Johannes Berg, maintainer of Linux kernel wireless networking subsystems (802.11, mac80211, WWAN, rfkill), has announced a strict new policy for handling AI/LLM-generated patches. Any patch that doesn't pass a "three second review" as obviously correct will be ignored outright. He refuses to engage in back-and-forth with LLM-generated code, arguing that human contributors must genuinely understand the semantics of the code before submitting patches. In response, Syzbot — which pre-reviews AI patches with a human in the loop — has been disabled from sending AI-assisted patches to the wireless subsystem. Berg's core concern is scalability: AI-generated targeted fixes often ignore broader code architecture, creating long-term maintenance burdens and consuming reviewer bandwidth.
