The AI assistant runs on a full instruction set: a defined r…
The AI assistant runs on a full instruction set: a defined role, a clear goal, relevant context, a set of constraints, a reasoning step, and a format specification. For a multi-item cost estimate or a sensitive client situation, this setup earns its keep. Then an owner emails a simple question: “Do you board cats?” A colleague points out that running the full setup on a one-line factual question seems like a lot. She’s right. What’s the soundest way to scale the instruction set down?