The most striking thing about LLMs, to me, is that they’re doing exactly what they think they've been asked to do. Not what I asked — what they perceived I asked. That gap between intent and interpretation shows up constantly in development work, and it’s what pushed me to explore a better way to work with these models.