PinnedRobert Hjertmann Christiansen·15h agoA rubber duck and a critic for AI coding agentsI let AI agents fix planted bugs 2,680 times, with and without two helper models. One helper raised fully correct fixes from 62% to 85%…
Robert Hjertmann Christiansen·15h agoA rubber duck and a critic for AI coding agents: the deep diveHow the harness, the helpers and the 2,680-run experiment were set up, and the full results: every hypothesis, every variation and the…
