1 paper · 1 filter
Om Shankar Tiwari, Tangi Vass, Gagan Deep Singh
AI coding agents fail in two coupled ways. They spend most of their context window rediscovering where things live, and they assert success without evidence when the work gets hard…