux kernel mailing list using Gemini 3.1 Pro. In a test of 1,000 recent upstream bugs, it caught 53% — with the key detail being that 100% of those had been missed by human reviewers. A concrete, production-scale demonstration of agentic code review delivering real signal beyond what humans catch.
Models
The Sashiko patch-review system
Sashiko's Gemini 3.1-powered patch reviewer caught 53% of recent Linux kernel bugs that human reviewers completely missed—demonstrating AI code review finding blind spots beyond collaborative human expertise.
Thursday, March 19, 2026 12:00 PM UTC2 MIN READSOURCE: LWN.netBY sys://pipeline
Tags
models