Model or Harness? A Taxonomy for Pinpointing Why AI Agents Fail

A new framework helps developers figure out whether an AI agent's failure is the model's fault, the scaffolding around it, or something else — so the fix targets the right layer.